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, |
f3ea38da JS |
41 | struct btrfs_path *path, int level, int slot, |
42 | int tree_mod_log); | |
f230475e JS |
43 | static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
44 | struct extent_buffer *eb); | |
45 | struct extent_buffer *read_old_tree_block(struct btrfs_root *root, u64 bytenr, | |
46 | u32 blocksize, u64 parent_transid, | |
47 | u64 time_seq); | |
48 | struct extent_buffer *btrfs_find_old_tree_block(struct btrfs_root *root, | |
49 | u64 bytenr, u32 blocksize, | |
50 | u64 time_seq); | |
d97e63b6 | 51 | |
df24a2b9 | 52 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 53 | { |
df24a2b9 | 54 | struct btrfs_path *path; |
e00f7308 | 55 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 56 | return path; |
2c90e5d6 CM |
57 | } |
58 | ||
b4ce94de CM |
59 | /* |
60 | * set all locked nodes in the path to blocking locks. This should | |
61 | * be done before scheduling | |
62 | */ | |
63 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
64 | { | |
65 | int i; | |
66 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
67 | if (!p->nodes[i] || !p->locks[i]) |
68 | continue; | |
69 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
70 | if (p->locks[i] == BTRFS_READ_LOCK) | |
71 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
72 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
73 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
74 | } |
75 | } | |
76 | ||
77 | /* | |
78 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
79 | * |
80 | * held is used to keep lockdep happy, when lockdep is enabled | |
81 | * we set held to a blocking lock before we go around and | |
82 | * retake all the spinlocks in the path. You can safely use NULL | |
83 | * for held | |
b4ce94de | 84 | */ |
4008c04a | 85 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 86 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
87 | { |
88 | int i; | |
4008c04a CM |
89 | |
90 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
91 | /* lockdep really cares that we take all of these spinlocks | |
92 | * in the right order. If any of the locks in the path are not | |
93 | * currently blocking, it is going to complain. So, make really | |
94 | * really sure by forcing the path to blocking before we clear | |
95 | * the path blocking. | |
96 | */ | |
bd681513 CM |
97 | if (held) { |
98 | btrfs_set_lock_blocking_rw(held, held_rw); | |
99 | if (held_rw == BTRFS_WRITE_LOCK) | |
100 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
101 | else if (held_rw == BTRFS_READ_LOCK) | |
102 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
103 | } | |
4008c04a CM |
104 | btrfs_set_path_blocking(p); |
105 | #endif | |
106 | ||
107 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
108 | if (p->nodes[i] && p->locks[i]) { |
109 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
110 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
111 | p->locks[i] = BTRFS_WRITE_LOCK; | |
112 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
113 | p->locks[i] = BTRFS_READ_LOCK; | |
114 | } | |
b4ce94de | 115 | } |
4008c04a CM |
116 | |
117 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
118 | if (held) | |
bd681513 | 119 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 120 | #endif |
b4ce94de CM |
121 | } |
122 | ||
d352ac68 | 123 | /* this also releases the path */ |
df24a2b9 | 124 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 125 | { |
ff175d57 JJ |
126 | if (!p) |
127 | return; | |
b3b4aa74 | 128 | btrfs_release_path(p); |
df24a2b9 | 129 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
130 | } |
131 | ||
d352ac68 CM |
132 | /* |
133 | * path release drops references on the extent buffers in the path | |
134 | * and it drops any locks held by this path | |
135 | * | |
136 | * It is safe to call this on paths that no locks or extent buffers held. | |
137 | */ | |
b3b4aa74 | 138 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
139 | { |
140 | int i; | |
a2135011 | 141 | |
234b63a0 | 142 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 143 | p->slots[i] = 0; |
eb60ceac | 144 | if (!p->nodes[i]) |
925baedd CM |
145 | continue; |
146 | if (p->locks[i]) { | |
bd681513 | 147 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
148 | p->locks[i] = 0; |
149 | } | |
5f39d397 | 150 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 151 | p->nodes[i] = NULL; |
eb60ceac CM |
152 | } |
153 | } | |
154 | ||
d352ac68 CM |
155 | /* |
156 | * safely gets a reference on the root node of a tree. A lock | |
157 | * is not taken, so a concurrent writer may put a different node | |
158 | * at the root of the tree. See btrfs_lock_root_node for the | |
159 | * looping required. | |
160 | * | |
161 | * The extent buffer returned by this has a reference taken, so | |
162 | * it won't disappear. It may stop being the root of the tree | |
163 | * at any time because there are no locks held. | |
164 | */ | |
925baedd CM |
165 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
166 | { | |
167 | struct extent_buffer *eb; | |
240f62c8 | 168 | |
3083ee2e JB |
169 | while (1) { |
170 | rcu_read_lock(); | |
171 | eb = rcu_dereference(root->node); | |
172 | ||
173 | /* | |
174 | * RCU really hurts here, we could free up the root node because | |
175 | * it was cow'ed but we may not get the new root node yet so do | |
176 | * the inc_not_zero dance and if it doesn't work then | |
177 | * synchronize_rcu and try again. | |
178 | */ | |
179 | if (atomic_inc_not_zero(&eb->refs)) { | |
180 | rcu_read_unlock(); | |
181 | break; | |
182 | } | |
183 | rcu_read_unlock(); | |
184 | synchronize_rcu(); | |
185 | } | |
925baedd CM |
186 | return eb; |
187 | } | |
188 | ||
d352ac68 CM |
189 | /* loop around taking references on and locking the root node of the |
190 | * tree until you end up with a lock on the root. A locked buffer | |
191 | * is returned, with a reference held. | |
192 | */ | |
925baedd CM |
193 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
194 | { | |
195 | struct extent_buffer *eb; | |
196 | ||
d397712b | 197 | while (1) { |
925baedd CM |
198 | eb = btrfs_root_node(root); |
199 | btrfs_tree_lock(eb); | |
240f62c8 | 200 | if (eb == root->node) |
925baedd | 201 | break; |
925baedd CM |
202 | btrfs_tree_unlock(eb); |
203 | free_extent_buffer(eb); | |
204 | } | |
205 | return eb; | |
206 | } | |
207 | ||
bd681513 CM |
208 | /* loop around taking references on and locking the root node of the |
209 | * tree until you end up with a lock on the root. A locked buffer | |
210 | * is returned, with a reference held. | |
211 | */ | |
212 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) | |
213 | { | |
214 | struct extent_buffer *eb; | |
215 | ||
216 | while (1) { | |
217 | eb = btrfs_root_node(root); | |
218 | btrfs_tree_read_lock(eb); | |
219 | if (eb == root->node) | |
220 | break; | |
221 | btrfs_tree_read_unlock(eb); | |
222 | free_extent_buffer(eb); | |
223 | } | |
224 | return eb; | |
225 | } | |
226 | ||
d352ac68 CM |
227 | /* cowonly root (everything not a reference counted cow subvolume), just get |
228 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
229 | * get properly updated on disk. | |
230 | */ | |
0b86a832 CM |
231 | static void add_root_to_dirty_list(struct btrfs_root *root) |
232 | { | |
e5846fc6 | 233 | spin_lock(&root->fs_info->trans_lock); |
0b86a832 CM |
234 | if (root->track_dirty && list_empty(&root->dirty_list)) { |
235 | list_add(&root->dirty_list, | |
236 | &root->fs_info->dirty_cowonly_roots); | |
237 | } | |
e5846fc6 | 238 | spin_unlock(&root->fs_info->trans_lock); |
0b86a832 CM |
239 | } |
240 | ||
d352ac68 CM |
241 | /* |
242 | * used by snapshot creation to make a copy of a root for a tree with | |
243 | * a given objectid. The buffer with the new root node is returned in | |
244 | * cow_ret, and this func returns zero on success or a negative error code. | |
245 | */ | |
be20aa9d CM |
246 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
247 | struct btrfs_root *root, | |
248 | struct extent_buffer *buf, | |
249 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
250 | { | |
251 | struct extent_buffer *cow; | |
be20aa9d CM |
252 | int ret = 0; |
253 | int level; | |
5d4f98a2 | 254 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
255 | |
256 | WARN_ON(root->ref_cows && trans->transid != | |
257 | root->fs_info->running_transaction->transid); | |
258 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
259 | ||
260 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
261 | if (level == 0) |
262 | btrfs_item_key(buf, &disk_key, 0); | |
263 | else | |
264 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 265 | |
5d4f98a2 YZ |
266 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
267 | new_root_objectid, &disk_key, level, | |
5581a51a | 268 | buf->start, 0); |
5d4f98a2 | 269 | if (IS_ERR(cow)) |
be20aa9d CM |
270 | return PTR_ERR(cow); |
271 | ||
272 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
273 | btrfs_set_header_bytenr(cow, cow->start); | |
274 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
275 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
276 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
277 | BTRFS_HEADER_FLAG_RELOC); | |
278 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
279 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
280 | else | |
281 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 282 | |
2b82032c YZ |
283 | write_extent_buffer(cow, root->fs_info->fsid, |
284 | (unsigned long)btrfs_header_fsid(cow), | |
285 | BTRFS_FSID_SIZE); | |
286 | ||
be20aa9d | 287 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 288 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 289 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 290 | else |
66d7e7f0 | 291 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 292 | |
be20aa9d CM |
293 | if (ret) |
294 | return ret; | |
295 | ||
296 | btrfs_mark_buffer_dirty(cow); | |
297 | *cow_ret = cow; | |
298 | return 0; | |
299 | } | |
300 | ||
bd989ba3 JS |
301 | enum mod_log_op { |
302 | MOD_LOG_KEY_REPLACE, | |
303 | MOD_LOG_KEY_ADD, | |
304 | MOD_LOG_KEY_REMOVE, | |
305 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
306 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
307 | MOD_LOG_MOVE_KEYS, | |
308 | MOD_LOG_ROOT_REPLACE, | |
309 | }; | |
310 | ||
311 | struct tree_mod_move { | |
312 | int dst_slot; | |
313 | int nr_items; | |
314 | }; | |
315 | ||
316 | struct tree_mod_root { | |
317 | u64 logical; | |
318 | u8 level; | |
319 | }; | |
320 | ||
321 | struct tree_mod_elem { | |
322 | struct rb_node node; | |
323 | u64 index; /* shifted logical */ | |
324 | struct seq_list elem; | |
325 | enum mod_log_op op; | |
326 | ||
327 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
328 | int slot; | |
329 | ||
330 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
331 | u64 generation; | |
332 | ||
333 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
334 | struct btrfs_disk_key key; | |
335 | u64 blockptr; | |
336 | ||
337 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
338 | struct tree_mod_move move; | |
339 | ||
340 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
341 | struct tree_mod_root old_root; | |
342 | }; | |
343 | ||
344 | static inline void | |
345 | __get_tree_mod_seq(struct btrfs_fs_info *fs_info, struct seq_list *elem) | |
346 | { | |
347 | elem->seq = atomic_inc_return(&fs_info->tree_mod_seq); | |
348 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); | |
349 | } | |
350 | ||
351 | void btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
352 | struct seq_list *elem) | |
353 | { | |
354 | elem->flags = 1; | |
355 | spin_lock(&fs_info->tree_mod_seq_lock); | |
356 | __get_tree_mod_seq(fs_info, elem); | |
357 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
358 | } | |
359 | ||
360 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
361 | struct seq_list *elem) | |
362 | { | |
363 | struct rb_root *tm_root; | |
364 | struct rb_node *node; | |
365 | struct rb_node *next; | |
366 | struct seq_list *cur_elem; | |
367 | struct tree_mod_elem *tm; | |
368 | u64 min_seq = (u64)-1; | |
369 | u64 seq_putting = elem->seq; | |
370 | ||
371 | if (!seq_putting) | |
372 | return; | |
373 | ||
374 | BUG_ON(!(elem->flags & 1)); | |
375 | spin_lock(&fs_info->tree_mod_seq_lock); | |
376 | list_del(&elem->list); | |
377 | ||
378 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
379 | if ((cur_elem->flags & 1) && cur_elem->seq < min_seq) { | |
380 | if (seq_putting > cur_elem->seq) { | |
381 | /* | |
382 | * blocker with lower sequence number exists, we | |
383 | * cannot remove anything from the log | |
384 | */ | |
385 | goto out; | |
386 | } | |
387 | min_seq = cur_elem->seq; | |
388 | } | |
389 | } | |
390 | ||
391 | /* | |
392 | * anything that's lower than the lowest existing (read: blocked) | |
393 | * sequence number can be removed from the tree. | |
394 | */ | |
395 | write_lock(&fs_info->tree_mod_log_lock); | |
396 | tm_root = &fs_info->tree_mod_log; | |
397 | for (node = rb_first(tm_root); node; node = next) { | |
398 | next = rb_next(node); | |
399 | tm = container_of(node, struct tree_mod_elem, node); | |
400 | if (tm->elem.seq > min_seq) | |
401 | continue; | |
402 | rb_erase(node, tm_root); | |
403 | list_del(&tm->elem.list); | |
404 | kfree(tm); | |
405 | } | |
406 | write_unlock(&fs_info->tree_mod_log_lock); | |
407 | out: | |
408 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
409 | } | |
410 | ||
411 | /* | |
412 | * key order of the log: | |
413 | * index -> sequence | |
414 | * | |
415 | * the index is the shifted logical of the *new* root node for root replace | |
416 | * operations, or the shifted logical of the affected block for all other | |
417 | * operations. | |
418 | */ | |
419 | static noinline int | |
420 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
421 | { | |
422 | struct rb_root *tm_root; | |
423 | struct rb_node **new; | |
424 | struct rb_node *parent = NULL; | |
425 | struct tree_mod_elem *cur; | |
426 | int ret = 0; | |
427 | ||
428 | BUG_ON(!tm || !tm->elem.seq); | |
429 | ||
430 | write_lock(&fs_info->tree_mod_log_lock); | |
431 | tm_root = &fs_info->tree_mod_log; | |
432 | new = &tm_root->rb_node; | |
433 | while (*new) { | |
434 | cur = container_of(*new, struct tree_mod_elem, node); | |
435 | parent = *new; | |
436 | if (cur->index < tm->index) | |
437 | new = &((*new)->rb_left); | |
438 | else if (cur->index > tm->index) | |
439 | new = &((*new)->rb_right); | |
440 | else if (cur->elem.seq < tm->elem.seq) | |
441 | new = &((*new)->rb_left); | |
442 | else if (cur->elem.seq > tm->elem.seq) | |
443 | new = &((*new)->rb_right); | |
444 | else { | |
445 | kfree(tm); | |
446 | ret = -EEXIST; | |
447 | goto unlock; | |
448 | } | |
449 | } | |
450 | ||
451 | rb_link_node(&tm->node, parent, new); | |
452 | rb_insert_color(&tm->node, tm_root); | |
453 | unlock: | |
454 | write_unlock(&fs_info->tree_mod_log_lock); | |
455 | return ret; | |
456 | } | |
457 | ||
458 | int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags, | |
459 | struct tree_mod_elem **tm_ret) | |
460 | { | |
461 | struct tree_mod_elem *tm; | |
462 | u64 seq = 0; | |
463 | ||
464 | smp_mb(); | |
465 | if (list_empty(&fs_info->tree_mod_seq_list)) | |
466 | return 0; | |
467 | ||
468 | tm = *tm_ret = kzalloc(sizeof(*tm), flags); | |
469 | if (!tm) | |
470 | return -ENOMEM; | |
471 | ||
472 | __get_tree_mod_seq(fs_info, &tm->elem); | |
473 | seq = tm->elem.seq; | |
474 | tm->elem.flags = 0; | |
475 | ||
476 | return seq; | |
477 | } | |
478 | ||
479 | static noinline int | |
480 | tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info, | |
481 | struct extent_buffer *eb, int slot, | |
482 | enum mod_log_op op, gfp_t flags) | |
483 | { | |
484 | struct tree_mod_elem *tm; | |
485 | int ret; | |
486 | ||
487 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
488 | if (ret <= 0) | |
489 | return ret; | |
490 | ||
491 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
492 | if (op != MOD_LOG_KEY_ADD) { | |
493 | btrfs_node_key(eb, &tm->key, slot); | |
494 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
495 | } | |
496 | tm->op = op; | |
497 | tm->slot = slot; | |
498 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
499 | ||
500 | return __tree_mod_log_insert(fs_info, tm); | |
501 | } | |
502 | ||
503 | static noinline int | |
504 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, | |
505 | int slot, enum mod_log_op op) | |
506 | { | |
507 | return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS); | |
508 | } | |
509 | ||
510 | static noinline int | |
511 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
512 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
513 | int nr_items, gfp_t flags) | |
514 | { | |
515 | struct tree_mod_elem *tm; | |
516 | int ret; | |
517 | int i; | |
518 | ||
519 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
520 | if (ret <= 0) | |
521 | return ret; | |
522 | ||
523 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
524 | ret = tree_mod_log_insert_key(fs_info, eb, i + dst_slot, | |
525 | MOD_LOG_KEY_REMOVE_WHILE_MOVING); | |
526 | BUG_ON(ret < 0); | |
527 | } | |
528 | ||
529 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
530 | tm->slot = src_slot; | |
531 | tm->move.dst_slot = dst_slot; | |
532 | tm->move.nr_items = nr_items; | |
533 | tm->op = MOD_LOG_MOVE_KEYS; | |
534 | ||
535 | return __tree_mod_log_insert(fs_info, tm); | |
536 | } | |
537 | ||
538 | static noinline int | |
539 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
540 | struct extent_buffer *old_root, | |
541 | struct extent_buffer *new_root, gfp_t flags) | |
542 | { | |
543 | struct tree_mod_elem *tm; | |
544 | int ret; | |
545 | ||
546 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
547 | if (ret <= 0) | |
548 | return ret; | |
549 | ||
550 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
551 | tm->old_root.logical = old_root->start; | |
552 | tm->old_root.level = btrfs_header_level(old_root); | |
553 | tm->generation = btrfs_header_generation(old_root); | |
554 | tm->op = MOD_LOG_ROOT_REPLACE; | |
555 | ||
556 | return __tree_mod_log_insert(fs_info, tm); | |
557 | } | |
558 | ||
559 | static struct tree_mod_elem * | |
560 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
561 | int smallest) | |
562 | { | |
563 | struct rb_root *tm_root; | |
564 | struct rb_node *node; | |
565 | struct tree_mod_elem *cur = NULL; | |
566 | struct tree_mod_elem *found = NULL; | |
567 | u64 index = start >> PAGE_CACHE_SHIFT; | |
568 | ||
569 | read_lock(&fs_info->tree_mod_log_lock); | |
570 | tm_root = &fs_info->tree_mod_log; | |
571 | node = tm_root->rb_node; | |
572 | while (node) { | |
573 | cur = container_of(node, struct tree_mod_elem, node); | |
574 | if (cur->index < index) { | |
575 | node = node->rb_left; | |
576 | } else if (cur->index > index) { | |
577 | node = node->rb_right; | |
578 | } else if (cur->elem.seq < min_seq) { | |
579 | node = node->rb_left; | |
580 | } else if (!smallest) { | |
581 | /* we want the node with the highest seq */ | |
582 | if (found) | |
583 | BUG_ON(found->elem.seq > cur->elem.seq); | |
584 | found = cur; | |
585 | node = node->rb_left; | |
586 | } else if (cur->elem.seq > min_seq) { | |
587 | /* we want the node with the smallest seq */ | |
588 | if (found) | |
589 | BUG_ON(found->elem.seq < cur->elem.seq); | |
590 | found = cur; | |
591 | node = node->rb_right; | |
592 | } else { | |
593 | found = cur; | |
594 | break; | |
595 | } | |
596 | } | |
597 | read_unlock(&fs_info->tree_mod_log_lock); | |
598 | ||
599 | return found; | |
600 | } | |
601 | ||
602 | /* | |
603 | * this returns the element from the log with the smallest time sequence | |
604 | * value that's in the log (the oldest log item). any element with a time | |
605 | * sequence lower than min_seq will be ignored. | |
606 | */ | |
607 | static struct tree_mod_elem * | |
608 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
609 | u64 min_seq) | |
610 | { | |
611 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
612 | } | |
613 | ||
614 | /* | |
615 | * this returns the element from the log with the largest time sequence | |
616 | * value that's in the log (the most recent log item). any element with | |
617 | * a time sequence lower than min_seq will be ignored. | |
618 | */ | |
619 | static struct tree_mod_elem * | |
620 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
621 | { | |
622 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
623 | } | |
624 | ||
625 | static inline void | |
626 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
627 | struct extent_buffer *src, unsigned long dst_offset, | |
628 | unsigned long src_offset, int nr_items) | |
629 | { | |
630 | int ret; | |
631 | int i; | |
632 | ||
633 | smp_mb(); | |
634 | if (list_empty(&fs_info->tree_mod_seq_list)) | |
635 | return; | |
636 | ||
637 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) | |
638 | return; | |
639 | ||
640 | /* speed this up by single seq for all operations? */ | |
641 | for (i = 0; i < nr_items; i++) { | |
642 | ret = tree_mod_log_insert_key(fs_info, src, i + src_offset, | |
643 | MOD_LOG_KEY_REMOVE); | |
644 | BUG_ON(ret < 0); | |
645 | ret = tree_mod_log_insert_key(fs_info, dst, i + dst_offset, | |
646 | MOD_LOG_KEY_ADD); | |
647 | BUG_ON(ret < 0); | |
648 | } | |
649 | } | |
650 | ||
651 | static inline void | |
652 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
653 | int dst_offset, int src_offset, int nr_items) | |
654 | { | |
655 | int ret; | |
656 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
657 | nr_items, GFP_NOFS); | |
658 | BUG_ON(ret < 0); | |
659 | } | |
660 | ||
661 | static inline void | |
662 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, | |
663 | struct extent_buffer *eb, | |
664 | struct btrfs_disk_key *disk_key, int slot, int atomic) | |
665 | { | |
666 | int ret; | |
667 | ||
668 | ret = tree_mod_log_insert_key_mask(fs_info, eb, slot, | |
669 | MOD_LOG_KEY_REPLACE, | |
670 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
671 | BUG_ON(ret < 0); | |
672 | } | |
673 | ||
674 | static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
675 | struct extent_buffer *eb) | |
676 | { | |
677 | int i; | |
678 | int ret; | |
679 | u32 nritems; | |
680 | ||
681 | smp_mb(); | |
682 | if (list_empty(&fs_info->tree_mod_seq_list)) | |
683 | return; | |
684 | ||
685 | if (btrfs_header_level(eb) == 0) | |
686 | return; | |
