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, |
e089f05c | 41 | struct btrfs_path *path, int level, int slot); |
d97e63b6 | 42 | |
df24a2b9 | 43 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 44 | { |
df24a2b9 | 45 | struct btrfs_path *path; |
e00f7308 | 46 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 47 | return path; |
2c90e5d6 CM |
48 | } |
49 | ||
b4ce94de CM |
50 | /* |
51 | * set all locked nodes in the path to blocking locks. This should | |
52 | * be done before scheduling | |
53 | */ | |
54 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
55 | { | |
56 | int i; | |
57 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
58 | if (!p->nodes[i] || !p->locks[i]) |
59 | continue; | |
60 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
61 | if (p->locks[i] == BTRFS_READ_LOCK) | |
62 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
63 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
64 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
65 | } |
66 | } | |
67 | ||
68 | /* | |
69 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
70 | * |
71 | * held is used to keep lockdep happy, when lockdep is enabled | |
72 | * we set held to a blocking lock before we go around and | |
73 | * retake all the spinlocks in the path. You can safely use NULL | |
74 | * for held | |
b4ce94de | 75 | */ |
4008c04a | 76 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 77 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
78 | { |
79 | int i; | |
4008c04a CM |
80 | |
81 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
82 | /* lockdep really cares that we take all of these spinlocks | |
83 | * in the right order. If any of the locks in the path are not | |
84 | * currently blocking, it is going to complain. So, make really | |
85 | * really sure by forcing the path to blocking before we clear | |
86 | * the path blocking. | |
87 | */ | |
bd681513 CM |
88 | if (held) { |
89 | btrfs_set_lock_blocking_rw(held, held_rw); | |
90 | if (held_rw == BTRFS_WRITE_LOCK) | |
91 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
92 | else if (held_rw == BTRFS_READ_LOCK) | |
93 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
94 | } | |
4008c04a CM |
95 | btrfs_set_path_blocking(p); |
96 | #endif | |
97 | ||
98 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
99 | if (p->nodes[i] && p->locks[i]) { |
100 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
101 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
102 | p->locks[i] = BTRFS_WRITE_LOCK; | |
103 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
104 | p->locks[i] = BTRFS_READ_LOCK; | |
105 | } | |
b4ce94de | 106 | } |
4008c04a CM |
107 | |
108 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
109 | if (held) | |
bd681513 | 110 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 111 | #endif |
b4ce94de CM |
112 | } |
113 | ||
d352ac68 | 114 | /* this also releases the path */ |
df24a2b9 | 115 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 116 | { |
ff175d57 JJ |
117 | if (!p) |
118 | return; | |
b3b4aa74 | 119 | btrfs_release_path(p); |
df24a2b9 | 120 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
121 | } |
122 | ||
d352ac68 CM |
123 | /* |
124 | * path release drops references on the extent buffers in the path | |
125 | * and it drops any locks held by this path | |
126 | * | |
127 | * It is safe to call this on paths that no locks or extent buffers held. | |
128 | */ | |
b3b4aa74 | 129 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
130 | { |
131 | int i; | |
a2135011 | 132 | |
234b63a0 | 133 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 134 | p->slots[i] = 0; |
eb60ceac | 135 | if (!p->nodes[i]) |
925baedd CM |
136 | continue; |
137 | if (p->locks[i]) { | |
bd681513 | 138 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
139 | p->locks[i] = 0; |
140 | } | |
5f39d397 | 141 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 142 | p->nodes[i] = NULL; |
eb60ceac CM |
143 | } |
144 | } | |
145 | ||
d352ac68 CM |
146 | /* |
147 | * safely gets a reference on the root node of a tree. A lock | |
148 | * is not taken, so a concurrent writer may put a different node | |
149 | * at the root of the tree. See btrfs_lock_root_node for the | |
150 | * looping required. | |
151 | * | |
152 | * The extent buffer returned by this has a reference taken, so | |
153 | * it won't disappear. It may stop being the root of the tree | |
154 | * at any time because there are no locks held. | |
155 | */ | |
925baedd CM |
156 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
157 | { | |
158 | struct extent_buffer *eb; | |
240f62c8 | 159 | |
3083ee2e JB |
160 | while (1) { |
161 | rcu_read_lock(); | |
162 | eb = rcu_dereference(root->node); | |
163 | ||
164 | /* | |
165 | * RCU really hurts here, we could free up the root node because | |
166 | * it was cow'ed but we may not get the new root node yet so do | |
167 | * the inc_not_zero dance and if it doesn't work then | |
168 | * synchronize_rcu and try again. | |
169 | */ | |
170 | if (atomic_inc_not_zero(&eb->refs)) { | |
171 | rcu_read_unlock(); | |
172 | break; | |
173 | } | |
174 | rcu_read_unlock(); | |
175 | synchronize_rcu(); | |
176 | } | |
925baedd CM |
177 | return eb; |
178 | } | |
179 | ||
d352ac68 CM |
180 | /* loop around taking references on and locking the root node of the |
181 | * tree until you end up with a lock on the root. A locked buffer | |
182 | * is returned, with a reference held. | |
183 | */ | |
925baedd CM |
184 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
185 | { | |
186 | struct extent_buffer *eb; | |
187 | ||
d397712b | 188 | while (1) { |
925baedd CM |
189 | eb = btrfs_root_node(root); |
190 | btrfs_tree_lock(eb); | |
240f62c8 | 191 | if (eb == root->node) |
925baedd | 192 | break; |
925baedd CM |
193 | btrfs_tree_unlock(eb); |
194 | free_extent_buffer(eb); | |
195 | } | |
196 | return eb; | |
197 | } | |
198 | ||
bd681513 CM |
199 | /* loop around taking references on and locking the root node of the |
200 | * tree until you end up with a lock on the root. A locked buffer | |
201 | * is returned, with a reference held. | |
202 | */ | |
203 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) | |
204 | { | |
205 | struct extent_buffer *eb; | |
206 | ||
207 | while (1) { | |
208 | eb = btrfs_root_node(root); | |
209 | btrfs_tree_read_lock(eb); | |
210 | if (eb == root->node) | |
211 | break; | |
212 | btrfs_tree_read_unlock(eb); | |
213 | free_extent_buffer(eb); | |
214 | } | |
215 | return eb; | |
216 | } | |
217 | ||
d352ac68 CM |
218 | /* cowonly root (everything not a reference counted cow subvolume), just get |
219 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
220 | * get properly updated on disk. | |
221 | */ | |
0b86a832 CM |
222 | static void add_root_to_dirty_list(struct btrfs_root *root) |
223 | { | |
e5846fc6 | 224 | spin_lock(&root->fs_info->trans_lock); |
0b86a832 CM |
225 | if (root->track_dirty && list_empty(&root->dirty_list)) { |
226 | list_add(&root->dirty_list, | |
227 | &root->fs_info->dirty_cowonly_roots); | |
228 | } | |
e5846fc6 | 229 | spin_unlock(&root->fs_info->trans_lock); |
0b86a832 CM |
230 | } |
231 | ||
d352ac68 CM |
232 | /* |
233 | * used by snapshot creation to make a copy of a root for a tree with | |
234 | * a given objectid. The buffer with the new root node is returned in | |
235 | * cow_ret, and this func returns zero on success or a negative error code. | |
236 | */ | |
be20aa9d CM |
237 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
238 | struct btrfs_root *root, | |
239 | struct extent_buffer *buf, | |
240 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
241 | { | |
242 | struct extent_buffer *cow; | |
be20aa9d CM |
243 | int ret = 0; |
244 | int level; | |
5d4f98a2 | 245 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
246 | |
247 | WARN_ON(root->ref_cows && trans->transid != | |
248 | root->fs_info->running_transaction->transid); | |
249 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
250 | ||
251 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
252 | if (level == 0) |
253 | btrfs_item_key(buf, &disk_key, 0); | |
254 | else | |
255 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 256 | |
5d4f98a2 YZ |
257 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
258 | new_root_objectid, &disk_key, level, | |
5581a51a | 259 | buf->start, 0); |
5d4f98a2 | 260 | if (IS_ERR(cow)) |
be20aa9d CM |
261 | return PTR_ERR(cow); |
262 | ||
263 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
264 | btrfs_set_header_bytenr(cow, cow->start); | |
265 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
266 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
267 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
268 | BTRFS_HEADER_FLAG_RELOC); | |
269 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
270 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
271 | else | |
272 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 273 | |
2b82032c YZ |
274 | write_extent_buffer(cow, root->fs_info->fsid, |
275 | (unsigned long)btrfs_header_fsid(cow), | |
276 | BTRFS_FSID_SIZE); | |
277 | ||
be20aa9d | 278 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 279 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 280 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 281 | else |
66d7e7f0 | 282 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 283 | |
be20aa9d CM |
284 | if (ret) |
285 | return ret; | |
286 | ||
287 | btrfs_mark_buffer_dirty(cow); | |
288 | *cow_ret = cow; | |
289 | return 0; | |
290 | } | |
291 | ||
bd989ba3 JS |
292 | enum mod_log_op { |
293 | MOD_LOG_KEY_REPLACE, | |
294 | MOD_LOG_KEY_ADD, | |
295 | MOD_LOG_KEY_REMOVE, | |
296 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
297 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
298 | MOD_LOG_MOVE_KEYS, | |
299 | MOD_LOG_ROOT_REPLACE, | |
300 | }; | |
301 | ||
302 | struct tree_mod_move { | |
303 | int dst_slot; | |
304 | int nr_items; | |
305 | }; | |
306 | ||
307 | struct tree_mod_root { | |
308 | u64 logical; | |
309 | u8 level; | |
310 | }; | |
311 | ||
312 | struct tree_mod_elem { | |
313 | struct rb_node node; | |
314 | u64 index; /* shifted logical */ | |
315 | struct seq_list elem; | |
316 | enum mod_log_op op; | |
317 | ||
318 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
319 | int slot; | |
320 | ||
321 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
322 | u64 generation; | |
323 | ||
324 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
325 | struct btrfs_disk_key key; | |
326 | u64 blockptr; | |
327 | ||
328 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
329 | struct tree_mod_move move; | |
330 | ||
331 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
332 | struct tree_mod_root old_root; | |
333 | }; | |
334 | ||
335 | static inline void | |
336 | __get_tree_mod_seq(struct btrfs_fs_info *fs_info, struct seq_list *elem) | |
337 | { | |
338 | elem->seq = atomic_inc_return(&fs_info->tree_mod_seq); | |
339 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); | |
340 | } | |
341 | ||
342 | void btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
343 | struct seq_list *elem) | |
344 | { | |
345 | elem->flags = 1; | |
346 | spin_lock(&fs_info->tree_mod_seq_lock); | |
347 | __get_tree_mod_seq(fs_info, elem); | |
348 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
349 | } | |
350 | ||
351 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
352 | struct seq_list *elem) | |
353 | { | |
354 | struct rb_root *tm_root; | |
355 | struct rb_node *node; | |
356 | struct rb_node *next; | |
357 | struct seq_list *cur_elem; | |
358 | struct tree_mod_elem *tm; | |
359 | u64 min_seq = (u64)-1; | |
360 | u64 seq_putting = elem->seq; | |
361 | ||
362 | if (!seq_putting) | |
363 | return; | |
364 | ||
365 | BUG_ON(!(elem->flags & 1)); | |
366 | spin_lock(&fs_info->tree_mod_seq_lock); | |
367 | list_del(&elem->list); | |
368 | ||
369 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
370 | if ((cur_elem->flags & 1) && cur_elem->seq < min_seq) { | |
371 | if (seq_putting > cur_elem->seq) { | |
372 | /* | |
373 | * blocker with lower sequence number exists, we | |
374 | * cannot remove anything from the log | |
375 | */ | |
376 | goto out; | |
377 | } | |
378 | min_seq = cur_elem->seq; | |
379 | } | |
380 | } | |
381 | ||
382 | /* | |
383 | * anything that's lower than the lowest existing (read: blocked) | |
384 | * sequence number can be removed from the tree. | |
385 | */ | |
386 | write_lock(&fs_info->tree_mod_log_lock); | |
387 | tm_root = &fs_info->tree_mod_log; | |
388 | for (node = rb_first(tm_root); node; node = next) { | |
389 | next = rb_next(node); | |
390 | tm = container_of(node, struct tree_mod_elem, node); | |
391 | if (tm->elem.seq > min_seq) | |
392 | continue; | |
393 | rb_erase(node, tm_root); | |
394 | list_del(&tm->elem.list); | |
395 | kfree(tm); | |
396 | } | |
397 | write_unlock(&fs_info->tree_mod_log_lock); | |
398 | out: | |
399 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
400 | } | |
401 | ||
402 | /* | |
403 | * key order of the log: | |
404 | * index -> sequence | |
405 | * | |
406 | * the index is the shifted logical of the *new* root node for root replace | |
407 | * operations, or the shifted logical of the affected block for all other | |
408 | * operations. | |
409 | */ | |
410 | static noinline int | |
411 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
412 | { | |
413 | struct rb_root *tm_root; | |
414 | struct rb_node **new; | |
415 | struct rb_node *parent = NULL; | |
416 | struct tree_mod_elem *cur; | |
417 | int ret = 0; | |
418 | ||
419 | BUG_ON(!tm || !tm->elem.seq); | |
420 | ||
421 | write_lock(&fs_info->tree_mod_log_lock); | |
422 | tm_root = &fs_info->tree_mod_log; | |
423 | new = &tm_root->rb_node; | |
424 | while (*new) { | |
425 | cur = container_of(*new, struct tree_mod_elem, node); | |
426 | parent = *new; | |
427 | if (cur->index < tm->index) | |
428 | new = &((*new)->rb_left); | |
429 | else if (cur->index > tm->index) | |
430 | new = &((*new)->rb_right); | |
431 | else if (cur->elem.seq < tm->elem.seq) | |
432 | new = &((*new)->rb_left); | |
433 | else if (cur->elem.seq > tm->elem.seq) | |
434 | new = &((*new)->rb_right); | |
435 | else { | |
436 | kfree(tm); | |
437 | ret = -EEXIST; | |
438 | goto unlock; | |
439 | } | |
440 | } | |
441 | ||
442 | rb_link_node(&tm->node, parent, new); | |
443 | rb_insert_color(&tm->node, tm_root); | |
444 | unlock: | |
445 | write_unlock(&fs_info->tree_mod_log_lock); | |
446 | return ret; | |
447 | } | |
448 | ||
449 | int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags, | |
450 | struct tree_mod_elem **tm_ret) | |
451 | { | |
452 | struct tree_mod_elem *tm; | |
453 | u64 seq = 0; | |
454 | ||
455 | smp_mb(); | |
456 | if (list_empty(&fs_info->tree_mod_seq_list)) | |
457 | return 0; | |
458 | ||
459 | tm = *tm_ret = kzalloc(sizeof(*tm), flags); | |
460 | if (!tm) | |
461 | return -ENOMEM; | |
462 | ||
463 | __get_tree_mod_seq(fs_info, &tm->elem); | |
464 | seq = tm->elem.seq; | |
465 | tm->elem.flags = 0; | |
466 | ||
467 | return seq; | |
468 | } | |
469 | ||
470 | static noinline int | |
471 | tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info, | |
472 | struct extent_buffer *eb, int slot, | |
473 | enum mod_log_op op, gfp_t flags) | |
474 | { | |
475 | struct tree_mod_elem *tm; | |
476 | int ret; | |
477 | ||
478 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
479 | if (ret <= 0) | |
480 | return ret; | |
481 | ||
482 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
483 | if (op != MOD_LOG_KEY_ADD) { | |
484 | btrfs_node_key(eb, &tm->key, slot); | |
485 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
486 | } | |
487 | tm->op = op; | |
488 | tm->slot = slot; | |
489 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
490 | ||
491 | return __tree_mod_log_insert(fs_info, tm); | |
492 | } | |
493 | ||
494 | static noinline int | |
495 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, | |
496 | int slot, enum mod_log_op op) | |
497 | { | |
498 | return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS); | |
499 | } | |
500 | ||
501 | static noinline int | |
502 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
503 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
504 | int nr_items, gfp_t flags) | |
505 | { | |
506 | struct tree_mod_elem *tm; | |
507 | int ret; | |
508 | int i; | |
509 | ||
510 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
511 | if (ret <= 0) | |
512 | return ret; | |
513 | ||
514 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
515 | ret = tree_mod_log_insert_key(fs_info, eb, i + dst_slot, | |
516 | MOD_LOG_KEY_REMOVE_WHILE_MOVING); | |
517 | BUG_ON(ret < 0); | |
518 | } | |
519 | ||
520 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
521 | tm->slot = src_slot; | |
522 | tm->move.dst_slot = dst_slot; | |
523 | tm->move.nr_items = nr_items; | |
524 | tm->op = MOD_LOG_MOVE_KEYS; | |
525 | ||
526 | return __tree_mod_log_insert(fs_info, tm); | |
527 | } | |
528 | ||
529 | static noinline int | |
530 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
531 | struct extent_buffer *old_root, | |
532 | struct extent_buffer *new_root, gfp_t flags) | |
533 | { | |
534 | struct tree_mod_elem *tm; | |
535 | int ret; | |
536 | ||
537 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
538 | if (ret <= 0) | |
539 | return ret; | |
540 | ||
541 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
542 | tm->old_root.logical = old_root->start; | |
543 | tm->old_root.level = btrfs_header_level(old_root); | |
544 | tm->generation = btrfs_header_generation(old_root); | |
545 | tm->op = MOD_LOG_ROOT_REPLACE; | |
546 | ||
547 | return __tree_mod_log_insert(fs_info, tm); | |
548 | } | |
549 | ||
550 | static struct tree_mod_elem * | |
551 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
552 | int smallest) | |
553 | { | |
554 | struct rb_root *tm_root; | |
555 | struct rb_node *node; | |
556 | struct tree_mod_elem *cur = NULL; | |
557 | struct tree_mod_elem *found = NULL; | |
558 | u64 index = start >> PAGE_CACHE_SHIFT; | |
559 | ||
560 | read_lock(&fs_info->tree_mod_log_lock); | |
561 | tm_root = &fs_info->tree_mod_log; | |
562 | node = tm_root->rb_node; | |
563 | while (node) { | |
564 | cur = container_of(node, struct tree_mod_elem, node); | |
565 | if (cur->index < index) { | |
566 | node = node->rb_left; | |
567 | } else if (cur->index > index) { | |
568 | node = node->rb_right; | |
569 | } else if (cur->elem.seq < min_seq) { | |
570 | node = node->rb_left; | |
571 | } else if (!smallest) { | |
572 | /* we want the node with the highest seq */ | |
573 | if (found) | |
574 | BUG_ON(found->elem.seq > cur->elem.seq); | |
575 | found = cur; | |
576 | node = node->rb_left; | |
577 | } else if (cur->elem.seq > min_seq) { | |
578 | /* we want the node with the smallest seq */ | |
579 | if (found) | |
580 | BUG_ON(found->elem.seq < cur->elem.seq); | |
581 | found = cur; | |
582 | node = node->rb_right; | |
583 | } else { | |
584 | found = cur; | |
585 | break; | |
586 | } | |
587 | } | |
588 | read_unlock(&fs_info->tree_mod_log_lock); | |
589 | ||
590 | return found; | |
591 | } | |
592 | ||
593 | /* | |
594 | * this returns the element from the log with the smallest time sequence | |
595 | * value that's in the log (the oldest log item). any element with a time | |
596 | * sequence lower than min_seq will be ignored. | |
597 | */ | |
598 | static struct tree_mod_elem * | |
599 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
600 | u64 min_seq) | |
601 | { | |
602 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
603 | } | |
604 | ||
605 | /* | |
606 | * this returns the element from the log with the largest time sequence | |
607 | * value that's in the log (the most recent log item). any element with | |
608 | * a time sequence lower than min_seq will be ignored. | |
609 | */ | |
610 | static struct tree_mod_elem * | |
611 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
612 | { | |
613 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
614 | } | |
615 | ||
616 | static inline void | |
617 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
618 | struct extent_buffer *src, unsigned long dst_offset, | |
619 | unsigned long src_offset, int nr_items) | |
620 | { | |
621 | int ret; | |
622 | int i; | |
623 | ||
624 | smp_mb(); | |
625 | if (list_empty(&fs_info->tree_mod_seq_list)) | |
626 | return; | |
627 | ||
628 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) | |
629 | return; | |
630 | ||
631 | /* speed this up by single seq for all operations? */ | |
632 | for (i = 0; i < nr_items; i++) { | |
633 | ret = tree_mod_log_insert_key(fs_info, src, i + src_offset, | |
634 | MOD_LOG_KEY_REMOVE); | |
635 | BUG_ON(ret < 0); | |
636 | ret = tree_mod_log_insert_key(fs_info, dst, i + dst_offset, | |
637 | MOD_LOG_KEY_ADD); | |
638 | BUG_ON(ret < 0); | |
639 | } | |
640 | } | |
641 | ||
642 | static inline void | |
643 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
644 | int dst_offset, int src_offset, int nr_items) | |
645 | { | |
646 | int ret; | |
647 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
648 | nr_items, GFP_NOFS); | |
649 | BUG_ON(ret < 0); | |
650 | } | |
651 | ||
652 | static inline void | |
653 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, | |
654 | struct extent_buffer *eb, | |
655 | struct btrfs_disk_key *disk_key, int slot, int atomic) | |
656 | { | |
657 | int ret; | |
658 | ||
659 | ret = tree_mod_log_insert_key_mask(fs_info, eb, slot, | |
660 | MOD_LOG_KEY_REPLACE, | |
661 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
662 | BUG_ON(ret < 0); | |
663 | } | |
664 | ||
665 | static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
666 | struct extent_buffer *eb) | |
667 | { | |
668 | int i; | |
669 | int ret; | |
670 | u32 nritems; | |
671 | ||
672 | smp_mb(); | |
673 | if (list_empty(&fs_info->tree_mod_seq_list)) | |
674 | return; | |
675 | ||
676 | if (btrfs_header_level(eb) == 0) | |
677 | return; | |
678 | ||
679 | nritems = btrfs_header_nritems(eb); | |
680 | for (i = nritems - 1; i >= 0; i--) { | |