687 | ||
688 | nritems = btrfs_header_nritems(eb); | |
689 | for (i = nritems - 1; i >= 0; i--) { | |
690 | ret = tree_mod_log_insert_key(fs_info, eb, i, | |
691 | MOD_LOG_KEY_REMOVE_WHILE_FREEING); | |
692 | BUG_ON(ret < 0); | |
693 | } | |
694 | } | |
695 | ||
696 | static inline void | |
697 | tree_mod_log_set_root_pointer(struct btrfs_root *root, | |
698 | struct extent_buffer *new_root_node) | |
699 | { | |
700 | int ret; | |
701 | tree_mod_log_free_eb(root->fs_info, root->node); | |
702 | ret = tree_mod_log_insert_root(root->fs_info, root->node, | |
703 | new_root_node, GFP_NOFS); | |
704 | BUG_ON(ret < 0); | |
705 | } | |
706 | ||
5d4f98a2 YZ |
707 | /* |
708 | * check if the tree block can be shared by multiple trees | |
709 | */ | |
710 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
711 | struct extent_buffer *buf) | |
712 | { | |
713 | /* | |
714 | * Tree blocks not in refernece counted trees and tree roots | |
715 | * are never shared. If a block was allocated after the last | |
716 | * snapshot and the block was not allocated by tree relocation, | |
717 | * we know the block is not shared. | |
718 | */ | |
719 | if (root->ref_cows && | |
720 | buf != root->node && buf != root->commit_root && | |
721 | (btrfs_header_generation(buf) <= | |
722 | btrfs_root_last_snapshot(&root->root_item) || | |
723 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
724 | return 1; | |
725 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
726 | if (root->ref_cows && | |
727 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
728 | return 1; | |
729 | #endif | |
730 | return 0; | |
731 | } | |
732 | ||
733 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
734 | struct btrfs_root *root, | |
735 | struct extent_buffer *buf, | |
f0486c68 YZ |
736 | struct extent_buffer *cow, |
737 | int *last_ref) | |
5d4f98a2 YZ |
738 | { |
739 | u64 refs; | |
740 | u64 owner; | |
741 | u64 flags; | |
742 | u64 new_flags = 0; | |
743 | int ret; | |
744 | ||
745 | /* | |
746 | * Backrefs update rules: | |
747 | * | |
748 | * Always use full backrefs for extent pointers in tree block | |
749 | * allocated by tree relocation. | |
750 | * | |
751 | * If a shared tree block is no longer referenced by its owner | |
752 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
753 | * use full backrefs for extent pointers in tree block. | |
754 | * | |
755 | * If a tree block is been relocating | |
756 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
757 | * use full backrefs for extent pointers in tree block. | |
758 | * The reason for this is some operations (such as drop tree) | |
759 | * are only allowed for blocks use full backrefs. | |
760 | */ | |
761 | ||
762 | if (btrfs_block_can_be_shared(root, buf)) { | |
763 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
764 | buf->len, &refs, &flags); | |
be1a5564 MF |
765 | if (ret) |
766 | return ret; | |
e5df9573 MF |
767 | if (refs == 0) { |
768 | ret = -EROFS; | |
769 | btrfs_std_error(root->fs_info, ret); | |
770 | return ret; | |
771 | } | |
5d4f98a2 YZ |
772 | } else { |
773 | refs = 1; | |
774 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
775 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
776 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
777 | else | |
778 | flags = 0; | |
779 | } | |
780 | ||
781 | owner = btrfs_header_owner(buf); | |
782 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
783 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
784 | ||
785 | if (refs > 1) { | |
786 | if ((owner == root->root_key.objectid || | |
787 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
788 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 789 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 790 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
791 | |
792 | if (root->root_key.objectid == | |
793 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 794 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 795 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 796 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 797 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
798 | } |
799 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
800 | } else { | |
801 | ||
802 | if (root->root_key.objectid == | |
803 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 804 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 805 | else |
66d7e7f0 | 806 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 807 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
808 | } |
809 | if (new_flags != 0) { | |
810 | ret = btrfs_set_disk_extent_flags(trans, root, | |
811 | buf->start, | |
812 | buf->len, | |
813 | new_flags, 0); | |
be1a5564 MF |
814 | if (ret) |
815 | return ret; | |
5d4f98a2 YZ |
816 | } |
817 | } else { | |
818 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
819 | if (root->root_key.objectid == | |
820 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 821 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 822 | else |
66d7e7f0 | 823 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 824 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 825 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 826 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 | 827 | } |
f230475e JS |
828 | /* |
829 | * don't log freeing in case we're freeing the root node, this | |
830 | * is done by tree_mod_log_set_root_pointer later | |
831 | */ | |
832 | if (buf != root->node && btrfs_header_level(buf) != 0) | |
833 | tree_mod_log_free_eb(root->fs_info, buf); | |
5d4f98a2 | 834 | clean_tree_block(trans, root, buf); |
f0486c68 | 835 | *last_ref = 1; |
5d4f98a2 YZ |
836 | } |
837 | return 0; | |
838 | } | |
839 | ||
d352ac68 | 840 | /* |
d397712b CM |
841 | * does the dirty work in cow of a single block. The parent block (if |
842 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
843 | * dirty and returned locked. If you modify the block it needs to be marked | |
844 | * dirty again. | |
d352ac68 CM |
845 | * |
846 | * search_start -- an allocation hint for the new block | |
847 | * | |
d397712b CM |
848 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
849 | * bytes the allocator should try to find free next to the block it returns. | |
850 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 851 | */ |
d397712b | 852 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
853 | struct btrfs_root *root, |
854 | struct extent_buffer *buf, | |
855 | struct extent_buffer *parent, int parent_slot, | |
856 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 857 | u64 search_start, u64 empty_size) |
02217ed2 | 858 | { |
5d4f98a2 | 859 | struct btrfs_disk_key disk_key; |
5f39d397 | 860 | struct extent_buffer *cow; |
be1a5564 | 861 | int level, ret; |
f0486c68 | 862 | int last_ref = 0; |
925baedd | 863 | int unlock_orig = 0; |
5d4f98a2 | 864 | u64 parent_start; |
7bb86316 | 865 | |
925baedd CM |
866 | if (*cow_ret == buf) |
867 | unlock_orig = 1; | |
868 | ||
b9447ef8 | 869 | btrfs_assert_tree_locked(buf); |
925baedd | 870 | |
7bb86316 CM |
871 | WARN_ON(root->ref_cows && trans->transid != |
872 | root->fs_info->running_transaction->transid); | |
6702ed49 | 873 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 874 | |
7bb86316 | 875 | level = btrfs_header_level(buf); |
31840ae1 | 876 | |
5d4f98a2 YZ |
877 | if (level == 0) |
878 | btrfs_item_key(buf, &disk_key, 0); | |
879 | else | |
880 | btrfs_node_key(buf, &disk_key, 0); | |
881 | ||
882 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
883 | if (parent) | |
884 | parent_start = parent->start; | |
885 | else | |
886 | parent_start = 0; | |
887 | } else | |
888 | parent_start = 0; | |
889 | ||
890 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
891 | root->root_key.objectid, &disk_key, | |
5581a51a | 892 | level, search_start, empty_size); |
54aa1f4d CM |
893 | if (IS_ERR(cow)) |
894 | return PTR_ERR(cow); | |
6702ed49 | 895 | |
b4ce94de CM |
896 | /* cow is set to blocking by btrfs_init_new_buffer */ |
897 | ||
5f39d397 | 898 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 899 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 900 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
901 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
902 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
903 | BTRFS_HEADER_FLAG_RELOC); | |
904 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
905 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
906 | else | |
907 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 908 | |
2b82032c YZ |
909 | write_extent_buffer(cow, root->fs_info->fsid, |
910 | (unsigned long)btrfs_header_fsid(cow), | |
911 | BTRFS_FSID_SIZE); | |
912 | ||
be1a5564 | 913 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 914 | if (ret) { |
79787eaa | 915 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
916 | return ret; |
917 | } | |
1a40e23b | 918 | |
3fd0a558 YZ |
919 | if (root->ref_cows) |
920 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
921 | ||
02217ed2 | 922 | if (buf == root->node) { |
925baedd | 923 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
924 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
925 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
926 | parent_start = buf->start; | |
927 | else | |
928 | parent_start = 0; | |
925baedd | 929 | |
5f39d397 | 930 | extent_buffer_get(cow); |
f230475e | 931 | tree_mod_log_set_root_pointer(root, cow); |
240f62c8 | 932 | rcu_assign_pointer(root->node, cow); |
925baedd | 933 | |
f0486c68 | 934 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 935 | last_ref); |
5f39d397 | 936 | free_extent_buffer(buf); |
0b86a832 | 937 | add_root_to_dirty_list(root); |
02217ed2 | 938 | } else { |
5d4f98a2 YZ |
939 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
940 | parent_start = parent->start; | |
941 | else | |
942 | parent_start = 0; | |
943 | ||
944 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e JS |
945 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
946 | MOD_LOG_KEY_REPLACE); | |
5f39d397 | 947 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 948 | cow->start); |
74493f7a CM |
949 | btrfs_set_node_ptr_generation(parent, parent_slot, |
950 | trans->transid); | |
d6025579 | 951 | btrfs_mark_buffer_dirty(parent); |
f0486c68 | 952 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 953 | last_ref); |
02217ed2 | 954 | } |
925baedd CM |
955 | if (unlock_orig) |
956 | btrfs_tree_unlock(buf); | |
3083ee2e | 957 | free_extent_buffer_stale(buf); |
ccd467d6 | 958 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 959 | *cow_ret = cow; |
02217ed2 CM |
960 | return 0; |
961 | } | |
962 | ||
5d4f98a2 YZ |
963 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
964 | struct btrfs_root *root, | |
965 | struct extent_buffer *buf) | |
966 | { | |
f1ebcc74 LB |
967 | /* ensure we can see the force_cow */ |
968 | smp_rmb(); | |
969 | ||
970 | /* | |
971 | * We do not need to cow a block if | |
972 | * 1) this block is not created or changed in this transaction; | |
973 | * 2) this block does not belong to TREE_RELOC tree; | |
974 | * 3) the root is not forced COW. | |
975 | * | |
976 | * What is forced COW: | |
977 | * when we create snapshot during commiting the transaction, | |
978 | * after we've finished coping src root, we must COW the shared | |
979 | * block to ensure the metadata consistency. | |
980 | */ | |
5d4f98a2 YZ |
981 | if (btrfs_header_generation(buf) == trans->transid && |
982 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
983 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
984 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
985 | !root->force_cow) | |
5d4f98a2 YZ |
986 | return 0; |
987 | return 1; | |
988 | } | |
989 | ||
d352ac68 CM |
990 | /* |
991 | * cows a single block, see __btrfs_cow_block for the real work. | |
992 | * This version of it has extra checks so that a block isn't cow'd more than | |
993 | * once per transaction, as long as it hasn't been written yet | |
994 | */ | |
d397712b | 995 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
996 | struct btrfs_root *root, struct extent_buffer *buf, |
997 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 998 | struct extent_buffer **cow_ret) |
6702ed49 CM |
999 | { |
1000 | u64 search_start; | |
f510cfec | 1001 | int ret; |
dc17ff8f | 1002 | |
6702ed49 | 1003 | if (trans->transaction != root->fs_info->running_transaction) { |
d397712b CM |
1004 | printk(KERN_CRIT "trans %llu running %llu\n", |
1005 | (unsigned long long)trans->transid, | |
1006 | (unsigned long long) | |
6702ed49 CM |
1007 | root->fs_info->running_transaction->transid); |
1008 | WARN_ON(1); | |
1009 | } | |
1010 | if (trans->transid != root->fs_info->generation) { | |
d397712b CM |
1011 | printk(KERN_CRIT "trans %llu running %llu\n", |
1012 | (unsigned long long)trans->transid, | |
1013 | (unsigned long long)root->fs_info->generation); | |
6702ed49 CM |
1014 | WARN_ON(1); |
1015 | } | |
dc17ff8f | 1016 | |
5d4f98a2 | 1017 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1018 | *cow_ret = buf; |
1019 | return 0; | |
1020 | } | |
c487685d | 1021 | |
0b86a832 | 1022 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1023 | |
1024 | if (parent) | |
1025 | btrfs_set_lock_blocking(parent); | |
1026 | btrfs_set_lock_blocking(buf); | |
1027 | ||
f510cfec | 1028 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1029 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1030 | |
1031 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1032 | ||
f510cfec | 1033 | return ret; |
6702ed49 CM |
1034 | } |
1035 | ||
d352ac68 CM |
1036 | /* |
1037 | * helper function for defrag to decide if two blocks pointed to by a | |
1038 | * node are actually close by | |
1039 | */ | |
6b80053d | 1040 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1041 | { |
6b80053d | 1042 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1043 | return 1; |
6b80053d | 1044 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1045 | return 1; |
1046 | return 0; | |
1047 | } | |
1048 | ||
081e9573 CM |
1049 | /* |
1050 | * compare two keys in a memcmp fashion | |
1051 | */ | |
1052 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1053 | { | |
1054 | struct btrfs_key k1; | |
1055 | ||
1056 | btrfs_disk_key_to_cpu(&k1, disk); | |
1057 | ||
20736aba | 1058 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1059 | } |
1060 | ||
f3465ca4 JB |
1061 | /* |
1062 | * same as comp_keys only with two btrfs_key's | |
1063 | */ | |
5d4f98a2 | 1064 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1065 | { |
1066 | if (k1->objectid > k2->objectid) | |
1067 | return 1; | |
1068 | if (k1->objectid < k2->objectid) | |
1069 | return -1; | |
1070 | if (k1->type > k2->type) | |
1071 | return 1; | |
1072 | if (k1->type < k2->type) | |
1073 | return -1; | |
1074 | if (k1->offset > k2->offset) | |
1075 | return 1; | |
1076 | if (k1->offset < k2->offset) | |
1077 | return -1; | |
1078 | return 0; | |
1079 | } | |
081e9573 | 1080 | |
d352ac68 CM |
1081 | /* |
1082 | * this is used by the defrag code to go through all the | |
1083 | * leaves pointed to by a node and reallocate them so that | |
1084 | * disk order is close to key order | |
1085 | */ | |
6702ed49 | 1086 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1087 | struct btrfs_root *root, struct extent_buffer *parent, |
a6b6e75e CM |
1088 | int start_slot, int cache_only, u64 *last_ret, |
1089 | struct btrfs_key *progress) | |
6702ed49 | 1090 | { |
6b80053d | 1091 | struct extent_buffer *cur; |
6702ed49 | 1092 | u64 blocknr; |
ca7a79ad | 1093 | u64 gen; |
e9d0b13b CM |
1094 | u64 search_start = *last_ret; |
1095 | u64 last_block = 0; | |
6702ed49 CM |
1096 | u64 other; |
1097 | u32 parent_nritems; | |
6702ed49 CM |
1098 | int end_slot; |
1099 | int i; | |
1100 | int err = 0; | |
f2183bde | 1101 | int parent_level; |
6b80053d CM |
1102 | int uptodate; |
1103 | u32 blocksize; | |
081e9573 CM |
1104 | int progress_passed = 0; |
1105 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1106 | |
5708b959 CM |
1107 | parent_level = btrfs_header_level(parent); |
1108 | if (cache_only && parent_level != 1) | |
1109 | return 0; | |
1110 | ||
d397712b | 1111 | if (trans->transaction != root->fs_info->running_transaction) |
6702ed49 | 1112 | WARN_ON(1); |
d397712b | 1113 | if (trans->transid != root->fs_info->generation) |
6702ed49 | 1114 | WARN_ON(1); |
86479a04 | 1115 | |
6b80053d | 1116 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1117 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1118 | end_slot = parent_nritems; |
1119 | ||
1120 | if (parent_nritems == 1) | |
1121 | return 0; | |
1122 | ||
b4ce94de CM |
1123 | btrfs_set_lock_blocking(parent); |
1124 | ||
6702ed49 CM |
1125 | for (i = start_slot; i < end_slot; i++) { |
1126 | int close = 1; | |
a6b6e75e | 1127 | |
081e9573 CM |
1128 | btrfs_node_key(parent, &disk_key, i); |
1129 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1130 | continue; | |
1131 | ||
1132 | progress_passed = 1; | |
6b80053d | 1133 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1134 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1135 | if (last_block == 0) |
1136 | last_block = blocknr; | |
5708b959 | 1137 | |
6702ed49 | 1138 | if (i > 0) { |
6b80053d CM |
1139 | other = btrfs_node_blockptr(parent, i - 1); |
1140 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1141 | } |
0ef3e66b | 1142 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1143 | other = btrfs_node_blockptr(parent, i + 1); |
1144 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1145 | } |
e9d0b13b CM |
1146 | if (close) { |
1147 | last_block = blocknr; | |
6702ed49 | 1148 | continue; |
e9d0b13b | 1149 | } |
6702ed49 | 1150 | |
6b80053d CM |
1151 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1152 | if (cur) | |
b9fab919 | 1153 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1154 | else |
1155 | uptodate = 0; | |
5708b959 | 1156 | if (!cur || !uptodate) { |
6702ed49 | 1157 | if (cache_only) { |
6b80053d | 1158 | free_extent_buffer(cur); |
6702ed49 CM |
1159 | continue; |
1160 | } | |
6b80053d CM |
1161 | if (!cur) { |
1162 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1163 | blocksize, gen); |
97d9a8a4 TI |
1164 | if (!cur) |
1165 | return -EIO; | |
6b80053d | 1166 | } else if (!uptodate) { |
ca7a79ad | 1167 | btrfs_read_buffer(cur, gen); |
f2183bde | 1168 | } |
6702ed49 | 1169 | } |
e9d0b13b | 1170 | if (search_start == 0) |
6b80053d | 1171 | search_start = last_block; |
e9d0b13b | 1172 | |
e7a84565 | 1173 | btrfs_tree_lock(cur); |
b4ce94de | 1174 | btrfs_set_lock_blocking(cur); |
6b80053d | 1175 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1176 | &cur, search_start, |
6b80053d | 1177 | min(16 * blocksize, |
9fa8cfe7 | 1178 | (end_slot - i) * blocksize)); |
252c38f0 | 1179 | if (err) { |
e7a84565 | 1180 | btrfs_tree_unlock(cur); |
6b80053d | 1181 | free_extent_buffer(cur); |
6702ed49 | 1182 | break; |
252c38f0 | 1183 | } |
e7a84565 CM |
1184 | search_start = cur->start; |
1185 | last_block = cur->start; | |
f2183bde | 1186 | *last_ret = search_start; |
e7a84565 CM |
1187 | btrfs_tree_unlock(cur); |
1188 | free_extent_buffer(cur); | |
6702ed49 CM |
1189 | } |
1190 | return err; | |
1191 | } | |
1192 | ||
74123bd7 CM |
1193 | /* |
1194 | * The leaf data grows from end-to-front in the node. | |
1195 | * this returns the address of the start of the last item, | |
1196 | * which is the stop of the leaf data stack | |
1197 | */ | |
123abc88 | 1198 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1199 | struct extent_buffer *leaf) |
be0e5c09 | 1200 | { |
5f39d397 | 1201 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1202 | if (nr == 0) |
123abc88 | 1203 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1204 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1205 | } |
1206 | ||
aa5d6bed | 1207 | |
74123bd7 | 1208 | /* |
5f39d397 CM |
1209 | * search for key in the extent_buffer. The items start at offset p, |
1210 | * and they are item_size apart. There are 'max' items in p. | |
1211 | * | |
74123bd7 CM |
1212 | * the slot in the array is returned via slot, and it points to |
1213 | * the place where you would insert key if it is not found in | |
1214 | * the array. | |
1215 | * | |
1216 | * slot may point to max if the key is bigger than all of the keys | |
1217 | */ | |
e02119d5 CM |
1218 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1219 | unsigned long p, | |
1220 | int item_size, struct btrfs_key *key, | |
1221 | int max, int *slot) | |
be0e5c09 CM |
1222 | { |
1223 | int low = 0; | |
1224 | int high = max; | |
1225 | int mid; | |
1226 | int ret; | |
479965d6 | 1227 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1228 | struct btrfs_disk_key unaligned; |
1229 | unsigned long offset; | |
5f39d397 CM |
1230 | char *kaddr = NULL; |
1231 | unsigned long map_start = 0; | |
1232 | unsigned long map_len = 0; | |
479965d6 | 1233 | int err; |
be0e5c09 | 1234 | |
d397712b | 1235 | while (low < high) { |
be0e5c09 | 1236 | mid = (low + high) / 2; |
5f39d397 CM |
1237 | offset = p + mid * item_size; |
1238 | ||
a6591715 | 1239 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1240 | (offset + sizeof(struct btrfs_disk_key)) > |
1241 | map_start + map_len) { | |
934d375b CM |
1242 | |
1243 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1244 | sizeof(struct btrfs_disk_key), |
a6591715 | 1245 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1246 | |
1247 | if (!err) { | |
1248 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1249 | map_start); | |
1250 | } else { | |
1251 | read_extent_buffer(eb, &unaligned, | |
1252 | offset, sizeof(unaligned)); | |
1253 | tmp = &unaligned; | |
1254 | } | |
5f39d397 | 1255 | |
5f39d397 CM |
1256 | } else { |
1257 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1258 | map_start); | |
1259 | } | |
be0e5c09 CM |
1260 | ret = comp_keys(tmp, key); |
1261 | ||
1262 | if (ret < 0) | |
1263 | low = mid + 1; | |
1264 | else if (ret > 0) | |
1265 | high = mid; | |
1266 | else { | |
1267 | *slot = mid; | |
1268 | return 0; | |
1269 | } | |
1270 | } | |
1271 | *slot = low; | |
1272 | return 1; | |
1273 | } | |
1274 | ||
97571fd0 CM |
1275 | /* |
1276 | * simple bin_search frontend that does the right thing for | |
1277 | * leaves vs nodes | |
1278 | */ | |
5f39d397 CM |
1279 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1280 | int level, int *slot) | |
be0e5c09 | 1281 | { |
5f39d397 CM |
1282 | if (level == 0) { |
1283 | return generic_bin_search(eb, | |
1284 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1285 | sizeof(struct btrfs_item), |
5f39d397 | 1286 | key, btrfs_header_nritems(eb), |
7518a238 | 1287 | slot); |
be0e5c09 | 1288 | } else { |
5f39d397 CM |
1289 | return generic_bin_search(eb, |
1290 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1291 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1292 | key, btrfs_header_nritems(eb), |
7518a238 | 1293 | slot); |
be0e5c09 CM |
1294 | } |
1295 | return -1; | |
1296 | } | |
1297 | ||
5d4f98a2 YZ |
1298 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1299 | int level, int *slot) | |
1300 | { | |
1301 | return bin_search(eb, key, level, slot); | |
1302 | } | |
1303 | ||
f0486c68 YZ |
1304 | static void root_add_used(struct btrfs_root *root, u32 size) |
1305 | { | |
1306 | spin_lock(&root->accounting_lock); | |
1307 | btrfs_set_root_used(&root->root_item, | |
1308 | btrfs_root_used(&root->root_item) + size); | |
1309 | spin_unlock(&root->accounting_lock); | |
1310 | } | |
1311 | ||
1312 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1313 | { | |
1314 | spin_lock(&root->accounting_lock); | |
1315 | btrfs_set_root_used(&root->root_item, | |
1316 | btrfs_root_used(&root->root_item) - size); | |
1317 | spin_unlock(&root->accounting_lock); | |
1318 | } | |
1319 | ||
d352ac68 CM |
1320 | /* given a node and slot number, this reads the blocks it points to. The |
1321 | * extent buffer is returned with a reference taken (but unlocked). | |
1322 | * NULL is returned on error. | |
1323 | */ | |
e02119d5 | 1324 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1325 | struct extent_buffer *parent, int slot) |
bb803951 | 1326 | { |
ca7a79ad | 1327 | int level = btrfs_header_level(parent); |
bb803951 CM |
1328 | if (slot < 0) |
1329 | return NULL; | |
5f39d397 | 1330 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1331 | return NULL; |
ca7a79ad CM |
1332 | |
1333 | BUG_ON(level == 0); | |
1334 | ||
db94535d | 1335 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
1336 | btrfs_level_size(root, level - 1), |
1337 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
1338 | } |
1339 | ||
d352ac68 CM |
1340 | /* |
1341 | * node level balancing, used to make sure nodes are in proper order for | |
1342 | * item deletion. We balance from the top down, so we have to make sure | |
1343 | * that a deletion won't leave an node completely empty later on. | |
1344 | */ | |
e02119d5 | 1345 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1346 | struct btrfs_root *root, |
1347 | struct btrfs_path *path, int level) | |
bb803951 | 1348 | { |
5f39d397 CM |
1349 | struct extent_buffer *right = NULL; |
1350 | struct extent_buffer *mid; | |
1351 | struct extent_buffer *left = NULL; | |
1352 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1353 | int ret = 0; |
1354 | int wret; | |
1355 | int pslot; | |
bb803951 | 1356 | int orig_slot = path->slots[level]; |
79f95c82 | 1357 | u64 orig_ptr; |
bb803951 CM |
1358 | |
1359 | if (level == 0) | |
1360 | return 0; | |
1361 | ||
5f39d397 | 1362 | mid = path->nodes[level]; |
b4ce94de | 1363 | |
bd681513 CM |
1364 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1365 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1366 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1367 | ||
1d4f8a0c | 1368 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1369 | |
a05a9bb1 | 1370 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1371 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1372 | pslot = path->slots[level + 1]; |
1373 | } | |
bb803951 | 1374 | |
40689478 CM |
1375 | /* |
1376 | * deal with the case where there is only one pointer in the root | |
1377 | * by promoting the node below to a root | |
1378 | */ | |
5f39d397 CM |
1379 | if (!parent) { |
1380 | struct extent_buffer *child; | |
bb803951 | 1381 | |
5f39d397 | 1382 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1383 | return 0; |
1384 | ||
1385 | /* promote the child to a root */ | |
5f39d397 | 1386 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1387 | if (!child) { |
1388 | ret = -EROFS; | |
1389 | btrfs_std_error(root->fs_info, ret); | |
1390 | goto enospc; | |
1391 | } | |
1392 | ||
925baedd | 1393 | btrfs_tree_lock(child); |
b4ce94de | 1394 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1395 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1396 | if (ret) { |
1397 | btrfs_tree_unlock(child); | |
1398 | free_extent_buffer(child); | |
1399 | goto enospc; | |
1400 | } | |
2f375ab9 | 1401 | |
f230475e | 1402 | tree_mod_log_set_root_pointer(root, child); |
240f62c8 | 1403 | rcu_assign_pointer(root->node, child); |
925baedd | 1404 | |
0b86a832 | 1405 | add_root_to_dirty_list(root); |
925baedd | 1406 | btrfs_tree_unlock(child); |
b4ce94de | 1407 | |
925baedd | 1408 | path->locks[level] = 0; |
bb803951 | 1409 | path->nodes[level] = NULL; |
5f39d397 | 1410 | clean_tree_block(trans, root, mid); |
925baedd | 1411 | btrfs_tree_unlock(mid); |
bb803951 | 1412 | /* once for the path */ |
5f39d397 | 1413 | free_extent_buffer(mid); |
f0486c68 YZ |
1414 | |
1415 | root_sub_used(root, mid->len); | |
5581a51a | 1416 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1417 | /* once for the root ptr */ |
3083ee2e | 1418 | free_extent_buffer_stale(mid); |
f0486c68 | 1419 | return 0; |
bb803951 | 1420 | } |
5f39d397 | 1421 | if (btrfs_header_nritems(mid) > |
123abc88 | 1422 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1423 | return 0; |
1424 | ||
559af821 | 1425 | btrfs_header_nritems(mid); |
54aa1f4d | 1426 | |
5f39d397 CM |
1427 | left = read_node_slot(root, parent, pslot - 1); |
1428 | if (left) { | |
925baedd | 1429 | btrfs_tree_lock(left); |
b4ce94de | 1430 | btrfs_set_lock_blocking(left); |
5f39d397 | 1431 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1432 | parent, pslot - 1, &left); |
54aa1f4d CM |
1433 | if (wret) { |
1434 | ret = wret; | |
1435 | goto enospc; | |
1436 | } | |
2cc58cf2 | 1437 | } |
5f39d397 CM |
1438 | right = read_node_slot(root, parent, pslot + 1); |
1439 | if (right) { | |
925baedd | 1440 | btrfs_tree_lock(right); |
b4ce94de | 1441 | btrfs_set_lock_blocking(right); |
5f39d397 | 1442 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1443 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1444 | if (wret) { |
1445 | ret = wret; | |
1446 | goto enospc; | |
1447 | } | |
1448 | } | |
1449 | ||
1450 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1451 | if (left) { |
1452 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1453 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1454 | if (wret < 0) |
1455 | ret = wret; | |
559af821 | 1456 | btrfs_header_nritems(mid); |
bb803951 | 1457 | } |
79f95c82 CM |
1458 | |
1459 | /* | |
1460 | * then try to empty the right most buffer into the middle | |
1461 | */ | |
5f39d397 | 1462 | if (right) { |
971a1f66 | 1463 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1464 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1465 | ret = wret; |
5f39d397 | 1466 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1467 | clean_tree_block(trans, root, right); |
925baedd | 1468 | btrfs_tree_unlock(right); |
f3ea38da | 1469 | del_ptr(trans, root, path, level + 1, pslot + 1, 1); |
f0486c68 | 1470 | root_sub_used(root, right->len); |
5581a51a | 1471 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1472 | free_extent_buffer_stale(right); |
f0486c68 | 1473 | right = NULL; |
bb803951 | 1474 | } else { |
5f39d397 CM |
1475 | struct btrfs_disk_key right_key; |
1476 | btrfs_node_key(right, &right_key, 0); | |
f230475e JS |
1477 | tree_mod_log_set_node_key(root->fs_info, parent, |
1478 | &right_key, pslot + 1, 0); | |
5f39d397 CM |
1479 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1480 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1481 | } |
1482 | } | |
5f39d397 | 1483 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1484 | /* |
1485 | * we're not allowed to leave a node with one item in the | |
1486 | * tree during a delete. A deletion from lower in the tree | |
1487 | * could try to delete the only pointer in this node. | |
1488 | * So, pull some keys from the left. | |
1489 | * There has to be a left pointer at this point because | |
1490 | * otherwise we would have pulled some pointers from the | |
1491 | * right | |
1492 | */ | |
305a26af MF |
1493 | if (!left) { |
1494 | ret = -EROFS; | |
1495 | btrfs_std_error(root->fs_info, ret); | |
1496 | goto enospc; | |
1497 | } | |
5f39d397 | 1498 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1499 | if (wret < 0) { |
79f95c82 | 1500 | ret = wret; |
54aa1f4d CM |
1501 | goto enospc; |
1502 | } | |
bce4eae9 CM |
1503 | if (wret == 1) { |
1504 | wret = push_node_left(trans, root, left, mid, 1); | |
1505 | if (wret < 0) | |
1506 | ret = wret; | |
1507 | } | |
79f95c82 CM |
1508 | BUG_ON(wret == 1); |
1509 | } | |
5f39d397 | 1510 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1511 | clean_tree_block(trans, root, mid); |
925baedd | 1512 | btrfs_tree_unlock(mid); |
f3ea38da | 1513 | del_ptr(trans, root, path, level + 1, pslot, 1); |
f0486c68 | 1514 | root_sub_used(root, mid->len); |
5581a51a | 1515 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1516 | free_extent_buffer_stale(mid); |
f0486c68 | 1517 | mid = NULL; |
79f95c82 CM |
1518 | } else { |
1519 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1520 | struct btrfs_disk_key mid_key; |
1521 | btrfs_node_key(mid, &mid_key, 0); | |
f230475e JS |
1522 | tree_mod_log_set_node_key(root->fs_info, parent, &mid_key, |
1523 | pslot, 0); | |
5f39d397 CM |
1524 | btrfs_set_node_key(parent, &mid_key, pslot); |
1525 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1526 | } |
bb803951 | 1527 | |
79f95c82 | 1528 | /* update the path */ |
5f39d397 CM |
1529 | if (left) { |
1530 | if (btrfs_header_nritems(left) > orig_slot) { | |
1531 | extent_buffer_get(left); | |
925baedd | 1532 | /* left was locked after cow */ |
5f39d397 | 1533 | path->nodes[level] = left; |
bb803951 CM |
1534 | path->slots[level + 1] -= 1; |
1535 | path->slots[level] = orig_slot; | |
925baedd CM |
1536 | if (mid) { |
1537 | btrfs_tree_unlock(mid); | |
5f39d397 | 1538 | free_extent_buffer(mid); |
925baedd | 1539 | } |
bb803951 | 1540 | } else { |
5f39d397 | 1541 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1542 | path->slots[level] = orig_slot; |
1543 | } | |
1544 | } | |
79f95c82 | 1545 | /* double check we haven't messed things up */ |
e20d96d6 | 1546 | if (orig_ptr != |
5f39d397 | 1547 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1548 | BUG(); |
54aa1f4d | 1549 | enospc: |
925baedd CM |
1550 | if (right) { |
1551 | btrfs_tree_unlock(right); | |
5f39d397 | 1552 | free_extent_buffer(right); |
925baedd CM |
1553 | } |
1554 | if (left) { | |
1555 | if (path->nodes[level] != left) | |
1556 | btrfs_tree_unlock(left); | |
5f39d397 | 1557 | free_extent_buffer(left); |
925baedd | 1558 | } |
bb803951 CM |
1559 | return ret; |
1560 | } | |
1561 | ||
d352ac68 CM |
1562 | /* Node balancing for insertion. Here we only split or push nodes around |
1563 | * when they are completely full. This is also done top down, so we | |
1564 | * have to be pessimistic. | |
1565 | */ | |
d397712b | 1566 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1567 | struct btrfs_root *root, |
1568 | struct btrfs_path *path, int level) | |
e66f709b | 1569 | { |
5f39d397 CM |
1570 | struct extent_buffer *right = NULL; |
1571 | struct extent_buffer *mid; | |
1572 | struct extent_buffer *left = NULL; | |
1573 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1574 | int ret = 0; |
1575 | int wret; | |
1576 | int pslot; | |
1577 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1578 | |
1579 | if (level == 0) | |
1580 | return 1; | |
1581 | ||
5f39d397 | 1582 | mid = path->nodes[level]; |
7bb86316 | 1583 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1584 | |
a05a9bb1 | 1585 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1586 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1587 | pslot = path->slots[level + 1]; |
1588 | } | |
e66f709b | 1589 | |
5f39d397 | 1590 | if (!parent) |
e66f709b | 1591 | return 1; |
e66f709b | 1592 | |
5f39d397 | 1593 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1594 | |
1595 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1596 | if (left) { |
e66f709b | 1597 | u32 left_nr; |
925baedd CM |
1598 | |
1599 | btrfs_tree_lock(left); | |
b4ce94de CM |
1600 | btrfs_set_lock_blocking(left); |
1601 | ||
5f39d397 | 1602 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1603 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1604 | wret = 1; | |
1605 | } else { | |
5f39d397 | 1606 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1607 | pslot - 1, &left); |
54aa1f4d CM |
1608 | if (ret) |
1609 | wret = 1; | |
1610 | else { | |
54aa1f4d | 1611 | wret = push_node_left(trans, root, |
971a1f66 | 1612 | left, mid, 0); |
54aa1f4d | 1613 | } |
33ade1f8 | 1614 | } |
e66f709b CM |
1615 | if (wret < 0) |
1616 | ret = wret; | |
1617 | if (wret == 0) { | |
5f39d397 | 1618 | struct btrfs_disk_key disk_key; |
e66f709b | 1619 | orig_slot += left_nr; |
5f39d397 | 1620 | btrfs_node_key(mid, &disk_key, 0); |
f230475e JS |
1621 | tree_mod_log_set_node_key(root->fs_info, parent, |
1622 | &disk_key, pslot, 0); | |
5f39d397 CM |
1623 | btrfs_set_node_key(parent, &disk_key, pslot); |
1624 | btrfs_mark_buffer_dirty(parent); | |
1625 | if (btrfs_header_nritems(left) > orig_slot) { | |
1626 | path->nodes[level] = left; | |
e66f709b CM |
1627 | path->slots[level + 1] -= 1; |
1628 | path->slots[level] = orig_slot; | |
925baedd | 1629 | btrfs_tree_unlock(mid); |
5f39d397 | 1630 | free_extent_buffer(mid); |
e66f709b CM |
1631 | } else { |
1632 | orig_slot -= | |
5f39d397 | 1633 | btrfs_header_nritems(left); |
e66f709b | 1634 | path->slots[level] = orig_slot; |
925baedd | 1635 | btrfs_tree_unlock(left); |
5f39d397 | 1636 | free_extent_buffer(left); |
e66f709b | 1637 | } |
e66f709b CM |
1638 | return 0; |
1639 | } | |
925baedd | 1640 | btrfs_tree_unlock(left); |
5f39d397 | 1641 | free_extent_buffer(left); |
e66f709b | 1642 | } |
925baedd | 1643 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1644 | |
1645 | /* | |
1646 | * then try to empty the right most buffer into the middle | |
1647 | */ | |
5f39d397 | 1648 | if (right) { |
33ade1f8 | 1649 | u32 right_nr; |
b4ce94de | 1650 | |
925baedd | 1651 | btrfs_tree_lock(right); |
b4ce94de CM |
1652 | btrfs_set_lock_blocking(right); |
1653 | ||
5f39d397 | 1654 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
1655 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1656 | wret = 1; | |
1657 | } else { | |
5f39d397 CM |
1658 | ret = btrfs_cow_block(trans, root, right, |
1659 | parent, pslot + 1, | |
9fa8cfe7 | 1660 | &right); |
54aa1f4d CM |
1661 | if (ret) |
1662 | wret = 1; | |
1663 | else { | |
54aa1f4d | 1664 | wret = balance_node_right(trans, root, |
5f39d397 | 1665 | right, mid); |
54aa1f4d | 1666 | } |
33ade1f8 | 1667 | } |
e66f709b CM |
1668 | if (wret < 0) |
1669 | ret = wret; | |
1670 | if (wret == 0) { | |
5f39d397 CM |
1671 | struct btrfs_disk_key disk_key; |
1672 | ||
1673 | btrfs_node_key(right, &disk_key, 0); | |
f230475e JS |
1674 | tree_mod_log_set_node_key(root->fs_info, parent, |
1675 | &disk_key, pslot + 1, 0); | |
5f39d397 CM |
1676 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
1677 | btrfs_mark_buffer_dirty(parent); | |
1678 | ||
1679 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1680 | path->nodes[level] = right; | |
e66f709b CM |
1681 | path->slots[level + 1] += 1; |
1682 | path->slots[level] = orig_slot - | |
5f39d397 | 1683 | btrfs_header_nritems(mid); |
925baedd | 1684 | btrfs_tree_unlock(mid); |
5f39d397 | 1685 | free_extent_buffer(mid); |
e66f709b | 1686 | } else { |
925baedd | 1687 | btrfs_tree_unlock(right); |
5f39d397 | 1688 | free_extent_buffer(right); |
e66f709b | 1689 | } |
e66f709b CM |
1690 | return 0; |
1691 | } | |
925baedd | 1692 | btrfs_tree_unlock(right); |
5f39d397 | 1693 | free_extent_buffer(right); |
e66f709b | 1694 | } |
e66f709b CM |
1695 | return 1; |
1696 | } | |
1697 | ||
3c69faec | 1698 | /* |
d352ac68 CM |
1699 | * readahead one full node of leaves, finding things that are close |
1700 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1701 | */ |
c8c42864 CM |
1702 | static void reada_for_search(struct btrfs_root *root, |
1703 | struct btrfs_path *path, | |
1704 | int level, int slot, u64 objectid) | |
3c69faec | 1705 | { |
5f39d397 | 1706 | struct extent_buffer *node; |
01f46658 | 1707 | struct btrfs_disk_key disk_key; |
3c69faec | 1708 | u32 nritems; |
3c69faec | 1709 | u64 search; |
a7175319 | 1710 | u64 target; |
6b80053d | 1711 | u64 nread = 0; |
cb25c2ea | 1712 | u64 gen; |
3c69faec | 1713 | int direction = path->reada; |
5f39d397 | 1714 | struct extent_buffer *eb; |
6b80053d CM |
1715 | u32 nr; |
1716 | u32 blocksize; | |
1717 | u32 nscan = 0; | |
db94535d | 1718 | |
a6b6e75e | 1719 | if (level != 1) |
6702ed49 CM |
1720 | return; |
1721 | ||
1722 | if (!path->nodes[level]) | |
3c69faec CM |
1723 | return; |
1724 | ||
5f39d397 | 1725 | node = path->nodes[level]; |
925baedd | 1726 | |
3c69faec | 1727 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
1728 | blocksize = btrfs_level_size(root, level - 1); |
1729 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
1730 | if (eb) { |
1731 | free_extent_buffer(eb); | |
3c69faec CM |
1732 | return; |
1733 | } | |
1734 | ||
a7175319 | 1735 | target = search; |
6b80053d | 1736 | |
5f39d397 | 1737 | nritems = btrfs_header_nritems(node); |
6b80053d | 1738 | nr = slot; |
25b8b936 | 1739 | |
d397712b | 1740 | while (1) { |
6b80053d CM |
1741 | if (direction < 0) { |
1742 | if (nr == 0) | |
1743 | break; | |
1744 | nr--; | |
1745 | } else if (direction > 0) { | |
1746 | nr++; | |
1747 | if (nr >= nritems) | |
1748 | break; | |
3c69faec | 1749 | } |
01f46658 CM |
1750 | if (path->reada < 0 && objectid) { |
1751 | btrfs_node_key(node, &disk_key, nr); | |
1752 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1753 | break; | |
1754 | } | |
6b80053d | 1755 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
1756 | if ((search <= target && target - search <= 65536) || |
1757 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 1758 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 1759 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
1760 | nread += blocksize; |
1761 | } | |
1762 | nscan++; | |
a7175319 | 1763 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 1764 | break; |
3c69faec CM |
1765 | } |
1766 | } | |
925baedd | 1767 | |
b4ce94de CM |
1768 | /* |
1769 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
1770 | * cache | |
1771 | */ | |
1772 | static noinline int reada_for_balance(struct btrfs_root *root, | |
1773 | struct btrfs_path *path, int level) | |
1774 | { | |
1775 | int slot; | |
1776 | int nritems; | |
1777 | struct extent_buffer *parent; | |
1778 | struct extent_buffer *eb; | |
1779 | u64 gen; | |
1780 | u64 block1 = 0; | |
1781 | u64 block2 = 0; | |
1782 | int ret = 0; | |
1783 | int blocksize; | |
1784 | ||
8c594ea8 | 1785 | parent = path->nodes[level + 1]; |
b4ce94de CM |
1786 | if (!parent) |
1787 | return 0; | |
1788 | ||
1789 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1790 | slot = path->slots[level + 1]; |
b4ce94de CM |
1791 | blocksize = btrfs_level_size(root, level); |
1792 | ||
1793 | if (slot > 0) { | |
1794 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
1795 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
1796 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
1797 | /* |
1798 | * if we get -eagain from btrfs_buffer_uptodate, we | |
1799 | * don't want to return eagain here. That will loop | |
1800 | * forever | |
1801 | */ | |
1802 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
1803 | block1 = 0; |
1804 | free_extent_buffer(eb); | |
1805 | } | |
8c594ea8 | 1806 | if (slot + 1 < nritems) { |
b4ce94de CM |
1807 | block2 = btrfs_node_blockptr(parent, slot + 1); |
1808 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
1809 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 1810 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
1811 | block2 = 0; |
1812 | free_extent_buffer(eb); | |
1813 | } | |
1814 | if (block1 || block2) { | |
1815 | ret = -EAGAIN; | |
8c594ea8 CM |
1816 | |
1817 | /* release the whole path */ | |
b3b4aa74 | 1818 | btrfs_release_path(path); |
8c594ea8 CM |
1819 | |
1820 | /* read the blocks */ | |
b4ce94de CM |
1821 | if (block1) |
1822 | readahead_tree_block(root, block1, blocksize, 0); | |
1823 | if (block2) | |
1824 | readahead_tree_block(root, block2, blocksize, 0); | |
1825 | ||
1826 | if (block1) { | |
1827 | eb = read_tree_block(root, block1, blocksize, 0); | |
1828 | free_extent_buffer(eb); | |
1829 | } | |
8c594ea8 | 1830 | if (block2) { |
b4ce94de CM |
1831 | eb = read_tree_block(root, block2, blocksize, 0); |
1832 | free_extent_buffer(eb); | |
1833 | } | |
1834 | } | |
1835 | return ret; | |
1836 | } | |
1837 | ||
1838 | ||
d352ac68 | 1839 | /* |
d397712b CM |
1840 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1841 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1842 | * operations on the tree might require changing key pointers higher up in the | |
1843 | * tree. | |
d352ac68 | 1844 | * |
d397712b CM |
1845 | * callers might also have set path->keep_locks, which tells this code to keep |
1846 | * the lock if the path points to the last slot in the block. This is part of | |
1847 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1848 | * |
d397712b CM |
1849 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1850 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1851 | */ |
e02119d5 | 1852 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
1853 | int lowest_unlock, int min_write_lock_level, |
1854 | int *write_lock_level) | |
925baedd CM |
1855 | { |
1856 | int i; | |
1857 | int skip_level = level; | |
051e1b9f | 1858 | int no_skips = 0; |
925baedd CM |
1859 | struct extent_buffer *t; |
1860 | ||
1861 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1862 | if (!path->nodes[i]) | |
1863 | break; | |
1864 | if (!path->locks[i]) | |
1865 | break; | |
051e1b9f | 1866 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1867 | skip_level = i + 1; |
1868 | continue; | |
1869 | } | |
051e1b9f | 1870 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1871 | u32 nritems; |
1872 | t = path->nodes[i]; | |