681 | ret = tree_mod_log_insert_key(fs_info, eb, i, | |
682 | MOD_LOG_KEY_REMOVE_WHILE_FREEING); | |
683 | BUG_ON(ret < 0); | |
684 | } | |
685 | } | |
686 | ||
687 | static inline void | |
688 | tree_mod_log_set_root_pointer(struct btrfs_root *root, | |
689 | struct extent_buffer *new_root_node) | |
690 | { | |
691 | int ret; | |
692 | tree_mod_log_free_eb(root->fs_info, root->node); | |
693 | ret = tree_mod_log_insert_root(root->fs_info, root->node, | |
694 | new_root_node, GFP_NOFS); | |
695 | BUG_ON(ret < 0); | |
696 | } | |
697 | ||
5d4f98a2 YZ |
698 | /* |
699 | * check if the tree block can be shared by multiple trees | |
700 | */ | |
701 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
702 | struct extent_buffer *buf) | |
703 | { | |
704 | /* | |
705 | * Tree blocks not in refernece counted trees and tree roots | |
706 | * are never shared. If a block was allocated after the last | |
707 | * snapshot and the block was not allocated by tree relocation, | |
708 | * we know the block is not shared. | |
709 | */ | |
710 | if (root->ref_cows && | |
711 | buf != root->node && buf != root->commit_root && | |
712 | (btrfs_header_generation(buf) <= | |
713 | btrfs_root_last_snapshot(&root->root_item) || | |
714 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
715 | return 1; | |
716 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
717 | if (root->ref_cows && | |
718 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
719 | return 1; | |
720 | #endif | |
721 | return 0; | |
722 | } | |
723 | ||
724 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
725 | struct btrfs_root *root, | |
726 | struct extent_buffer *buf, | |
f0486c68 YZ |
727 | struct extent_buffer *cow, |
728 | int *last_ref) | |
5d4f98a2 YZ |
729 | { |
730 | u64 refs; | |
731 | u64 owner; | |
732 | u64 flags; | |
733 | u64 new_flags = 0; | |
734 | int ret; | |
735 | ||
736 | /* | |
737 | * Backrefs update rules: | |
738 | * | |
739 | * Always use full backrefs for extent pointers in tree block | |
740 | * allocated by tree relocation. | |
741 | * | |
742 | * If a shared tree block is no longer referenced by its owner | |
743 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
744 | * use full backrefs for extent pointers in tree block. | |
745 | * | |
746 | * If a tree block is been relocating | |
747 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
748 | * use full backrefs for extent pointers in tree block. | |
749 | * The reason for this is some operations (such as drop tree) | |
750 | * are only allowed for blocks use full backrefs. | |
751 | */ | |
752 | ||
753 | if (btrfs_block_can_be_shared(root, buf)) { | |
754 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
755 | buf->len, &refs, &flags); | |
be1a5564 MF |
756 | if (ret) |
757 | return ret; | |
e5df9573 MF |
758 | if (refs == 0) { |
759 | ret = -EROFS; | |
760 | btrfs_std_error(root->fs_info, ret); | |
761 | return ret; | |
762 | } | |
5d4f98a2 YZ |
763 | } else { |
764 | refs = 1; | |
765 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
766 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
767 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
768 | else | |
769 | flags = 0; | |
770 | } | |
771 | ||
772 | owner = btrfs_header_owner(buf); | |
773 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
774 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
775 | ||
776 | if (refs > 1) { | |
777 | if ((owner == root->root_key.objectid || | |
778 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
779 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 780 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 781 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
782 | |
783 | if (root->root_key.objectid == | |
784 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 785 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 786 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 787 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 788 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
789 | } |
790 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
791 | } else { | |
792 | ||
793 | if (root->root_key.objectid == | |
794 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 795 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 796 | else |
66d7e7f0 | 797 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 798 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
799 | } |
800 | if (new_flags != 0) { | |
801 | ret = btrfs_set_disk_extent_flags(trans, root, | |
802 | buf->start, | |
803 | buf->len, | |
804 | new_flags, 0); | |
be1a5564 MF |
805 | if (ret) |
806 | return ret; | |
5d4f98a2 YZ |
807 | } |
808 | } else { | |
809 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
810 | if (root->root_key.objectid == | |
811 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 812 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 813 | else |
66d7e7f0 | 814 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 815 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 816 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 817 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
818 | } |
819 | clean_tree_block(trans, root, buf); | |
f0486c68 | 820 | *last_ref = 1; |
5d4f98a2 YZ |
821 | } |
822 | return 0; | |
823 | } | |
824 | ||
d352ac68 | 825 | /* |
d397712b CM |
826 | * does the dirty work in cow of a single block. The parent block (if |
827 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
828 | * dirty and returned locked. If you modify the block it needs to be marked | |
829 | * dirty again. | |
d352ac68 CM |
830 | * |
831 | * search_start -- an allocation hint for the new block | |
832 | * | |
d397712b CM |
833 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
834 | * bytes the allocator should try to find free next to the block it returns. | |
835 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 836 | */ |
d397712b | 837 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
838 | struct btrfs_root *root, |
839 | struct extent_buffer *buf, | |
840 | struct extent_buffer *parent, int parent_slot, | |
841 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 842 | u64 search_start, u64 empty_size) |
02217ed2 | 843 | { |
5d4f98a2 | 844 | struct btrfs_disk_key disk_key; |
5f39d397 | 845 | struct extent_buffer *cow; |
be1a5564 | 846 | int level, ret; |
f0486c68 | 847 | int last_ref = 0; |
925baedd | 848 | int unlock_orig = 0; |
5d4f98a2 | 849 | u64 parent_start; |
7bb86316 | 850 | |
925baedd CM |
851 | if (*cow_ret == buf) |
852 | unlock_orig = 1; | |
853 | ||
b9447ef8 | 854 | btrfs_assert_tree_locked(buf); |
925baedd | 855 | |
7bb86316 CM |
856 | WARN_ON(root->ref_cows && trans->transid != |
857 | root->fs_info->running_transaction->transid); | |
6702ed49 | 858 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 859 | |
7bb86316 | 860 | level = btrfs_header_level(buf); |
31840ae1 | 861 | |
5d4f98a2 YZ |
862 | if (level == 0) |
863 | btrfs_item_key(buf, &disk_key, 0); | |
864 | else | |
865 | btrfs_node_key(buf, &disk_key, 0); | |
866 | ||
867 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
868 | if (parent) | |
869 | parent_start = parent->start; | |
870 | else | |
871 | parent_start = 0; | |
872 | } else | |
873 | parent_start = 0; | |
874 | ||
875 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
876 | root->root_key.objectid, &disk_key, | |
5581a51a | 877 | level, search_start, empty_size); |
54aa1f4d CM |
878 | if (IS_ERR(cow)) |
879 | return PTR_ERR(cow); | |
6702ed49 | 880 | |
b4ce94de CM |
881 | /* cow is set to blocking by btrfs_init_new_buffer */ |
882 | ||
5f39d397 | 883 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 884 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 885 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
886 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
887 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
888 | BTRFS_HEADER_FLAG_RELOC); | |
889 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
890 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
891 | else | |
892 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 893 | |
2b82032c YZ |
894 | write_extent_buffer(cow, root->fs_info->fsid, |
895 | (unsigned long)btrfs_header_fsid(cow), | |
896 | BTRFS_FSID_SIZE); | |
897 | ||
be1a5564 | 898 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 899 | if (ret) { |
79787eaa | 900 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
901 | return ret; |
902 | } | |
1a40e23b | 903 | |
3fd0a558 YZ |
904 | if (root->ref_cows) |
905 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
906 | ||
02217ed2 | 907 | if (buf == root->node) { |
925baedd | 908 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
909 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
910 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
911 | parent_start = buf->start; | |
912 | else | |
913 | parent_start = 0; | |
925baedd | 914 | |
5f39d397 | 915 | extent_buffer_get(cow); |
240f62c8 | 916 | rcu_assign_pointer(root->node, cow); |
925baedd | 917 | |
f0486c68 | 918 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 919 | last_ref); |
5f39d397 | 920 | free_extent_buffer(buf); |
0b86a832 | 921 | add_root_to_dirty_list(root); |
02217ed2 | 922 | } else { |
5d4f98a2 YZ |
923 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
924 | parent_start = parent->start; | |
925 | else | |
926 | parent_start = 0; | |
927 | ||
928 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
5f39d397 | 929 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 930 | cow->start); |
74493f7a CM |
931 | btrfs_set_node_ptr_generation(parent, parent_slot, |
932 | trans->transid); | |
d6025579 | 933 | btrfs_mark_buffer_dirty(parent); |
f0486c68 | 934 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 935 | last_ref); |
02217ed2 | 936 | } |
925baedd CM |
937 | if (unlock_orig) |
938 | btrfs_tree_unlock(buf); | |
3083ee2e | 939 | free_extent_buffer_stale(buf); |
ccd467d6 | 940 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 941 | *cow_ret = cow; |
02217ed2 CM |
942 | return 0; |
943 | } | |
944 | ||
5d4f98a2 YZ |
945 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
946 | struct btrfs_root *root, | |
947 | struct extent_buffer *buf) | |
948 | { | |
f1ebcc74 LB |
949 | /* ensure we can see the force_cow */ |
950 | smp_rmb(); | |
951 | ||
952 | /* | |
953 | * We do not need to cow a block if | |
954 | * 1) this block is not created or changed in this transaction; | |
955 | * 2) this block does not belong to TREE_RELOC tree; | |
956 | * 3) the root is not forced COW. | |
957 | * | |
958 | * What is forced COW: | |
959 | * when we create snapshot during commiting the transaction, | |
960 | * after we've finished coping src root, we must COW the shared | |
961 | * block to ensure the metadata consistency. | |
962 | */ | |
5d4f98a2 YZ |
963 | if (btrfs_header_generation(buf) == trans->transid && |
964 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
965 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
966 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
967 | !root->force_cow) | |
5d4f98a2 YZ |
968 | return 0; |
969 | return 1; | |
970 | } | |
971 | ||
d352ac68 CM |
972 | /* |
973 | * cows a single block, see __btrfs_cow_block for the real work. | |
974 | * This version of it has extra checks so that a block isn't cow'd more than | |
975 | * once per transaction, as long as it hasn't been written yet | |
976 | */ | |
d397712b | 977 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
978 | struct btrfs_root *root, struct extent_buffer *buf, |
979 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 980 | struct extent_buffer **cow_ret) |
6702ed49 CM |
981 | { |
982 | u64 search_start; | |
f510cfec | 983 | int ret; |
dc17ff8f | 984 | |
6702ed49 | 985 | if (trans->transaction != root->fs_info->running_transaction) { |
d397712b CM |
986 | printk(KERN_CRIT "trans %llu running %llu\n", |
987 | (unsigned long long)trans->transid, | |
988 | (unsigned long long) | |
6702ed49 CM |
989 | root->fs_info->running_transaction->transid); |
990 | WARN_ON(1); | |
991 | } | |
992 | if (trans->transid != root->fs_info->generation) { | |
d397712b CM |
993 | printk(KERN_CRIT "trans %llu running %llu\n", |
994 | (unsigned long long)trans->transid, | |
995 | (unsigned long long)root->fs_info->generation); | |
6702ed49 CM |
996 | WARN_ON(1); |
997 | } | |
dc17ff8f | 998 | |
5d4f98a2 | 999 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1000 | *cow_ret = buf; |
1001 | return 0; | |
1002 | } | |
c487685d | 1003 | |
0b86a832 | 1004 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1005 | |
1006 | if (parent) | |
1007 | btrfs_set_lock_blocking(parent); | |
1008 | btrfs_set_lock_blocking(buf); | |
1009 | ||
f510cfec | 1010 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1011 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1012 | |
1013 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1014 | ||
f510cfec | 1015 | return ret; |
6702ed49 CM |
1016 | } |
1017 | ||
d352ac68 CM |
1018 | /* |
1019 | * helper function for defrag to decide if two blocks pointed to by a | |
1020 | * node are actually close by | |
1021 | */ | |
6b80053d | 1022 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1023 | { |
6b80053d | 1024 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1025 | return 1; |
6b80053d | 1026 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1027 | return 1; |
1028 | return 0; | |
1029 | } | |
1030 | ||
081e9573 CM |
1031 | /* |
1032 | * compare two keys in a memcmp fashion | |
1033 | */ | |
1034 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1035 | { | |
1036 | struct btrfs_key k1; | |
1037 | ||
1038 | btrfs_disk_key_to_cpu(&k1, disk); | |
1039 | ||
20736aba | 1040 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1041 | } |
1042 | ||
f3465ca4 JB |
1043 | /* |
1044 | * same as comp_keys only with two btrfs_key's | |
1045 | */ | |
5d4f98a2 | 1046 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1047 | { |
1048 | if (k1->objectid > k2->objectid) | |
1049 | return 1; | |
1050 | if (k1->objectid < k2->objectid) | |
1051 | return -1; | |
1052 | if (k1->type > k2->type) | |
1053 | return 1; | |
1054 | if (k1->type < k2->type) | |
1055 | return -1; | |
1056 | if (k1->offset > k2->offset) | |
1057 | return 1; | |
1058 | if (k1->offset < k2->offset) | |
1059 | return -1; | |
1060 | return 0; | |
1061 | } | |
081e9573 | 1062 | |
d352ac68 CM |
1063 | /* |
1064 | * this is used by the defrag code to go through all the | |
1065 | * leaves pointed to by a node and reallocate them so that | |
1066 | * disk order is close to key order | |
1067 | */ | |
6702ed49 | 1068 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1069 | struct btrfs_root *root, struct extent_buffer *parent, |
a6b6e75e CM |
1070 | int start_slot, int cache_only, u64 *last_ret, |
1071 | struct btrfs_key *progress) | |
6702ed49 | 1072 | { |
6b80053d | 1073 | struct extent_buffer *cur; |
6702ed49 | 1074 | u64 blocknr; |
ca7a79ad | 1075 | u64 gen; |
e9d0b13b CM |
1076 | u64 search_start = *last_ret; |
1077 | u64 last_block = 0; | |
6702ed49 CM |
1078 | u64 other; |
1079 | u32 parent_nritems; | |
6702ed49 CM |
1080 | int end_slot; |
1081 | int i; | |
1082 | int err = 0; | |
f2183bde | 1083 | int parent_level; |
6b80053d CM |
1084 | int uptodate; |
1085 | u32 blocksize; | |
081e9573 CM |
1086 | int progress_passed = 0; |
1087 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1088 | |
5708b959 CM |
1089 | parent_level = btrfs_header_level(parent); |
1090 | if (cache_only && parent_level != 1) | |
1091 | return 0; | |
1092 | ||
d397712b | 1093 | if (trans->transaction != root->fs_info->running_transaction) |
6702ed49 | 1094 | WARN_ON(1); |
d397712b | 1095 | if (trans->transid != root->fs_info->generation) |
6702ed49 | 1096 | WARN_ON(1); |
86479a04 | 1097 | |
6b80053d | 1098 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1099 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1100 | end_slot = parent_nritems; |
1101 | ||
1102 | if (parent_nritems == 1) | |
1103 | return 0; | |
1104 | ||
b4ce94de CM |
1105 | btrfs_set_lock_blocking(parent); |
1106 | ||
6702ed49 CM |
1107 | for (i = start_slot; i < end_slot; i++) { |
1108 | int close = 1; | |
a6b6e75e | 1109 | |
081e9573 CM |
1110 | btrfs_node_key(parent, &disk_key, i); |
1111 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1112 | continue; | |
1113 | ||
1114 | progress_passed = 1; | |
6b80053d | 1115 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1116 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1117 | if (last_block == 0) |
1118 | last_block = blocknr; | |
5708b959 | 1119 | |
6702ed49 | 1120 | if (i > 0) { |
6b80053d CM |
1121 | other = btrfs_node_blockptr(parent, i - 1); |
1122 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1123 | } |
0ef3e66b | 1124 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1125 | other = btrfs_node_blockptr(parent, i + 1); |
1126 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1127 | } |
e9d0b13b CM |
1128 | if (close) { |
1129 | last_block = blocknr; | |
6702ed49 | 1130 | continue; |
e9d0b13b | 1131 | } |
6702ed49 | 1132 | |
6b80053d CM |
1133 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1134 | if (cur) | |
b9fab919 | 1135 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1136 | else |
1137 | uptodate = 0; | |
5708b959 | 1138 | if (!cur || !uptodate) { |
6702ed49 | 1139 | if (cache_only) { |
6b80053d | 1140 | free_extent_buffer(cur); |
6702ed49 CM |
1141 | continue; |
1142 | } | |
6b80053d CM |
1143 | if (!cur) { |
1144 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1145 | blocksize, gen); |
97d9a8a4 TI |
1146 | if (!cur) |
1147 | return -EIO; | |
6b80053d | 1148 | } else if (!uptodate) { |
ca7a79ad | 1149 | btrfs_read_buffer(cur, gen); |
f2183bde | 1150 | } |
6702ed49 | 1151 | } |
e9d0b13b | 1152 | if (search_start == 0) |
6b80053d | 1153 | search_start = last_block; |
e9d0b13b | 1154 | |
e7a84565 | 1155 | btrfs_tree_lock(cur); |
b4ce94de | 1156 | btrfs_set_lock_blocking(cur); |
6b80053d | 1157 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1158 | &cur, search_start, |
6b80053d | 1159 | min(16 * blocksize, |
9fa8cfe7 | 1160 | (end_slot - i) * blocksize)); |
252c38f0 | 1161 | if (err) { |
e7a84565 | 1162 | btrfs_tree_unlock(cur); |
6b80053d | 1163 | free_extent_buffer(cur); |
6702ed49 | 1164 | break; |
252c38f0 | 1165 | } |
e7a84565 CM |
1166 | search_start = cur->start; |
1167 | last_block = cur->start; | |
f2183bde | 1168 | *last_ret = search_start; |
e7a84565 CM |
1169 | btrfs_tree_unlock(cur); |
1170 | free_extent_buffer(cur); | |
6702ed49 CM |
1171 | } |
1172 | return err; | |
1173 | } | |
1174 | ||
74123bd7 CM |
1175 | /* |
1176 | * The leaf data grows from end-to-front in the node. | |
1177 | * this returns the address of the start of the last item, | |
1178 | * which is the stop of the leaf data stack | |
1179 | */ | |
123abc88 | 1180 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1181 | struct extent_buffer *leaf) |
be0e5c09 | 1182 | { |
5f39d397 | 1183 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1184 | if (nr == 0) |
123abc88 | 1185 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1186 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1187 | } |
1188 | ||
aa5d6bed | 1189 | |
74123bd7 | 1190 | /* |
5f39d397 CM |
1191 | * search for key in the extent_buffer. The items start at offset p, |
1192 | * and they are item_size apart. There are 'max' items in p. | |
1193 | * | |
74123bd7 CM |
1194 | * the slot in the array is returned via slot, and it points to |
1195 | * the place where you would insert key if it is not found in | |
1196 | * the array. | |
1197 | * | |
1198 | * slot may point to max if the key is bigger than all of the keys | |
1199 | */ | |
e02119d5 CM |
1200 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1201 | unsigned long p, | |
1202 | int item_size, struct btrfs_key *key, | |
1203 | int max, int *slot) | |
be0e5c09 CM |
1204 | { |
1205 | int low = 0; | |
1206 | int high = max; | |
1207 | int mid; | |
1208 | int ret; | |
479965d6 | 1209 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1210 | struct btrfs_disk_key unaligned; |
1211 | unsigned long offset; | |
5f39d397 CM |
1212 | char *kaddr = NULL; |
1213 | unsigned long map_start = 0; | |
1214 | unsigned long map_len = 0; | |
479965d6 | 1215 | int err; |
be0e5c09 | 1216 | |
d397712b | 1217 | while (low < high) { |
be0e5c09 | 1218 | mid = (low + high) / 2; |
5f39d397 CM |
1219 | offset = p + mid * item_size; |
1220 | ||
a6591715 | 1221 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1222 | (offset + sizeof(struct btrfs_disk_key)) > |
1223 | map_start + map_len) { | |
934d375b CM |
1224 | |
1225 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1226 | sizeof(struct btrfs_disk_key), |
a6591715 | 1227 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1228 | |
1229 | if (!err) { | |
1230 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1231 | map_start); | |
1232 | } else { | |
1233 | read_extent_buffer(eb, &unaligned, | |
1234 | offset, sizeof(unaligned)); | |
1235 | tmp = &unaligned; | |
1236 | } | |
5f39d397 | 1237 | |
5f39d397 CM |
1238 | } else { |
1239 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1240 | map_start); | |
1241 | } | |
be0e5c09 CM |
1242 | ret = comp_keys(tmp, key); |
1243 | ||
1244 | if (ret < 0) | |
1245 | low = mid + 1; | |
1246 | else if (ret > 0) | |
1247 | high = mid; | |
1248 | else { | |
1249 | *slot = mid; | |
1250 | return 0; | |
1251 | } | |
1252 | } | |
1253 | *slot = low; | |
1254 | return 1; | |
1255 | } | |
1256 | ||
97571fd0 CM |
1257 | /* |
1258 | * simple bin_search frontend that does the right thing for | |
1259 | * leaves vs nodes | |
1260 | */ | |
5f39d397 CM |
1261 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1262 | int level, int *slot) | |
be0e5c09 | 1263 | { |
5f39d397 CM |
1264 | if (level == 0) { |
1265 | return generic_bin_search(eb, | |
1266 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1267 | sizeof(struct btrfs_item), |
5f39d397 | 1268 | key, btrfs_header_nritems(eb), |
7518a238 | 1269 | slot); |