1873 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1874 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1875 | skip_level = i + 1; |
1876 | continue; | |
1877 | } | |
1878 | } | |
051e1b9f CM |
1879 | if (skip_level < i && i >= lowest_unlock) |
1880 | no_skips = 1; | |
1881 | ||
925baedd CM |
1882 | t = path->nodes[i]; |
1883 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 1884 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 1885 | path->locks[i] = 0; |
f7c79f30 CM |
1886 | if (write_lock_level && |
1887 | i > min_write_lock_level && | |
1888 | i <= *write_lock_level) { | |
1889 | *write_lock_level = i - 1; | |
1890 | } | |
925baedd CM |
1891 | } |
1892 | } | |
1893 | } | |
1894 | ||
b4ce94de CM |
1895 | /* |
1896 | * This releases any locks held in the path starting at level and | |
1897 | * going all the way up to the root. | |
1898 | * | |
1899 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
1900 | * corner cases, such as COW of the block at slot zero in the node. This | |
1901 | * ignores those rules, and it should only be called when there are no | |
1902 | * more updates to be done higher up in the tree. | |
1903 | */ | |
1904 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
1905 | { | |
1906 | int i; | |
1907 | ||
5d4f98a2 | 1908 | if (path->keep_locks) |
b4ce94de CM |
1909 | return; |
1910 | ||
1911 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1912 | if (!path->nodes[i]) | |
12f4dacc | 1913 | continue; |
b4ce94de | 1914 | if (!path->locks[i]) |
12f4dacc | 1915 | continue; |
bd681513 | 1916 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
1917 | path->locks[i] = 0; |
1918 | } | |
1919 | } | |
1920 | ||
c8c42864 CM |
1921 | /* |
1922 | * helper function for btrfs_search_slot. The goal is to find a block | |
1923 | * in cache without setting the path to blocking. If we find the block | |
1924 | * we return zero and the path is unchanged. | |
1925 | * | |
1926 | * If we can't find the block, we set the path blocking and do some | |
1927 | * reada. -EAGAIN is returned and the search must be repeated. | |
1928 | */ | |
1929 | static int | |
1930 | read_block_for_search(struct btrfs_trans_handle *trans, | |
1931 | struct btrfs_root *root, struct btrfs_path *p, | |
1932 | struct extent_buffer **eb_ret, int level, int slot, | |
1933 | struct btrfs_key *key) | |
1934 | { | |
1935 | u64 blocknr; | |
1936 | u64 gen; | |
1937 | u32 blocksize; | |
1938 | struct extent_buffer *b = *eb_ret; | |
1939 | struct extent_buffer *tmp; | |
76a05b35 | 1940 | int ret; |
c8c42864 CM |
1941 | |
1942 | blocknr = btrfs_node_blockptr(b, slot); | |
1943 | gen = btrfs_node_ptr_generation(b, slot); | |
1944 | blocksize = btrfs_level_size(root, level - 1); | |
1945 | ||
1946 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 1947 | if (tmp) { |
b9fab919 CM |
1948 | /* first we do an atomic uptodate check */ |
1949 | if (btrfs_buffer_uptodate(tmp, 0, 1) > 0) { | |
1950 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { | |
cb44921a CM |
1951 | /* |
1952 | * we found an up to date block without | |
1953 | * sleeping, return | |
1954 | * right away | |
1955 | */ | |
1956 | *eb_ret = tmp; | |
1957 | return 0; | |
1958 | } | |
1959 | /* the pages were up to date, but we failed | |
1960 | * the generation number check. Do a full | |
1961 | * read for the generation number that is correct. | |
1962 | * We must do this without dropping locks so | |
1963 | * we can trust our generation number | |
1964 | */ | |
1965 | free_extent_buffer(tmp); | |
bd681513 CM |
1966 | btrfs_set_path_blocking(p); |
1967 | ||
b9fab919 | 1968 | /* now we're allowed to do a blocking uptodate check */ |
cb44921a | 1969 | tmp = read_tree_block(root, blocknr, blocksize, gen); |
b9fab919 | 1970 | if (tmp && btrfs_buffer_uptodate(tmp, gen, 0) > 0) { |
cb44921a CM |
1971 | *eb_ret = tmp; |
1972 | return 0; | |
1973 | } | |
1974 | free_extent_buffer(tmp); | |
b3b4aa74 | 1975 | btrfs_release_path(p); |
cb44921a CM |
1976 | return -EIO; |
1977 | } | |
c8c42864 CM |
1978 | } |
1979 | ||
1980 | /* | |
1981 | * reduce lock contention at high levels | |
1982 | * of the btree by dropping locks before | |
76a05b35 CM |
1983 | * we read. Don't release the lock on the current |
1984 | * level because we need to walk this node to figure | |
1985 | * out which blocks to read. | |
c8c42864 | 1986 | */ |
8c594ea8 CM |
1987 | btrfs_unlock_up_safe(p, level + 1); |
1988 | btrfs_set_path_blocking(p); | |
1989 | ||
cb44921a | 1990 | free_extent_buffer(tmp); |
c8c42864 CM |
1991 | if (p->reada) |
1992 | reada_for_search(root, p, level, slot, key->objectid); | |
1993 | ||
b3b4aa74 | 1994 | btrfs_release_path(p); |
76a05b35 CM |
1995 | |
1996 | ret = -EAGAIN; | |
5bdd3536 | 1997 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
1998 | if (tmp) { |
1999 | /* | |
2000 | * If the read above didn't mark this buffer up to date, | |
2001 | * it will never end up being up to date. Set ret to EIO now | |
2002 | * and give up so that our caller doesn't loop forever | |
2003 | * on our EAGAINs. | |
2004 | */ | |
b9fab919 | 2005 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2006 | ret = -EIO; |
c8c42864 | 2007 | free_extent_buffer(tmp); |
76a05b35 CM |
2008 | } |
2009 | return ret; | |
c8c42864 CM |
2010 | } |
2011 | ||
2012 | /* | |
2013 | * helper function for btrfs_search_slot. This does all of the checks | |
2014 | * for node-level blocks and does any balancing required based on | |
2015 | * the ins_len. | |
2016 | * | |
2017 | * If no extra work was required, zero is returned. If we had to | |
2018 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2019 | * start over | |
2020 | */ | |
2021 | static int | |
2022 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2023 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2024 | struct extent_buffer *b, int level, int ins_len, |
2025 | int *write_lock_level) | |
c8c42864 CM |
2026 | { |
2027 | int ret; | |
2028 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2029 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2030 | int sret; | |
2031 | ||
bd681513 CM |
2032 | if (*write_lock_level < level + 1) { |
2033 | *write_lock_level = level + 1; | |
2034 | btrfs_release_path(p); | |
2035 | goto again; | |
2036 | } | |
2037 | ||
c8c42864 CM |
2038 | sret = reada_for_balance(root, p, level); |
2039 | if (sret) | |
2040 | goto again; | |
2041 | ||
2042 | btrfs_set_path_blocking(p); | |
2043 | sret = split_node(trans, root, p, level); | |
bd681513 | 2044 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2045 | |
2046 | BUG_ON(sret > 0); | |
2047 | if (sret) { | |
2048 | ret = sret; | |
2049 | goto done; | |
2050 | } | |
2051 | b = p->nodes[level]; | |
2052 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2053 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2054 | int sret; |
2055 | ||
bd681513 CM |
2056 | if (*write_lock_level < level + 1) { |
2057 | *write_lock_level = level + 1; | |
2058 | btrfs_release_path(p); | |
2059 | goto again; | |
2060 | } | |
2061 | ||
c8c42864 CM |
2062 | sret = reada_for_balance(root, p, level); |
2063 | if (sret) | |
2064 | goto again; | |
2065 | ||
2066 | btrfs_set_path_blocking(p); | |
2067 | sret = balance_level(trans, root, p, level); | |
bd681513 | 2068 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2069 | |
2070 | if (sret) { | |
2071 | ret = sret; | |
2072 | goto done; | |
2073 | } | |
2074 | b = p->nodes[level]; | |
2075 | if (!b) { | |
b3b4aa74 | 2076 | btrfs_release_path(p); |
c8c42864 CM |
2077 | goto again; |
2078 | } | |
2079 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2080 | } | |
2081 | return 0; | |
2082 | ||
2083 | again: | |
2084 | ret = -EAGAIN; | |
2085 | done: | |
2086 | return ret; | |
2087 | } | |
2088 | ||
74123bd7 CM |
2089 | /* |
2090 | * look for key in the tree. path is filled in with nodes along the way | |
2091 | * if key is found, we return zero and you can find the item in the leaf | |
2092 | * level of the path (level 0) | |
2093 | * | |
2094 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2095 | * be inserted, and 1 is returned. If there are other errors during the |
2096 | * search a negative error number is returned. | |
97571fd0 CM |
2097 | * |
2098 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2099 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2100 | * possible) | |
74123bd7 | 2101 | */ |
e089f05c CM |
2102 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2103 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2104 | ins_len, int cow) | |
be0e5c09 | 2105 | { |
5f39d397 | 2106 | struct extent_buffer *b; |
be0e5c09 CM |
2107 | int slot; |
2108 | int ret; | |
33c66f43 | 2109 | int err; |
be0e5c09 | 2110 | int level; |
925baedd | 2111 | int lowest_unlock = 1; |
bd681513 CM |
2112 | int root_lock; |
2113 | /* everything at write_lock_level or lower must be write locked */ | |
2114 | int write_lock_level = 0; | |
9f3a7427 | 2115 | u8 lowest_level = 0; |
f7c79f30 | 2116 | int min_write_lock_level; |
9f3a7427 | 2117 | |
6702ed49 | 2118 | lowest_level = p->lowest_level; |
323ac95b | 2119 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2120 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2121 | |
bd681513 | 2122 | if (ins_len < 0) { |
925baedd | 2123 | lowest_unlock = 2; |
65b51a00 | 2124 | |
bd681513 CM |
2125 | /* when we are removing items, we might have to go up to level |
2126 | * two as we update tree pointers Make sure we keep write | |
2127 | * for those levels as well | |
2128 | */ | |
2129 | write_lock_level = 2; | |
2130 | } else if (ins_len > 0) { | |
2131 | /* | |
2132 | * for inserting items, make sure we have a write lock on | |
2133 | * level 1 so we can update keys | |
2134 | */ | |
2135 | write_lock_level = 1; | |
2136 | } | |
2137 | ||
2138 | if (!cow) | |
2139 | write_lock_level = -1; | |
2140 | ||
2141 | if (cow && (p->keep_locks || p->lowest_level)) | |
2142 | write_lock_level = BTRFS_MAX_LEVEL; | |
2143 | ||
f7c79f30 CM |
2144 | min_write_lock_level = write_lock_level; |
2145 | ||
bb803951 | 2146 | again: |
bd681513 CM |
2147 | /* |
2148 | * we try very hard to do read locks on the root | |
2149 | */ | |
2150 | root_lock = BTRFS_READ_LOCK; | |
2151 | level = 0; | |
5d4f98a2 | 2152 | if (p->search_commit_root) { |
bd681513 CM |
2153 | /* |
2154 | * the commit roots are read only | |
2155 | * so we always do read locks | |
2156 | */ | |
5d4f98a2 YZ |
2157 | b = root->commit_root; |
2158 | extent_buffer_get(b); | |
bd681513 | 2159 | level = btrfs_header_level(b); |
5d4f98a2 | 2160 | if (!p->skip_locking) |
bd681513 | 2161 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2162 | } else { |
bd681513 | 2163 | if (p->skip_locking) { |
5d4f98a2 | 2164 | b = btrfs_root_node(root); |
bd681513 CM |
2165 | level = btrfs_header_level(b); |
2166 | } else { | |
2167 | /* we don't know the level of the root node | |
2168 | * until we actually have it read locked | |
2169 | */ | |
2170 | b = btrfs_read_lock_root_node(root); | |
2171 | level = btrfs_header_level(b); | |
2172 | if (level <= write_lock_level) { | |
2173 | /* whoops, must trade for write lock */ | |
2174 | btrfs_tree_read_unlock(b); | |
2175 | free_extent_buffer(b); | |
2176 | b = btrfs_lock_root_node(root); | |
2177 | root_lock = BTRFS_WRITE_LOCK; | |
2178 | ||
2179 | /* the level might have changed, check again */ | |
2180 | level = btrfs_header_level(b); | |
2181 | } | |
2182 | } | |
5d4f98a2 | 2183 | } |
bd681513 CM |
2184 | p->nodes[level] = b; |
2185 | if (!p->skip_locking) | |
2186 | p->locks[level] = root_lock; | |
925baedd | 2187 | |
eb60ceac | 2188 | while (b) { |
5f39d397 | 2189 | level = btrfs_header_level(b); |
65b51a00 CM |
2190 | |
2191 | /* | |
2192 | * setup the path here so we can release it under lock | |
2193 | * contention with the cow code | |
2194 | */ | |
02217ed2 | 2195 | if (cow) { |
c8c42864 CM |
2196 | /* |
2197 | * if we don't really need to cow this block | |
2198 | * then we don't want to set the path blocking, | |
2199 | * so we test it here | |
2200 | */ | |
5d4f98a2 | 2201 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2202 | goto cow_done; |
5d4f98a2 | 2203 | |
b4ce94de CM |
2204 | btrfs_set_path_blocking(p); |
2205 | ||
bd681513 CM |
2206 | /* |
2207 | * must have write locks on this node and the | |
2208 | * parent | |
2209 | */ | |
2210 | if (level + 1 > write_lock_level) { | |
2211 | write_lock_level = level + 1; | |
2212 | btrfs_release_path(p); | |
2213 | goto again; | |
2214 | } | |
2215 | ||
33c66f43 YZ |
2216 | err = btrfs_cow_block(trans, root, b, |
2217 | p->nodes[level + 1], | |
2218 | p->slots[level + 1], &b); | |
2219 | if (err) { | |
33c66f43 | 2220 | ret = err; |
65b51a00 | 2221 | goto done; |
54aa1f4d | 2222 | } |
02217ed2 | 2223 | } |
65b51a00 | 2224 | cow_done: |
02217ed2 | 2225 | BUG_ON(!cow && ins_len); |
65b51a00 | 2226 | |
eb60ceac | 2227 | p->nodes[level] = b; |
bd681513 | 2228 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2229 | |
2230 | /* | |
2231 | * we have a lock on b and as long as we aren't changing | |
2232 | * the tree, there is no way to for the items in b to change. | |
2233 | * It is safe to drop the lock on our parent before we | |
2234 | * go through the expensive btree search on b. | |
2235 | * | |
2236 | * If cow is true, then we might be changing slot zero, | |
2237 | * which may require changing the parent. So, we can't | |
2238 | * drop the lock until after we know which slot we're | |
2239 | * operating on. | |
2240 | */ | |
2241 | if (!cow) | |
2242 | btrfs_unlock_up_safe(p, level + 1); | |
2243 | ||
5f39d397 | 2244 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 2245 | |
5f39d397 | 2246 | if (level != 0) { |
33c66f43 YZ |
2247 | int dec = 0; |
2248 | if (ret && slot > 0) { | |
2249 | dec = 1; | |
be0e5c09 | 2250 | slot -= 1; |
33c66f43 | 2251 | } |
be0e5c09 | 2252 | p->slots[level] = slot; |
33c66f43 | 2253 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2254 | ins_len, &write_lock_level); |
33c66f43 | 2255 | if (err == -EAGAIN) |
c8c42864 | 2256 | goto again; |
33c66f43 YZ |
2257 | if (err) { |
2258 | ret = err; | |
c8c42864 | 2259 | goto done; |
33c66f43 | 2260 | } |
c8c42864 CM |
2261 | b = p->nodes[level]; |
2262 | slot = p->slots[level]; | |
b4ce94de | 2263 | |
bd681513 CM |
2264 | /* |
2265 | * slot 0 is special, if we change the key | |
2266 | * we have to update the parent pointer | |
2267 | * which means we must have a write lock | |
2268 | * on the parent | |
2269 | */ | |
2270 | if (slot == 0 && cow && | |
2271 | write_lock_level < level + 1) { | |
2272 | write_lock_level = level + 1; | |
2273 | btrfs_release_path(p); | |
2274 | goto again; | |
2275 | } | |
2276 | ||
f7c79f30 CM |
2277 | unlock_up(p, level, lowest_unlock, |
2278 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2279 | |
925baedd | 2280 | if (level == lowest_level) { |
33c66f43 YZ |
2281 | if (dec) |
2282 | p->slots[level]++; | |
5b21f2ed | 2283 | goto done; |
925baedd | 2284 | } |
ca7a79ad | 2285 | |
33c66f43 | 2286 | err = read_block_for_search(trans, root, p, |
c8c42864 | 2287 | &b, level, slot, key); |
33c66f43 | 2288 | if (err == -EAGAIN) |
c8c42864 | 2289 | goto again; |
33c66f43 YZ |
2290 | if (err) { |
2291 | ret = err; | |
76a05b35 | 2292 | goto done; |
33c66f43 | 2293 | } |
76a05b35 | 2294 | |
b4ce94de | 2295 | if (!p->skip_locking) { |
bd681513 CM |
2296 | level = btrfs_header_level(b); |
2297 | if (level <= write_lock_level) { | |
2298 | err = btrfs_try_tree_write_lock(b); | |
2299 | if (!err) { | |
2300 | btrfs_set_path_blocking(p); | |
2301 | btrfs_tree_lock(b); | |
2302 | btrfs_clear_path_blocking(p, b, | |
2303 | BTRFS_WRITE_LOCK); | |
2304 | } | |
2305 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2306 | } else { | |
2307 | err = btrfs_try_tree_read_lock(b); | |
2308 | if (!err) { | |
2309 | btrfs_set_path_blocking(p); | |
2310 | btrfs_tree_read_lock(b); | |
2311 | btrfs_clear_path_blocking(p, b, | |
2312 | BTRFS_READ_LOCK); | |
2313 | } | |
2314 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2315 | } |
bd681513 | 2316 | p->nodes[level] = b; |
b4ce94de | 2317 | } |
be0e5c09 CM |
2318 | } else { |
2319 | p->slots[level] = slot; | |
87b29b20 YZ |
2320 | if (ins_len > 0 && |
2321 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2322 | if (write_lock_level < 1) { |
2323 | write_lock_level = 1; | |
2324 | btrfs_release_path(p); | |
2325 | goto again; | |
2326 | } | |
2327 | ||
b4ce94de | 2328 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2329 | err = split_leaf(trans, root, key, |
2330 | p, ins_len, ret == 0); | |
bd681513 | 2331 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2332 | |
33c66f43 YZ |
2333 | BUG_ON(err > 0); |
2334 | if (err) { | |
2335 | ret = err; | |
65b51a00 CM |
2336 | goto done; |
2337 | } | |
5c680ed6 | 2338 | } |
459931ec | 2339 | if (!p->search_for_split) |
f7c79f30 CM |
2340 | unlock_up(p, level, lowest_unlock, |
2341 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2342 | goto done; |
be0e5c09 CM |
2343 | } |
2344 | } | |
65b51a00 CM |
2345 | ret = 1; |
2346 | done: | |
b4ce94de CM |
2347 | /* |
2348 | * we don't really know what they plan on doing with the path | |
2349 | * from here on, so for now just mark it as blocking | |
2350 | */ | |
b9473439 CM |
2351 | if (!p->leave_spinning) |
2352 | btrfs_set_path_blocking(p); | |
76a05b35 | 2353 | if (ret < 0) |
b3b4aa74 | 2354 | btrfs_release_path(p); |
65b51a00 | 2355 | return ret; |
be0e5c09 CM |
2356 | } |
2357 | ||
74123bd7 CM |
2358 | /* |
2359 | * adjust the pointers going up the tree, starting at level | |
2360 | * making sure the right key of each node is points to 'key'. | |
2361 | * This is used after shifting pointers to the left, so it stops | |
2362 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2363 | * higher levels | |
aa5d6bed | 2364 | * |
74123bd7 | 2365 | */ |
143bede5 JM |
2366 | static void fixup_low_keys(struct btrfs_trans_handle *trans, |
2367 | struct btrfs_root *root, struct btrfs_path *path, | |
2368 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
2369 | { |
2370 | int i; | |
5f39d397 CM |
2371 | struct extent_buffer *t; |
2372 | ||
234b63a0 | 2373 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2374 | int tslot = path->slots[i]; |
eb60ceac | 2375 | if (!path->nodes[i]) |
be0e5c09 | 2376 | break; |
5f39d397 | 2377 | t = path->nodes[i]; |
f230475e | 2378 | tree_mod_log_set_node_key(root->fs_info, t, key, tslot, 1); |
5f39d397 | 2379 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2380 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2381 | if (tslot != 0) |
2382 | break; | |
2383 | } | |
2384 | } | |
2385 | ||
31840ae1 ZY |
2386 | /* |
2387 | * update item key. | |
2388 | * | |
2389 | * This function isn't completely safe. It's the caller's responsibility | |
2390 | * that the new key won't break the order | |
2391 | */ | |
143bede5 JM |
2392 | void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, |
2393 | struct btrfs_root *root, struct btrfs_path *path, | |
2394 | struct btrfs_key *new_key) | |
31840ae1 ZY |
2395 | { |
2396 | struct btrfs_disk_key disk_key; | |
2397 | struct extent_buffer *eb; | |
2398 | int slot; | |
2399 | ||
2400 | eb = path->nodes[0]; | |
2401 | slot = path->slots[0]; | |
2402 | if (slot > 0) { | |
2403 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 2404 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
2405 | } |
2406 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2407 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 2408 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
2409 | } |
2410 | ||
2411 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2412 | btrfs_set_item_key(eb, &disk_key, slot); | |
2413 | btrfs_mark_buffer_dirty(eb); | |
2414 | if (slot == 0) | |
2415 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
31840ae1 ZY |
2416 | } |
2417 | ||
74123bd7 CM |
2418 | /* |
2419 | * try to push data from one node into the next node left in the | |
79f95c82 | 2420 | * tree. |
aa5d6bed CM |
2421 | * |
2422 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2423 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2424 | */ |
98ed5174 CM |
2425 | static int push_node_left(struct btrfs_trans_handle *trans, |
2426 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 2427 | struct extent_buffer *src, int empty) |
be0e5c09 | 2428 | { |
be0e5c09 | 2429 | int push_items = 0; |
bb803951 CM |
2430 | int src_nritems; |
2431 | int dst_nritems; | |
aa5d6bed | 2432 | int ret = 0; |
be0e5c09 | 2433 | |
5f39d397 CM |
2434 | src_nritems = btrfs_header_nritems(src); |
2435 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2436 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
2437 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2438 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2439 | |
bce4eae9 | 2440 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2441 | return 1; |
2442 | ||
d397712b | 2443 | if (push_items <= 0) |
be0e5c09 CM |
2444 | return 1; |
2445 | ||
bce4eae9 | 2446 | if (empty) { |
971a1f66 | 2447 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2448 | if (push_items < src_nritems) { |
2449 | /* leave at least 8 pointers in the node if | |
2450 | * we aren't going to empty it | |
2451 | */ | |
2452 | if (src_nritems - push_items < 8) { | |
2453 | if (push_items <= 8) | |
2454 | return 1; | |
2455 | push_items -= 8; | |
2456 | } | |
2457 | } | |
2458 | } else | |
2459 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2460 | |
f230475e JS |
2461 | tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
2462 | push_items); | |
5f39d397 CM |
2463 | copy_extent_buffer(dst, src, |
2464 | btrfs_node_key_ptr_offset(dst_nritems), | |
2465 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2466 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2467 | |
bb803951 | 2468 | if (push_items < src_nritems) { |
f230475e JS |
2469 | tree_mod_log_eb_move(root->fs_info, src, 0, push_items, |
2470 | src_nritems - push_items); | |
5f39d397 CM |
2471 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2472 | btrfs_node_key_ptr_offset(push_items), | |
2473 | (src_nritems - push_items) * | |
2474 | sizeof(struct btrfs_key_ptr)); | |
2475 | } | |
2476 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2477 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
2478 | btrfs_mark_buffer_dirty(src); | |
2479 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2480 | |
79f95c82 CM |
2481 | return ret; |
2482 | } | |
2483 | ||
2484 | /* | |
2485 | * try to push data from one node into the next node right in the | |
2486 | * tree. | |
2487 | * | |
2488 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2489 | * error, and > 0 if there was no room in the right hand block. | |
2490 | * | |
2491 | * this will only push up to 1/2 the contents of the left node over | |
2492 | */ | |
5f39d397 CM |
2493 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2494 | struct btrfs_root *root, | |
2495 | struct extent_buffer *dst, | |
2496 | struct extent_buffer *src) | |
79f95c82 | 2497 | { |
79f95c82 CM |
2498 | int push_items = 0; |
2499 | int max_push; | |
2500 | int src_nritems; | |
2501 | int dst_nritems; | |
2502 | int ret = 0; | |
79f95c82 | 2503 | |
7bb86316 CM |
2504 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2505 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2506 | ||
5f39d397 CM |
2507 | src_nritems = btrfs_header_nritems(src); |
2508 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2509 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 2510 | if (push_items <= 0) |
79f95c82 | 2511 | return 1; |
bce4eae9 | 2512 | |
d397712b | 2513 | if (src_nritems < 4) |
bce4eae9 | 2514 | return 1; |
79f95c82 CM |
2515 | |
2516 | max_push = src_nritems / 2 + 1; | |
2517 | /* don't try to empty the node */ | |
d397712b | 2518 | if (max_push >= src_nritems) |
79f95c82 | 2519 | return 1; |
252c38f0 | 2520 | |
79f95c82 CM |
2521 | if (max_push < push_items) |
2522 | push_items = max_push; | |
2523 | ||
f230475e | 2524 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
2525 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2526 | btrfs_node_key_ptr_offset(0), | |
2527 | (dst_nritems) * | |
2528 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2529 | |
f230475e JS |
2530 | tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
2531 | src_nritems - push_items, push_items); | |
5f39d397 CM |
2532 | copy_extent_buffer(dst, src, |
2533 | btrfs_node_key_ptr_offset(0), | |
2534 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2535 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2536 | |
5f39d397 CM |
2537 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2538 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2539 | |
5f39d397 CM |
2540 | btrfs_mark_buffer_dirty(src); |
2541 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2542 | |
aa5d6bed | 2543 | return ret; |
be0e5c09 CM |
2544 | } |
2545 | ||
97571fd0 CM |
2546 | /* |
2547 | * helper function to insert a new root level in the tree. | |
2548 | * A new node is allocated, and a single item is inserted to | |
2549 | * point to the existing root | |
aa5d6bed CM |
2550 | * |
2551 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2552 | */ |
d397712b | 2553 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2554 | struct btrfs_root *root, |
2555 | struct btrfs_path *path, int level) | |
5c680ed6 | 2556 | { |
7bb86316 | 2557 | u64 lower_gen; |
5f39d397 CM |
2558 | struct extent_buffer *lower; |
2559 | struct extent_buffer *c; | |
925baedd | 2560 | struct extent_buffer *old; |
5f39d397 | 2561 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
2562 | |
2563 | BUG_ON(path->nodes[level]); | |
2564 | BUG_ON(path->nodes[level-1] != root->node); | |
2565 | ||
7bb86316 CM |
2566 | lower = path->nodes[level-1]; |
2567 | if (level == 1) | |
2568 | btrfs_item_key(lower, &lower_key, 0); | |
2569 | else | |
2570 | btrfs_node_key(lower, &lower_key, 0); | |
2571 | ||
31840ae1 | 2572 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 2573 | root->root_key.objectid, &lower_key, |
5581a51a | 2574 | level, root->node->start, 0); |
5f39d397 CM |
2575 | if (IS_ERR(c)) |
2576 | return PTR_ERR(c); | |
925baedd | 2577 | |
f0486c68 YZ |
2578 | root_add_used(root, root->nodesize); |
2579 | ||
5d4f98a2 | 2580 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
2581 | btrfs_set_header_nritems(c, 1); |
2582 | btrfs_set_header_level(c, level); | |
db94535d | 2583 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 2584 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 2585 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 2586 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
2587 | |
2588 | write_extent_buffer(c, root->fs_info->fsid, | |
2589 | (unsigned long)btrfs_header_fsid(c), | |
2590 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2591 | |
2592 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
2593 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
2594 | BTRFS_UUID_SIZE); | |
2595 | ||
5f39d397 | 2596 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2597 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2598 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2599 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2600 | |
2601 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2602 | |
5f39d397 | 2603 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2604 | |
925baedd | 2605 | old = root->node; |
f230475e | 2606 | tree_mod_log_set_root_pointer(root, c); |
240f62c8 | 2607 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2608 | |
2609 | /* the super has an extra ref to root->node */ | |
2610 | free_extent_buffer(old); | |
2611 | ||
0b86a832 | 2612 | add_root_to_dirty_list(root); |
5f39d397 CM |
2613 | extent_buffer_get(c); |
2614 | path->nodes[level] = c; | |
bd681513 | 2615 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2616 | path->slots[level] = 0; |
2617 | return 0; | |
2618 | } | |
2619 | ||
74123bd7 CM |
2620 | /* |
2621 | * worker function to insert a single pointer in a node. | |
2622 | * the node should have enough room for the pointer already | |
97571fd0 | 2623 | * |
74123bd7 CM |
2624 | * slot and level indicate where you want the key to go, and |
2625 | * blocknr is the block the key points to. | |
2626 | */ | |
143bede5 JM |
2627 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2628 | struct btrfs_root *root, struct btrfs_path *path, | |
2629 | struct btrfs_disk_key *key, u64 bytenr, | |
f3ea38da | 2630 | int slot, int level, int tree_mod_log) |
74123bd7 | 2631 | { |
5f39d397 | 2632 | struct extent_buffer *lower; |
74123bd7 | 2633 | int nritems; |
f3ea38da | 2634 | int ret; |
5c680ed6 CM |
2635 | |
2636 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 2637 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
2638 | lower = path->nodes[level]; |
2639 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2640 | BUG_ON(slot > nritems); |
143bede5 | 2641 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 2642 | if (slot != nritems) { |
f3ea38da JS |
2643 | if (tree_mod_log && level) |
2644 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, | |
2645 | slot, nritems - slot); | |
5f39d397 CM |
2646 | memmove_extent_buffer(lower, |
2647 | btrfs_node_key_ptr_offset(slot + 1), | |
2648 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2649 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2650 | } |
f3ea38da JS |
2651 | if (tree_mod_log && level) { |
2652 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, | |
2653 | MOD_LOG_KEY_ADD); | |
2654 | BUG_ON(ret < 0); | |
2655 | } | |
5f39d397 | 2656 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2657 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2658 | WARN_ON(trans->transid == 0); |
2659 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2660 | btrfs_set_header_nritems(lower, nritems + 1); |
2661 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2662 | } |
2663 | ||
97571fd0 CM |
2664 | /* |
2665 | * split the node at the specified level in path in two. | |
2666 | * The path is corrected to point to the appropriate node after the split | |
2667 | * | |
2668 | * Before splitting this tries to make some room in the node by pushing | |
2669 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2670 | * |
2671 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2672 | */ |
e02119d5 CM |
2673 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2674 | struct btrfs_root *root, | |
2675 | struct btrfs_path *path, int level) | |
be0e5c09 | 2676 | { |
5f39d397 CM |
2677 | struct extent_buffer *c; |
2678 | struct extent_buffer *split; | |
2679 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2680 | int mid; |
5c680ed6 | 2681 | int ret; |
7518a238 | 2682 | u32 c_nritems; |
eb60ceac | 2683 | |
5f39d397 | 2684 | c = path->nodes[level]; |
7bb86316 | 2685 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2686 | if (c == root->node) { |
5c680ed6 | 2687 | /* trying to split the root, lets make a new one */ |
e089f05c | 2688 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2689 | if (ret) |
2690 | return ret; | |
b3612421 | 2691 | } else { |
e66f709b | 2692 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2693 | c = path->nodes[level]; |
2694 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 2695 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 2696 | return 0; |
54aa1f4d CM |
2697 | if (ret < 0) |
2698 | return ret; | |
be0e5c09 | 2699 | } |
e66f709b | 2700 | |
5f39d397 | 2701 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2702 | mid = (c_nritems + 1) / 2; |
2703 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2704 | |
5d4f98a2 | 2705 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 2706 | root->root_key.objectid, |
5581a51a | 2707 | &disk_key, level, c->start, 0); |
5f39d397 CM |
2708 | if (IS_ERR(split)) |
2709 | return PTR_ERR(split); | |
2710 | ||
f0486c68 YZ |
2711 | root_add_used(root, root->nodesize); |
2712 | ||
5d4f98a2 | 2713 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 2714 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 2715 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 2716 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 2717 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
2718 | btrfs_set_header_owner(split, root->root_key.objectid); |
2719 | write_extent_buffer(split, root->fs_info->fsid, | |
2720 | (unsigned long)btrfs_header_fsid(split), | |
2721 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
2722 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
2723 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
2724 | BTRFS_UUID_SIZE); | |
54aa1f4d | 2725 | |
f230475e | 2726 | tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid); |
5f39d397 CM |
2727 | copy_extent_buffer(split, c, |
2728 | btrfs_node_key_ptr_offset(0), | |
2729 | btrfs_node_key_ptr_offset(mid), | |
2730 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2731 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2732 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
2733 | ret = 0; |
2734 | ||
5f39d397 CM |
2735 | btrfs_mark_buffer_dirty(c); |
2736 | btrfs_mark_buffer_dirty(split); | |
2737 | ||
143bede5 | 2738 | insert_ptr(trans, root, path, &disk_key, split->start, |
f3ea38da | 2739 | path->slots[level + 1] + 1, level + 1, 1); |
aa5d6bed | 2740 | |
5de08d7d | 2741 | if (path->slots[level] >= mid) { |
5c680ed6 | 2742 | path->slots[level] -= mid; |
925baedd | 2743 | btrfs_tree_unlock(c); |
5f39d397 CM |
2744 | free_extent_buffer(c); |
2745 | path->nodes[level] = split; | |
5c680ed6 CM |
2746 | path->slots[level + 1] += 1; |
2747 | } else { | |
925baedd | 2748 | btrfs_tree_unlock(split); |
5f39d397 | 2749 | free_extent_buffer(split); |
be0e5c09 | 2750 | } |
aa5d6bed | 2751 | return ret; |
be0e5c09 CM |
2752 | } |
2753 | ||
74123bd7 CM |
2754 | /* |
2755 | * how many bytes are required to store the items in a leaf. start | |
2756 | * and nr indicate which items in the leaf to check. This totals up the | |
2757 | * space used both by the item structs and the item data | |
2758 | */ | |
5f39d397 | 2759 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2760 | { |
2761 | int data_len; | |
5f39d397 | 2762 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2763 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2764 | |
2765 | if (!nr) | |
2766 | return 0; | |
5f39d397 CM |
2767 | data_len = btrfs_item_end_nr(l, start); |
2768 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2769 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2770 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2771 | return data_len; |
2772 | } | |
2773 | ||
d4dbff95 CM |
2774 | /* |
2775 | * The space between the end of the leaf items and | |
2776 | * the start of the leaf data. IOW, how much room | |
2777 | * the leaf has left for both items and data | |
2778 | */ | |
d397712b | 2779 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 2780 | struct extent_buffer *leaf) |
d4dbff95 | 2781 | { |
5f39d397 CM |
2782 | int nritems = btrfs_header_nritems(leaf); |
2783 | int ret; | |
2784 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
2785 | if (ret < 0) { | |
d397712b CM |
2786 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
2787 | "used %d nritems %d\n", | |
ae2f5411 | 2788 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
2789 | leaf_space_used(leaf, 0, nritems), nritems); |
2790 | } | |
2791 | return ret; | |
d4dbff95 CM |
2792 | } |
2793 | ||
99d8f83c CM |
2794 | /* |
2795 | * min slot controls the lowest index we're willing to push to the | |
2796 | * right. We'll push up to and including min_slot, but no lower | |
2797 | */ | |
44871b1b CM |
2798 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
2799 | struct btrfs_root *root, | |
2800 | struct btrfs_path *path, | |
2801 | int data_size, int empty, | |
2802 | struct extent_buffer *right, | |
99d8f83c CM |
2803 | int free_space, u32 left_nritems, |
2804 | u32 min_slot) | |
00ec4c51 | 2805 | { |
5f39d397 | 2806 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2807 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 2808 | struct btrfs_map_token token; |
5f39d397 | 2809 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2810 | int slot; |
34a38218 | 2811 | u32 i; |
00ec4c51 CM |
2812 | int push_space = 0; |
2813 | int push_items = 0; | |
0783fcfc | 2814 | struct btrfs_item *item; |
34a38218 | 2815 | u32 nr; |
7518a238 | 2816 | u32 right_nritems; |
5f39d397 | 2817 | u32 data_end; |
db94535d | 2818 | u32 this_item_size; |
00ec4c51 | 2819 | |
cfed81a0 CM |
2820 | btrfs_init_map_token(&token); |
2821 | ||
34a38218 CM |
2822 | if (empty) |
2823 | nr = 0; | |
2824 | else | |
99d8f83c | 2825 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2826 | |
31840ae1 | 2827 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2828 | push_space += data_size; |
31840ae1 | 2829 | |
44871b1b | 2830 | slot = path->slots[1]; |
34a38218 CM |
2831 | i = left_nritems - 1; |
2832 | while (i >= nr) { | |
5f39d397 | 2833 | item = btrfs_item_nr(left, i); |
db94535d | 2834 | |
31840ae1 ZY |
2835 | if (!empty && push_items > 0) { |
2836 | if (path->slots[0] > i) | |
2837 | break; | |
2838 | if (path->slots[0] == i) { | |
2839 | int space = btrfs_leaf_free_space(root, left); | |
2840 | if (space + push_space * 2 > free_space) | |
2841 | break; | |
2842 | } | |
2843 | } | |
2844 | ||
00ec4c51 | 2845 | if (path->slots[0] == i) |
87b29b20 | 2846 | push_space += data_size; |
db94535d | 2847 | |
db94535d CM |
2848 | this_item_size = btrfs_item_size(left, item); |
2849 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 2850 | break; |
31840ae1 | 2851 | |
00ec4c51 | 2852 | push_items++; |
db94535d | 2853 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2854 | if (i == 0) |
2855 | break; | |
2856 | i--; | |
db94535d | 2857 | } |
5f39d397 | 2858 | |
925baedd CM |
2859 | if (push_items == 0) |
2860 | goto out_unlock; | |
5f39d397 | 2861 | |
34a38218 | 2862 | if (!empty && push_items == left_nritems) |
a429e513 | 2863 | WARN_ON(1); |
5f39d397 | 2864 | |
00ec4c51 | 2865 | /* push left to right */ |
5f39d397 | 2866 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2867 | |
5f39d397 | 2868 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 2869 | push_space -= leaf_data_end(root, left); |
5f39d397 | 2870 | |
00ec4c51 | 2871 | /* make room in the right data area */ |
5f39d397 CM |
2872 | data_end = leaf_data_end(root, right); |
2873 | memmove_extent_buffer(right, | |
2874 | btrfs_leaf_data(right) + data_end - push_space, | |
2875 | btrfs_leaf_data(right) + data_end, | |
2876 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
2877 | ||
00ec4c51 | 2878 | /* copy from the left data area */ |
5f39d397 | 2879 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
2880 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
2881 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
2882 | push_space); | |
5f39d397 CM |
2883 | |
2884 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2885 | btrfs_item_nr_offset(0), | |
2886 | right_nritems * sizeof(struct btrfs_item)); | |
2887 | ||
00ec4c51 | 2888 | /* copy the items from left to right */ |
5f39d397 CM |
2889 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2890 | btrfs_item_nr_offset(left_nritems - push_items), | |
2891 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2892 | |
2893 | /* update the item pointers */ | |
7518a238 | 2894 | right_nritems += push_items; |
5f39d397 | 2895 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 2896 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 2897 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 2898 | item = btrfs_item_nr(right, i); |
cfed81a0 CM |
2899 | push_space -= btrfs_token_item_size(right, item, &token); |
2900 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
2901 | } |
2902 | ||
7518a238 | 2903 | left_nritems -= push_items; |
5f39d397 | 2904 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2905 | |
34a38218 CM |
2906 | if (left_nritems) |
2907 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
2908 | else |
2909 | clean_tree_block(trans, root, left); | |
2910 | ||
5f39d397 | 2911 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2912 | |
5f39d397 CM |
2913 | btrfs_item_key(right, &disk_key, 0); |
2914 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2915 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2916 | |
00ec4c51 | 2917 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2918 | if (path->slots[0] >= left_nritems) { |
2919 | path->slots[0] -= left_nritems; | |
925baedd CM |
2920 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
2921 | clean_tree_block(trans, root, path->nodes[0]); | |
2922 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
2923 | free_extent_buffer(path->nodes[0]); |
2924 | path->nodes[0] = right; | |
00ec4c51 CM |
2925 | path->slots[1] += 1; |
2926 | } else { | |
925baedd | 2927 | btrfs_tree_unlock(right); |
5f39d397 | 2928 | free_extent_buffer(right); |
00ec4c51 CM |
2929 | } |
2930 | return 0; | |
925baedd CM |
2931 | |
2932 | out_unlock: | |
2933 | btrfs_tree_unlock(right); | |
2934 | free_extent_buffer(right); | |
2935 | return 1; | |
00ec4c51 | 2936 | } |
925baedd | 2937 | |
44871b1b CM |
2938 | /* |
2939 | * push some data in the path leaf to the right, trying to free up at | |
2940 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2941 | * | |
2942 | * returns 1 if the push failed because the other node didn't have enough | |
2943 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
2944 | * |
2945 | * this will push starting from min_slot to the end of the leaf. It won't | |
2946 | * push any slot lower than min_slot | |
44871b1b CM |
2947 | */ |
2948 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2949 | *root, struct btrfs_path *path, |
2950 | int min_data_size, int data_size, | |
2951 | int empty, u32 min_slot) | |
44871b1b CM |
2952 | { |
2953 | struct extent_buffer *left = path->nodes[0]; | |
2954 | struct extent_buffer *right; | |
2955 | struct extent_buffer *upper; | |
2956 | int slot; | |
2957 | int free_space; | |
2958 | u32 left_nritems; | |
2959 | int ret; | |
2960 | ||
2961 | if (!path->nodes[1]) | |
2962 | return 1; | |
2963 | ||
2964 | slot = path->slots[1]; | |
2965 | upper = path->nodes[1]; | |
2966 | if (slot >= btrfs_header_nritems(upper) - 1) | |
2967 | return 1; | |
2968 | ||
2969 | btrfs_assert_tree_locked(path->nodes[1]); | |
2970 | ||
2971 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
2972 | if (right == NULL) |
2973 | return 1; | |
2974 | ||
44871b1b CM |
2975 | btrfs_tree_lock(right); |
2976 | btrfs_set_lock_blocking(right); | |
2977 | ||
2978 | free_space = btrfs_leaf_free_space(root, right); | |
2979 | if (free_space < data_size) | |
2980 | goto out_unlock; | |
2981 | ||
2982 | /* cow and double check */ | |
2983 | ret = btrfs_cow_block(trans, root, right, upper, | |
2984 | slot + 1, &right); | |
2985 | if (ret) | |
2986 | goto out_unlock; | |
2987 | ||
2988 | free_space = btrfs_leaf_free_space(root, right); | |
2989 | if (free_space < data_size) | |
2990 | goto out_unlock; | |
2991 | ||
2992 | left_nritems = btrfs_header_nritems(left); | |
2993 | if (left_nritems == 0) | |
2994 | goto out_unlock; | |
2995 | ||
99d8f83c CM |
2996 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
2997 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
2998 | out_unlock: |
2999 | btrfs_tree_unlock(right); | |
3000 | free_extent_buffer(right); | |
3001 | return 1; | |
3002 | } | |
3003 | ||
74123bd7 CM |
3004 | /* |
3005 | * push some data in the path leaf to the left, trying to free up at | |
3006 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3007 | * |
3008 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3009 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3010 | * items | |
74123bd7 | 3011 | */ |
44871b1b CM |
3012 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3013 | struct btrfs_root *root, | |
3014 | struct btrfs_path *path, int data_size, | |
3015 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3016 | int free_space, u32 right_nritems, |
3017 | u32 max_slot) | |
be0e5c09 | 3018 | { |
5f39d397 CM |
3019 | struct btrfs_disk_key disk_key; |
3020 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3021 | int i; |
be0e5c09 CM |
3022 | int push_space = 0; |
3023 | int push_items = 0; | |
0783fcfc | 3024 | struct btrfs_item *item; |
7518a238 | 3025 | u32 old_left_nritems; |
34a38218 | 3026 | u32 nr; |
aa5d6bed | 3027 | int ret = 0; |
db94535d CM |
3028 | u32 this_item_size; |
3029 | u32 old_left_item_size; | |
cfed81a0 CM |
3030 | struct btrfs_map_token token; |
3031 | ||
3032 | btrfs_init_map_token(&token); | |
be0e5c09 | 3033 | |
34a38218 | 3034 | if (empty) |
99d8f83c | 3035 | nr = min(right_nritems, max_slot); |
34a38218 | 3036 | else |
99d8f83c | 3037 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3038 | |
3039 | for (i = 0; i < nr; i++) { | |
5f39d397 | 3040 | item = btrfs_item_nr(right, i); |
db94535d | 3041 | |
31840ae1 ZY |