be0e5c09 | 1270 | } else { |
5f39d397 CM |
1271 | return generic_bin_search(eb, |
1272 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1273 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1274 | key, btrfs_header_nritems(eb), |
7518a238 | 1275 | slot); |
be0e5c09 CM |
1276 | } |
1277 | return -1; | |
1278 | } | |
1279 | ||
5d4f98a2 YZ |
1280 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1281 | int level, int *slot) | |
1282 | { | |
1283 | return bin_search(eb, key, level, slot); | |
1284 | } | |
1285 | ||
f0486c68 YZ |
1286 | static void root_add_used(struct btrfs_root *root, u32 size) |
1287 | { | |
1288 | spin_lock(&root->accounting_lock); | |
1289 | btrfs_set_root_used(&root->root_item, | |
1290 | btrfs_root_used(&root->root_item) + size); | |
1291 | spin_unlock(&root->accounting_lock); | |
1292 | } | |
1293 | ||
1294 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1295 | { | |
1296 | spin_lock(&root->accounting_lock); | |
1297 | btrfs_set_root_used(&root->root_item, | |
1298 | btrfs_root_used(&root->root_item) - size); | |
1299 | spin_unlock(&root->accounting_lock); | |
1300 | } | |
1301 | ||
d352ac68 CM |
1302 | /* given a node and slot number, this reads the blocks it points to. The |
1303 | * extent buffer is returned with a reference taken (but unlocked). | |
1304 | * NULL is returned on error. | |
1305 | */ | |
e02119d5 | 1306 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1307 | struct extent_buffer *parent, int slot) |
bb803951 | 1308 | { |
ca7a79ad | 1309 | int level = btrfs_header_level(parent); |
bb803951 CM |
1310 | if (slot < 0) |
1311 | return NULL; | |
5f39d397 | 1312 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1313 | return NULL; |
ca7a79ad CM |
1314 | |
1315 | BUG_ON(level == 0); | |
1316 | ||
db94535d | 1317 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
1318 | btrfs_level_size(root, level - 1), |
1319 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
1320 | } |
1321 | ||
d352ac68 CM |
1322 | /* |
1323 | * node level balancing, used to make sure nodes are in proper order for | |
1324 | * item deletion. We balance from the top down, so we have to make sure | |
1325 | * that a deletion won't leave an node completely empty later on. | |
1326 | */ | |
e02119d5 | 1327 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1328 | struct btrfs_root *root, |
1329 | struct btrfs_path *path, int level) | |
bb803951 | 1330 | { |
5f39d397 CM |
1331 | struct extent_buffer *right = NULL; |
1332 | struct extent_buffer *mid; | |
1333 | struct extent_buffer *left = NULL; | |
1334 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1335 | int ret = 0; |
1336 | int wret; | |
1337 | int pslot; | |
bb803951 | 1338 | int orig_slot = path->slots[level]; |
79f95c82 | 1339 | u64 orig_ptr; |
bb803951 CM |
1340 | |
1341 | if (level == 0) | |
1342 | return 0; | |
1343 | ||
5f39d397 | 1344 | mid = path->nodes[level]; |
b4ce94de | 1345 | |
bd681513 CM |
1346 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1347 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1348 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1349 | ||
1d4f8a0c | 1350 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1351 | |
a05a9bb1 | 1352 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1353 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1354 | pslot = path->slots[level + 1]; |
1355 | } | |
bb803951 | 1356 | |
40689478 CM |
1357 | /* |
1358 | * deal with the case where there is only one pointer in the root | |
1359 | * by promoting the node below to a root | |
1360 | */ | |
5f39d397 CM |
1361 | if (!parent) { |
1362 | struct extent_buffer *child; | |
bb803951 | 1363 | |
5f39d397 | 1364 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1365 | return 0; |
1366 | ||
1367 | /* promote the child to a root */ | |
5f39d397 | 1368 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1369 | if (!child) { |
1370 | ret = -EROFS; | |
1371 | btrfs_std_error(root->fs_info, ret); | |
1372 | goto enospc; | |
1373 | } | |
1374 | ||
925baedd | 1375 | btrfs_tree_lock(child); |
b4ce94de | 1376 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1377 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1378 | if (ret) { |
1379 | btrfs_tree_unlock(child); | |
1380 | free_extent_buffer(child); | |
1381 | goto enospc; | |
1382 | } | |
2f375ab9 | 1383 | |
240f62c8 | 1384 | rcu_assign_pointer(root->node, child); |
925baedd | 1385 | |
0b86a832 | 1386 | add_root_to_dirty_list(root); |
925baedd | 1387 | btrfs_tree_unlock(child); |
b4ce94de | 1388 | |
925baedd | 1389 | path->locks[level] = 0; |
bb803951 | 1390 | path->nodes[level] = NULL; |
5f39d397 | 1391 | clean_tree_block(trans, root, mid); |
925baedd | 1392 | btrfs_tree_unlock(mid); |
bb803951 | 1393 | /* once for the path */ |
5f39d397 | 1394 | free_extent_buffer(mid); |
f0486c68 YZ |
1395 | |
1396 | root_sub_used(root, mid->len); | |
5581a51a | 1397 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1398 | /* once for the root ptr */ |
3083ee2e | 1399 | free_extent_buffer_stale(mid); |
f0486c68 | 1400 | return 0; |
bb803951 | 1401 | } |
5f39d397 | 1402 | if (btrfs_header_nritems(mid) > |
123abc88 | 1403 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1404 | return 0; |
1405 | ||
559af821 | 1406 | btrfs_header_nritems(mid); |
54aa1f4d | 1407 | |
5f39d397 CM |
1408 | left = read_node_slot(root, parent, pslot - 1); |
1409 | if (left) { | |
925baedd | 1410 | btrfs_tree_lock(left); |
b4ce94de | 1411 | btrfs_set_lock_blocking(left); |
5f39d397 | 1412 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1413 | parent, pslot - 1, &left); |
54aa1f4d CM |
1414 | if (wret) { |
1415 | ret = wret; | |
1416 | goto enospc; | |
1417 | } | |
2cc58cf2 | 1418 | } |
5f39d397 CM |
1419 | right = read_node_slot(root, parent, pslot + 1); |
1420 | if (right) { | |
925baedd | 1421 | btrfs_tree_lock(right); |
b4ce94de | 1422 | btrfs_set_lock_blocking(right); |
5f39d397 | 1423 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1424 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1425 | if (wret) { |
1426 | ret = wret; | |
1427 | goto enospc; | |
1428 | } | |
1429 | } | |
1430 | ||
1431 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1432 | if (left) { |
1433 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1434 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1435 | if (wret < 0) |
1436 | ret = wret; | |
559af821 | 1437 | btrfs_header_nritems(mid); |
bb803951 | 1438 | } |
79f95c82 CM |
1439 | |
1440 | /* | |
1441 | * then try to empty the right most buffer into the middle | |
1442 | */ | |
5f39d397 | 1443 | if (right) { |
971a1f66 | 1444 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1445 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1446 | ret = wret; |
5f39d397 | 1447 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1448 | clean_tree_block(trans, root, right); |
925baedd | 1449 | btrfs_tree_unlock(right); |
143bede5 | 1450 | del_ptr(trans, root, path, level + 1, pslot + 1); |
f0486c68 | 1451 | root_sub_used(root, right->len); |
5581a51a | 1452 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1453 | free_extent_buffer_stale(right); |
f0486c68 | 1454 | right = NULL; |
bb803951 | 1455 | } else { |
5f39d397 CM |
1456 | struct btrfs_disk_key right_key; |
1457 | btrfs_node_key(right, &right_key, 0); | |
1458 | btrfs_set_node_key(parent, &right_key, pslot + 1); | |
1459 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1460 | } |
1461 | } | |
5f39d397 | 1462 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1463 | /* |
1464 | * we're not allowed to leave a node with one item in the | |
1465 | * tree during a delete. A deletion from lower in the tree | |
1466 | * could try to delete the only pointer in this node. | |
1467 | * So, pull some keys from the left. | |
1468 | * There has to be a left pointer at this point because | |
1469 | * otherwise we would have pulled some pointers from the | |
1470 | * right | |
1471 | */ | |
305a26af MF |
1472 | if (!left) { |
1473 | ret = -EROFS; | |
1474 | btrfs_std_error(root->fs_info, ret); | |
1475 | goto enospc; | |
1476 | } | |
5f39d397 | 1477 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1478 | if (wret < 0) { |
79f95c82 | 1479 | ret = wret; |
54aa1f4d CM |
1480 | goto enospc; |
1481 | } | |
bce4eae9 CM |
1482 | if (wret == 1) { |
1483 | wret = push_node_left(trans, root, left, mid, 1); | |
1484 | if (wret < 0) | |
1485 | ret = wret; | |
1486 | } | |
79f95c82 CM |
1487 | BUG_ON(wret == 1); |
1488 | } | |
5f39d397 | 1489 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1490 | clean_tree_block(trans, root, mid); |
925baedd | 1491 | btrfs_tree_unlock(mid); |
143bede5 | 1492 | del_ptr(trans, root, path, level + 1, pslot); |
f0486c68 | 1493 | root_sub_used(root, mid->len); |
5581a51a | 1494 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1495 | free_extent_buffer_stale(mid); |
f0486c68 | 1496 | mid = NULL; |
79f95c82 CM |
1497 | } else { |
1498 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1499 | struct btrfs_disk_key mid_key; |
1500 | btrfs_node_key(mid, &mid_key, 0); | |
1501 | btrfs_set_node_key(parent, &mid_key, pslot); | |
1502 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1503 | } |
bb803951 | 1504 | |
79f95c82 | 1505 | /* update the path */ |
5f39d397 CM |
1506 | if (left) { |
1507 | if (btrfs_header_nritems(left) > orig_slot) { | |
1508 | extent_buffer_get(left); | |
925baedd | 1509 | /* left was locked after cow */ |
5f39d397 | 1510 | path->nodes[level] = left; |
bb803951 CM |
1511 | path->slots[level + 1] -= 1; |
1512 | path->slots[level] = orig_slot; | |
925baedd CM |
1513 | if (mid) { |
1514 | btrfs_tree_unlock(mid); | |
5f39d397 | 1515 | free_extent_buffer(mid); |
925baedd | 1516 | } |
bb803951 | 1517 | } else { |
5f39d397 | 1518 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1519 | path->slots[level] = orig_slot; |
1520 | } | |
1521 | } | |
79f95c82 | 1522 | /* double check we haven't messed things up */ |
e20d96d6 | 1523 | if (orig_ptr != |
5f39d397 | 1524 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1525 | BUG(); |
54aa1f4d | 1526 | enospc: |
925baedd CM |
1527 | if (right) { |
1528 | btrfs_tree_unlock(right); | |
5f39d397 | 1529 | free_extent_buffer(right); |
925baedd CM |
1530 | } |
1531 | if (left) { | |
1532 | if (path->nodes[level] != left) | |
1533 | btrfs_tree_unlock(left); | |
5f39d397 | 1534 | free_extent_buffer(left); |
925baedd | 1535 | } |
bb803951 CM |
1536 | return ret; |
1537 | } | |
1538 | ||
d352ac68 CM |
1539 | /* Node balancing for insertion. Here we only split or push nodes around |
1540 | * when they are completely full. This is also done top down, so we | |
1541 | * have to be pessimistic. | |
1542 | */ | |
d397712b | 1543 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1544 | struct btrfs_root *root, |
1545 | struct btrfs_path *path, int level) | |
e66f709b | 1546 | { |
5f39d397 CM |
1547 | struct extent_buffer *right = NULL; |
1548 | struct extent_buffer *mid; | |
1549 | struct extent_buffer *left = NULL; | |
1550 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1551 | int ret = 0; |
1552 | int wret; | |
1553 | int pslot; | |
1554 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1555 | |
1556 | if (level == 0) | |
1557 | return 1; | |
1558 | ||
5f39d397 | 1559 | mid = path->nodes[level]; |
7bb86316 | 1560 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1561 | |
a05a9bb1 | 1562 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1563 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1564 | pslot = path->slots[level + 1]; |
1565 | } | |
e66f709b | 1566 | |
5f39d397 | 1567 | if (!parent) |
e66f709b | 1568 | return 1; |
e66f709b | 1569 | |
5f39d397 | 1570 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1571 | |
1572 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1573 | if (left) { |
e66f709b | 1574 | u32 left_nr; |
925baedd CM |
1575 | |
1576 | btrfs_tree_lock(left); | |
b4ce94de CM |
1577 | btrfs_set_lock_blocking(left); |
1578 | ||
5f39d397 | 1579 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1580 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1581 | wret = 1; | |
1582 | } else { | |
5f39d397 | 1583 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1584 | pslot - 1, &left); |
54aa1f4d CM |
1585 | if (ret) |
1586 | wret = 1; | |
1587 | else { | |
54aa1f4d | 1588 | wret = push_node_left(trans, root, |
971a1f66 | 1589 | left, mid, 0); |
54aa1f4d | 1590 | } |
33ade1f8 | 1591 | } |
e66f709b CM |
1592 | if (wret < 0) |
1593 | ret = wret; | |
1594 | if (wret == 0) { | |
5f39d397 | 1595 | struct btrfs_disk_key disk_key; |
e66f709b | 1596 | orig_slot += left_nr; |
5f39d397 CM |
1597 | btrfs_node_key(mid, &disk_key, 0); |
1598 | btrfs_set_node_key(parent, &disk_key, pslot); | |
1599 | btrfs_mark_buffer_dirty(parent); | |
1600 | if (btrfs_header_nritems(left) > orig_slot) { | |
1601 | path->nodes[level] = left; | |
e66f709b CM |
1602 | path->slots[level + 1] -= 1; |
1603 | path->slots[level] = orig_slot; | |
925baedd | 1604 | btrfs_tree_unlock(mid); |
5f39d397 | 1605 | free_extent_buffer(mid); |
e66f709b CM |
1606 | } else { |
1607 | orig_slot -= | |
5f39d397 | 1608 | btrfs_header_nritems(left); |
e66f709b | 1609 | path->slots[level] = orig_slot; |
925baedd | 1610 | btrfs_tree_unlock(left); |
5f39d397 | 1611 | free_extent_buffer(left); |
e66f709b | 1612 | } |
e66f709b CM |
1613 | return 0; |
1614 | } | |
925baedd | 1615 | btrfs_tree_unlock(left); |
5f39d397 | 1616 | free_extent_buffer(left); |
e66f709b | 1617 | } |
925baedd | 1618 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1619 | |
1620 | /* | |
1621 | * then try to empty the right most buffer into the middle | |
1622 | */ | |
5f39d397 | 1623 | if (right) { |
33ade1f8 | 1624 | u32 right_nr; |
b4ce94de | 1625 | |
925baedd | 1626 | btrfs_tree_lock(right); |
b4ce94de CM |
1627 | btrfs_set_lock_blocking(right); |
1628 | ||
5f39d397 | 1629 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
1630 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1631 | wret = 1; | |
1632 | } else { | |
5f39d397 CM |
1633 | ret = btrfs_cow_block(trans, root, right, |
1634 | parent, pslot + 1, | |
9fa8cfe7 | 1635 | &right); |
54aa1f4d CM |
1636 | if (ret) |
1637 | wret = 1; | |
1638 | else { | |
54aa1f4d | 1639 | wret = balance_node_right(trans, root, |
5f39d397 | 1640 | right, mid); |
54aa1f4d | 1641 | } |
33ade1f8 | 1642 | } |
e66f709b CM |
1643 | if (wret < 0) |
1644 | ret = wret; | |
1645 | if (wret == 0) { | |
5f39d397 CM |
1646 | struct btrfs_disk_key disk_key; |
1647 | ||
1648 | btrfs_node_key(right, &disk_key, 0); | |
1649 | btrfs_set_node_key(parent, &disk_key, pslot + 1); | |
1650 | btrfs_mark_buffer_dirty(parent); | |
1651 | ||
1652 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1653 | path->nodes[level] = right; | |
e66f709b CM |
1654 | path->slots[level + 1] += 1; |
1655 | path->slots[level] = orig_slot - | |
5f39d397 | 1656 | btrfs_header_nritems(mid); |
925baedd | 1657 | btrfs_tree_unlock(mid); |
5f39d397 | 1658 | free_extent_buffer(mid); |
e66f709b | 1659 | } else { |
925baedd | 1660 | btrfs_tree_unlock(right); |
5f39d397 | 1661 | free_extent_buffer(right); |
e66f709b | 1662 | } |
e66f709b CM |
1663 | return 0; |
1664 | } | |
925baedd | 1665 | btrfs_tree_unlock(right); |
5f39d397 | 1666 | free_extent_buffer(right); |
e66f709b | 1667 | } |
e66f709b CM |
1668 | return 1; |
1669 | } | |
1670 | ||
3c69faec | 1671 | /* |
d352ac68 CM |
1672 | * readahead one full node of leaves, finding things that are close |
1673 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1674 | */ |
c8c42864 CM |
1675 | static void reada_for_search(struct btrfs_root *root, |
1676 | struct btrfs_path *path, | |
1677 | int level, int slot, u64 objectid) | |
3c69faec | 1678 | { |
5f39d397 | 1679 | struct extent_buffer *node; |
01f46658 | 1680 | struct btrfs_disk_key disk_key; |
3c69faec | 1681 | u32 nritems; |
3c69faec | 1682 | u64 search; |
a7175319 | 1683 | u64 target; |
6b80053d | 1684 | u64 nread = 0; |
cb25c2ea | 1685 | u64 gen; |
3c69faec | 1686 | int direction = path->reada; |
5f39d397 | 1687 | struct extent_buffer *eb; |
6b80053d CM |
1688 | u32 nr; |
1689 | u32 blocksize; | |
1690 | u32 nscan = 0; | |
db94535d | 1691 | |
a6b6e75e | 1692 | if (level != 1) |
6702ed49 CM |
1693 | return; |
1694 | ||
1695 | if (!path->nodes[level]) | |
3c69faec CM |
1696 | return; |
1697 | ||
5f39d397 | 1698 | node = path->nodes[level]; |
925baedd | 1699 | |
3c69faec | 1700 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
1701 | blocksize = btrfs_level_size(root, level - 1); |
1702 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
1703 | if (eb) { |
1704 | free_extent_buffer(eb); | |
3c69faec CM |
1705 | return; |
1706 | } | |
1707 | ||
a7175319 | 1708 | target = search; |
6b80053d | 1709 | |
5f39d397 | 1710 | nritems = btrfs_header_nritems(node); |
6b80053d | 1711 | nr = slot; |
25b8b936 | 1712 | |
d397712b | 1713 | while (1) { |
6b80053d CM |
1714 | if (direction < 0) { |
1715 | if (nr == 0) | |
1716 | break; | |
1717 | nr--; | |
1718 | } else if (direction > 0) { | |
1719 | nr++; | |
1720 | if (nr >= nritems) | |
1721 | break; | |
3c69faec | 1722 | } |
01f46658 CM |
1723 | if (path->reada < 0 && objectid) { |
1724 | btrfs_node_key(node, &disk_key, nr); | |
1725 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1726 | break; | |
1727 | } | |
6b80053d | 1728 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
1729 | if ((search <= target && target - search <= 65536) || |
1730 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 1731 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 1732 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
1733 | nread += blocksize; |
1734 | } | |
1735 | nscan++; | |
a7175319 | 1736 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 1737 | break; |
3c69faec CM |
1738 | } |
1739 | } | |
925baedd | 1740 | |
b4ce94de CM |
1741 | /* |
1742 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
1743 | * cache | |
1744 | */ | |
1745 | static noinline int reada_for_balance(struct btrfs_root *root, | |
1746 | struct btrfs_path *path, int level) | |
1747 | { | |
1748 | int slot; | |
1749 | int nritems; | |
1750 | struct extent_buffer *parent; | |
1751 | struct extent_buffer *eb; | |
1752 | u64 gen; | |
1753 | u64 block1 = 0; | |
1754 | u64 block2 = 0; | |
1755 | int ret = 0; | |
1756 | int blocksize; | |
1757 | ||
8c594ea8 | 1758 | parent = path->nodes[level + 1]; |
b4ce94de CM |
1759 | if (!parent) |
1760 | return 0; | |
1761 | ||
1762 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1763 | slot = path->slots[level + 1]; |
b4ce94de CM |
1764 | blocksize = btrfs_level_size(root, level); |
1765 | ||
1766 | if (slot > 0) { | |
1767 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
1768 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
1769 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
1770 | /* |
1771 | * if we get -eagain from btrfs_buffer_uptodate, we | |
1772 | * don't want to return eagain here. That will loop | |
1773 | * forever | |
1774 | */ | |
1775 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
1776 | block1 = 0; |
1777 | free_extent_buffer(eb); | |
1778 | } | |
8c594ea8 | 1779 | if (slot + 1 < nritems) { |
b4ce94de CM |
1780 | block2 = btrfs_node_blockptr(parent, slot + 1); |
1781 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
1782 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 1783 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
1784 | block2 = 0; |
1785 | free_extent_buffer(eb); | |
1786 | } | |
1787 | if (block1 || block2) { | |
1788 | ret = -EAGAIN; | |
8c594ea8 CM |
1789 | |
1790 | /* release the whole path */ | |
b3b4aa74 | 1791 | btrfs_release_path(path); |
8c594ea8 CM |
1792 | |
1793 | /* read the blocks */ | |
b4ce94de CM |
1794 | if (block1) |
1795 | readahead_tree_block(root, block1, blocksize, 0); | |
1796 | if (block2) | |
1797 | readahead_tree_block(root, block2, blocksize, 0); | |
1798 | ||
1799 | if (block1) { | |
1800 | eb = read_tree_block(root, block1, blocksize, 0); | |
1801 | free_extent_buffer(eb); | |
1802 | } | |
8c594ea8 | 1803 | if (block2) { |
b4ce94de CM |
1804 | eb = read_tree_block(root, block2, blocksize, 0); |
1805 | free_extent_buffer(eb); | |
1806 | } | |
1807 | } | |
1808 | return ret; | |
1809 | } | |
1810 | ||
1811 | ||
d352ac68 | 1812 | /* |
d397712b CM |
1813 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1814 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1815 | * operations on the tree might require changing key pointers higher up in the | |
1816 | * tree. | |
d352ac68 | 1817 | * |
d397712b CM |
1818 | * callers might also have set path->keep_locks, which tells this code to keep |
1819 | * the lock if the path points to the last slot in the block. This is part of | |
1820 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1821 | * |
d397712b CM |
1822 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1823 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1824 | */ |
e02119d5 | 1825 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
1826 | int lowest_unlock, int min_write_lock_level, |
1827 | int *write_lock_level) | |
925baedd CM |
1828 | { |
1829 | int i; | |
1830 | int skip_level = level; | |
051e1b9f | 1831 | int no_skips = 0; |
925baedd CM |
1832 | struct extent_buffer *t; |
1833 | ||
1834 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1835 | if (!path->nodes[i]) | |
1836 | break; | |
1837 | if (!path->locks[i]) | |
1838 | break; | |
051e1b9f | 1839 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1840 | skip_level = i + 1; |
1841 | continue; | |
1842 | } | |
051e1b9f | 1843 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1844 | u32 nritems; |
1845 | t = path->nodes[i]; | |
1846 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1847 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1848 | skip_level = i + 1; |
1849 | continue; | |
1850 | } | |
1851 | } | |
051e1b9f CM |
1852 | if (skip_level < i && i >= lowest_unlock) |
1853 | no_skips = 1; | |
1854 | ||
925baedd CM |
1855 | t = path->nodes[i]; |
1856 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 1857 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 1858 | path->locks[i] = 0; |
f7c79f30 CM |
1859 | if (write_lock_level && |
1860 | i > min_write_lock_level && | |
1861 | i <= *write_lock_level) { | |
1862 | *write_lock_level = i - 1; | |
1863 | } | |
925baedd CM |
1864 | } |
1865 | } | |
1866 | } | |
1867 | ||
b4ce94de CM |
1868 | /* |
1869 | * This releases any locks held in the path starting at level and | |
1870 | * going all the way up to the root. | |
1871 | * | |
1872 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
1873 | * corner cases, such as COW of the block at slot zero in the node. This | |
1874 | * ignores those rules, and it should only be called when there are no | |
1875 | * more updates to be done higher up in the tree. | |
1876 | */ | |
1877 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
1878 | { | |
1879 | int i; | |
1880 | ||
5d4f98a2 | 1881 | if (path->keep_locks) |
b4ce94de CM |
1882 | return; |
1883 | ||
1884 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1885 | if (!path->nodes[i]) | |
12f4dacc | 1886 | continue; |
b4ce94de | 1887 | if (!path->locks[i]) |
12f4dacc | 1888 | continue; |
bd681513 | 1889 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
1890 | path->locks[i] = 0; |
1891 | } | |
1892 | } | |
1893 | ||
c8c42864 CM |
1894 | /* |
1895 | * helper function for btrfs_search_slot. The goal is to find a block | |
1896 | * in cache without setting the path to blocking. If we find the block | |
1897 | * we return zero and the path is unchanged. | |
1898 | * | |
1899 | * If we can't find the block, we set the path blocking and do some | |
1900 | * reada. -EAGAIN is returned and the search must be repeated. | |
1901 | */ | |
1902 | static int | |
1903 | read_block_for_search(struct btrfs_trans_handle *trans, | |
1904 | struct btrfs_root *root, struct btrfs_path *p, | |
1905 | struct extent_buffer **eb_ret, int level, int slot, | |
1906 | struct btrfs_key *key) | |
1907 | { | |
1908 | u64 blocknr; | |
1909 | u64 gen; | |
1910 | u32 blocksize; | |
1911 | struct extent_buffer *b = *eb_ret; | |
1912 | struct extent_buffer *tmp; | |
76a05b35 | 1913 | int ret; |
c8c42864 CM |
1914 | |
1915 | blocknr = btrfs_node_blockptr(b, slot); | |
1916 | gen = btrfs_node_ptr_generation(b, slot); | |
1917 | blocksize = btrfs_level_size(root, level - 1); | |
1918 | ||
1919 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 1920 | if (tmp) { |
b9fab919 CM |
1921 | /* first we do an atomic uptodate check */ |
1922 | if (btrfs_buffer_uptodate(tmp, 0, 1) > 0) { | |
1923 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { | |
cb44921a CM |
1924 | /* |
1925 | * we found an up to date block without | |
1926 | * sleeping, return | |
1927 | * right away | |
1928 | */ | |
1929 | *eb_ret = tmp; | |
1930 | return 0; | |
1931 | } | |
1932 | /* the pages were up to date, but we failed | |
1933 | * the generation number check. Do a full | |
1934 | * read for the generation number that is correct. | |
1935 | * We must do this without dropping locks so | |
1936 | * we can trust our generation number | |
1937 | */ | |
1938 | free_extent_buffer(tmp); | |
bd681513 CM |
1939 | btrfs_set_path_blocking(p); |
1940 | ||
b9fab919 | 1941 | /* now we're allowed to do a blocking uptodate check */ |
cb44921a | 1942 | tmp = read_tree_block(root, blocknr, blocksize, gen); |
b9fab919 | 1943 | if (tmp && btrfs_buffer_uptodate(tmp, gen, 0) > 0) { |
cb44921a CM |
1944 | *eb_ret = tmp; |
1945 | return 0; | |
1946 | } | |
1947 | free_extent_buffer(tmp); | |
b3b4aa74 | 1948 | btrfs_release_path(p); |
cb44921a CM |
1949 | return -EIO; |
1950 | } | |
c8c42864 CM |
1951 | } |
1952 | ||
1953 | /* | |
1954 | * reduce lock contention at high levels | |
1955 | * of the btree by dropping locks before | |
76a05b35 CM |
1956 | * we read. Don't release the lock on the current |
1957 | * level because we need to walk this node to figure | |
1958 | * out which blocks to read. | |
c8c42864 | 1959 | */ |
8c594ea8 CM |
1960 | btrfs_unlock_up_safe(p, level + 1); |
1961 | btrfs_set_path_blocking(p); | |
1962 | ||
cb44921a | 1963 | free_extent_buffer(tmp); |
c8c42864 CM |
1964 | if (p->reada) |
1965 | reada_for_search(root, p, level, slot, key->objectid); | |
1966 | ||
b3b4aa74 | 1967 | btrfs_release_path(p); |
76a05b35 CM |
1968 | |
1969 | ret = -EAGAIN; | |
5bdd3536 | 1970 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
1971 | if (tmp) { |
1972 | /* | |
1973 | * If the read above didn't mark this buffer up to date, | |
1974 | * it will never end up being up to date. Set ret to EIO now | |
1975 | * and give up so that our caller doesn't loop forever | |
1976 | * on our EAGAINs. | |
1977 | */ | |
b9fab919 | 1978 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 1979 | ret = -EIO; |
c8c42864 | 1980 | free_extent_buffer(tmp); |
76a05b35 CM |
1981 | } |
1982 | return ret; | |
c8c42864 CM |
1983 | } |
1984 | ||
1985 | /* | |
1986 | * helper function for btrfs_search_slot. This does all of the checks | |
1987 | * for node-level blocks and does any balancing required based on | |
1988 | * the ins_len. | |
1989 | * | |
1990 | * If no extra work was required, zero is returned. If we had to | |
1991 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1992 | * start over | |
1993 | */ | |
1994 | static int | |
1995 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1996 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1997 | struct extent_buffer *b, int level, int ins_len, |
1998 | int *write_lock_level) | |
c8c42864 CM |
1999 | { |
2000 | int ret; | |
2001 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2002 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2003 | int sret; | |
2004 | ||
bd681513 CM |
2005 | if (*write_lock_level < level + 1) { |
2006 | *write_lock_level = level + 1; | |
2007 | btrfs_release_path(p); | |
2008 | goto again; | |
2009 | } | |
2010 | ||
c8c42864 CM |
2011 | sret = reada_for_balance(root, p, level); |
2012 | if (sret) | |
2013 | goto again; | |
2014 | ||
2015 | btrfs_set_path_blocking(p); | |
2016 | sret = split_node(trans, root, p, level); | |
bd681513 | 2017 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2018 | |
2019 | BUG_ON(sret > 0); | |
2020 | if (sret) { | |
2021 | ret = sret; | |
2022 | goto done; | |
2023 | } | |
2024 | b = p->nodes[level]; | |
2025 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2026 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2027 | int sret; |
2028 | ||
bd681513 CM |
2029 | if (*write_lock_level < level + 1) { |
2030 | *write_lock_level = level + 1; | |
2031 | btrfs_release_path(p); | |
2032 | goto again; | |
2033 | } | |
2034 | ||
c8c42864 CM |
2035 | sret = reada_for_balance(root, p, level); |
2036 | if (sret) | |
2037 | goto again; | |
2038 | ||
2039 | btrfs_set_path_blocking(p); | |
2040 | sret = balance_level(trans, root, p, level); | |
bd681513 | 2041 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2042 | |
2043 | if (sret) { | |
2044 | ret = sret; | |
2045 | goto done; | |
2046 | } | |
2047 | b = p->nodes[level]; | |
2048 | if (!b) { | |
b3b4aa74 | 2049 | btrfs_release_path(p); |
c8c42864 CM |
2050 | goto again; |
2051 | } | |
2052 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2053 | } | |
2054 | return 0; | |
2055 | ||
2056 | again: | |
2057 | ret = -EAGAIN; | |
2058 | done: | |
2059 | return ret; | |
2060 | } | |
2061 | ||
74123bd7 CM |
2062 | /* |
2063 | * look for key in the tree. path is filled in with nodes along the way | |
2064 | * if key is found, we return zero and you can find the item in the leaf | |
2065 | * level of the path (level 0) | |
2066 | * | |
2067 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2068 | * be inserted, and 1 is returned. If there are other errors during the |
2069 | * search a negative error number is returned. | |
97571fd0 CM |
2070 | * |
2071 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2072 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2073 | * possible) | |
74123bd7 | 2074 | */ |
e089f05c CM |
2075 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2076 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2077 | ins_len, int cow) | |
be0e5c09 | 2078 | { |
5f39d397 | 2079 | struct extent_buffer *b; |
be0e5c09 CM |
2080 | int slot; |
2081 | int ret; | |
33c66f43 | 2082 | int err; |
be0e5c09 | 2083 | int level; |
925baedd | 2084 | int lowest_unlock = 1; |
bd681513 CM |
2085 | int root_lock; |
2086 | /* everything at write_lock_level or lower must be write locked */ | |
2087 | int write_lock_level = 0; | |
9f3a7427 | 2088 | u8 lowest_level = 0; |
f7c79f30 | 2089 | int min_write_lock_level; |
9f3a7427 | 2090 | |
6702ed49 | 2091 | lowest_level = p->lowest_level; |
323ac95b | 2092 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2093 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2094 | |
bd681513 | 2095 | if (ins_len < 0) { |
925baedd | 2096 | lowest_unlock = 2; |
65b51a00 | 2097 | |
bd681513 CM |
2098 | /* when we are removing items, we might have to go up to level |
2099 | * two as we update tree pointers Make sure we keep write | |
2100 | * for those levels as well | |
2101 | */ | |
2102 | write_lock_level = 2; | |
2103 | } else if (ins_len > 0) { | |
2104 | /* | |
2105 | * for inserting items, make sure we have a write lock on | |
2106 | * level 1 so we can update keys | |
2107 | */ | |
2108 | write_lock_level = 1; | |
2109 | } | |
2110 | ||
2111 | if (!cow) | |
2112 | write_lock_level = -1; | |
2113 | ||
2114 | if (cow && (p->keep_locks || p->lowest_level)) | |
2115 | write_lock_level = BTRFS_MAX_LEVEL; | |
2116 | ||
f7c79f30 CM |
2117 | min_write_lock_level = write_lock_level; |
2118 | ||
bb803951 | 2119 | again: |
bd681513 CM |
2120 | /* |
2121 | * we try very hard to do read locks on the root | |
2122 | */ | |
2123 | root_lock = BTRFS_READ_LOCK; | |
2124 | level = 0; | |
5d4f98a2 | 2125 | if (p->search_commit_root) { |
bd681513 CM |
2126 | /* |
2127 | * the commit roots are read only | |
2128 | * so we always do read locks | |
2129 | */ | |
5d4f98a2 YZ |
2130 | b = root->commit_root; |
2131 | extent_buffer_get(b); | |
bd681513 | 2132 | level = btrfs_header_level(b); |
5d4f98a2 | 2133 | if (!p->skip_locking) |
bd681513 | 2134 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2135 | } else { |
bd681513 | 2136 | if (p->skip_locking) { |
5d4f98a2 | 2137 | b = btrfs_root_node(root); |
bd681513 CM |
2138 | level = btrfs_header_level(b); |
2139 | } else { | |
2140 | /* we don't know the level of the root node | |
2141 | * until we actually have it read locked | |
2142 | */ | |
2143 | b = btrfs_read_lock_root_node(root); | |
2144 | level = btrfs_header_level(b); | |
2145 | if (level <= write_lock_level) { | |
2146 | /* whoops, must trade for write lock */ | |
2147 | btrfs_tree_read_unlock(b); | |
2148 | free_extent_buffer(b); | |
2149 | b = btrfs_lock_root_node(root); | |
2150 | root_lock = BTRFS_WRITE_LOCK; | |
2151 | ||
2152 | /* the level might have changed, check again */ | |
2153 | level = btrfs_header_level(b); | |
2154 | } | |
2155 | } | |
5d4f98a2 | 2156 | } |
bd681513 CM |
2157 | p->nodes[level] = b; |
2158 | if (!p->skip_locking) | |
2159 | p->locks[level] = root_lock; | |
925baedd | 2160 | |
eb60ceac | 2161 | while (b) { |
5f39d397 | 2162 | level = btrfs_header_level(b); |
65b51a00 CM |
2163 | |
2164 | /* | |
2165 | * setup the path here so we can release it under lock | |
2166 | * contention with the cow code | |
2167 | */ | |
02217ed2 | 2168 | if (cow) { |
c8c42864 CM |
2169 | /* |
2170 | * if we don't really need to cow this block | |
2171 | * then we don't want to set the path blocking, | |
2172 | * so we test it here | |
2173 | */ | |
5d4f98a2 | 2174 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2175 | goto cow_done; |
5d4f98a2 | 2176 | |
b4ce94de CM |
2177 | btrfs_set_path_blocking(p); |
2178 | ||
bd681513 CM |
2179 | /* |
2180 | * must have write locks on this node and the | |
2181 | * parent | |
2182 | */ | |
2183 | if (level + 1 > write_lock_level) { | |
2184 | write_lock_level = level + 1; | |
2185 | btrfs_release_path(p); | |
2186 | goto again; | |
2187 | } | |
2188 | ||
33c66f43 YZ |
2189 | err = btrfs_cow_block(trans, root, b, |
2190 | p->nodes[level + 1], | |
2191 | p->slots[level + 1], &b); | |
2192 | if (err) { | |
33c66f43 | 2193 | ret = err; |
65b51a00 | 2194 | goto done; |
54aa1f4d | 2195 | } |
02217ed2 | 2196 | } |
65b51a00 | 2197 | cow_done: |
02217ed2 | 2198 | BUG_ON(!cow && ins_len); |
65b51a00 | 2199 | |
eb60ceac | 2200 | p->nodes[level] = b; |
bd681513 | 2201 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2202 | |
2203 | /* | |
2204 | * we have a lock on b and as long as we aren't changing | |
2205 | * the tree, there is no way to for the items in b to change. | |
2206 | * It is safe to drop the lock on our parent before we | |
2207 | * go through the expensive btree search on b. | |
2208 | * | |
2209 | * If cow is true, then we might be changing slot zero, | |
2210 | * which may require changing the parent. So, we can't | |
2211 | * drop the lock until after we know which slot we're | |
2212 | * operating on. | |
2213 | */ | |
2214 | if (!cow) | |
2215 | btrfs_unlock_up_safe(p, level + 1); | |
2216 | ||
5f39d397 | 2217 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 2218 | |
5f39d397 | 2219 | if (level != 0) { |
33c66f43 YZ |
2220 | int dec = 0; |
2221 | if (ret && slot > 0) { | |
2222 | dec = 1; | |
be0e5c09 | 2223 | slot -= 1; |
33c66f43 | 2224 | } |
be0e5c09 | 2225 | p->slots[level] = slot; |
33c66f43 | 2226 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2227 | ins_len, &write_lock_level); |
33c66f43 | 2228 | if (err == -EAGAIN) |
c8c42864 | 2229 | goto again; |
33c66f43 YZ |
2230 | if (err) { |
2231 | ret = err; | |
c8c42864 | 2232 | goto done; |
33c66f43 | 2233 | } |
c8c42864 CM |
2234 | b = p->nodes[level]; |
2235 | slot = p->slots[level]; | |
b4ce94de | 2236 | |
bd681513 CM |
2237 | /* |
2238 | * slot 0 is special, if we change the key | |
2239 | * we have to update the parent pointer | |
2240 | * which means we must have a write lock | |
2241 | * on the parent | |
2242 | */ | |
2243 | if (slot == 0 && cow && | |
2244 | write_lock_level < level + 1) { | |
2245 | write_lock_level = level + 1; | |
2246 | btrfs_release_path(p); | |
2247 | goto again; | |
2248 | } | |
2249 | ||
f7c79f30 CM |
2250 | unlock_up(p, level, lowest_unlock, |
2251 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2252 | |
925baedd | 2253 | if (level == lowest_level) { |
33c66f43 YZ |
2254 | if (dec) |
2255 | p->slots[level]++; | |
5b21f2ed | 2256 | goto done; |
925baedd | 2257 | } |
ca7a79ad | 2258 | |
33c66f43 | 2259 | err = read_block_for_search(trans, root, p, |
c8c42864 | 2260 | &b, level, slot, key); |
33c66f43 | 2261 | if (err == -EAGAIN) |
c8c42864 | 2262 | goto again; |
33c66f43 YZ |
2263 | if (err) { |
2264 | ret = err; | |
76a05b35 | 2265 | goto done; |
33c66f43 | 2266 | } |
76a05b35 | 2267 | |
b4ce94de | 2268 | if (!p->skip_locking) { |
bd681513 CM |
2269 | level = btrfs_header_level(b); |
2270 | if (level <= write_lock_level) { | |
2271 | err = btrfs_try_tree_write_lock(b); | |
2272 | if (!err) { | |
2273 | btrfs_set_path_blocking(p); | |
2274 | btrfs_tree_lock(b); | |
2275 | btrfs_clear_path_blocking(p, b, | |
2276 | BTRFS_WRITE_LOCK); | |
2277 | } | |
2278 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2279 | } else { | |
2280 | err = btrfs_try_tree_read_lock(b); | |
2281 | if (!err) { | |
2282 | btrfs_set_path_blocking(p); | |
2283 | btrfs_tree_read_lock(b); | |
2284 | btrfs_clear_path_blocking(p, b, | |
2285 | BTRFS_READ_LOCK); | |
2286 | } | |
2287 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2288 | } |
bd681513 | 2289 | p->nodes[level] = b; |
b4ce94de | 2290 | } |
be0e5c09 CM |
2291 | } else { |
2292 | p->slots[level] = slot; | |
87b29b20 YZ |
2293 | if (ins_len > 0 && |
2294 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2295 | if (write_lock_level < 1) { |
2296 | write_lock_level = 1; | |
2297 | btrfs_release_path(p); | |
2298 | goto again; | |
2299 | } | |
2300 | ||
b4ce94de | 2301 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2302 | err = split_leaf(trans, root, key, |
2303 | p, ins_len, ret == 0); | |
bd681513 | 2304 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2305 | |
33c66f43 YZ |
2306 | BUG_ON(err > 0); |
2307 | if (err) { | |
2308 | ret = err; | |
65b51a00 CM |
2309 | goto done; |
2310 | } | |
5c680ed6 | 2311 | } |
459931ec | 2312 | if (!p->search_for_split) |
f7c79f30 CM |
2313 | unlock_up(p, level, lowest_unlock, |
2314 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2315 | goto done; |
be0e5c09 CM |
2316 | } |
2317 | } | |
65b51a00 CM |
2318 | ret = 1; |
2319 | done: | |
b4ce94de CM |
2320 | /* |
2321 | * we don't really know what they plan on doing with the path | |
2322 | * from here on, so for now just mark it as blocking | |
2323 | */ | |
b9473439 CM |
2324 | if (!p->leave_spinning) |
2325 | btrfs_set_path_blocking(p); | |
76a05b35 | 2326 | if (ret < 0) |
b3b4aa74 | 2327 | btrfs_release_path(p); |
65b51a00 | 2328 | return ret; |
be0e5c09 CM |
2329 | } |
2330 | ||
74123bd7 CM |
2331 | /* |
2332 | * adjust the pointers going up the tree, starting at level | |
2333 | * making sure the right key of each node is points to 'key'. | |
2334 | * This is used after shifting pointers to the left, so it stops | |
2335 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2336 | * higher levels | |
aa5d6bed | 2337 | * |
74123bd7 | 2338 | */ |
143bede5 JM |
2339 | static void fixup_low_keys(struct btrfs_trans_handle *trans, |
2340 | struct btrfs_root *root, struct btrfs_path *path, | |
2341 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
2342 | { |
2343 | int i; | |
5f39d397 CM |
2344 | struct extent_buffer *t; |
2345 | ||
234b63a0 | 2346 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2347 | int tslot = path->slots[i]; |
eb60ceac | 2348 | if (!path->nodes[i]) |
be0e5c09 | 2349 | break; |
5f39d397 CM |
2350 | t = path->nodes[i]; |
2351 | btrfs_set_node_key(t, key, tslot); | |
d6025579 | 2352 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2353 | if (tslot != 0) |
2354 | break; | |
2355 | } | |
2356 | } | |
2357 | ||
31840ae1 ZY |
2358 | /* |
2359 | * update item key. | |
2360 | * | |
2361 | * This function isn't completely safe. It's the caller's responsibility | |
2362 | * that the new key won't break the order | |
2363 | */ | |
143bede5 JM |
2364 | void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, |
2365 | struct btrfs_root *root, struct btrfs_path *path, | |
2366 | struct btrfs_key *new_key) | |
31840ae1 ZY |
2367 | { |
2368 | struct btrfs_disk_key disk_key; | |
2369 | struct extent_buffer *eb; | |
2370 | int slot; | |
2371 | ||
2372 | eb = path->nodes[0]; | |
2373 | slot = path->slots[0]; | |
2374 | if (slot > 0) { | |
2375 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 2376 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
2377 | } |
2378 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2379 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 2380 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
2381 | } |
2382 | ||
2383 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2384 | btrfs_set_item_key(eb, &disk_key, slot); | |
2385 | btrfs_mark_buffer_dirty(eb); | |
2386 | if (slot == 0) | |
2387 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
31840ae1 ZY |
2388 | } |
2389 | ||
74123bd7 CM |
2390 | /* |
2391 | * try to push data from one node into the next node left in the | |
79f95c82 | 2392 | * tree. |
aa5d6bed CM |
2393 | * |
2394 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2395 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2396 | */ |
98ed5174 CM |
2397 | static int push_node_left(struct btrfs_trans_handle *trans, |
2398 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 2399 | struct extent_buffer *src, int empty) |
be0e5c09 | 2400 | { |
be0e5c09 | 2401 | int push_items = 0; |
bb803951 CM |
2402 | int src_nritems; |
2403 | int dst_nritems; | |
aa5d6bed | 2404 | int ret = 0; |
be0e5c09 | 2405 | |
5f39d397 CM |
2406 | src_nritems = btrfs_header_nritems(src); |
2407 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2408 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
2409 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2410 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2411 | |
bce4eae9 | 2412 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2413 | return 1; |
2414 | ||
d397712b | 2415 | if (push_items <= 0) |
be0e5c09 CM |
2416 | return 1; |
2417 | ||
bce4eae9 | 2418 | if (empty) { |
971a1f66 | 2419 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2420 | if (push_items < src_nritems) { |
2421 | /* leave at least 8 pointers in the node if | |
2422 | * we aren't going to empty it | |
2423 | */ | |
2424 | if (src_nritems - push_items < 8) { | |
2425 | if (push_items <= 8) | |
2426 | return 1; | |
2427 | push_items -= 8; | |
2428 | } | |
2429 | } | |
2430 | } else | |
2431 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2432 | |
5f39d397 CM |
2433 | copy_extent_buffer(dst, src, |
2434 | btrfs_node_key_ptr_offset(dst_nritems), | |
2435 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2436 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2437 | |
bb803951 | 2438 | if (push_items < src_nritems) { |
5f39d397 CM |
2439 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2440 | btrfs_node_key_ptr_offset(push_items), | |
2441 | (src_nritems - push_items) * | |
2442 | sizeof(struct btrfs_key_ptr)); | |
2443 | } | |
2444 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2445 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
2446 | btrfs_mark_buffer_dirty(src); | |
2447 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2448 | |
79f95c82 CM |
2449 | return ret; |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * try to push data from one node into the next node right in the | |
2454 | * tree. | |
2455 | * | |
2456 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2457 | * error, and > 0 if there was no room in the right hand block. | |
2458 | * | |
2459 | * this will only push up to 1/2 the contents of the left node over | |
2460 | */ | |
5f39d397 CM |