3042 | if (!empty && push_items > 0) { |
3043 | if (path->slots[0] < i) | |
3044 | break; | |
3045 | if (path->slots[0] == i) { | |
3046 | int space = btrfs_leaf_free_space(root, right); | |
3047 | if (space + push_space * 2 > free_space) | |
3048 | break; | |
3049 | } | |
3050 | } | |
3051 | ||
be0e5c09 | 3052 | if (path->slots[0] == i) |
87b29b20 | 3053 | push_space += data_size; |
db94535d CM |
3054 | |
3055 | this_item_size = btrfs_item_size(right, item); | |
3056 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3057 | break; |
db94535d | 3058 | |
be0e5c09 | 3059 | push_items++; |
db94535d CM |
3060 | push_space += this_item_size + sizeof(*item); |
3061 | } | |
3062 | ||
be0e5c09 | 3063 | if (push_items == 0) { |
925baedd CM |
3064 | ret = 1; |
3065 | goto out; | |
be0e5c09 | 3066 | } |
34a38218 | 3067 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 3068 | WARN_ON(1); |
5f39d397 | 3069 | |
be0e5c09 | 3070 | /* push data from right to left */ |
5f39d397 CM |
3071 | copy_extent_buffer(left, right, |
3072 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3073 | btrfs_item_nr_offset(0), | |
3074 | push_items * sizeof(struct btrfs_item)); | |
3075 | ||
123abc88 | 3076 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3077 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3078 | |
3079 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3080 | leaf_data_end(root, left) - push_space, |
3081 | btrfs_leaf_data(right) + | |
5f39d397 | 3082 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3083 | push_space); |
5f39d397 | 3084 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3085 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3086 | |
db94535d | 3087 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3088 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3089 | u32 ioff; |
db94535d | 3090 | |
5f39d397 | 3091 | item = btrfs_item_nr(left, i); |
db94535d | 3092 | |
cfed81a0 CM |
3093 | ioff = btrfs_token_item_offset(left, item, &token); |
3094 | btrfs_set_token_item_offset(left, item, | |
3095 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3096 | &token); | |
be0e5c09 | 3097 | } |
5f39d397 | 3098 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3099 | |
3100 | /* fixup right node */ | |
34a38218 | 3101 | if (push_items > right_nritems) { |
d397712b CM |
3102 | printk(KERN_CRIT "push items %d nr %u\n", push_items, |
3103 | right_nritems); | |
34a38218 CM |
3104 | WARN_ON(1); |
3105 | } | |
3106 | ||
3107 | if (push_items < right_nritems) { | |
3108 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3109 | leaf_data_end(root, right); | |
3110 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3111 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3112 | btrfs_leaf_data(right) + | |
3113 | leaf_data_end(root, right), push_space); | |
3114 | ||
3115 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3116 | btrfs_item_nr_offset(push_items), |
3117 | (btrfs_header_nritems(right) - push_items) * | |
3118 | sizeof(struct btrfs_item)); | |
34a38218 | 3119 | } |
eef1c494 Y |
3120 | right_nritems -= push_items; |
3121 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3122 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
3123 | for (i = 0; i < right_nritems; i++) { |
3124 | item = btrfs_item_nr(right, i); | |
db94535d | 3125 | |
cfed81a0 CM |
3126 | push_space = push_space - btrfs_token_item_size(right, |
3127 | item, &token); | |
3128 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3129 | } |
eb60ceac | 3130 | |
5f39d397 | 3131 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3132 | if (right_nritems) |
3133 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3134 | else |
3135 | clean_tree_block(trans, root, right); | |
098f59c2 | 3136 | |
5f39d397 | 3137 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3138 | fixup_low_keys(trans, root, path, &disk_key, 1); |
be0e5c09 CM |
3139 | |
3140 | /* then fixup the leaf pointer in the path */ | |
3141 | if (path->slots[0] < push_items) { | |
3142 | path->slots[0] += old_left_nritems; | |
925baedd | 3143 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3144 | free_extent_buffer(path->nodes[0]); |
3145 | path->nodes[0] = left; | |
be0e5c09 CM |
3146 | path->slots[1] -= 1; |
3147 | } else { | |
925baedd | 3148 | btrfs_tree_unlock(left); |
5f39d397 | 3149 | free_extent_buffer(left); |
be0e5c09 CM |
3150 | path->slots[0] -= push_items; |
3151 | } | |
eb60ceac | 3152 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3153 | return ret; |
925baedd CM |
3154 | out: |
3155 | btrfs_tree_unlock(left); | |
3156 | free_extent_buffer(left); | |
3157 | return ret; | |
be0e5c09 CM |
3158 | } |
3159 | ||
44871b1b CM |
3160 | /* |
3161 | * push some data in the path leaf to the left, trying to free up at | |
3162 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3163 | * |
3164 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3165 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3166 | * items | |
44871b1b CM |
3167 | */ |
3168 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3169 | *root, struct btrfs_path *path, int min_data_size, |
3170 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3171 | { |
3172 | struct extent_buffer *right = path->nodes[0]; | |
3173 | struct extent_buffer *left; | |
3174 | int slot; | |
3175 | int free_space; | |
3176 | u32 right_nritems; | |
3177 | int ret = 0; | |
3178 | ||
3179 | slot = path->slots[1]; | |
3180 | if (slot == 0) | |
3181 | return 1; | |
3182 | if (!path->nodes[1]) | |
3183 | return 1; | |
3184 | ||
3185 | right_nritems = btrfs_header_nritems(right); | |
3186 | if (right_nritems == 0) | |
3187 | return 1; | |
3188 | ||
3189 | btrfs_assert_tree_locked(path->nodes[1]); | |
3190 | ||
3191 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
3192 | if (left == NULL) |
3193 | return 1; | |
3194 | ||
44871b1b CM |
3195 | btrfs_tree_lock(left); |
3196 | btrfs_set_lock_blocking(left); | |
3197 | ||
3198 | free_space = btrfs_leaf_free_space(root, left); | |
3199 | if (free_space < data_size) { | |
3200 | ret = 1; | |
3201 | goto out; | |
3202 | } | |
3203 | ||
3204 | /* cow and double check */ | |
3205 | ret = btrfs_cow_block(trans, root, left, | |
3206 | path->nodes[1], slot - 1, &left); | |
3207 | if (ret) { | |
3208 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3209 | if (ret == -ENOSPC) |
3210 | ret = 1; | |
44871b1b CM |
3211 | goto out; |
3212 | } | |
3213 | ||
3214 | free_space = btrfs_leaf_free_space(root, left); | |
3215 | if (free_space < data_size) { | |
3216 | ret = 1; | |
3217 | goto out; | |
3218 | } | |
3219 | ||
99d8f83c CM |
3220 | return __push_leaf_left(trans, root, path, min_data_size, |
3221 | empty, left, free_space, right_nritems, | |
3222 | max_slot); | |
44871b1b CM |
3223 | out: |
3224 | btrfs_tree_unlock(left); | |
3225 | free_extent_buffer(left); | |
3226 | return ret; | |
3227 | } | |
3228 | ||
3229 | /* | |
3230 | * split the path's leaf in two, making sure there is at least data_size | |
3231 | * available for the resulting leaf level of the path. | |
44871b1b | 3232 | */ |
143bede5 JM |
3233 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
3234 | struct btrfs_root *root, | |
3235 | struct btrfs_path *path, | |
3236 | struct extent_buffer *l, | |
3237 | struct extent_buffer *right, | |
3238 | int slot, int mid, int nritems) | |
44871b1b CM |
3239 | { |
3240 | int data_copy_size; | |
3241 | int rt_data_off; | |
3242 | int i; | |
44871b1b | 3243 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3244 | struct btrfs_map_token token; |
3245 | ||
3246 | btrfs_init_map_token(&token); | |
44871b1b CM |
3247 | |
3248 | nritems = nritems - mid; | |
3249 | btrfs_set_header_nritems(right, nritems); | |
3250 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
3251 | ||
3252 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3253 | btrfs_item_nr_offset(mid), | |
3254 | nritems * sizeof(struct btrfs_item)); | |
3255 | ||
3256 | copy_extent_buffer(right, l, | |
3257 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
3258 | data_copy_size, btrfs_leaf_data(l) + | |
3259 | leaf_data_end(root, l), data_copy_size); | |
3260 | ||
3261 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
3262 | btrfs_item_end_nr(l, mid); | |
3263 | ||
3264 | for (i = 0; i < nritems; i++) { | |
3265 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
3266 | u32 ioff; | |
3267 | ||
cfed81a0 CM |
3268 | ioff = btrfs_token_item_offset(right, item, &token); |
3269 | btrfs_set_token_item_offset(right, item, | |
3270 | ioff + rt_data_off, &token); | |
44871b1b CM |
3271 | } |
3272 | ||
44871b1b | 3273 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3274 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3275 | insert_ptr(trans, root, path, &disk_key, right->start, |
f3ea38da | 3276 | path->slots[1] + 1, 1, 0); |
44871b1b CM |
3277 | |
3278 | btrfs_mark_buffer_dirty(right); | |
3279 | btrfs_mark_buffer_dirty(l); | |
3280 | BUG_ON(path->slots[0] != slot); | |
3281 | ||
44871b1b CM |
3282 | if (mid <= slot) { |
3283 | btrfs_tree_unlock(path->nodes[0]); | |
3284 | free_extent_buffer(path->nodes[0]); | |
3285 | path->nodes[0] = right; | |
3286 | path->slots[0] -= mid; | |
3287 | path->slots[1] += 1; | |
3288 | } else { | |
3289 | btrfs_tree_unlock(right); | |
3290 | free_extent_buffer(right); | |
3291 | } | |
3292 | ||
3293 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3294 | } |
3295 | ||
99d8f83c CM |
3296 | /* |
3297 | * double splits happen when we need to insert a big item in the middle | |
3298 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3299 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3300 | * A B C | |
3301 | * | |
3302 | * We avoid this by trying to push the items on either side of our target | |
3303 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3304 | * completely. | |
3305 | */ | |
3306 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3307 | struct btrfs_root *root, | |
3308 | struct btrfs_path *path, | |
3309 | int data_size) | |
3310 | { | |
3311 | int ret; | |
3312 | int progress = 0; | |
3313 | int slot; | |
3314 | u32 nritems; | |
3315 | ||
3316 | slot = path->slots[0]; | |
3317 | ||
3318 | /* | |
3319 | * try to push all the items after our slot into the | |
3320 | * right leaf | |
3321 | */ | |
3322 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
3323 | if (ret < 0) | |
3324 | return ret; | |
3325 | ||
3326 | if (ret == 0) | |
3327 | progress++; | |
3328 | ||
3329 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3330 | /* | |
3331 | * our goal is to get our slot at the start or end of a leaf. If | |
3332 | * we've done so we're done | |
3333 | */ | |
3334 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3335 | return 0; | |
3336 | ||
3337 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3338 | return 0; | |
3339 | ||
3340 | /* try to push all the items before our slot into the next leaf */ | |
3341 | slot = path->slots[0]; | |
3342 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
3343 | if (ret < 0) | |
3344 | return ret; | |
3345 | ||
3346 | if (ret == 0) | |
3347 | progress++; | |
3348 | ||
3349 | if (progress) | |
3350 | return 0; | |
3351 | return 1; | |
3352 | } | |
3353 | ||
74123bd7 CM |
3354 | /* |
3355 | * split the path's leaf in two, making sure there is at least data_size | |
3356 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3357 | * |
3358 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3359 | */ |
e02119d5 CM |
3360 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3361 | struct btrfs_root *root, | |
3362 | struct btrfs_key *ins_key, | |
3363 | struct btrfs_path *path, int data_size, | |
3364 | int extend) | |
be0e5c09 | 3365 | { |
5d4f98a2 | 3366 | struct btrfs_disk_key disk_key; |
5f39d397 | 3367 | struct extent_buffer *l; |
7518a238 | 3368 | u32 nritems; |
eb60ceac CM |
3369 | int mid; |
3370 | int slot; | |
5f39d397 | 3371 | struct extent_buffer *right; |
d4dbff95 | 3372 | int ret = 0; |
aa5d6bed | 3373 | int wret; |
5d4f98a2 | 3374 | int split; |
cc0c5538 | 3375 | int num_doubles = 0; |
99d8f83c | 3376 | int tried_avoid_double = 0; |
aa5d6bed | 3377 | |
a5719521 YZ |
3378 | l = path->nodes[0]; |
3379 | slot = path->slots[0]; | |
3380 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
3381 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
3382 | return -EOVERFLOW; | |
3383 | ||
40689478 | 3384 | /* first try to make some room by pushing left and right */ |
99d8f83c CM |
3385 | if (data_size) { |
3386 | wret = push_leaf_right(trans, root, path, data_size, | |
3387 | data_size, 0, 0); | |
d397712b | 3388 | if (wret < 0) |
eaee50e8 | 3389 | return wret; |
3685f791 | 3390 | if (wret) { |
99d8f83c CM |
3391 | wret = push_leaf_left(trans, root, path, data_size, |
3392 | data_size, 0, (u32)-1); | |
3685f791 CM |
3393 | if (wret < 0) |
3394 | return wret; | |
3395 | } | |
3396 | l = path->nodes[0]; | |
aa5d6bed | 3397 | |
3685f791 | 3398 | /* did the pushes work? */ |
87b29b20 | 3399 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 3400 | return 0; |
3326d1b0 | 3401 | } |
aa5d6bed | 3402 | |
5c680ed6 | 3403 | if (!path->nodes[1]) { |
e089f05c | 3404 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3405 | if (ret) |
3406 | return ret; | |
3407 | } | |
cc0c5538 | 3408 | again: |
5d4f98a2 | 3409 | split = 1; |
cc0c5538 | 3410 | l = path->nodes[0]; |
eb60ceac | 3411 | slot = path->slots[0]; |
5f39d397 | 3412 | nritems = btrfs_header_nritems(l); |
d397712b | 3413 | mid = (nritems + 1) / 2; |
54aa1f4d | 3414 | |
5d4f98a2 YZ |
3415 | if (mid <= slot) { |
3416 | if (nritems == 1 || | |
3417 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
3418 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3419 | if (slot >= nritems) { | |
3420 | split = 0; | |
3421 | } else { | |
3422 | mid = slot; | |
3423 | if (mid != nritems && | |
3424 | leaf_space_used(l, mid, nritems - mid) + | |
3425 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3426 | if (data_size && !tried_avoid_double) |
3427 | goto push_for_double; | |
5d4f98a2 YZ |
3428 | split = 2; |
3429 | } | |
3430 | } | |
3431 | } | |
3432 | } else { | |
3433 | if (leaf_space_used(l, 0, mid) + data_size > | |
3434 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3435 | if (!extend && data_size && slot == 0) { | |
3436 | split = 0; | |
3437 | } else if ((extend || !data_size) && slot == 0) { | |
3438 | mid = 1; | |
3439 | } else { | |
3440 | mid = slot; | |
3441 | if (mid != nritems && | |
3442 | leaf_space_used(l, mid, nritems - mid) + | |
3443 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3444 | if (data_size && !tried_avoid_double) |
3445 | goto push_for_double; | |
5d4f98a2 YZ |
3446 | split = 2 ; |
3447 | } | |
3448 | } | |
3449 | } | |
3450 | } | |
3451 | ||
3452 | if (split == 0) | |
3453 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3454 | else | |
3455 | btrfs_item_key(l, &disk_key, mid); | |
3456 | ||
3457 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 3458 | root->root_key.objectid, |
5581a51a | 3459 | &disk_key, 0, l->start, 0); |
f0486c68 | 3460 | if (IS_ERR(right)) |
5f39d397 | 3461 | return PTR_ERR(right); |
f0486c68 YZ |
3462 | |
3463 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
3464 | |
3465 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 3466 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 3467 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 3468 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3469 | btrfs_set_header_owner(right, root->root_key.objectid); |
3470 | btrfs_set_header_level(right, 0); | |
3471 | write_extent_buffer(right, root->fs_info->fsid, | |
3472 | (unsigned long)btrfs_header_fsid(right), | |
3473 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3474 | |
3475 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
3476 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
3477 | BTRFS_UUID_SIZE); | |
44871b1b | 3478 | |
5d4f98a2 YZ |
3479 | if (split == 0) { |
3480 | if (mid <= slot) { | |
3481 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 3482 | insert_ptr(trans, root, path, &disk_key, right->start, |
f3ea38da | 3483 | path->slots[1] + 1, 1, 0); |
5d4f98a2 YZ |
3484 | btrfs_tree_unlock(path->nodes[0]); |
3485 | free_extent_buffer(path->nodes[0]); | |
3486 | path->nodes[0] = right; | |
3487 | path->slots[0] = 0; | |
3488 | path->slots[1] += 1; | |
3489 | } else { | |
3490 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 3491 | insert_ptr(trans, root, path, &disk_key, right->start, |
f3ea38da | 3492 | path->slots[1], 1, 0); |
5d4f98a2 YZ |
3493 | btrfs_tree_unlock(path->nodes[0]); |
3494 | free_extent_buffer(path->nodes[0]); | |
3495 | path->nodes[0] = right; | |
3496 | path->slots[0] = 0; | |
143bede5 JM |
3497 | if (path->slots[1] == 0) |
3498 | fixup_low_keys(trans, root, path, | |
3499 | &disk_key, 1); | |
d4dbff95 | 3500 | } |
5d4f98a2 YZ |
3501 | btrfs_mark_buffer_dirty(right); |
3502 | return ret; | |
d4dbff95 | 3503 | } |
74123bd7 | 3504 | |
143bede5 | 3505 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 3506 | |
5d4f98a2 | 3507 | if (split == 2) { |
cc0c5538 CM |
3508 | BUG_ON(num_doubles != 0); |
3509 | num_doubles++; | |
3510 | goto again; | |
a429e513 | 3511 | } |
44871b1b | 3512 | |
143bede5 | 3513 | return 0; |
99d8f83c CM |
3514 | |
3515 | push_for_double: | |
3516 | push_for_double_split(trans, root, path, data_size); | |
3517 | tried_avoid_double = 1; | |
3518 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3519 | return 0; | |
3520 | goto again; | |
be0e5c09 CM |
3521 | } |
3522 | ||
ad48fd75 YZ |
3523 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3524 | struct btrfs_root *root, | |
3525 | struct btrfs_path *path, int ins_len) | |
459931ec | 3526 | { |
ad48fd75 | 3527 | struct btrfs_key key; |
459931ec | 3528 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3529 | struct btrfs_file_extent_item *fi; |
3530 | u64 extent_len = 0; | |
3531 | u32 item_size; | |
3532 | int ret; | |
459931ec CM |
3533 | |
3534 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3535 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3536 | ||
3537 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3538 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3539 | ||
3540 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
3541 | return 0; | |
459931ec CM |
3542 | |
3543 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
3544 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3545 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3546 | struct btrfs_file_extent_item); | |
3547 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3548 | } | |
b3b4aa74 | 3549 | btrfs_release_path(path); |
459931ec | 3550 | |
459931ec | 3551 | path->keep_locks = 1; |
ad48fd75 YZ |
3552 | path->search_for_split = 1; |
3553 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3554 | path->search_for_split = 0; |
ad48fd75 YZ |
3555 | if (ret < 0) |
3556 | goto err; | |
459931ec | 3557 | |
ad48fd75 YZ |
3558 | ret = -EAGAIN; |
3559 | leaf = path->nodes[0]; | |
459931ec | 3560 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
3561 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
3562 | goto err; | |
3563 | ||
109f6aef CM |
3564 | /* the leaf has changed, it now has room. return now */ |
3565 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
3566 | goto err; | |
3567 | ||
ad48fd75 YZ |
3568 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3569 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3570 | struct btrfs_file_extent_item); | |
3571 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3572 | goto err; | |
459931ec CM |
3573 | } |
3574 | ||
b9473439 | 3575 | btrfs_set_path_blocking(path); |
ad48fd75 | 3576 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3577 | if (ret) |
3578 | goto err; | |
459931ec | 3579 | |
ad48fd75 | 3580 | path->keep_locks = 0; |
b9473439 | 3581 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3582 | return 0; |
3583 | err: | |
3584 | path->keep_locks = 0; | |
3585 | return ret; | |
3586 | } | |
3587 | ||
3588 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
3589 | struct btrfs_root *root, | |
3590 | struct btrfs_path *path, | |
3591 | struct btrfs_key *new_key, | |
3592 | unsigned long split_offset) | |
3593 | { | |
3594 | struct extent_buffer *leaf; | |
3595 | struct btrfs_item *item; | |
3596 | struct btrfs_item *new_item; | |
3597 | int slot; | |
3598 | char *buf; | |
3599 | u32 nritems; | |
3600 | u32 item_size; | |
3601 | u32 orig_offset; | |
3602 | struct btrfs_disk_key disk_key; | |
3603 | ||
b9473439 CM |
3604 | leaf = path->nodes[0]; |
3605 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
3606 | ||
b4ce94de CM |
3607 | btrfs_set_path_blocking(path); |
3608 | ||
459931ec CM |
3609 | item = btrfs_item_nr(leaf, path->slots[0]); |
3610 | orig_offset = btrfs_item_offset(leaf, item); | |
3611 | item_size = btrfs_item_size(leaf, item); | |
3612 | ||
459931ec | 3613 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3614 | if (!buf) |
3615 | return -ENOMEM; | |
3616 | ||
459931ec CM |
3617 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3618 | path->slots[0]), item_size); | |
459931ec | 3619 | |
ad48fd75 | 3620 | slot = path->slots[0] + 1; |
459931ec | 3621 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3622 | if (slot != nritems) { |
3623 | /* shift the items */ | |
3624 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3625 | btrfs_item_nr_offset(slot), |
3626 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3627 | } |
3628 | ||
3629 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3630 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3631 | ||
3632 | new_item = btrfs_item_nr(leaf, slot); | |
3633 | ||
3634 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
3635 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
3636 | ||
3637 | btrfs_set_item_offset(leaf, item, | |
3638 | orig_offset + item_size - split_offset); | |
3639 | btrfs_set_item_size(leaf, item, split_offset); | |
3640 | ||
3641 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3642 | ||
3643 | /* write the data for the start of the original item */ | |
3644 | write_extent_buffer(leaf, buf, | |
3645 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3646 | split_offset); | |
3647 | ||
3648 | /* write the data for the new item */ | |
3649 | write_extent_buffer(leaf, buf + split_offset, | |
3650 | btrfs_item_ptr_offset(leaf, slot), | |
3651 | item_size - split_offset); | |
3652 | btrfs_mark_buffer_dirty(leaf); | |