2461 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2462 | struct btrfs_root *root, | |
2463 | struct extent_buffer *dst, | |
2464 | struct extent_buffer *src) | |
79f95c82 | 2465 | { |
79f95c82 CM |
2466 | int push_items = 0; |
2467 | int max_push; | |
2468 | int src_nritems; | |
2469 | int dst_nritems; | |
2470 | int ret = 0; | |
79f95c82 | 2471 | |
7bb86316 CM |
2472 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2473 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2474 | ||
5f39d397 CM |
2475 | src_nritems = btrfs_header_nritems(src); |
2476 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2477 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 2478 | if (push_items <= 0) |
79f95c82 | 2479 | return 1; |
bce4eae9 | 2480 | |
d397712b | 2481 | if (src_nritems < 4) |
bce4eae9 | 2482 | return 1; |
79f95c82 CM |
2483 | |
2484 | max_push = src_nritems / 2 + 1; | |
2485 | /* don't try to empty the node */ | |
d397712b | 2486 | if (max_push >= src_nritems) |
79f95c82 | 2487 | return 1; |
252c38f0 | 2488 | |
79f95c82 CM |
2489 | if (max_push < push_items) |
2490 | push_items = max_push; | |
2491 | ||
5f39d397 CM |
2492 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2493 | btrfs_node_key_ptr_offset(0), | |
2494 | (dst_nritems) * | |
2495 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2496 | |
5f39d397 CM |
2497 | copy_extent_buffer(dst, src, |
2498 | btrfs_node_key_ptr_offset(0), | |
2499 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2500 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2501 | |
5f39d397 CM |
2502 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2503 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2504 | |
5f39d397 CM |
2505 | btrfs_mark_buffer_dirty(src); |
2506 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2507 | |
aa5d6bed | 2508 | return ret; |
be0e5c09 CM |
2509 | } |
2510 | ||
97571fd0 CM |
2511 | /* |
2512 | * helper function to insert a new root level in the tree. | |
2513 | * A new node is allocated, and a single item is inserted to | |
2514 | * point to the existing root | |
aa5d6bed CM |
2515 | * |
2516 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2517 | */ |
d397712b | 2518 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2519 | struct btrfs_root *root, |
2520 | struct btrfs_path *path, int level) | |
5c680ed6 | 2521 | { |
7bb86316 | 2522 | u64 lower_gen; |
5f39d397 CM |
2523 | struct extent_buffer *lower; |
2524 | struct extent_buffer *c; | |
925baedd | 2525 | struct extent_buffer *old; |
5f39d397 | 2526 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
2527 | |
2528 | BUG_ON(path->nodes[level]); | |
2529 | BUG_ON(path->nodes[level-1] != root->node); | |
2530 | ||
7bb86316 CM |
2531 | lower = path->nodes[level-1]; |
2532 | if (level == 1) | |
2533 | btrfs_item_key(lower, &lower_key, 0); | |
2534 | else | |
2535 | btrfs_node_key(lower, &lower_key, 0); | |
2536 | ||
31840ae1 | 2537 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 2538 | root->root_key.objectid, &lower_key, |
5581a51a | 2539 | level, root->node->start, 0); |
5f39d397 CM |
2540 | if (IS_ERR(c)) |
2541 | return PTR_ERR(c); | |
925baedd | 2542 | |
f0486c68 YZ |
2543 | root_add_used(root, root->nodesize); |
2544 | ||
5d4f98a2 | 2545 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
2546 | btrfs_set_header_nritems(c, 1); |
2547 | btrfs_set_header_level(c, level); | |
db94535d | 2548 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 2549 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 2550 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 2551 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
2552 | |
2553 | write_extent_buffer(c, root->fs_info->fsid, | |
2554 | (unsigned long)btrfs_header_fsid(c), | |
2555 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2556 | |
2557 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
2558 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
2559 | BTRFS_UUID_SIZE); | |
2560 | ||
5f39d397 | 2561 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2562 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2563 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2564 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2565 | |
2566 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2567 | |
5f39d397 | 2568 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2569 | |
925baedd | 2570 | old = root->node; |
240f62c8 | 2571 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2572 | |
2573 | /* the super has an extra ref to root->node */ | |
2574 | free_extent_buffer(old); | |
2575 | ||
0b86a832 | 2576 | add_root_to_dirty_list(root); |
5f39d397 CM |
2577 | extent_buffer_get(c); |
2578 | path->nodes[level] = c; | |
bd681513 | 2579 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2580 | path->slots[level] = 0; |
2581 | return 0; | |
2582 | } | |
2583 | ||
74123bd7 CM |
2584 | /* |
2585 | * worker function to insert a single pointer in a node. | |
2586 | * the node should have enough room for the pointer already | |
97571fd0 | 2587 | * |
74123bd7 CM |
2588 | * slot and level indicate where you want the key to go, and |
2589 | * blocknr is the block the key points to. | |
2590 | */ | |
143bede5 JM |
2591 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2592 | struct btrfs_root *root, struct btrfs_path *path, | |
2593 | struct btrfs_disk_key *key, u64 bytenr, | |
2594 | int slot, int level) | |
74123bd7 | 2595 | { |
5f39d397 | 2596 | struct extent_buffer *lower; |
74123bd7 | 2597 | int nritems; |
5c680ed6 CM |
2598 | |
2599 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 2600 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
2601 | lower = path->nodes[level]; |
2602 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2603 | BUG_ON(slot > nritems); |
143bede5 | 2604 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 2605 | if (slot != nritems) { |
5f39d397 CM |
2606 | memmove_extent_buffer(lower, |
2607 | btrfs_node_key_ptr_offset(slot + 1), | |
2608 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2609 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2610 | } |
5f39d397 | 2611 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2612 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2613 | WARN_ON(trans->transid == 0); |
2614 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2615 | btrfs_set_header_nritems(lower, nritems + 1); |
2616 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2617 | } |
2618 | ||
97571fd0 CM |
2619 | /* |
2620 | * split the node at the specified level in path in two. | |
2621 | * The path is corrected to point to the appropriate node after the split | |
2622 | * | |
2623 | * Before splitting this tries to make some room in the node by pushing | |
2624 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2625 | * |
2626 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2627 | */ |
e02119d5 CM |
2628 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2629 | struct btrfs_root *root, | |
2630 | struct btrfs_path *path, int level) | |
be0e5c09 | 2631 | { |
5f39d397 CM |
2632 | struct extent_buffer *c; |
2633 | struct extent_buffer *split; | |
2634 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2635 | int mid; |
5c680ed6 | 2636 | int ret; |
7518a238 | 2637 | u32 c_nritems; |
eb60ceac | 2638 | |
5f39d397 | 2639 | c = path->nodes[level]; |
7bb86316 | 2640 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2641 | if (c == root->node) { |
5c680ed6 | 2642 | /* trying to split the root, lets make a new one */ |
e089f05c | 2643 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2644 | if (ret) |
2645 | return ret; | |
b3612421 | 2646 | } else { |
e66f709b | 2647 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2648 | c = path->nodes[level]; |
2649 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 2650 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 2651 | return 0; |
54aa1f4d CM |
2652 | if (ret < 0) |
2653 | return ret; | |
be0e5c09 | 2654 | } |
e66f709b | 2655 | |
5f39d397 | 2656 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2657 | mid = (c_nritems + 1) / 2; |
2658 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2659 | |
5d4f98a2 | 2660 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 2661 | root->root_key.objectid, |
5581a51a | 2662 | &disk_key, level, c->start, 0); |
5f39d397 CM |
2663 | if (IS_ERR(split)) |
2664 | return PTR_ERR(split); | |
2665 | ||
f0486c68 YZ |
2666 | root_add_used(root, root->nodesize); |
2667 | ||
5d4f98a2 | 2668 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 2669 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 2670 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 2671 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 2672 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
2673 | btrfs_set_header_owner(split, root->root_key.objectid); |
2674 | write_extent_buffer(split, root->fs_info->fsid, | |
2675 | (unsigned long)btrfs_header_fsid(split), | |
2676 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2677 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
2678 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
2679 | BTRFS_UUID_SIZE); | |
54aa1f4d | 2680 | |
5f39d397 CM |
2681 | copy_extent_buffer(split, c, |
2682 | btrfs_node_key_ptr_offset(0), | |
2683 | btrfs_node_key_ptr_offset(mid), | |
2684 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2685 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2686 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
2687 | ret = 0; |
2688 | ||
5f39d397 CM |
2689 | btrfs_mark_buffer_dirty(c); |
2690 | btrfs_mark_buffer_dirty(split); | |
2691 | ||
143bede5 JM |
2692 | insert_ptr(trans, root, path, &disk_key, split->start, |
2693 | path->slots[level + 1] + 1, level + 1); | |
aa5d6bed | 2694 | |
5de08d7d | 2695 | if (path->slots[level] >= mid) { |
5c680ed6 | 2696 | path->slots[level] -= mid; |
925baedd | 2697 | btrfs_tree_unlock(c); |
5f39d397 CM |
2698 | free_extent_buffer(c); |
2699 | path->nodes[level] = split; | |
5c680ed6 CM |
2700 | path->slots[level + 1] += 1; |
2701 | } else { | |
925baedd | 2702 | btrfs_tree_unlock(split); |
5f39d397 | 2703 | free_extent_buffer(split); |
be0e5c09 | 2704 | } |
aa5d6bed | 2705 | return ret; |
be0e5c09 CM |
2706 | } |
2707 | ||
74123bd7 CM |
2708 | /* |
2709 | * how many bytes are required to store the items in a leaf. start | |
2710 | * and nr indicate which items in the leaf to check. This totals up the | |
2711 | * space used both by the item structs and the item data | |
2712 | */ | |
5f39d397 | 2713 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2714 | { |
2715 | int data_len; | |
5f39d397 | 2716 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2717 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2718 | |
2719 | if (!nr) | |
2720 | return 0; | |
5f39d397 CM |
2721 | data_len = btrfs_item_end_nr(l, start); |
2722 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2723 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2724 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2725 | return data_len; |
2726 | } | |
2727 | ||
d4dbff95 CM |
2728 | /* |
2729 | * The space between the end of the leaf items and | |
2730 | * the start of the leaf data. IOW, how much room | |
2731 | * the leaf has left for both items and data | |
2732 | */ | |
d397712b | 2733 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 2734 | struct extent_buffer *leaf) |
d4dbff95 | 2735 | { |
5f39d397 CM |
2736 | int nritems = btrfs_header_nritems(leaf); |
2737 | int ret; | |
2738 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
2739 | if (ret < 0) { | |
d397712b CM |
2740 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
2741 | "used %d nritems %d\n", | |
ae2f5411 | 2742 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
2743 | leaf_space_used(leaf, 0, nritems), nritems); |
2744 | } | |
2745 | return ret; | |
d4dbff95 CM |
2746 | } |
2747 | ||
99d8f83c CM |
2748 | /* |
2749 | * min slot controls the lowest index we're willing to push to the | |
2750 | * right. We'll push up to and including min_slot, but no lower | |
2751 | */ | |
44871b1b CM |
2752 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
2753 | struct btrfs_root *root, | |
2754 | struct btrfs_path *path, | |
2755 | int data_size, int empty, | |
2756 | struct extent_buffer *right, | |
99d8f83c CM |
2757 | int free_space, u32 left_nritems, |
2758 | u32 min_slot) | |
00ec4c51 | 2759 | { |
5f39d397 | 2760 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2761 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 2762 | struct btrfs_map_token token; |
5f39d397 | 2763 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2764 | int slot; |
34a38218 | 2765 | u32 i; |
00ec4c51 CM |
2766 | int push_space = 0; |
2767 | int push_items = 0; | |
0783fcfc | 2768 | struct btrfs_item *item; |
34a38218 | 2769 | u32 nr; |
7518a238 | 2770 | u32 right_nritems; |
5f39d397 | 2771 | u32 data_end; |
db94535d | 2772 | u32 this_item_size; |
00ec4c51 | 2773 | |
cfed81a0 CM |
2774 | btrfs_init_map_token(&token); |
2775 | ||
34a38218 CM |
2776 | if (empty) |
2777 | nr = 0; | |
2778 | else | |
99d8f83c | 2779 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2780 | |
31840ae1 | 2781 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2782 | push_space += data_size; |
31840ae1 | 2783 | |
44871b1b | 2784 | slot = path->slots[1]; |
34a38218 CM |
2785 | i = left_nritems - 1; |
2786 | while (i >= nr) { | |
5f39d397 | 2787 | item = btrfs_item_nr(left, i); |
db94535d | 2788 | |
31840ae1 ZY |
2789 | if (!empty && push_items > 0) { |
2790 | if (path->slots[0] > i) | |
2791 | break; | |
2792 | if (path->slots[0] == i) { | |
2793 | int space = btrfs_leaf_free_space(root, left); | |
2794 | if (space + push_space * 2 > free_space) | |
2795 | break; | |
2796 | } | |
2797 | } | |
2798 | ||
00ec4c51 | 2799 | if (path->slots[0] == i) |
87b29b20 | 2800 | push_space += data_size; |
db94535d | 2801 | |
db94535d CM |
2802 | this_item_size = btrfs_item_size(left, item); |
2803 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 2804 | break; |
31840ae1 | 2805 | |
00ec4c51 | 2806 | push_items++; |
db94535d | 2807 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2808 | if (i == 0) |
2809 | break; | |
2810 | i--; | |
db94535d | 2811 | } |
5f39d397 | 2812 | |
925baedd CM |
2813 | if (push_items == 0) |
2814 | goto out_unlock; | |
5f39d397 | 2815 | |
34a38218 | 2816 | if (!empty && push_items == left_nritems) |
a429e513 | 2817 | WARN_ON(1); |
5f39d397 | 2818 | |
00ec4c51 | 2819 | /* push left to right */ |
5f39d397 | 2820 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2821 | |
5f39d397 | 2822 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 2823 | push_space -= leaf_data_end(root, left); |
5f39d397 | 2824 | |
00ec4c51 | 2825 | /* make room in the right data area */ |
5f39d397 CM |
2826 | data_end = leaf_data_end(root, right); |
2827 | memmove_extent_buffer(right, | |
2828 | btrfs_leaf_data(right) + data_end - push_space, | |
2829 | btrfs_leaf_data(right) + data_end, | |
2830 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
2831 | ||
00ec4c51 | 2832 | /* copy from the left data area */ |
5f39d397 | 2833 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
2834 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
2835 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
2836 | push_space); | |
5f39d397 CM |
2837 | |
2838 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2839 | btrfs_item_nr_offset(0), | |
2840 | right_nritems * sizeof(struct btrfs_item)); | |
2841 | ||
00ec4c51 | 2842 | /* copy the items from left to right */ |
5f39d397 CM |
2843 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2844 | btrfs_item_nr_offset(left_nritems - push_items), | |
2845 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2846 | |
2847 | /* update the item pointers */ | |
7518a238 | 2848 | right_nritems += push_items; |
5f39d397 | 2849 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 2850 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 2851 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 2852 | item = btrfs_item_nr(right, i); |
cfed81a0 CM |
2853 | push_space -= btrfs_token_item_size(right, item, &token); |
2854 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
2855 | } |
2856 | ||
7518a238 | 2857 | left_nritems -= push_items; |
5f39d397 | 2858 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2859 | |
34a38218 CM |
2860 | if (left_nritems) |
2861 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
2862 | else |
2863 | clean_tree_block(trans, root, left); | |
2864 | ||
5f39d397 | 2865 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2866 | |
5f39d397 CM |
2867 | btrfs_item_key(right, &disk_key, 0); |
2868 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2869 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2870 | |
00ec4c51 | 2871 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2872 | if (path->slots[0] >= left_nritems) { |
2873 | path->slots[0] -= left_nritems; | |
925baedd CM |
2874 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2875 | clean_tree_block(trans, root, path->nodes[0]); | |
2876 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2877 | free_extent_buffer(path->nodes[0]); |
2878 | path->nodes[0] = right; | |
00ec4c51 CM |
2879 | path->slots[1] += 1; |
2880 | } else { | |
925baedd | 2881 | btrfs_tree_unlock(right); |
5f39d397 | 2882 | free_extent_buffer(right); |
00ec4c51 CM |
2883 | } |
2884 | return 0; | |
925baedd CM |
2885 | |
2886 | out_unlock: | |
2887 | btrfs_tree_unlock(right); | |
2888 | free_extent_buffer(right); | |
2889 | return 1; | |
00ec4c51 | 2890 | } |
925baedd | 2891 | |
44871b1b CM |
2892 | /* |
2893 | * push some data in the path leaf to the right, trying to free up at | |
2894 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2895 | * | |
2896 | * returns 1 if the push failed because the other node didn't have enough | |
2897 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
2898 | * |
2899 | * this will push starting from min_slot to the end of the leaf. It won't | |
2900 | * push any slot lower than min_slot | |
44871b1b CM |
2901 | */ |
2902 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2903 | *root, struct btrfs_path *path, |
2904 | int min_data_size, int data_size, | |
2905 | int empty, u32 min_slot) | |
44871b1b CM |
2906 | { |
2907 | struct extent_buffer *left = path->nodes[0]; | |
2908 | struct extent_buffer *right; | |
2909 | struct extent_buffer *upper; | |
2910 | int slot; | |
2911 | int free_space; | |
2912 | u32 left_nritems; | |
2913 | int ret; | |
2914 | ||
2915 | if (!path->nodes[1]) | |
2916 | return 1; | |
2917 | ||
2918 | slot = path->slots[1]; | |
2919 | upper = path->nodes[1]; | |
2920 | if (slot >= btrfs_header_nritems(upper) - 1) | |
2921 | return 1; | |
2922 | ||
2923 | btrfs_assert_tree_locked(path->nodes[1]); | |
2924 | ||
2925 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
2926 | if (right == NULL) |
2927 | return 1; | |
2928 | ||
44871b1b CM |
2929 | btrfs_tree_lock(right); |
2930 | btrfs_set_lock_blocking(right); | |
2931 | ||
2932 | free_space = btrfs_leaf_free_space(root, right); | |
2933 | if (free_space < data_size) | |
2934 | goto out_unlock; | |
2935 | ||
2936 | /* cow and double check */ | |
2937 | ret = btrfs_cow_block(trans, root, right, upper, | |
2938 | slot + 1, &right); | |
2939 | if (ret) | |
2940 | goto out_unlock; | |
2941 | ||
2942 | free_space = btrfs_leaf_free_space(root, right); | |
2943 | if (free_space < data_size) | |
2944 | goto out_unlock; | |
2945 | ||
2946 | left_nritems = btrfs_header_nritems(left); | |
2947 | if (left_nritems == 0) | |
2948 | goto out_unlock; | |
2949 | ||
99d8f83c CM |
2950 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
2951 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
2952 | out_unlock: |
2953 | btrfs_tree_unlock(right); | |
2954 | free_extent_buffer(right); | |
2955 | return 1; | |
2956 | } | |
2957 | ||
74123bd7 CM |
2958 | /* |
2959 | * push some data in the path leaf to the left, trying to free up at | |
2960 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2961 | * |
2962 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2963 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
2964 | * items | |
74123bd7 | 2965 | */ |
44871b1b CM |
2966 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
2967 | struct btrfs_root *root, | |
2968 | struct btrfs_path *path, int data_size, | |
2969 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
2970 | int free_space, u32 right_nritems, |
2971 | u32 max_slot) | |
be0e5c09 | 2972 | { |
5f39d397 CM |
2973 | struct btrfs_disk_key disk_key; |
2974 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 2975 | int i; |
be0e5c09 CM |
2976 | int push_space = 0; |
2977 | int push_items = 0; | |
0783fcfc | 2978 | struct btrfs_item *item; |
7518a238 | 2979 | u32 old_left_nritems; |
34a38218 | 2980 | u32 nr; |
aa5d6bed | 2981 | int ret = 0; |
db94535d CM |
2982 | u32 this_item_size; |
2983 | u32 old_left_item_size; | |
cfed81a0 CM |
2984 | struct btrfs_map_token token; |
2985 | ||
2986 | btrfs_init_map_token(&token); | |
be0e5c09 | 2987 | |
34a38218 | 2988 | if (empty) |
99d8f83c | 2989 | nr = min(right_nritems, max_slot); |
34a38218 | 2990 | else |
99d8f83c | 2991 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
2992 | |
2993 | for (i = 0; i < nr; i++) { | |
5f39d397 | 2994 | item = btrfs_item_nr(right, i); |
db94535d | 2995 | |
31840ae1 ZY |
2996 | if (!empty && push_items > 0) { |
2997 | if (path->slots[0] < i) | |
2998 | break; | |
2999 | if (path->slots[0] == i) { | |
3000 | int space = btrfs_leaf_free_space(root, right); | |
3001 | if (space + push_space * 2 > free_space) | |
3002 | break; | |
3003 | } | |
3004 | } | |
3005 | ||
be0e5c09 | 3006 | if (path->slots[0] == i) |
87b29b20 | 3007 | push_space += data_size; |
db94535d CM |
3008 | |
3009 | this_item_size = btrfs_item_size(right, item); | |
3010 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3011 | break; |
db94535d | 3012 | |
be0e5c09 | 3013 | push_items++; |
db94535d CM |
3014 | push_space += this_item_size + sizeof(*item); |
3015 | } | |
3016 | ||
be0e5c09 | 3017 | if (push_items == 0) { |
925baedd CM |
3018 | ret = 1; |
3019 | goto out; | |
be0e5c09 | 3020 | } |
34a38218 | 3021 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 3022 | WARN_ON(1); |
5f39d397 | 3023 | |
be0e5c09 | 3024 | /* push data from right to left */ |
5f39d397 CM |
3025 | copy_extent_buffer(left, right, |
3026 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3027 | btrfs_item_nr_offset(0), | |
3028 | push_items * sizeof(struct btrfs_item)); | |
3029 | ||
123abc88 | 3030 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3031 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3032 | |
3033 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3034 | leaf_data_end(root, left) - push_space, |
3035 | btrfs_leaf_data(right) + | |
5f39d397 | 3036 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3037 | push_space); |
5f39d397 | 3038 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3039 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3040 | |
db94535d | 3041 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3042 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3043 | u32 ioff; |
db94535d | 3044 | |
5f39d397 | 3045 | item = btrfs_item_nr(left, i); |
db94535d | 3046 | |
cfed81a0 CM |
3047 | ioff = btrfs_token_item_offset(left, item, &token); |
3048 | btrfs_set_token_item_offset(left, item, | |
3049 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3050 | &token); | |
be0e5c09 | 3051 | } |
5f39d397 | 3052 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3053 | |
3054 | /* fixup right node */ | |
34a38218 | 3055 | if (push_items > right_nritems) { |
d397712b CM |
3056 | printk(KERN_CRIT "push items %d nr %u\n", push_items, |
3057 | right_nritems); | |
34a38218 CM |
3058 | WARN_ON(1); |
3059 | } | |
3060 | ||
3061 | if (push_items < right_nritems) { | |
3062 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3063 | leaf_data_end(root, right); | |
3064 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3065 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3066 | btrfs_leaf_data(right) + | |
3067 | leaf_data_end(root, right), push_space); | |
3068 | ||
3069 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3070 | btrfs_item_nr_offset(push_items), |
3071 | (btrfs_header_nritems(right) - push_items) * | |
3072 | sizeof(struct btrfs_item)); | |
34a38218 | 3073 | } |
eef1c494 Y |
3074 | right_nritems -= push_items; |
3075 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3076 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
3077 | for (i = 0; i < right_nritems; i++) { |
3078 | item = btrfs_item_nr(right, i); | |
db94535d | 3079 | |
cfed81a0 CM |
3080 | push_space = push_space - btrfs_token_item_size(right, |
3081 | item, &token); | |
3082 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3083 | } |
eb60ceac | 3084 | |
5f39d397 | 3085 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3086 | if (right_nritems) |
3087 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3088 | else |
3089 | clean_tree_block(trans, root, right); | |
098f59c2 | 3090 | |
5f39d397 | 3091 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3092 | fixup_low_keys(trans, root, path, &disk_key, 1); |
be0e5c09 CM |
3093 | |
3094 | /* then fixup the leaf pointer in the path */ | |
3095 | if (path->slots[0] < push_items) { | |
3096 | path->slots[0] += old_left_nritems; | |
925baedd | 3097 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3098 | free_extent_buffer(path->nodes[0]); |
3099 | path->nodes[0] = left; | |
be0e5c09 CM |
3100 | path->slots[1] -= 1; |
3101 | } else { | |
925baedd | 3102 | btrfs_tree_unlock(left); |
5f39d397 | 3103 | free_extent_buffer(left); |
be0e5c09 CM |
3104 | path->slots[0] -= push_items; |
3105 | } | |
eb60ceac | 3106 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3107 | return ret; |
925baedd CM |
3108 | out: |
3109 | btrfs_tree_unlock(left); | |
3110 | free_extent_buffer(left); | |
3111 | return ret; | |
be0e5c09 CM |
3112 | } |
3113 | ||
44871b1b CM |
3114 | /* |
3115 | * push some data in the path leaf to the left, trying to free up at | |
3116 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3117 | * |
3118 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3119 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3120 | * items | |
44871b1b CM |
3121 | */ |
3122 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3123 | *root, struct btrfs_path *path, int min_data_size, |
3124 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3125 | { |
3126 | struct extent_buffer *right = path->nodes[0]; | |
3127 | struct extent_buffer *left; | |
3128 | int slot; | |
3129 | int free_space; | |
3130 | u32 right_nritems; | |
3131 | int ret = 0; | |
3132 | ||
3133 | slot = path->slots[1]; | |
3134 | if (slot == 0) | |
3135 | return 1; | |
3136 | if (!path->nodes[1]) | |
3137 | return 1; | |
3138 | ||
3139 | right_nritems = btrfs_header_nritems(right); | |
3140 | if (right_nritems == 0) | |
3141 | return 1; | |
3142 | ||
3143 | btrfs_assert_tree_locked(path->nodes[1]); | |
3144 | ||
3145 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
3146 | if (left == NULL) |
3147 | return 1; | |
3148 | ||
44871b1b CM |
3149 | btrfs_tree_lock(left); |
3150 | btrfs_set_lock_blocking(left); | |
3151 | ||
3152 | free_space = btrfs_leaf_free_space(root, left); | |
3153 | if (free_space < data_size) { | |
3154 | ret = 1; | |
3155 | goto out; | |
3156 | } | |
3157 | ||
3158 | /* cow and double check */ | |
3159 | ret = btrfs_cow_block(trans, root, left, | |
3160 | path->nodes[1], slot - 1, &left); | |
3161 | if (ret) { | |
3162 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3163 | if (ret == -ENOSPC) |
3164 | ret = 1; | |
44871b1b CM |
3165 | goto out; |
3166 | } | |
3167 | ||
3168 | free_space = btrfs_leaf_free_space(root, left); | |
3169 | if (free_space < data_size) { | |
3170 | ret = 1; | |
3171 | goto out; | |
3172 | } | |
3173 | ||
99d8f83c CM |
3174 | return __push_leaf_left(trans, root, path, min_data_size, |
3175 | empty, left, free_space, right_nritems, | |
3176 | max_slot); | |
44871b1b CM |
3177 | out: |
3178 | btrfs_tree_unlock(left); | |
3179 | free_extent_buffer(left); | |
3180 | return ret; | |
3181 | } | |
3182 | ||
3183 | /* | |
3184 | * split the path's leaf in two, making sure there is at least data_size | |
3185 | * available for the resulting leaf level of the path. | |
44871b1b | 3186 | */ |
143bede5 JM |
3187 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
3188 | struct btrfs_root *root, | |
3189 | struct btrfs_path *path, | |
3190 | struct extent_buffer *l, | |
3191 | struct extent_buffer *right, | |
3192 | int slot, int mid, int nritems) | |
44871b1b CM |
3193 | { |
3194 | int data_copy_size; | |
3195 | int rt_data_off; | |
3196 | int i; | |
44871b1b | 3197 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3198 | struct btrfs_map_token token; |
3199 | ||
3200 | btrfs_init_map_token(&token); | |
44871b1b CM |
3201 | |
3202 | nritems = nritems - mid; | |
3203 | btrfs_set_header_nritems(right, nritems); | |
3204 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
3205 | ||
3206 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3207 | btrfs_item_nr_offset(mid), | |
3208 | nritems * sizeof(struct btrfs_item)); | |
3209 | ||
3210 | copy_extent_buffer(right, l, | |
3211 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
3212 | data_copy_size, btrfs_leaf_data(l) + | |
3213 | leaf_data_end(root, l), data_copy_size); | |
3214 | ||
3215 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
3216 | btrfs_item_end_nr(l, mid); | |
3217 | ||
3218 | for (i = 0; i < nritems; i++) { | |
3219 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
3220 | u32 ioff; | |
3221 | ||
cfed81a0 CM |
3222 | ioff = btrfs_token_item_offset(right, item, &token); |
3223 | btrfs_set_token_item_offset(right, item, | |
3224 | ioff + rt_data_off, &token); | |
44871b1b CM |
3225 | } |
3226 | ||
44871b1b | 3227 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3228 | btrfs_item_key(right, &disk_key, 0); |
143bede5 JM |
3229 | insert_ptr(trans, root, path, &disk_key, right->start, |
3230 | path->slots[1] + 1, 1); | |
44871b1b CM |
3231 | |
3232 | btrfs_mark_buffer_dirty(right); | |
3233 | btrfs_mark_buffer_dirty(l); | |
3234 | BUG_ON(path->slots[0] != slot); | |
3235 | ||
44871b1b CM |
3236 | if (mid <= slot) { |
3237 | btrfs_tree_unlock(path->nodes[0]); | |
3238 | free_extent_buffer(path->nodes[0]); | |
3239 | path->nodes[0] = right; | |
3240 | path->slots[0] -= mid; | |
3241 | path->slots[1] += 1; | |
3242 | } else { | |
3243 | btrfs_tree_unlock(right); | |
3244 | free_extent_buffer(right); | |
3245 | } | |
3246 | ||
3247 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3248 | } |
3249 | ||
99d8f83c CM |
3250 | /* |
3251 | * double splits happen when we need to insert a big item in the middle | |
3252 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3253 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3254 | * A B C | |
3255 | * | |
3256 | * We avoid this by trying to push the items on either side of our target | |
3257 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3258 | * completely. | |
3259 | */ | |
3260 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3261 | struct btrfs_root *root, | |
3262 | struct btrfs_path *path, | |
3263 | int data_size) | |
3264 | { | |
3265 | int ret; | |
3266 | int progress = 0; | |
3267 | int slot; | |
3268 | u32 nritems; | |
3269 | ||
3270 | slot = path->slots[0]; | |
3271 | ||
3272 | /* | |
3273 | * try to push all the items after our slot into the | |
3274 | * right leaf | |
3275 | */ | |
3276 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
3277 | if (ret < 0) | |
3278 | return ret; | |
3279 | ||
3280 | if (ret == 0) | |
3281 | progress++; | |
3282 | ||
3283 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3284 | /* | |
3285 | * our goal is to get our slot at the start or end of a leaf. If | |
3286 | * we've done so we're done | |
3287 | */ | |
3288 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3289 | return 0; | |
3290 | ||
3291 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3292 | return 0; | |
3293 | ||
3294 | /* try to push all the items before our slot into the next leaf */ | |
3295 | slot = path->slots[0]; | |
3296 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
3297 | if (ret < 0) | |
3298 | return ret; | |
3299 | ||
3300 | if (ret == 0) | |
3301 | progress++; | |
3302 | ||
3303 | if (progress) | |
3304 | return 0; | |
3305 | return 1; | |
3306 | } | |
3307 | ||
74123bd7 CM |
3308 | /* |
3309 | * split the path's leaf in two, making sure there is at least data_size | |
3310 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3311 | * |
3312 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3313 | */ |
e02119d5 CM |
3314 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3315 | struct btrfs_root *root, | |
3316 | struct btrfs_key *ins_key, | |
3317 | struct btrfs_path *path, int data_size, | |
3318 | int extend) | |
be0e5c09 | 3319 | { |
5d4f98a2 | 3320 | struct btrfs_disk_key disk_key; |
5f39d397 | 3321 | struct extent_buffer *l; |
7518a238 | 3322 | u32 nritems; |
eb60ceac CM |
3323 | int mid; |
3324 | int slot; | |
5f39d397 | 3325 | struct extent_buffer *right; |
d4dbff95 | 3326 | int ret = 0; |
aa5d6bed | 3327 | int wret; |
5d4f98a2 | 3328 | int split; |
cc0c5538 | 3329 | int num_doubles = 0; |
99d8f83c | 3330 | int tried_avoid_double = 0; |
aa5d6bed | 3331 | |
a5719521 YZ |
3332 | l = path->nodes[0]; |
3333 | slot = path->slots[0]; | |
3334 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
3335 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
3336 | return -EOVERFLOW; | |
3337 | ||
40689478 | 3338 | /* first try to make some room by pushing left and right */ |
99d8f83c CM |
3339 | if (data_size) { |
3340 | wret = push_leaf_right(trans, root, path, data_size, | |
3341 | data_size, 0, 0); | |
d397712b | 3342 | if (wret < 0) |
eaee50e8 | 3343 | return wret; |
3685f791 | 3344 | if (wret) { |
99d8f83c CM |
3345 | wret = push_leaf_left(trans, root, path, data_size, |
3346 | data_size, 0, (u32)-1); | |
3685f791 CM |
3347 | if (wret < 0) |
3348 | return wret; | |
3349 | } | |
3350 | l = path->nodes[0]; | |
aa5d6bed | 3351 | |
3685f791 | 3352 | /* did the pushes work? */ |
87b29b20 | 3353 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 3354 | return 0; |
3326d1b0 | 3355 | } |
aa5d6bed | 3356 | |
5c680ed6 | 3357 | if (!path->nodes[1]) { |
e089f05c | 3358 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3359 | if (ret) |
3360 | return ret; | |
3361 | } | |
cc0c5538 | 3362 | again: |
5d4f98a2 | 3363 | split = 1; |
cc0c5538 | 3364 | l = path->nodes[0]; |
eb60ceac | 3365 | slot = path->slots[0]; |
5f39d397 | 3366 | nritems = btrfs_header_nritems(l); |
d397712b | 3367 | mid = (nritems + 1) / 2; |
54aa1f4d | 3368 | |
5d4f98a2 YZ |
3369 | if (mid <= slot) { |
3370 | if (nritems == 1 || | |
3371 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
3372 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3373 | if (slot >= nritems) { | |
3374 | split = 0; | |
3375 | } else { | |
3376 | mid = slot; | |
3377 | if (mid != nritems && | |
3378 | leaf_space_used(l, mid, nritems - mid) + | |
3379 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3380 | if (data_size && !tried_avoid_double) |
3381 | goto push_for_double; | |
5d4f98a2 YZ |
3382 | split = 2; |
3383 | } | |
3384 | } | |
3385 | } | |
3386 | } else { | |
3387 | if (leaf_space_used(l, 0, mid) + data_size > | |
3388 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3389 | if (!extend && data_size && slot == 0) { | |
3390 | split = 0; | |
3391 | } else if ((extend || !data_size) && slot == 0) { | |
3392 | mid = 1; | |
3393 | } else { | |
3394 | mid = slot; | |
3395 | if (mid != nritems && | |
3396 | leaf_space_used(l, mid, nritems - mid) + | |
3397 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3398 | if (data_size && !tried_avoid_double) |
3399 | goto push_for_double; | |
5d4f98a2 YZ |
3400 | split = 2 ; |
3401 | } | |
3402 | } | |
3403 | } | |
3404 | } | |
3405 | ||
3406 | if (split == 0) | |
3407 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3408 | else | |
3409 | btrfs_item_key(l, &disk_key, mid); | |
3410 | ||
3411 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 3412 | root->root_key.objectid, |
5581a51a | 3413 | &disk_key, 0, l->start, 0); |
f0486c68 | 3414 | if (IS_ERR(right)) |
5f39d397 | 3415 | return PTR_ERR(right); |
f0486c68 YZ |
3416 | |
3417 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
3418 | |
3419 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 3420 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 3421 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 3422 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3423 | btrfs_set_header_owner(right, root->root_key.objectid); |
3424 | btrfs_set_header_level(right, 0); | |
3425 | write_extent_buffer(right, root->fs_info->fsid, | |
3426 | (unsigned long)btrfs_header_fsid(right), | |
3427 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3428 | |
3429 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
3430 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
3431 | BTRFS_UUID_SIZE); | |
44871b1b | 3432 | |
5d4f98a2 YZ |
3433 | if (split == 0) { |
3434 | if (mid <= slot) { | |
3435 | btrfs_set_header_nritems(right, 0); | |
143bede5 JM |
3436 | insert_ptr(trans, root, path, &disk_key, right->start, |
3437 | path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
3438 | btrfs_tree_unlock(path->nodes[0]); |
3439 | free_extent_buffer(path->nodes[0]); | |
3440 | path->nodes[0] = right; | |
3441 | path->slots[0] = 0; | |
3442 | path->slots[1] += 1; | |
3443 | } else { | |
3444 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 3445 | insert_ptr(trans, root, path, &disk_key, right->start, |
5d4f98a2 | 3446 | path->slots[1], 1); |
5d4f98a2 YZ |
3447 | btrfs_tree_unlock(path->nodes[0]); |
3448 | free_extent_buffer(path->nodes[0]); | |
3449 | path->nodes[0] = right; | |
3450 | path->slots[0] = 0; | |
143bede5 JM |
3451 | if (path->slots[1] == 0) |
3452 | fixup_low_keys(trans, root, path, | |
3453 | &disk_key, 1); | |
d4dbff95 | 3454 | } |
5d4f98a2 YZ |
3455 | btrfs_mark_buffer_dirty(right); |
3456 | return ret; | |
d4dbff95 | 3457 | } |
74123bd7 | 3458 | |
143bede5 | 3459 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 3460 | |
5d4f98a2 | 3461 | if (split == 2) { |
cc0c5538 CM |
3462 | BUG_ON(num_doubles != 0); |
3463 | num_doubles++; | |
3464 | goto again; | |
a429e513 | 3465 | } |
44871b1b | 3466 | |
143bede5 | 3467 | return 0; |
99d8f83c CM |
3468 | |
3469 | push_for_double: | |
3470 | push_for_double_split(trans, root, path, data_size); | |
3471 | tried_avoid_double = 1; | |
3472 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3473 | return 0; | |
3474 | goto again; | |
be0e5c09 CM |
3475 | } |
3476 | ||
ad48fd75 YZ |
3477 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3478 | struct btrfs_root *root, | |
3479 | struct btrfs_path *path, int ins_len) | |
459931ec | 3480 | { |
ad48fd75 | 3481 | struct btrfs_key key; |
459931ec | 3482 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3483 | struct btrfs_file_extent_item *fi; |
3484 | u64 extent_len = 0; | |
3485 | u32 item_size; | |
3486 | int ret; | |
459931ec CM |
3487 | |
3488 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3489 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3490 | ||
3491 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3492 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3493 | ||
3494 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
3495 | return 0; | |
459931ec CM |
3496 | |
3497 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
3498 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3499 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3500 | struct btrfs_file_extent_item); | |
3501 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3502 | } | |
b3b4aa74 | 3503 | btrfs_release_path(path); |
459931ec | 3504 | |
459931ec | 3505 | path->keep_locks = 1; |
ad48fd75 YZ |
3506 | path->search_for_split = 1; |
3507 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3508 | path->search_for_split = 0; |
ad48fd75 YZ |
3509 | if (ret < 0) |
3510 | goto err; | |
459931ec | 3511 | |
ad48fd75 YZ |
3512 | ret = -EAGAIN; |
3513 | leaf = path->nodes[0]; | |
459931ec | 3514 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
3515 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
3516 | goto err; | |
3517 | ||
109f6aef CM |
3518 | /* the leaf has changed, it now has room. return now */ |
3519 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
3520 | goto err; | |
3521 | ||
ad48fd75 YZ |
3522 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3523 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3524 | struct btrfs_file_extent_item); | |
3525 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3526 | goto err; | |
459931ec CM |
3527 | } |
3528 | ||
b9473439 | 3529 | btrfs_set_path_blocking(path); |
ad48fd75 | 3530 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3531 | if (ret) |
3532 | goto err; | |
459931ec | 3533 | |
ad48fd75 | 3534 | path->keep_locks = 0; |
b9473439 | 3535 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3536 | return 0; |
3537 | err: | |
3538 | path->keep_locks = 0; | |
3539 | return ret; | |
3540 | } | |
3541 | ||
3542 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
3543 | struct btrfs_root *root, | |
3544 | struct btrfs_path *path, | |
3545 | struct btrfs_key *new_key, | |
3546 | unsigned long split_offset) | |
3547 | { | |
3548 | struct extent_buffer *leaf; | |
3549 | struct btrfs_item *item; | |
3550 | struct btrfs_item *new_item; | |
3551 | int slot; | |
3552 | char *buf; | |
3553 | u32 nritems; | |
3554 | u32 item_size; | |
3555 | u32 orig_offset; | |
3556 | struct btrfs_disk_key disk_key; | |
3557 | ||
b9473439 CM |
3558 | leaf = path->nodes[0]; |
3559 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
3560 | ||
b4ce94de CM |
3561 | btrfs_set_path_blocking(path); |
3562 | ||
459931ec CM |
3563 | item = btrfs_item_nr(leaf, path->slots[0]); |
3564 | orig_offset = btrfs_item_offset(leaf, item); | |
3565 | item_size = btrfs_item_size(leaf, item); | |
3566 | ||
459931ec | 3567 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3568 | if (!buf) |
3569 | return -ENOMEM; | |
3570 | ||
459931ec CM |
3571 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3572 | path->slots[0]), item_size); | |
459931ec | 3573 | |
ad48fd75 | 3574 | slot = path->slots[0] + 1; |
459931ec | 3575 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3576 | if (slot != nritems) { |
3577 | /* shift the items */ | |
3578 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3579 | btrfs_item_nr_offset(slot), |
3580 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3581 | } |
3582 | ||
3583 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3584 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3585 | ||
3586 | new_item = btrfs_item_nr(leaf, slot); | |
3587 | ||
3588 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
3589 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
3590 | ||
3591 | btrfs_set_item_offset(leaf, item, | |
3592 | orig_offset + item_size - split_offset); | |
3593 | btrfs_set_item_size(leaf, item, split_offset); | |
3594 | ||
3595 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3596 | ||
3597 | /* write the data for the start of the original item */ | |
3598 | write_extent_buffer(leaf, buf, | |
3599 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3600 | split_offset); | |
3601 | ||
3602 | /* write the data for the new item */ | |
3603 | write_extent_buffer(leaf, buf + split_offset, | |
3604 | btrfs_item_ptr_offset(leaf, slot), | |
3605 | item_size - split_offset); | |
3606 | btrfs_mark_buffer_dirty(leaf); | |
3607 | ||
ad48fd75 | 3608 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 3609 | kfree(buf); |
ad48fd75 YZ |
3610 | return 0; |
3611 | } | |
3612 | ||
3613 | /* | |
3614 | * This function splits a single item into two items, | |
3615 | * giving 'new_key' to the new item and splitting the | |
3616 | * old one at split_offset (from the start of the item). | |