3653 | ||
ad48fd75 | 3654 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 3655 | kfree(buf); |
ad48fd75 YZ |
3656 | return 0; |
3657 | } | |
3658 | ||
3659 | /* | |
3660 | * This function splits a single item into two items, | |
3661 | * giving 'new_key' to the new item and splitting the | |
3662 | * old one at split_offset (from the start of the item). | |
3663 | * | |
3664 | * The path may be released by this operation. After | |
3665 | * the split, the path is pointing to the old item. The | |
3666 | * new item is going to be in the same node as the old one. | |
3667 | * | |
3668 | * Note, the item being split must be smaller enough to live alone on | |
3669 | * a tree block with room for one extra struct btrfs_item | |
3670 | * | |
3671 | * This allows us to split the item in place, keeping a lock on the | |
3672 | * leaf the entire time. | |
3673 | */ | |
3674 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3675 | struct btrfs_root *root, | |
3676 | struct btrfs_path *path, | |
3677 | struct btrfs_key *new_key, | |
3678 | unsigned long split_offset) | |
3679 | { | |
3680 | int ret; | |
3681 | ret = setup_leaf_for_split(trans, root, path, | |
3682 | sizeof(struct btrfs_item)); | |
3683 | if (ret) | |
3684 | return ret; | |
3685 | ||
3686 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
3687 | return ret; |
3688 | } | |
3689 | ||
ad48fd75 YZ |
3690 | /* |
3691 | * This function duplicate a item, giving 'new_key' to the new item. | |
3692 | * It guarantees both items live in the same tree leaf and the new item | |
3693 | * is contiguous with the original item. | |
3694 | * | |
3695 | * This allows us to split file extent in place, keeping a lock on the | |
3696 | * leaf the entire time. | |
3697 | */ | |
3698 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
3699 | struct btrfs_root *root, | |
3700 | struct btrfs_path *path, | |
3701 | struct btrfs_key *new_key) | |
3702 | { | |
3703 | struct extent_buffer *leaf; | |
3704 | int ret; | |
3705 | u32 item_size; | |
3706 | ||
3707 | leaf = path->nodes[0]; | |
3708 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3709 | ret = setup_leaf_for_split(trans, root, path, | |
3710 | item_size + sizeof(struct btrfs_item)); | |
3711 | if (ret) | |
3712 | return ret; | |
3713 | ||
3714 | path->slots[0]++; | |
143bede5 JM |
3715 | setup_items_for_insert(trans, root, path, new_key, &item_size, |
3716 | item_size, item_size + | |
3717 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
3718 | leaf = path->nodes[0]; |
3719 | memcpy_extent_buffer(leaf, | |
3720 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3721 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
3722 | item_size); | |
3723 | return 0; | |
3724 | } | |
3725 | ||
d352ac68 CM |
3726 | /* |
3727 | * make the item pointed to by the path smaller. new_size indicates | |
3728 | * how small to make it, and from_end tells us if we just chop bytes | |
3729 | * off the end of the item or if we shift the item to chop bytes off | |
3730 | * the front. | |
3731 | */ | |
143bede5 JM |
3732 | void btrfs_truncate_item(struct btrfs_trans_handle *trans, |
3733 | struct btrfs_root *root, | |
3734 | struct btrfs_path *path, | |
3735 | u32 new_size, int from_end) | |
b18c6685 | 3736 | { |
b18c6685 | 3737 | int slot; |
5f39d397 CM |
3738 | struct extent_buffer *leaf; |
3739 | struct btrfs_item *item; | |
b18c6685 CM |
3740 | u32 nritems; |
3741 | unsigned int data_end; | |
3742 | unsigned int old_data_start; | |
3743 | unsigned int old_size; | |
3744 | unsigned int size_diff; | |
3745 | int i; | |
cfed81a0 CM |
3746 | struct btrfs_map_token token; |
3747 | ||
3748 | btrfs_init_map_token(&token); | |
b18c6685 | 3749 | |
5f39d397 | 3750 | leaf = path->nodes[0]; |
179e29e4 CM |
3751 | slot = path->slots[0]; |
3752 | ||
3753 | old_size = btrfs_item_size_nr(leaf, slot); | |
3754 | if (old_size == new_size) | |
143bede5 | 3755 | return; |
b18c6685 | 3756 | |
5f39d397 | 3757 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
3758 | data_end = leaf_data_end(root, leaf); |
3759 | ||
5f39d397 | 3760 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 3761 | |
b18c6685 CM |
3762 | size_diff = old_size - new_size; |
3763 | ||
3764 | BUG_ON(slot < 0); | |
3765 | BUG_ON(slot >= nritems); | |
3766 | ||
3767 | /* | |
3768 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3769 | */ | |
3770 | /* first correct the data pointers */ | |
3771 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3772 | u32 ioff; |
3773 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3774 | |
cfed81a0 CM |
3775 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3776 | btrfs_set_token_item_offset(leaf, item, | |
3777 | ioff + size_diff, &token); | |
b18c6685 | 3778 | } |
db94535d | 3779 | |
b18c6685 | 3780 | /* shift the data */ |
179e29e4 CM |
3781 | if (from_end) { |
3782 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3783 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3784 | data_end, old_data_start + new_size - data_end); | |
3785 | } else { | |
3786 | struct btrfs_disk_key disk_key; | |
3787 | u64 offset; | |
3788 | ||
3789 | btrfs_item_key(leaf, &disk_key, slot); | |
3790 | ||
3791 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3792 | unsigned long ptr; | |
3793 | struct btrfs_file_extent_item *fi; | |
3794 | ||
3795 | fi = btrfs_item_ptr(leaf, slot, | |
3796 | struct btrfs_file_extent_item); | |
3797 | fi = (struct btrfs_file_extent_item *)( | |
3798 | (unsigned long)fi - size_diff); | |
3799 | ||
3800 | if (btrfs_file_extent_type(leaf, fi) == | |
3801 | BTRFS_FILE_EXTENT_INLINE) { | |
3802 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3803 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
3804 | (unsigned long)fi, |
3805 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
3806 | disk_bytenr)); |
3807 | } | |
3808 | } | |
3809 | ||
3810 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
3811 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
3812 | data_end, old_data_start - data_end); | |
3813 | ||
3814 | offset = btrfs_disk_key_offset(&disk_key); | |
3815 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3816 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3817 | if (slot == 0) | |
3818 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
3819 | } | |
5f39d397 CM |
3820 | |
3821 | item = btrfs_item_nr(leaf, slot); | |
3822 | btrfs_set_item_size(leaf, item, new_size); | |
3823 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 3824 | |
5f39d397 CM |
3825 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3826 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 3827 | BUG(); |
5f39d397 | 3828 | } |
b18c6685 CM |
3829 | } |
3830 | ||
d352ac68 CM |
3831 | /* |
3832 | * make the item pointed to by the path bigger, data_size is the new size. | |
3833 | */ | |
143bede5 JM |
3834 | void btrfs_extend_item(struct btrfs_trans_handle *trans, |
3835 | struct btrfs_root *root, struct btrfs_path *path, | |
3836 | u32 data_size) | |
6567e837 | 3837 | { |
6567e837 | 3838 | int slot; |
5f39d397 CM |
3839 | struct extent_buffer *leaf; |
3840 | struct btrfs_item *item; | |
6567e837 CM |
3841 | u32 nritems; |
3842 | unsigned int data_end; | |
3843 | unsigned int old_data; | |
3844 | unsigned int old_size; | |
3845 | int i; | |
cfed81a0 CM |
3846 | struct btrfs_map_token token; |
3847 | ||
3848 | btrfs_init_map_token(&token); | |
6567e837 | 3849 | |
5f39d397 | 3850 | leaf = path->nodes[0]; |
6567e837 | 3851 | |
5f39d397 | 3852 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
3853 | data_end = leaf_data_end(root, leaf); |
3854 | ||
5f39d397 CM |
3855 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
3856 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3857 | BUG(); |
5f39d397 | 3858 | } |
6567e837 | 3859 | slot = path->slots[0]; |
5f39d397 | 3860 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
3861 | |
3862 | BUG_ON(slot < 0); | |
3326d1b0 CM |
3863 | if (slot >= nritems) { |
3864 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
3865 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
3866 | slot, nritems); | |
3326d1b0 CM |
3867 | BUG_ON(1); |
3868 | } | |
6567e837 CM |
3869 | |
3870 | /* | |
3871 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3872 | */ | |
3873 | /* first correct the data pointers */ | |
3874 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
3875 | u32 ioff; |
3876 | item = btrfs_item_nr(leaf, i); | |
db94535d | 3877 | |
cfed81a0 CM |
3878 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3879 | btrfs_set_token_item_offset(leaf, item, | |
3880 | ioff - data_size, &token); | |
6567e837 | 3881 | } |
5f39d397 | 3882 | |
6567e837 | 3883 | /* shift the data */ |
5f39d397 | 3884 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
3885 | data_end - data_size, btrfs_leaf_data(leaf) + |
3886 | data_end, old_data - data_end); | |
5f39d397 | 3887 | |
6567e837 | 3888 | data_end = old_data; |
5f39d397 CM |
3889 | old_size = btrfs_item_size_nr(leaf, slot); |
3890 | item = btrfs_item_nr(leaf, slot); | |
3891 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
3892 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 3893 | |
5f39d397 CM |
3894 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
3895 | btrfs_print_leaf(root, leaf); | |
6567e837 | 3896 | BUG(); |
5f39d397 | 3897 | } |
6567e837 CM |
3898 | } |
3899 | ||
f3465ca4 JB |
3900 | /* |
3901 | * Given a key and some data, insert items into the tree. | |
3902 | * This does all the path init required, making room in the tree if needed. | |
3903 | * Returns the number of keys that were inserted. | |
3904 | */ | |
3905 | int btrfs_insert_some_items(struct btrfs_trans_handle *trans, | |
3906 | struct btrfs_root *root, | |
3907 | struct btrfs_path *path, | |
3908 | struct btrfs_key *cpu_key, u32 *data_size, | |
3909 | int nr) | |
3910 | { | |
3911 | struct extent_buffer *leaf; | |
3912 | struct btrfs_item *item; | |
3913 | int ret = 0; | |
3914 | int slot; | |
f3465ca4 JB |
3915 | int i; |
3916 | u32 nritems; | |
3917 | u32 total_data = 0; | |
3918 | u32 total_size = 0; | |
3919 | unsigned int data_end; | |
3920 | struct btrfs_disk_key disk_key; | |
3921 | struct btrfs_key found_key; | |
cfed81a0 CM |
3922 | struct btrfs_map_token token; |
3923 | ||
3924 | btrfs_init_map_token(&token); | |
f3465ca4 | 3925 | |
87b29b20 YZ |
3926 | for (i = 0; i < nr; i++) { |
3927 | if (total_size + data_size[i] + sizeof(struct btrfs_item) > | |
3928 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3929 | break; | |
3930 | nr = i; | |
3931 | } | |
f3465ca4 | 3932 | total_data += data_size[i]; |
87b29b20 YZ |
3933 | total_size += data_size[i] + sizeof(struct btrfs_item); |
3934 | } | |
3935 | BUG_ON(nr == 0); | |
f3465ca4 | 3936 | |
f3465ca4 JB |
3937 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); |
3938 | if (ret == 0) | |
3939 | return -EEXIST; | |
3940 | if (ret < 0) | |
3941 | goto out; | |
3942 | ||
f3465ca4 JB |
3943 | leaf = path->nodes[0]; |
3944 | ||
3945 | nritems = btrfs_header_nritems(leaf); | |
3946 | data_end = leaf_data_end(root, leaf); | |
3947 | ||
3948 | if (btrfs_leaf_free_space(root, leaf) < total_size) { | |
3949 | for (i = nr; i >= 0; i--) { | |
3950 | total_data -= data_size[i]; | |
3951 | total_size -= data_size[i] + sizeof(struct btrfs_item); | |
3952 | if (total_size < btrfs_leaf_free_space(root, leaf)) | |
3953 | break; | |
3954 | } | |
3955 | nr = i; | |
3956 | } | |
3957 | ||
3958 | slot = path->slots[0]; | |
3959 | BUG_ON(slot < 0); | |
3960 | ||
3961 | if (slot != nritems) { | |
3962 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); | |
3963 | ||
3964 | item = btrfs_item_nr(leaf, slot); | |
3965 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3966 | ||
3967 | /* figure out how many keys we can insert in here */ | |
3968 | total_data = data_size[0]; | |
3969 | for (i = 1; i < nr; i++) { | |
5d4f98a2 | 3970 | if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0) |
f3465ca4 JB |
3971 | break; |
3972 | total_data += data_size[i]; | |
3973 | } | |
3974 | nr = i; | |
3975 | ||
3976 | if (old_data < data_end) { | |
3977 | btrfs_print_leaf(root, leaf); | |
d397712b | 3978 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
f3465ca4 JB |
3979 | slot, old_data, data_end); |
3980 | BUG_ON(1); | |
3981 | } | |
3982 | /* | |
3983 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3984 | */ | |
3985 | /* first correct the data pointers */ | |
f3465ca4 JB |
3986 | for (i = slot; i < nritems; i++) { |
3987 | u32 ioff; | |
3988 | ||
3989 | item = btrfs_item_nr(leaf, i); | |
cfed81a0 CM |
3990 | ioff = btrfs_token_item_offset(leaf, item, &token); |
3991 | btrfs_set_token_item_offset(leaf, item, | |
3992 | ioff - total_data, &token); | |
f3465ca4 | 3993 | } |
f3465ca4 JB |
3994 | /* shift the items */ |
3995 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), | |
3996 | btrfs_item_nr_offset(slot), | |
3997 | (nritems - slot) * sizeof(struct btrfs_item)); | |
3998 | ||
3999 | /* shift the data */ | |
4000 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4001 | data_end - total_data, btrfs_leaf_data(leaf) + | |
4002 | data_end, old_data - data_end); | |
4003 | data_end = old_data; | |
4004 | } else { | |
4005 | /* | |
4006 | * this sucks but it has to be done, if we are inserting at | |
4007 | * the end of the leaf only insert 1 of the items, since we | |
4008 | * have no way of knowing whats on the next leaf and we'd have | |
4009 | * to drop our current locks to figure it out | |
4010 | */ | |
4011 | nr = 1; | |
4012 | } | |
4013 | ||
4014 | /* setup the item for the new data */ | |
4015 | for (i = 0; i < nr; i++) { | |
4016 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4017 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
4018 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
4019 | btrfs_set_token_item_offset(leaf, item, |
4020 | data_end - data_size[i], &token); | |
f3465ca4 | 4021 | data_end -= data_size[i]; |
cfed81a0 | 4022 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
f3465ca4 JB |
4023 | } |
4024 | btrfs_set_header_nritems(leaf, nritems + nr); | |
4025 | btrfs_mark_buffer_dirty(leaf); | |
4026 | ||
4027 | ret = 0; | |
4028 | if (slot == 0) { | |
4029 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 4030 | fixup_low_keys(trans, root, path, &disk_key, 1); |
f3465ca4 JB |
4031 | } |
4032 | ||
4033 | if (btrfs_leaf_free_space(root, leaf) < 0) { | |
4034 | btrfs_print_leaf(root, leaf); | |
4035 | BUG(); | |
4036 | } | |
4037 | out: | |
4038 | if (!ret) | |
4039 | ret = nr; | |
4040 | return ret; | |
4041 | } | |
4042 | ||
74123bd7 | 4043 | /* |
44871b1b CM |
4044 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4045 | * to save stack depth by doing the bulk of the work in a function | |
4046 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4047 | */ |
143bede5 JM |
4048 | void setup_items_for_insert(struct btrfs_trans_handle *trans, |
4049 | struct btrfs_root *root, struct btrfs_path *path, | |
4050 | struct btrfs_key *cpu_key, u32 *data_size, | |
4051 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4052 | { |
5f39d397 | 4053 | struct btrfs_item *item; |
9c58309d | 4054 | int i; |
7518a238 | 4055 | u32 nritems; |
be0e5c09 | 4056 | unsigned int data_end; |
e2fa7227 | 4057 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4058 | struct extent_buffer *leaf; |
4059 | int slot; | |
cfed81a0 CM |
4060 | struct btrfs_map_token token; |
4061 | ||
4062 | btrfs_init_map_token(&token); | |
e2fa7227 | 4063 | |
5f39d397 | 4064 | leaf = path->nodes[0]; |
44871b1b | 4065 | slot = path->slots[0]; |
74123bd7 | 4066 | |
5f39d397 | 4067 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4068 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4069 | |
f25956cc | 4070 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4071 | btrfs_print_leaf(root, leaf); |
d397712b | 4072 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 4073 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4074 | BUG(); |
d4dbff95 | 4075 | } |
5f39d397 | 4076 | |
be0e5c09 | 4077 | if (slot != nritems) { |
5f39d397 | 4078 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4079 | |
5f39d397 CM |
4080 | if (old_data < data_end) { |
4081 | btrfs_print_leaf(root, leaf); | |
d397712b | 4082 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
4083 | slot, old_data, data_end); |
4084 | BUG_ON(1); | |
4085 | } | |
be0e5c09 CM |
4086 | /* |
4087 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4088 | */ | |
4089 | /* first correct the data pointers */ | |
0783fcfc | 4090 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4091 | u32 ioff; |
db94535d | 4092 | |
5f39d397 | 4093 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4094 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4095 | btrfs_set_token_item_offset(leaf, item, | |
4096 | ioff - total_data, &token); | |
0783fcfc | 4097 | } |
be0e5c09 | 4098 | /* shift the items */ |
9c58309d | 4099 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4100 | btrfs_item_nr_offset(slot), |
d6025579 | 4101 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4102 | |
4103 | /* shift the data */ | |
5f39d397 | 4104 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4105 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4106 | data_end, old_data - data_end); |
be0e5c09 CM |
4107 | data_end = old_data; |
4108 | } | |
5f39d397 | 4109 | |
62e2749e | 4110 | /* setup the item for the new data */ |
9c58309d CM |
4111 | for (i = 0; i < nr; i++) { |
4112 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4113 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
4114 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
4115 | btrfs_set_token_item_offset(leaf, item, |
4116 | data_end - data_size[i], &token); | |
9c58309d | 4117 | data_end -= data_size[i]; |
cfed81a0 | 4118 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4119 | } |
44871b1b | 4120 | |
9c58309d | 4121 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4122 | |
5a01a2e3 CM |
4123 | if (slot == 0) { |
4124 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 4125 | fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 4126 | } |
b9473439 CM |
4127 | btrfs_unlock_up_safe(path, 1); |
4128 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4129 | |
5f39d397 CM |
4130 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4131 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4132 | BUG(); |
5f39d397 | 4133 | } |
44871b1b CM |
4134 | } |
4135 | ||
4136 | /* | |
4137 | * Given a key and some data, insert items into the tree. | |
4138 | * This does all the path init required, making room in the tree if needed. | |
4139 | */ | |
4140 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4141 | struct btrfs_root *root, | |
4142 | struct btrfs_path *path, | |
4143 | struct btrfs_key *cpu_key, u32 *data_size, | |
4144 | int nr) | |
4145 | { | |
44871b1b CM |
4146 | int ret = 0; |
4147 | int slot; | |
4148 | int i; | |
4149 | u32 total_size = 0; | |
4150 | u32 total_data = 0; | |
4151 | ||
4152 | for (i = 0; i < nr; i++) | |
4153 | total_data += data_size[i]; | |
4154 | ||
4155 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4156 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4157 | if (ret == 0) | |
4158 | return -EEXIST; | |
4159 | if (ret < 0) | |
143bede5 | 4160 | return ret; |
44871b1b | 4161 | |
44871b1b CM |
4162 | slot = path->slots[0]; |
4163 | BUG_ON(slot < 0); | |
4164 | ||
143bede5 | 4165 | setup_items_for_insert(trans, root, path, cpu_key, data_size, |
44871b1b | 4166 | total_data, total_size, nr); |
143bede5 | 4167 | return 0; |
62e2749e CM |
4168 | } |
4169 | ||
4170 | /* | |
4171 | * Given a key and some data, insert an item into the tree. | |
4172 | * This does all the path init required, making room in the tree if needed. | |
4173 | */ | |
e089f05c CM |
4174 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4175 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4176 | data_size) | |
62e2749e CM |
4177 | { |
4178 | int ret = 0; | |
2c90e5d6 | 4179 | struct btrfs_path *path; |
5f39d397 CM |
4180 | struct extent_buffer *leaf; |
4181 | unsigned long ptr; | |
62e2749e | 4182 | |
2c90e5d6 | 4183 | path = btrfs_alloc_path(); |
db5b493a TI |
4184 | if (!path) |
4185 | return -ENOMEM; | |
2c90e5d6 | 4186 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4187 | if (!ret) { |
5f39d397 CM |
4188 | leaf = path->nodes[0]; |
4189 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4190 | write_extent_buffer(leaf, data, ptr, data_size); | |
4191 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4192 | } |
2c90e5d6 | 4193 | btrfs_free_path(path); |
aa5d6bed | 4194 | return ret; |
be0e5c09 CM |
4195 | } |
4196 | ||
74123bd7 | 4197 | /* |
5de08d7d | 4198 | * delete the pointer from a given node. |
74123bd7 | 4199 | * |
d352ac68 CM |
4200 | * the tree should have been previously balanced so the deletion does not |
4201 | * empty a node. | |
74123bd7 | 4202 | */ |
143bede5 | 4203 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
f3ea38da JS |
4204 | struct btrfs_path *path, int level, int slot, |
4205 | int tree_mod_log) | |
be0e5c09 | 4206 | { |
5f39d397 | 4207 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4208 | u32 nritems; |
f3ea38da | 4209 | int ret; |
be0e5c09 | 4210 | |
5f39d397 | 4211 | nritems = btrfs_header_nritems(parent); |
d397712b | 4212 | if (slot != nritems - 1) { |
f3ea38da JS |
4213 | if (tree_mod_log && level) |
4214 | tree_mod_log_eb_move(root->fs_info, parent, slot, | |
4215 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4216 | memmove_extent_buffer(parent, |
4217 | btrfs_node_key_ptr_offset(slot), | |
4218 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4219 | sizeof(struct btrfs_key_ptr) * |
4220 | (nritems - slot - 1)); | |
bb803951 | 4221 | } |
f3ea38da JS |
4222 | |
4223 | if (tree_mod_log && level) { | |
4224 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
4225 | MOD_LOG_KEY_REMOVE); | |
4226 | BUG_ON(ret < 0); | |
4227 | } | |
4228 | ||
7518a238 | 4229 | nritems--; |
5f39d397 | 4230 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4231 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4232 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4233 | /* just turn the root into a leaf and break */ |
5f39d397 | 4234 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4235 | } else if (slot == 0) { |
5f39d397 CM |
4236 | struct btrfs_disk_key disk_key; |
4237 | ||
4238 | btrfs_node_key(parent, &disk_key, 0); | |