3617 | * | |
3618 | * The path may be released by this operation. After | |
3619 | * the split, the path is pointing to the old item. The | |
3620 | * new item is going to be in the same node as the old one. | |
3621 | * | |
3622 | * Note, the item being split must be smaller enough to live alone on | |
3623 | * a tree block with room for one extra struct btrfs_item | |
3624 | * | |
3625 | * This allows us to split the item in place, keeping a lock on the | |
3626 | * leaf the entire time. | |
3627 | */ | |
3628 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3629 | struct btrfs_root *root, | |
3630 | struct btrfs_path *path, | |
3631 | struct btrfs_key *new_key, | |
3632 | unsigned long split_offset) | |
3633 | { | |
3634 | int ret; | |
3635 | ret = setup_leaf_for_split(trans, root, path, | |
3636 | sizeof(struct btrfs_item)); | |
3637 | if (ret) | |
3638 | return ret; | |
3639 | ||
3640 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
3641 | return ret; |
3642 | } | |
3643 | ||
ad48fd75 YZ |
3644 | /* |
3645 | * This function duplicate a item, giving 'new_key' to the new item. | |
3646 | * It guarantees both items live in the same tree leaf and the new item | |
3647 | * is contiguous with the original item. | |
3648 | * | |
3649 | * This allows us to split file extent in place, keeping a lock on the | |
3650 | * leaf the entire time. | |
3651 | */ | |
3652 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
3653 | struct btrfs_root *root, | |
3654 | struct btrfs_path *path, | |
3655 | struct btrfs_key *new_key) | |
3656 | { | |
3657 | struct extent_buffer *leaf; | |
3658 | int ret; | |
3659 | u32 item_size; | |
3660 | ||
3661 | leaf = path->nodes[0]; | |
3662 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3663 | ret = setup_leaf_for_split(trans, root, path, | |
3664 | item_size + sizeof(struct btrfs_item)); | |
3665 | if (ret) | |
3666 | return ret; | |
3667 | ||
3668 | path->slots[0]++; | |
143bede5 JM |
3669 | setup_items_for_insert(trans, root, path, new_key, &item_size, |
3670 | item_size, item_size + | |
3671 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
3672 | leaf = path->nodes[0]; |
3673 | memcpy_extent_buffer(leaf, | |
3674 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3675 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
3676 | item_size); | |
3677 | return 0; | |
3678 | } | |
3679 | ||
d352ac68 CM |
3680 | /* |
3681 | * make the item pointed to by the path smaller. new_size indicates | |
3682 | * how small to make it, and from_end tells us if we just chop bytes | |
3683 | * off the end of the item or if we shift the item to chop bytes off | |
3684 | * the front. | |
3685 | */ | |
143bede5 JM |
3686 | void btrfs_truncate_item(struct btrfs_trans_handle *trans, |
3687 | struct btrfs_root *root, | |
3688 | struct btrfs_path *path, | |
3689 | u32 new_size, int from_end) | |
b18c6685 | 3690 | { |
b18c6685 | 3691 | int slot; |
5f39d397 CM |
3692 | struct extent_buffer *leaf; |
3693 | struct btrfs_item *item; | |
b18c6685 CM |
3694 | u32 nritems; |
3695 | unsigned int data_end; | |
3696 | unsigned int old_data_start; | |
3697 | unsigned int old_size; | |
3698 | unsigned int size_diff; | |
3699 | int i; | |
cfed81a0 CM |
3700 | struct btrfs_map_token token; |
3701 | ||
3702 | btrfs_init_map_token(&token); | |
b18c6685 | 3703 | |
5f39d397 | 3704 | leaf = path->nodes[0]; |
179e29e4 CM |
3705 | slot = path->slots[0]; |
3706 | ||
3707 | old_size = btrfs_item_size_nr(leaf, slot); | |
3708 | if (old_size == new_size) | |
143bede5 | 3709 | return; |
b18c6685 | 3710 | |
5f39d397 | 3711 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
3712 | data_end = leaf_data_end(root, leaf); |
3713 | ||
5f39d397 | 3714 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 3715 | |
b18c6685 CM |
3716 | size_diff = old_size - new_size; |
3717 | ||
3718 | BUG_ON(slot < 0); | |
3719 | BUG_ON(slot >= nritems); | |
3720 | ||
3721 | /* | |
3722 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3723 | */ | |
3724 | /* first correct the data pointers */ | |
3725 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3726 | u32 ioff; |
3727 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3728 | |
cfed81a0 CM |
3729 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3730 | btrfs_set_token_item_offset(leaf, item, | |
3731 | ioff + size_diff, &token); | |
b18c6685 | 3732 | } |
db94535d | 3733 | |
b18c6685 | 3734 | /* shift the data */ |
179e29e4 CM |
3735 | if (from_end) { |
3736 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3737 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3738 | data_end, old_data_start + new_size - data_end); | |
3739 | } else { | |
3740 | struct btrfs_disk_key disk_key; | |
3741 | u64 offset; | |
3742 | ||
3743 | btrfs_item_key(leaf, &disk_key, slot); | |
3744 | ||
3745 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3746 | unsigned long ptr; | |
3747 | struct btrfs_file_extent_item *fi; | |
3748 | ||
3749 | fi = btrfs_item_ptr(leaf, slot, | |
3750 | struct btrfs_file_extent_item); | |
3751 | fi = (struct btrfs_file_extent_item *)( | |
3752 | (unsigned long)fi - size_diff); | |
3753 | ||
3754 | if (btrfs_file_extent_type(leaf, fi) == | |
3755 | BTRFS_FILE_EXTENT_INLINE) { | |
3756 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3757 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
3758 | (unsigned long)fi, |
3759 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
3760 | disk_bytenr)); |
3761 | } | |
3762 | } | |
3763 | ||
3764 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3765 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3766 | data_end, old_data_start - data_end); | |
3767 | ||
3768 | offset = btrfs_disk_key_offset(&disk_key); | |
3769 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3770 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3771 | if (slot == 0) | |
3772 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
3773 | } | |
5f39d397 CM |
3774 | |
3775 | item = btrfs_item_nr(leaf, slot); | |
3776 | btrfs_set_item_size(leaf, item, new_size); | |
3777 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 3778 | |
5f39d397 CM |
3779 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3780 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 3781 | BUG(); |
5f39d397 | 3782 | } |
b18c6685 CM |
3783 | } |
3784 | ||
d352ac68 CM |
3785 | /* |
3786 | * make the item pointed to by the path bigger, data_size is the new size. | |
3787 | */ | |
143bede5 JM |
3788 | void btrfs_extend_item(struct btrfs_trans_handle *trans, |
3789 | struct btrfs_root *root, struct btrfs_path *path, | |
3790 | u32 data_size) | |
6567e837 | 3791 | { |
6567e837 | 3792 | int slot; |
5f39d397 CM |
3793 | struct extent_buffer *leaf; |
3794 | struct btrfs_item *item; | |
6567e837 CM |
3795 | u32 nritems; |
3796 | unsigned int data_end; | |
3797 | unsigned int old_data; | |
3798 | unsigned int old_size; | |
3799 | int i; | |
cfed81a0 CM |
3800 | struct btrfs_map_token token; |
3801 | ||
3802 | btrfs_init_map_token(&token); | |
6567e837 | 3803 | |
5f39d397 | 3804 | leaf = path->nodes[0]; |
6567e837 | 3805 | |
5f39d397 | 3806 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
3807 | data_end = leaf_data_end(root, leaf); |
3808 | ||
5f39d397 CM |
3809 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
3810 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3811 | BUG(); |
5f39d397 | 3812 | } |
6567e837 | 3813 | slot = path->slots[0]; |
5f39d397 | 3814 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
3815 | |
3816 | BUG_ON(slot < 0); | |
3326d1b0 CM |
3817 | if (slot >= nritems) { |
3818 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
3819 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
3820 | slot, nritems); | |
3326d1b0 CM |
3821 | BUG_ON(1); |
3822 | } | |
6567e837 CM |
3823 | |
3824 | /* | |
3825 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3826 | */ | |
3827 | /* first correct the data pointers */ | |
3828 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3829 | u32 ioff; |
3830 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3831 | |
cfed81a0 CM |
3832 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3833 | btrfs_set_token_item_offset(leaf, item, | |
3834 | ioff - data_size, &token); | |
6567e837 | 3835 | } |
5f39d397 | 3836 | |
6567e837 | 3837 | /* shift the data */ |
5f39d397 | 3838 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
3839 | data_end - data_size, btrfs_leaf_data(leaf) + |
3840 | data_end, old_data - data_end); | |
5f39d397 | 3841 | |
6567e837 | 3842 | data_end = old_data; |
5f39d397 CM |
3843 | old_size = btrfs_item_size_nr(leaf, slot); |
3844 | item = btrfs_item_nr(leaf, slot); | |
3845 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
3846 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 3847 | |
5f39d397 CM |
3848 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3849 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3850 | BUG(); |
5f39d397 | 3851 | } |
6567e837 CM |
3852 | } |
3853 | ||
f3465ca4 JB |
3854 | /* |
3855 | * Given a key and some data, insert items into the tree. | |
3856 | * This does all the path init required, making room in the tree if needed. | |
3857 | * Returns the number of keys that were inserted. | |
3858 | */ | |
3859 | int btrfs_insert_some_items(struct btrfs_trans_handle *trans, | |
3860 | struct btrfs_root *root, | |
3861 | struct btrfs_path *path, | |
3862 | struct btrfs_key *cpu_key, u32 *data_size, | |
3863 | int nr) | |
3864 | { | |
3865 | struct extent_buffer *leaf; | |
3866 | struct btrfs_item *item; | |
3867 | int ret = 0; | |
3868 | int slot; | |
f3465ca4 JB |
3869 | int i; |
3870 | u32 nritems; | |
3871 | u32 total_data = 0; | |
3872 | u32 total_size = 0; | |
3873 | unsigned int data_end; | |
3874 | struct btrfs_disk_key disk_key; | |
3875 | struct btrfs_key found_key; | |
cfed81a0 CM |
3876 | struct btrfs_map_token token; |
3877 | ||
3878 | btrfs_init_map_token(&token); | |
f3465ca4 | 3879 | |
87b29b20 YZ |
3880 | for (i = 0; i < nr; i++) { |
3881 | if (total_size + data_size[i] + sizeof(struct btrfs_item) > | |
3882 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3883 | break; | |
3884 | nr = i; | |
3885 | } | |
f3465ca4 | 3886 | total_data += data_size[i]; |
87b29b20 YZ |
3887 | total_size += data_size[i] + sizeof(struct btrfs_item); |
3888 | } | |
3889 | BUG_ON(nr == 0); | |
f3465ca4 | 3890 | |
f3465ca4 JB |
3891 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); |
3892 | if (ret == 0) | |
3893 | return -EEXIST; | |
3894 | if (ret < 0) | |
3895 | goto out; | |
3896 | ||
f3465ca4 JB |
3897 | leaf = path->nodes[0]; |
3898 | ||
3899 | nritems = btrfs_header_nritems(leaf); | |
3900 | data_end = leaf_data_end(root, leaf); | |
3901 | ||
3902 | if (btrfs_leaf_free_space(root, leaf) < total_size) { | |
3903 | for (i = nr; i >= 0; i--) { | |
3904 | total_data -= data_size[i]; | |
3905 | total_size -= data_size[i] + sizeof(struct btrfs_item); | |
3906 | if (total_size < btrfs_leaf_free_space(root, leaf)) | |
3907 | break; | |
3908 | } | |
3909 | nr = i; | |
3910 | } | |
3911 | ||
3912 | slot = path->slots[0]; | |
3913 | BUG_ON(slot < 0); | |
3914 | ||
3915 | if (slot != nritems) { | |
3916 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); | |
3917 | ||
3918 | item = btrfs_item_nr(leaf, slot); | |
3919 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3920 | ||
3921 | /* figure out how many keys we can insert in here */ | |
3922 | total_data = data_size[0]; | |
3923 | for (i = 1; i < nr; i++) { | |
5d4f98a2 | 3924 | if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0) |
f3465ca4 JB |
3925 | break; |
3926 | total_data += data_size[i]; | |
3927 | } | |
3928 | nr = i; | |
3929 | ||
3930 | if (old_data < data_end) { | |
3931 | btrfs_print_leaf(root, leaf); | |
d397712b | 3932 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
f3465ca4 JB |
3933 | slot, old_data, data_end); |
3934 | BUG_ON(1); | |
3935 | } | |
3936 | /* | |
3937 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3938 | */ | |
3939 | /* first correct the data pointers */ | |
f3465ca4 JB |
3940 | for (i = slot; i < nritems; i++) { |
3941 | u32 ioff; | |
3942 | ||
3943 | item = btrfs_item_nr(leaf, i); | |
cfed81a0 CM |
3944 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3945 | btrfs_set_token_item_offset(leaf, item, | |
3946 | ioff - total_data, &token); | |
f3465ca4 | 3947 | } |
f3465ca4 JB |
3948 | /* shift the items */ |
3949 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), | |
3950 | btrfs_item_nr_offset(slot), | |
3951 | (nritems - slot) * sizeof(struct btrfs_item)); | |
3952 | ||
3953 | /* shift the data */ | |
3954 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3955 | data_end - total_data, btrfs_leaf_data(leaf) + | |
3956 | data_end, old_data - data_end); | |
3957 | data_end = old_data; | |
3958 | } else { | |
3959 | /* | |
3960 | * this sucks but it has to be done, if we are inserting at | |
3961 | * the end of the leaf only insert 1 of the items, since we | |
3962 | * have no way of knowing whats on the next leaf and we'd have | |
3963 | * to drop our current locks to figure it out | |
3964 | */ | |
3965 | nr = 1; | |
3966 | } | |
3967 | ||
3968 | /* setup the item for the new data */ | |
3969 | for (i = 0; i < nr; i++) { | |
3970 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
3971 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
3972 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
3973 | btrfs_set_token_item_offset(leaf, item, |
3974 | data_end - data_size[i], &token); | |
f3465ca4 | 3975 | data_end -= data_size[i]; |
cfed81a0 | 3976 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
f3465ca4 JB |
3977 | } |
3978 | btrfs_set_header_nritems(leaf, nritems + nr); | |
3979 | btrfs_mark_buffer_dirty(leaf); | |
3980 | ||
3981 | ret = 0; | |
3982 | if (slot == 0) { | |
3983 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 3984 | fixup_low_keys(trans, root, path, &disk_key, 1); |
f3465ca4 JB |
3985 | } |
3986 | ||
3987 | if (btrfs_leaf_free_space(root, leaf) < 0) { | |
3988 | btrfs_print_leaf(root, leaf); | |
3989 | BUG(); | |
3990 | } | |
3991 | out: | |
3992 | if (!ret) | |
3993 | ret = nr; | |
3994 | return ret; | |
3995 | } | |
3996 | ||
74123bd7 | 3997 | /* |
44871b1b CM |
3998 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
3999 | * to save stack depth by doing the bulk of the work in a function | |
4000 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4001 | */ |
143bede5 JM |
4002 | void setup_items_for_insert(struct btrfs_trans_handle *trans, |
4003 | struct btrfs_root *root, struct btrfs_path *path, | |
4004 | struct btrfs_key *cpu_key, u32 *data_size, | |
4005 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4006 | { |
5f39d397 | 4007 | struct btrfs_item *item; |
9c58309d | 4008 | int i; |
7518a238 | 4009 | u32 nritems; |
be0e5c09 | 4010 | unsigned int data_end; |
e2fa7227 | 4011 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4012 | struct extent_buffer *leaf; |
4013 | int slot; | |
cfed81a0 CM |
4014 | struct btrfs_map_token token; |
4015 | ||
4016 | btrfs_init_map_token(&token); | |
e2fa7227 | 4017 | |
5f39d397 | 4018 | leaf = path->nodes[0]; |
44871b1b | 4019 | slot = path->slots[0]; |
74123bd7 | 4020 | |
5f39d397 | 4021 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4022 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4023 | |
f25956cc | 4024 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4025 | btrfs_print_leaf(root, leaf); |
d397712b | 4026 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 4027 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4028 | BUG(); |
d4dbff95 | 4029 | } |
5f39d397 | 4030 | |
be0e5c09 | 4031 | if (slot != nritems) { |
5f39d397 | 4032 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4033 | |
5f39d397 CM |
4034 | if (old_data < data_end) { |
4035 | btrfs_print_leaf(root, leaf); | |
d397712b | 4036 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
4037 | slot, old_data, data_end); |
4038 | BUG_ON(1); | |
4039 | } | |
be0e5c09 CM |
4040 | /* |
4041 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4042 | */ | |
4043 | /* first correct the data pointers */ | |
0783fcfc | 4044 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4045 | u32 ioff; |
db94535d | 4046 | |
5f39d397 | 4047 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4048 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4049 | btrfs_set_token_item_offset(leaf, item, | |
4050 | ioff - total_data, &token); | |
0783fcfc | 4051 | } |
be0e5c09 | 4052 | /* shift the items */ |
9c58309d | 4053 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4054 | btrfs_item_nr_offset(slot), |
d6025579 | 4055 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4056 | |
4057 | /* shift the data */ | |
5f39d397 | 4058 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4059 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4060 | data_end, old_data - data_end); |
be0e5c09 CM |
4061 | data_end = old_data; |
4062 | } | |
5f39d397 | 4063 | |
62e2749e | 4064 | /* setup the item for the new data */ |
9c58309d CM |
4065 | for (i = 0; i < nr; i++) { |
4066 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4067 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
4068 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
4069 | btrfs_set_token_item_offset(leaf, item, |
4070 | data_end - data_size[i], &token); | |
9c58309d | 4071 | data_end -= data_size[i]; |
cfed81a0 | 4072 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4073 | } |
44871b1b | 4074 | |
9c58309d | 4075 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4076 | |
5a01a2e3 CM |
4077 | if (slot == 0) { |
4078 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 4079 | fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 4080 | } |
b9473439 CM |
4081 | btrfs_unlock_up_safe(path, 1); |
4082 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4083 | |
5f39d397 CM |
4084 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4085 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4086 | BUG(); |
5f39d397 | 4087 | } |
44871b1b CM |
4088 | } |
4089 | ||
4090 | /* | |
4091 | * Given a key and some data, insert items into the tree. | |
4092 | * This does all the path init required, making room in the tree if needed. | |
4093 | */ | |
4094 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4095 | struct btrfs_root *root, | |
4096 | struct btrfs_path *path, | |
4097 | struct btrfs_key *cpu_key, u32 *data_size, | |
4098 | int nr) | |
4099 | { | |
44871b1b CM |
4100 | int ret = 0; |
4101 | int slot; | |
4102 | int i; | |
4103 | u32 total_size = 0; | |
4104 | u32 total_data = 0; | |
4105 | ||
4106 | for (i = 0; i < nr; i++) | |
4107 | total_data += data_size[i]; | |
4108 | ||
4109 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4110 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4111 | if (ret == 0) | |
4112 | return -EEXIST; | |
4113 | if (ret < 0) | |
143bede5 | 4114 | return ret; |
44871b1b | 4115 | |
44871b1b CM |
4116 | slot = path->slots[0]; |
4117 | BUG_ON(slot < 0); | |
4118 | ||
143bede5 | 4119 | setup_items_for_insert(trans, root, path, cpu_key, data_size, |
44871b1b | 4120 | total_data, total_size, nr); |
143bede5 | 4121 | return 0; |
62e2749e CM |
4122 | } |
4123 | ||
4124 | /* | |
4125 | * Given a key and some data, insert an item into the tree. | |
4126 | * This does all the path init required, making room in the tree if needed. | |
4127 | */ | |
e089f05c CM |
4128 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4129 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4130 | data_size) | |
62e2749e CM |
4131 | { |
4132 | int ret = 0; | |
2c90e5d6 | 4133 | struct btrfs_path *path; |
5f39d397 CM |
4134 | struct extent_buffer *leaf; |
4135 | unsigned long ptr; | |
62e2749e | 4136 | |
2c90e5d6 | 4137 | path = btrfs_alloc_path(); |
db5b493a TI |
4138 | if (!path) |
4139 | return -ENOMEM; | |
2c90e5d6 | 4140 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4141 | if (!ret) { |
5f39d397 CM |
4142 | leaf = path->nodes[0]; |
4143 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4144 | write_extent_buffer(leaf, data, ptr, data_size); | |
4145 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4146 | } |
2c90e5d6 | 4147 | btrfs_free_path(path); |
aa5d6bed | 4148 | return ret; |
be0e5c09 CM |
4149 | } |
4150 | ||
74123bd7 | 4151 | /* |
5de08d7d | 4152 | * delete the pointer from a given node. |
74123bd7 | 4153 | * |
d352ac68 CM |
4154 | * the tree should have been previously balanced so the deletion does not |
4155 | * empty a node. | |
74123bd7 | 4156 | */ |
143bede5 JM |
4157 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4158 | struct btrfs_path *path, int level, int slot) | |
be0e5c09 | 4159 | { |
5f39d397 | 4160 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4161 | u32 nritems; |
be0e5c09 | 4162 | |
5f39d397 | 4163 | nritems = btrfs_header_nritems(parent); |
d397712b | 4164 | if (slot != nritems - 1) { |
5f39d397 CM |
4165 | memmove_extent_buffer(parent, |
4166 | btrfs_node_key_ptr_offset(slot), | |
4167 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4168 | sizeof(struct btrfs_key_ptr) * |
4169 | (nritems - slot - 1)); | |
bb803951 | 4170 | } |
7518a238 | 4171 | nritems--; |
5f39d397 | 4172 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4173 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4174 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4175 | /* just turn the root into a leaf and break */ |
5f39d397 | 4176 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4177 | } else if (slot == 0) { |
5f39d397 CM |
4178 | struct btrfs_disk_key disk_key; |
4179 | ||
4180 | btrfs_node_key(parent, &disk_key, 0); | |
143bede5 | 4181 | fixup_low_keys(trans, root, path, &disk_key, level + 1); |
be0e5c09 | 4182 | } |
d6025579 | 4183 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4184 | } |
4185 | ||
323ac95b CM |
4186 | /* |
4187 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4188 | * path->nodes[1]. |
323ac95b CM |
4189 | * |
4190 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4191 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4192 | * | |
4193 | * The path must have already been setup for deleting the leaf, including | |
4194 | * all the proper balancing. path->nodes[1] must be locked. | |
4195 | */ | |
143bede5 JM |
4196 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4197 | struct btrfs_root *root, | |
4198 | struct btrfs_path *path, | |
4199 | struct extent_buffer *leaf) | |
323ac95b | 4200 | { |
5d4f98a2 | 4201 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
143bede5 | 4202 | del_ptr(trans, root, path, 1, path->slots[1]); |
323ac95b | 4203 | |
4d081c41 CM |
4204 | /* |
4205 | * btrfs_free_extent is expensive, we want to make sure we | |
4206 | * aren't holding any locks when we call it | |
4207 | */ | |
4208 | btrfs_unlock_up_safe(path, 0); | |
4209 | ||
f0486c68 YZ |
4210 | root_sub_used(root, leaf->len); |
4211 | ||
3083ee2e | 4212 | extent_buffer_get(leaf); |
5581a51a | 4213 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4214 | free_extent_buffer_stale(leaf); |
323ac95b | 4215 | } |
74123bd7 CM |
4216 | /* |
4217 | * delete the item at the leaf level in path. If that empties | |
4218 | * the leaf, remove it from the tree | |
4219 | */ | |
85e21bac CM |
4220 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4221 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4222 | { |
5f39d397 CM |
4223 | struct extent_buffer *leaf; |
4224 | struct btrfs_item *item; | |
85e21bac CM |
4225 | int last_off; |
4226 | int dsize = 0; | |
aa5d6bed CM |
4227 | int ret = 0; |
4228 | int wret; | |
85e21bac | 4229 | int i; |
7518a238 | 4230 | u32 nritems; |
cfed81a0 CM |
4231 | struct btrfs_map_token token; |
4232 | ||
4233 | btrfs_init_map_token(&token); | |
be0e5c09 | 4234 | |
5f39d397 | 4235 | leaf = path->nodes[0]; |
85e21bac CM |
4236 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4237 | ||
4238 | for (i = 0; i < nr; i++) | |
4239 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4240 | ||
5f39d397 | 4241 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4242 | |
85e21bac | 4243 | if (slot + nr != nritems) { |
123abc88 | 4244 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4245 | |
4246 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4247 | data_end + dsize, |
4248 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4249 | last_off - data_end); |
5f39d397 | 4250 | |
85e21bac | 4251 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4252 | u32 ioff; |
db94535d | 4253 | |
5f39d397 | 4254 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4255 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4256 | btrfs_set_token_item_offset(leaf, item, | |
4257 | ioff + dsize, &token); | |
0783fcfc | 4258 | } |
db94535d | 4259 | |
5f39d397 | 4260 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4261 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4262 | sizeof(struct btrfs_item) * |
85e21bac | 4263 | (nritems - slot - nr)); |
be0e5c09 | 4264 | } |
85e21bac CM |
4265 | btrfs_set_header_nritems(leaf, nritems - nr); |
4266 | nritems -= nr; | |
5f39d397 | 4267 | |
74123bd7 | 4268 | /* delete the leaf if we've emptied it */ |
7518a238 | 4269 | if (nritems == 0) { |
5f39d397 CM |
4270 | if (leaf == root->node) { |
4271 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4272 | } else { |
f0486c68 YZ |
4273 | btrfs_set_path_blocking(path); |
4274 | clean_tree_block(trans, root, leaf); | |
143bede5 | 4275 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4276 | } |
be0e5c09 | 4277 | } else { |
7518a238 | 4278 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4279 | if (slot == 0) { |
5f39d397 CM |
4280 | struct btrfs_disk_key disk_key; |
4281 | ||
4282 | btrfs_item_key(leaf, &disk_key, 0); | |
143bede5 | 4283 | fixup_low_keys(trans, root, path, &disk_key, 1); |
aa5d6bed | 4284 | } |
aa5d6bed | 4285 | |
74123bd7 | 4286 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 4287 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
4288 | /* push_leaf_left fixes the path. |
4289 | * make sure the path still points to our leaf | |
4290 | * for possible call to del_ptr below | |
4291 | */ | |
4920c9ac | 4292 | slot = path->slots[1]; |
5f39d397 CM |
4293 | extent_buffer_get(leaf); |
4294 | ||
b9473439 | 4295 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4296 | wret = push_leaf_left(trans, root, path, 1, 1, |
4297 | 1, (u32)-1); | |
54aa1f4d | 4298 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4299 | ret = wret; |
5f39d397 CM |
4300 | |
4301 | if (path->nodes[0] == leaf && | |
4302 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4303 | wret = push_leaf_right(trans, root, path, 1, |
4304 | 1, 1, 0); | |
54aa1f4d | 4305 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4306 | ret = wret; |
4307 | } | |
5f39d397 CM |
4308 | |
4309 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4310 | path->slots[1] = slot; |
143bede5 | 4311 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4312 | free_extent_buffer(leaf); |
143bede5 | 4313 | ret = 0; |
5de08d7d | 4314 | } else { |
925baedd CM |
4315 | /* if we're still in the path, make sure |
4316 | * we're dirty. Otherwise, one of the | |
4317 | * push_leaf functions must have already | |
4318 | * dirtied this buffer | |
4319 | */ | |
4320 | if (path->nodes[0] == leaf) | |
4321 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4322 | free_extent_buffer(leaf); |
be0e5c09 | 4323 | } |
d5719762 | 4324 | } else { |
5f39d397 | 4325 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4326 | } |
4327 | } | |
aa5d6bed | 4328 | return ret; |
be0e5c09 CM |
4329 | } |
4330 | ||
7bb86316 | 4331 | /* |
925baedd | 4332 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4333 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4334 | * returns < 0 on io errors. | |
d352ac68 CM |
4335 | * |
4336 | * This may release the path, and so you may lose any locks held at the | |
4337 | * time you call it. | |
7bb86316 CM |
4338 | */ |
4339 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
4340 | { | |
925baedd CM |
4341 | struct btrfs_key key; |
4342 | struct btrfs_disk_key found_key; | |
4343 | int ret; | |
7bb86316 | 4344 | |
925baedd | 4345 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4346 | |
925baedd CM |
4347 | if (key.offset > 0) |
4348 | key.offset--; | |
4349 | else if (key.type > 0) | |
4350 | key.type--; | |
4351 | else if (key.objectid > 0) | |
4352 | key.objectid--; | |
4353 | else | |
4354 | return 1; | |
7bb86316 | 4355 | |
b3b4aa74 | 4356 | btrfs_release_path(path); |
925baedd CM |
4357 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4358 | if (ret < 0) | |
4359 | return ret; | |
4360 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4361 | ret = comp_keys(&found_key, &key); | |
4362 | if (ret < 0) | |
4363 | return 0; | |
4364 | return 1; | |
7bb86316 CM |
4365 | } |
4366 | ||
3f157a2f CM |
4367 | /* |
4368 | * A helper function to walk down the tree starting at min_key, and looking | |
4369 | * for nodes or leaves that are either in cache or have a minimum | |
d352ac68 | 4370 | * transaction id. This is used by the btree defrag code, and tree logging |
3f157a2f CM |
4371 | * |
4372 | * This does not cow, but it does stuff the starting key it finds back | |
4373 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4374 | * key and get a writable path. | |
4375 | * | |
4376 | * This does lock as it descends, and path->keep_locks should be set | |
4377 | * to 1 by the caller. | |
4378 | * | |
4379 | * This honors path->lowest_level to prevent descent past a given level | |
4380 | * of the tree. | |
4381 | * | |
d352ac68 CM |
4382 | * min_trans indicates the oldest transaction that you are interested |
4383 | * in walking through. Any nodes or leaves older than min_trans are | |
4384 | * skipped over (without reading them). | |
4385 | * | |
3f157a2f CM |
4386 | * returns zero if something useful was found, < 0 on error and 1 if there |
4387 | * was nothing in the tree that matched the search criteria. | |
4388 | */ | |
4389 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 4390 | struct btrfs_key *max_key, |
3f157a2f CM |
4391 | struct btrfs_path *path, int cache_only, |
4392 | u64 min_trans) | |
4393 | { | |
4394 | struct extent_buffer *cur; | |
4395 | struct btrfs_key found_key; | |
4396 | int slot; | |
9652480b | 4397 | int sret; |
3f157a2f CM |
4398 | u32 nritems; |
4399 | int level; | |
4400 | int ret = 1; | |
4401 | ||
934d375b | 4402 | WARN_ON(!path->keep_locks); |
3f157a2f | 4403 | again: |
bd681513 | 4404 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4405 | level = btrfs_header_level(cur); |
e02119d5 | 4406 | WARN_ON(path->nodes[level]); |
3f157a2f | 4407 | path->nodes[level] = cur; |
bd681513 | 4408 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4409 | |
4410 | if (btrfs_header_generation(cur) < min_trans) { | |
4411 | ret = 1; | |
4412 | goto out; | |
4413 | } | |
d397712b | 4414 | while (1) { |
3f157a2f CM |
4415 | nritems = btrfs_header_nritems(cur); |
4416 | level = btrfs_header_level(cur); | |
9652480b | 4417 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4418 | |
323ac95b CM |
4419 | /* at the lowest level, we're done, setup the path and exit */ |
4420 | if (level == path->lowest_level) { | |
e02119d5 CM |
4421 | if (slot >= nritems) |
4422 | goto find_next_key; | |
3f157a2f CM |
4423 | ret = 0; |
4424 | path->slots[level] = slot; | |
4425 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4426 | goto out; | |
4427 | } | |
9652480b Y |
4428 | if (sret && slot > 0) |
4429 | slot--; | |
3f157a2f CM |
4430 | /* |
4431 | * check this node pointer against the cache_only and | |
4432 | * min_trans parameters. If it isn't in cache or is too | |
4433 | * old, skip to the next one. | |
4434 | */ | |
d397712b | 4435 | while (slot < nritems) { |
3f157a2f CM |
4436 | u64 blockptr; |
4437 | u64 gen; | |
4438 | struct extent_buffer *tmp; | |
e02119d5 CM |
4439 | struct btrfs_disk_key disk_key; |
4440 | ||
3f157a2f CM |
4441 | blockptr = btrfs_node_blockptr(cur, slot); |
4442 | gen = btrfs_node_ptr_generation(cur, slot); | |
4443 | if (gen < min_trans) { | |
4444 | slot++; | |
4445 | continue; | |
4446 | } | |
4447 | if (!cache_only) | |
4448 | break; | |
4449 | ||
e02119d5 CM |
4450 | if (max_key) { |
4451 | btrfs_node_key(cur, &disk_key, slot); | |
4452 | if (comp_keys(&disk_key, max_key) >= 0) { | |
4453 | ret = 1; | |
4454 | goto out; | |
4455 | } | |
4456 | } | |
4457 | ||
3f157a2f CM |
4458 | tmp = btrfs_find_tree_block(root, blockptr, |
4459 | btrfs_level_size(root, level - 1)); | |
4460 | ||
b9fab919 | 4461 | if (tmp && btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
3f157a2f CM |
4462 | free_extent_buffer(tmp); |
4463 | break; | |
4464 | } | |
4465 | if (tmp) | |
4466 | free_extent_buffer(tmp); | |
4467 | slot++; | |
4468 | } | |
e02119d5 | 4469 | find_next_key: |
3f157a2f CM |
4470 | /* |
4471 | * we didn't find a candidate key in this node, walk forward | |
4472 | * and find another one | |
4473 | */ | |
4474 | if (slot >= nritems) { | |
e02119d5 | 4475 | path->slots[level] = slot; |
b4ce94de | 4476 | btrfs_set_path_blocking(path); |
e02119d5 | 4477 | sret = btrfs_find_next_key(root, path, min_key, level, |
3f157a2f | 4478 | cache_only, min_trans); |
e02119d5 | 4479 | if (sret == 0) { |
b3b4aa74 | 4480 | btrfs_release_path(path); |
3f157a2f CM |
4481 | goto again; |
4482 | } else { | |
4483 | goto out; | |
4484 | } | |
4485 | } | |
4486 | /* save our key for returning back */ | |
4487 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4488 | path->slots[level] = slot; | |
4489 | if (level == path->lowest_level) { | |
4490 | ret = 0; | |
f7c79f30 | 4491 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4492 | goto out; |
4493 | } | |
b4ce94de | 4494 | btrfs_set_path_blocking(path); |
3f157a2f | 4495 | cur = read_node_slot(root, cur, slot); |
79787eaa | 4496 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 4497 | |
bd681513 | 4498 | btrfs_tree_read_lock(cur); |
b4ce94de | 4499 | |
bd681513 | 4500 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4501 | path->nodes[level - 1] = cur; |
f7c79f30 | 4502 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 4503 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4504 | } |
4505 | out: | |
4506 | if (ret == 0) | |
4507 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4508 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4509 | return ret; |
4510 | } | |
4511 | ||
4512 | /* | |
4513 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4514 | * and fixup the path. It looks for and returns the next key in the | |
4515 | * tree based on the current path and the cache_only and min_trans | |
4516 | * parameters. | |
4517 | * | |
4518 | * 0 is returned if another key is found, < 0 if there are any errors | |
4519 | * and 1 is returned if there are no higher keys in the tree | |
4520 | * | |
4521 | * path->keep_locks should be set to 1 on the search made before | |
4522 | * calling this function. | |
4523 | */ | |
e7a84565 | 4524 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
33c66f43 | 4525 | struct btrfs_key *key, int level, |
3f157a2f | 4526 | int cache_only, u64 min_trans) |
e7a84565 | 4527 | { |
e7a84565 CM |
4528 | int slot; |
4529 | struct extent_buffer *c; | |
4530 | ||
934d375b | 4531 | WARN_ON(!path->keep_locks); |
d397712b | 4532 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4533 | if (!path->nodes[level]) |
4534 | return 1; | |
4535 | ||
4536 | slot = path->slots[level] + 1; | |
4537 | c = path->nodes[level]; | |
3f157a2f | 4538 | next: |
e7a84565 | 4539 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4540 | int ret; |
4541 | int orig_lowest; | |
4542 | struct btrfs_key cur_key; | |
4543 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4544 | !path->nodes[level + 1]) | |
e7a84565 | 4545 | return 1; |
33c66f43 YZ |
4546 | |
4547 | if (path->locks[level + 1]) { | |
4548 | level++; | |
4549 | continue; | |
4550 | } | |
4551 | ||
4552 | slot = btrfs_header_nritems(c) - 1; | |
4553 | if (level == 0) | |
4554 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4555 | else | |
4556 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4557 | ||
4558 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4559 | btrfs_release_path(path); |
33c66f43 YZ |
4560 | path->lowest_level = level; |
4561 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4562 | 0, 0); | |
4563 | path->lowest_level = orig_lowest; | |
4564 | if (ret < 0) | |
4565 | return ret; | |
4566 | ||
4567 | c = path->nodes[level]; | |
4568 | slot = path->slots[level]; | |
4569 | if (ret == 0) | |
4570 | slot++; | |
4571 | goto next; | |
e7a84565 | 4572 | } |
33c66f43 | 4573 | |
e7a84565 CM |
4574 | if (level == 0) |
4575 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f CM |
4576 | else { |
4577 | u64 blockptr = btrfs_node_blockptr(c, slot); | |
4578 | u64 gen = btrfs_node_ptr_generation(c, slot); | |
4579 | ||
4580 | if (cache_only) { | |
4581 | struct extent_buffer *cur; | |
4582 | cur = btrfs_find_tree_block(root, blockptr, | |
4583 | btrfs_level_size(root, level - 1)); | |
b9fab919 CM |
4584 | if (!cur || |
4585 | btrfs_buffer_uptodate(cur, gen, 1) <= 0) { | |
3f157a2f CM |
4586 | slot++; |
4587 | if (cur) | |
4588 | free_extent_buffer(cur); | |
4589 | goto next; | |
4590 | } | |
4591 | free_extent_buffer(cur); | |
4592 | } | |
4593 | if (gen < min_trans) { | |
4594 | slot++; | |
4595 | goto next; | |
4596 | } | |
e7a84565 | 4597 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4598 | } |
e7a84565 CM |
4599 | return 0; |
4600 | } | |
4601 | return 1; | |
4602 | } | |
4603 | ||
97571fd0 | 4604 | /* |
925baedd | 4605 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
4606 | * returns 0 if it found something or 1 if there are no greater leaves. |
4607 | * returns < 0 on io errors. | |
97571fd0 | 4608 | */ |
234b63a0 | 4609 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
d97e63b6 CM |
4610 | { |
4611 | int slot; | |
8e73f275 | 4612 | int level; |
5f39d397 | 4613 | struct extent_buffer *c; |
8e73f275 | 4614 | struct extent_buffer *next; |
925baedd CM |
4615 | struct btrfs_key key; |
4616 | u32 nritems; | |
4617 | int ret; | |
8e73f275 | 4618 | int old_spinning = path->leave_spinning; |
bd681513 | 4619 | int next_rw_lock = 0; |
925baedd CM |
4620 | |
4621 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4622 | if (nritems == 0) |
925baedd | 4623 | return 1; |
925baedd | 4624 | |
8e73f275 CM |
4625 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4626 | again: | |
4627 | level = 1; | |
4628 | next = NULL; | |
bd681513 | 4629 | next_rw_lock = 0; |
b3b4aa74 | 4630 | btrfs_release_path(path); |
8e73f275 | 4631 | |
a2135011 | 4632 | path->keep_locks = 1; |
31533fb2 | 4633 | path->leave_spinning = 1; |
8e73f275 | 4634 | |
925baedd CM |
4635 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4636 | path->keep_locks = 0; | |
4637 | ||
4638 | if (ret < 0) | |
4639 | return ret; | |
4640 | ||
a2135011 | 4641 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4642 | /* |
4643 | * by releasing the path above we dropped all our locks. A balance | |
4644 | * could have added more items next to the key that used to be | |
4645 | * at the very end of the block. So, check again here and | |
4646 | * advance the path if there are now more items available. | |
4647 | */ | |
a2135011 | 4648 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4649 | if (ret == 0) |
4650 | path->slots[0]++; | |
8e73f275 | 4651 | ret = 0; |
925baedd CM |
4652 | goto done; |
4653 | } | |
d97e63b6 | 4654 | |
d397712b | 4655 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4656 | if (!path->nodes[level]) { |
4657 | ret = 1; | |
4658 | goto done; | |
4659 | } | |
5f39d397 | 4660 | |
d97e63b6 CM |
4661 | slot = path->slots[level] + 1; |
4662 | c = path->nodes[level]; | |
5f39d397 | 4663 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4664 | level++; |
8e73f275 CM |
4665 | if (level == BTRFS_MAX_LEVEL) { |
4666 | ret = 1; | |
4667 | goto done; | |
4668 | } | |
d97e63b6 CM |
4669 | continue; |
4670 | } | |
5f39d397 | 4671 | |
925baedd | 4672 | if (next) { |
bd681513 | 4673 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 4674 | free_extent_buffer(next); |
925baedd | 4675 | } |
5f39d397 | 4676 | |
8e73f275 | 4677 | next = c; |
bd681513 | 4678 | next_rw_lock = path->locks[level]; |
8e73f275 CM |
4679 | ret = read_block_for_search(NULL, root, path, &next, level, |
4680 | slot, &key); | |
4681 | if (ret == -EAGAIN) | |
4682 | goto again; | |
5f39d397 | 4683 | |
76a05b35 | 4684 | if (ret < 0) { |
b3b4aa74 | 4685 | btrfs_release_path(path); |
76a05b35 CM |
4686 | goto done; |
4687 | } | |
4688 | ||
5cd57b2c | 4689 | if (!path->skip_locking) { |
bd681513 | 4690 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4691 | if (!ret) { |
4692 | btrfs_set_path_blocking(path); | |
bd681513 | 4693 | btrfs_tree_read_lock(next); |
31533fb2 | 4694 | btrfs_clear_path_blocking(path, next, |
bd681513 | 4695 | BTRFS_READ_LOCK); |
8e73f275 | 4696 | } |
31533fb2 | 4697 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4698 | } |
d97e63b6 CM |
4699 | break; |
4700 | } | |
4701 | path->slots[level] = slot; | |
d397712b | 4702 | while (1) { |
d97e63b6 CM |
4703 | level--; |
4704 | c = path->nodes[level]; | |
925baedd | 4705 | if (path->locks[level]) |
bd681513 | 4706 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 4707 | |
5f39d397 | 4708 | free_extent_buffer(c); |
d97e63b6 CM |
4709 | path->nodes[level] = next; |
4710 | path->slots[level] = 0; | |
a74a4b97 | 4711 | if (!path->skip_locking) |
bd681513 | 4712 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
4713 | if (!level) |
4714 | break; | |
b4ce94de | 4715 | |
8e73f275 CM |
4716 | ret = read_block_for_search(NULL, root, path, &next, level, |
4717 | 0, &key); | |
4718 | if (ret == -EAGAIN) | |
4719 | goto again; | |
4720 | ||
76a05b35 | 4721 | if (ret < 0) { |
b3b4aa74 | 4722 | btrfs_release_path(path); |
76a05b35 CM |
4723 | goto done; |
4724 | } | |
4725 | ||
5cd57b2c | 4726 | if (!path->skip_locking) { |
bd681513 | 4727 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4728 | if (!ret) { |
4729 | btrfs_set_path_blocking(path); | |
bd681513 | 4730 | btrfs_tree_read_lock(next); |
31533fb2 | 4731 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
4732 | BTRFS_READ_LOCK); |
4733 | } | |
31533fb2 | 4734 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4735 | } |
d97e63b6 | 4736 | } |
8e73f275 | 4737 | ret = 0; |
925baedd | 4738 | done: |
f7c79f30 | 4739 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
4740 | path->leave_spinning = old_spinning; |
4741 | if (!old_spinning) | |
4742 | btrfs_set_path_blocking(path); | |
4743 | ||
4744 | return ret; | |
d97e63b6 | 4745 | } |
0b86a832 | 4746 | |
3f157a2f CM |
4747 | /* |
4748 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4749 | * searching until it gets past min_objectid or finds an item of 'type' | |
4750 | * | |
4751 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4752 | */ | |
0b86a832 CM |
4753 | int btrfs_previous_item(struct btrfs_root *root, |
4754 | struct btrfs_path *path, u64 min_objectid, | |
4755 | int type) | |
4756 | { | |
4757 | struct btrfs_key found_key; | |
4758 | struct extent_buffer *leaf; | |
e02119d5 | 4759 | u32 nritems; |
0b86a832 CM |
4760 | int ret; |
4761 | ||
d397712b | 4762 | while (1) { |
0b86a832 | 4763 | if (path->slots[0] == 0) { |
b4ce94de | 4764 | btrfs_set_path_blocking(path); |
0b86a832 CM |
4765 | ret = btrfs_prev_leaf(root, path); |
4766 | if (ret != 0) | |
4767 | return ret; | |
4768 | } else { | |
4769 | path->slots[0]--; | |
4770 | } | |
4771 | leaf = path->nodes[0]; | |
e02119d5 CM |
4772 | nritems = btrfs_header_nritems(leaf); |
4773 | if (nritems == 0) | |
4774 | return 1; | |
4775 | if (path->slots[0] == nritems) | |
4776 | path->slots[0]--; | |
4777 | ||
0b86a832 | 4778 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4779 | if (found_key.objectid < min_objectid) |
4780 | break; | |
0a4eefbb YZ |
4781 | if (found_key.type == type) |
4782 | return 0; | |
e02119d5 CM |
4783 | if (found_key.objectid == min_objectid && |
4784 | found_key.type < type) | |
4785 | break; | |
0b86a832 CM |
4786 | } |
4787 | return 1; | |
4788 | } |