143bede5 | 4239 | fixup_low_keys(trans, root, path, &disk_key, level + 1); |
be0e5c09 | 4240 | } |
d6025579 | 4241 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4242 | } |
4243 | ||
323ac95b CM |
4244 | /* |
4245 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4246 | * path->nodes[1]. |
323ac95b CM |
4247 | * |
4248 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4249 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4250 | * | |
4251 | * The path must have already been setup for deleting the leaf, including | |
4252 | * all the proper balancing. path->nodes[1] must be locked. | |
4253 | */ | |
143bede5 JM |
4254 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4255 | struct btrfs_root *root, | |
4256 | struct btrfs_path *path, | |
4257 | struct extent_buffer *leaf) | |
323ac95b | 4258 | { |
5d4f98a2 | 4259 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
f3ea38da | 4260 | del_ptr(trans, root, path, 1, path->slots[1], 1); |
323ac95b | 4261 | |
4d081c41 CM |
4262 | /* |
4263 | * btrfs_free_extent is expensive, we want to make sure we | |
4264 | * aren't holding any locks when we call it | |
4265 | */ | |
4266 | btrfs_unlock_up_safe(path, 0); | |
4267 | ||
f0486c68 YZ |
4268 | root_sub_used(root, leaf->len); |
4269 | ||
3083ee2e | 4270 | extent_buffer_get(leaf); |
5581a51a | 4271 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4272 | free_extent_buffer_stale(leaf); |
323ac95b | 4273 | } |
74123bd7 CM |
4274 | /* |
4275 | * delete the item at the leaf level in path. If that empties | |
4276 | * the leaf, remove it from the tree | |
4277 | */ | |
85e21bac CM |
4278 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4279 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4280 | { |
5f39d397 CM |
4281 | struct extent_buffer *leaf; |
4282 | struct btrfs_item *item; | |
85e21bac CM |
4283 | int last_off; |
4284 | int dsize = 0; | |
aa5d6bed CM |
4285 | int ret = 0; |
4286 | int wret; | |
85e21bac | 4287 | int i; |
7518a238 | 4288 | u32 nritems; |
cfed81a0 CM |
4289 | struct btrfs_map_token token; |
4290 | ||
4291 | btrfs_init_map_token(&token); | |
be0e5c09 | 4292 | |
5f39d397 | 4293 | leaf = path->nodes[0]; |
85e21bac CM |
4294 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4295 | ||
4296 | for (i = 0; i < nr; i++) | |
4297 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4298 | ||
5f39d397 | 4299 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4300 | |
85e21bac | 4301 | if (slot + nr != nritems) { |
123abc88 | 4302 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4303 | |
4304 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4305 | data_end + dsize, |
4306 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4307 | last_off - data_end); |
5f39d397 | 4308 | |
85e21bac | 4309 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4310 | u32 ioff; |
db94535d | 4311 | |
5f39d397 | 4312 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4313 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4314 | btrfs_set_token_item_offset(leaf, item, | |
4315 | ioff + dsize, &token); | |
0783fcfc | 4316 | } |
db94535d | 4317 | |
5f39d397 | 4318 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4319 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4320 | sizeof(struct btrfs_item) * |
85e21bac | 4321 | (nritems - slot - nr)); |
be0e5c09 | 4322 | } |
85e21bac CM |
4323 | btrfs_set_header_nritems(leaf, nritems - nr); |
4324 | nritems -= nr; | |
5f39d397 | 4325 | |
74123bd7 | 4326 | /* delete the leaf if we've emptied it */ |
7518a238 | 4327 | if (nritems == 0) { |
5f39d397 CM |
4328 | if (leaf == root->node) { |
4329 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4330 | } else { |
f0486c68 YZ |
4331 | btrfs_set_path_blocking(path); |
4332 | clean_tree_block(trans, root, leaf); | |
143bede5 | 4333 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4334 | } |
be0e5c09 | 4335 | } else { |
7518a238 | 4336 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4337 | if (slot == 0) { |
5f39d397 CM |
4338 | struct btrfs_disk_key disk_key; |
4339 | ||
4340 | btrfs_item_key(leaf, &disk_key, 0); | |
143bede5 | 4341 | fixup_low_keys(trans, root, path, &disk_key, 1); |
aa5d6bed | 4342 | } |
aa5d6bed | 4343 | |
74123bd7 | 4344 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 4345 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
4346 | /* push_leaf_left fixes the path. |
4347 | * make sure the path still points to our leaf | |
4348 | * for possible call to del_ptr below | |
4349 | */ | |
4920c9ac | 4350 | slot = path->slots[1]; |
5f39d397 CM |
4351 | extent_buffer_get(leaf); |
4352 | ||
b9473439 | 4353 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4354 | wret = push_leaf_left(trans, root, path, 1, 1, |
4355 | 1, (u32)-1); | |
54aa1f4d | 4356 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4357 | ret = wret; |
5f39d397 CM |
4358 | |
4359 | if (path->nodes[0] == leaf && | |
4360 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4361 | wret = push_leaf_right(trans, root, path, 1, |
4362 | 1, 1, 0); | |
54aa1f4d | 4363 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4364 | ret = wret; |
4365 | } | |
5f39d397 CM |
4366 | |
4367 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4368 | path->slots[1] = slot; |
143bede5 | 4369 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4370 | free_extent_buffer(leaf); |
143bede5 | 4371 | ret = 0; |
5de08d7d | 4372 | } else { |
925baedd CM |
4373 | /* if we're still in the path, make sure |
4374 | * we're dirty. Otherwise, one of the | |
4375 | * push_leaf functions must have already | |
4376 | * dirtied this buffer | |
4377 | */ | |
4378 | if (path->nodes[0] == leaf) | |
4379 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4380 | free_extent_buffer(leaf); |
be0e5c09 | 4381 | } |
d5719762 | 4382 | } else { |
5f39d397 | 4383 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4384 | } |
4385 | } | |
aa5d6bed | 4386 | return ret; |
be0e5c09 CM |
4387 | } |
4388 | ||
7bb86316 | 4389 | /* |
925baedd | 4390 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4391 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4392 | * returns < 0 on io errors. | |
d352ac68 CM |
4393 | * |
4394 | * This may release the path, and so you may lose any locks held at the | |
4395 | * time you call it. | |
7bb86316 CM |
4396 | */ |
4397 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
4398 | { | |
925baedd CM |
4399 | struct btrfs_key key; |
4400 | struct btrfs_disk_key found_key; | |
4401 | int ret; | |
7bb86316 | 4402 | |
925baedd | 4403 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4404 | |
925baedd CM |
4405 | if (key.offset > 0) |
4406 | key.offset--; | |
4407 | else if (key.type > 0) | |
4408 | key.type--; | |
4409 | else if (key.objectid > 0) | |
4410 | key.objectid--; | |
4411 | else | |
4412 | return 1; | |
7bb86316 | 4413 | |
b3b4aa74 | 4414 | btrfs_release_path(path); |
925baedd CM |
4415 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4416 | if (ret < 0) | |
4417 | return ret; | |
4418 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4419 | ret = comp_keys(&found_key, &key); | |
4420 | if (ret < 0) | |
4421 | return 0; | |
4422 | return 1; | |
7bb86316 CM |
4423 | } |
4424 | ||
3f157a2f CM |
4425 | /* |
4426 | * A helper function to walk down the tree starting at min_key, and looking | |
4427 | * for nodes or leaves that are either in cache or have a minimum | |
d352ac68 | 4428 | * transaction id. This is used by the btree defrag code, and tree logging |
3f157a2f CM |
4429 | * |
4430 | * This does not cow, but it does stuff the starting key it finds back | |
4431 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4432 | * key and get a writable path. | |
4433 | * | |
4434 | * This does lock as it descends, and path->keep_locks should be set | |
4435 | * to 1 by the caller. | |
4436 | * | |
4437 | * This honors path->lowest_level to prevent descent past a given level | |
4438 | * of the tree. | |
4439 | * | |
d352ac68 CM |
4440 | * min_trans indicates the oldest transaction that you are interested |
4441 | * in walking through. Any nodes or leaves older than min_trans are | |
4442 | * skipped over (without reading them). | |
4443 | * | |
3f157a2f CM |
4444 | * returns zero if something useful was found, < 0 on error and 1 if there |
4445 | * was nothing in the tree that matched the search criteria. | |
4446 | */ | |
4447 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 4448 | struct btrfs_key *max_key, |
3f157a2f CM |
4449 | struct btrfs_path *path, int cache_only, |
4450 | u64 min_trans) | |
4451 | { | |
4452 | struct extent_buffer *cur; | |
4453 | struct btrfs_key found_key; | |
4454 | int slot; | |
9652480b | 4455 | int sret; |
3f157a2f CM |
4456 | u32 nritems; |
4457 | int level; | |
4458 | int ret = 1; | |
4459 | ||
934d375b | 4460 | WARN_ON(!path->keep_locks); |
3f157a2f | 4461 | again: |
bd681513 | 4462 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4463 | level = btrfs_header_level(cur); |
e02119d5 | 4464 | WARN_ON(path->nodes[level]); |
3f157a2f | 4465 | path->nodes[level] = cur; |
bd681513 | 4466 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4467 | |
4468 | if (btrfs_header_generation(cur) < min_trans) { | |
4469 | ret = 1; | |
4470 | goto out; | |
4471 | } | |
d397712b | 4472 | while (1) { |
3f157a2f CM |
4473 | nritems = btrfs_header_nritems(cur); |
4474 | level = btrfs_header_level(cur); | |
9652480b | 4475 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4476 | |
323ac95b CM |
4477 | /* at the lowest level, we're done, setup the path and exit */ |
4478 | if (level == path->lowest_level) { | |
e02119d5 CM |
4479 | if (slot >= nritems) |
4480 | goto find_next_key; | |
3f157a2f CM |
4481 | ret = 0; |
4482 | path->slots[level] = slot; | |
4483 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4484 | goto out; | |
4485 | } | |
9652480b Y |
4486 | if (sret && slot > 0) |
4487 | slot--; | |
3f157a2f CM |
4488 | /* |
4489 | * check this node pointer against the cache_only and | |
4490 | * min_trans parameters. If it isn't in cache or is too | |
4491 | * old, skip to the next one. | |
4492 | */ | |
d397712b | 4493 | while (slot < nritems) { |
3f157a2f CM |
4494 | u64 blockptr; |
4495 | u64 gen; | |
4496 | struct extent_buffer *tmp; | |
e02119d5 CM |
4497 | struct btrfs_disk_key disk_key; |
4498 | ||
3f157a2f CM |
4499 | blockptr = btrfs_node_blockptr(cur, slot); |
4500 | gen = btrfs_node_ptr_generation(cur, slot); | |
4501 | if (gen < min_trans) { | |
4502 | slot++; | |
4503 | continue; | |
4504 | } | |
4505 | if (!cache_only) | |
4506 | break; | |
4507 | ||
e02119d5 CM |
4508 | if (max_key) { |
4509 | btrfs_node_key(cur, &disk_key, slot); | |
4510 | if (comp_keys(&disk_key, max_key) >= 0) { | |
4511 | ret = 1; | |
4512 | goto out; | |
4513 | } | |
4514 | } | |
4515 | ||
3f157a2f CM |
4516 | tmp = btrfs_find_tree_block(root, blockptr, |
4517 | btrfs_level_size(root, level - 1)); | |
4518 | ||
b9fab919 | 4519 | if (tmp && btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
3f157a2f CM |
4520 | free_extent_buffer(tmp); |
4521 | break; | |
4522 | } | |
4523 | if (tmp) | |
4524 | free_extent_buffer(tmp); | |
4525 | slot++; | |
4526 | } | |
e02119d5 | 4527 | find_next_key: |
3f157a2f CM |
4528 | /* |
4529 | * we didn't find a candidate key in this node, walk forward | |
4530 | * and find another one | |
4531 | */ | |
4532 | if (slot >= nritems) { | |
e02119d5 | 4533 | path->slots[level] = slot; |
b4ce94de | 4534 | btrfs_set_path_blocking(path); |
e02119d5 | 4535 | sret = btrfs_find_next_key(root, path, min_key, level, |
3f157a2f | 4536 | cache_only, min_trans); |
e02119d5 | 4537 | if (sret == 0) { |
b3b4aa74 | 4538 | btrfs_release_path(path); |
3f157a2f CM |
4539 | goto again; |
4540 | } else { | |
4541 | goto out; | |
4542 | } | |
4543 | } | |
4544 | /* save our key for returning back */ | |
4545 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4546 | path->slots[level] = slot; | |
4547 | if (level == path->lowest_level) { | |
4548 | ret = 0; | |
f7c79f30 | 4549 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4550 | goto out; |
4551 | } | |
b4ce94de | 4552 | btrfs_set_path_blocking(path); |
3f157a2f | 4553 | cur = read_node_slot(root, cur, slot); |
79787eaa | 4554 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 4555 | |
bd681513 | 4556 | btrfs_tree_read_lock(cur); |
b4ce94de | 4557 | |
bd681513 | 4558 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4559 | path->nodes[level - 1] = cur; |
f7c79f30 | 4560 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 4561 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4562 | } |
4563 | out: | |
4564 | if (ret == 0) | |
4565 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4566 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4567 | return ret; |
4568 | } | |
4569 | ||
4570 | /* | |
4571 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4572 | * and fixup the path. It looks for and returns the next key in the | |
4573 | * tree based on the current path and the cache_only and min_trans | |
4574 | * parameters. | |
4575 | * | |
4576 | * 0 is returned if another key is found, < 0 if there are any errors | |
4577 | * and 1 is returned if there are no higher keys in the tree | |
4578 | * | |
4579 | * path->keep_locks should be set to 1 on the search made before | |
4580 | * calling this function. | |
4581 | */ | |
e7a84565 | 4582 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
33c66f43 | 4583 | struct btrfs_key *key, int level, |
3f157a2f | 4584 | int cache_only, u64 min_trans) |
e7a84565 | 4585 | { |
e7a84565 CM |
4586 | int slot; |
4587 | struct extent_buffer *c; | |
4588 | ||
934d375b | 4589 | WARN_ON(!path->keep_locks); |
d397712b | 4590 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4591 | if (!path->nodes[level]) |
4592 | return 1; | |
4593 | ||
4594 | slot = path->slots[level] + 1; | |
4595 | c = path->nodes[level]; | |
3f157a2f | 4596 | next: |
e7a84565 | 4597 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4598 | int ret; |
4599 | int orig_lowest; | |
4600 | struct btrfs_key cur_key; | |
4601 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4602 | !path->nodes[level + 1]) | |
e7a84565 | 4603 | return 1; |
33c66f43 YZ |
4604 | |
4605 | if (path->locks[level + 1]) { | |
4606 | level++; | |
4607 | continue; | |
4608 | } | |
4609 | ||
4610 | slot = btrfs_header_nritems(c) - 1; | |
4611 | if (level == 0) | |
4612 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4613 | else | |
4614 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4615 | ||
4616 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4617 | btrfs_release_path(path); |
33c66f43 YZ |
4618 | path->lowest_level = level; |
4619 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4620 | 0, 0); | |
4621 | path->lowest_level = orig_lowest; | |
4622 | if (ret < 0) | |
4623 | return ret; | |
4624 | ||
4625 | c = path->nodes[level]; | |
4626 | slot = path->slots[level]; | |
4627 | if (ret == 0) | |
4628 | slot++; | |
4629 | goto next; | |
e7a84565 | 4630 | } |
33c66f43 | 4631 | |
e7a84565 CM |
4632 | if (level == 0) |
4633 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f CM |
4634 | else { |
4635 | u64 blockptr = btrfs_node_blockptr(c, slot); | |
4636 | u64 gen = btrfs_node_ptr_generation(c, slot); | |
4637 | ||
4638 | if (cache_only) { | |
4639 | struct extent_buffer *cur; | |
4640 | cur = btrfs_find_tree_block(root, blockptr, | |
4641 | btrfs_level_size(root, level - 1)); | |
b9fab919 CM |
4642 | if (!cur || |
4643 | btrfs_buffer_uptodate(cur, gen, 1) <= 0) { | |
3f157a2f CM |
4644 | slot++; |
4645 | if (cur) | |
4646 | free_extent_buffer(cur); | |
4647 | goto next; | |
4648 | } | |
4649 | free_extent_buffer(cur); | |
4650 | } | |
4651 | if (gen < min_trans) { | |
4652 | slot++; | |
4653 | goto next; | |
4654 | } | |
e7a84565 | 4655 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4656 | } |
e7a84565 CM |
4657 | return 0; |
4658 | } | |
4659 | return 1; | |
4660 | } | |
4661 | ||
97571fd0 | 4662 | /* |
925baedd | 4663 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
4664 | * returns 0 if it found something or 1 if there are no greater leaves. |
4665 | * returns < 0 on io errors. | |
97571fd0 | 4666 | */ |
234b63a0 | 4667 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
d97e63b6 CM |
4668 | { |
4669 | int slot; | |
8e73f275 | 4670 | int level; |
5f39d397 | 4671 | struct extent_buffer *c; |
8e73f275 | 4672 | struct extent_buffer *next; |
925baedd CM |
4673 | struct btrfs_key key; |
4674 | u32 nritems; | |
4675 | int ret; | |
8e73f275 | 4676 | int old_spinning = path->leave_spinning; |
bd681513 | 4677 | int next_rw_lock = 0; |
925baedd CM |
4678 | |
4679 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4680 | if (nritems == 0) |
925baedd | 4681 | return 1; |
925baedd | 4682 | |
8e73f275 CM |
4683 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4684 | again: | |
4685 | level = 1; | |
4686 | next = NULL; | |
bd681513 | 4687 | next_rw_lock = 0; |
b3b4aa74 | 4688 | btrfs_release_path(path); |
8e73f275 | 4689 | |
a2135011 | 4690 | path->keep_locks = 1; |
31533fb2 | 4691 | path->leave_spinning = 1; |
8e73f275 | 4692 | |
925baedd CM |
4693 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4694 | path->keep_locks = 0; | |
4695 | ||
4696 | if (ret < 0) | |
4697 | return ret; | |
4698 | ||
a2135011 | 4699 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4700 | /* |
4701 | * by releasing the path above we dropped all our locks. A balance | |
4702 | * could have added more items next to the key that used to be | |
4703 | * at the very end of the block. So, check again here and | |
4704 | * advance the path if there are now more items available. | |
4705 | */ | |
a2135011 | 4706 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4707 | if (ret == 0) |
4708 | path->slots[0]++; | |
8e73f275 | 4709 | ret = 0; |
925baedd CM |
4710 | goto done; |
4711 | } | |
d97e63b6 | 4712 | |
d397712b | 4713 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4714 | if (!path->nodes[level]) { |
4715 | ret = 1; | |
4716 | goto done; | |
4717 | } | |
5f39d397 | 4718 | |
d97e63b6 CM |
4719 | slot = path->slots[level] + 1; |
4720 | c = path->nodes[level]; | |
5f39d397 | 4721 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4722 | level++; |
8e73f275 CM |
4723 | if (level == BTRFS_MAX_LEVEL) { |
4724 | ret = 1; | |
4725 | goto done; | |
4726 | } | |
d97e63b6 CM |
4727 | continue; |
4728 | } | |
5f39d397 | 4729 | |
925baedd | 4730 | if (next) { |
bd681513 | 4731 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 4732 | free_extent_buffer(next); |
925baedd | 4733 | } |
5f39d397 | 4734 | |
8e73f275 | 4735 | next = c; |
bd681513 | 4736 | next_rw_lock = path->locks[level]; |
8e73f275 CM |
4737 | ret = read_block_for_search(NULL, root, path, &next, level, |
4738 | slot, &key); | |
4739 | if (ret == -EAGAIN) | |
4740 | goto again; | |
5f39d397 | 4741 | |
76a05b35 | 4742 | if (ret < 0) { |
b3b4aa74 | 4743 | btrfs_release_path(path); |
76a05b35 CM |
4744 | goto done; |
4745 | } | |
4746 | ||
5cd57b2c | 4747 | if (!path->skip_locking) { |
bd681513 | 4748 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4749 | if (!ret) { |
4750 | btrfs_set_path_blocking(path); | |
bd681513 | 4751 | btrfs_tree_read_lock(next); |
31533fb2 | 4752 | btrfs_clear_path_blocking(path, next, |
bd681513 | 4753 | BTRFS_READ_LOCK); |
8e73f275 | 4754 | } |
31533fb2 | 4755 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4756 | } |
d97e63b6 CM |
4757 | break; |
4758 | } | |
4759 | path->slots[level] = slot; | |
d397712b | 4760 | while (1) { |
d97e63b6 CM |
4761 | level--; |
4762 | c = path->nodes[level]; | |
925baedd | 4763 | if (path->locks[level]) |
bd681513 | 4764 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 4765 | |
5f39d397 | 4766 | free_extent_buffer(c); |
d97e63b6 CM |
4767 | path->nodes[level] = next; |
4768 | path->slots[level] = 0; | |
a74a4b97 | 4769 | if (!path->skip_locking) |
bd681513 | 4770 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
4771 | if (!level) |
4772 | break; | |
b4ce94de | 4773 | |
8e73f275 CM |
4774 | ret = read_block_for_search(NULL, root, path, &next, level, |
4775 | 0, &key); | |
4776 | if (ret == -EAGAIN) | |
4777 | goto again; | |
4778 | ||
76a05b35 | 4779 | if (ret < 0) { |
b3b4aa74 | 4780 | btrfs_release_path(path); |
76a05b35 CM |
4781 | goto done; |
4782 | } | |
4783 | ||
5cd57b2c | 4784 | if (!path->skip_locking) { |
bd681513 | 4785 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
4786 | if (!ret) { |
4787 | btrfs_set_path_blocking(path); | |
bd681513 | 4788 | btrfs_tree_read_lock(next); |
31533fb2 | 4789 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
4790 | BTRFS_READ_LOCK); |
4791 | } | |
31533fb2 | 4792 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 4793 | } |
d97e63b6 | 4794 | } |
8e73f275 | 4795 | ret = 0; |
925baedd | 4796 | done: |
f7c79f30 | 4797 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
4798 | path->leave_spinning = old_spinning; |
4799 | if (!old_spinning) | |
4800 | btrfs_set_path_blocking(path); | |
4801 | ||
4802 | return ret; | |
d97e63b6 | 4803 | } |
0b86a832 | 4804 | |
3f157a2f CM |
4805 | /* |
4806 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4807 | * searching until it gets past min_objectid or finds an item of 'type' | |
4808 | * | |
4809 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4810 | */ | |
0b86a832 CM |
4811 | int btrfs_previous_item(struct btrfs_root *root, |
4812 | struct btrfs_path *path, u64 min_objectid, | |
4813 | int type) | |
4814 | { | |
4815 | struct btrfs_key found_key; | |
4816 | struct extent_buffer *leaf; | |
e02119d5 | 4817 | u32 nritems; |
0b86a832 CM |
4818 | int ret; |
4819 | ||
d397712b | 4820 | while (1) { |
0b86a832 | 4821 | if (path->slots[0] == 0) { |
b4ce94de | 4822 | btrfs_set_path_blocking(path); |
0b86a832 CM |
4823 | ret = btrfs_prev_leaf(root, path); |
4824 | if (ret != 0) | |
4825 | return ret; | |
4826 | } else { | |
4827 | path->slots[0]--; | |
4828 | } | |
4829 | leaf = path->nodes[0]; | |
e02119d5 CM |
4830 | nritems = btrfs_header_nritems(leaf); |
4831 | if (nritems == 0) | |
4832 | return 1; | |
4833 | if (path->slots[0] == nritems) | |
4834 | path->slots[0]--; | |
4835 | ||
0b86a832 | 4836 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4837 | if (found_key.objectid < min_objectid) |
4838 | break; | |
0a4eefbb YZ |
4839 | if (found_key.type == type) |
4840 | return 0; | |
e02119d5 CM |
4841 | if (found_key.objectid == min_objectid && |
4842 | found_key.type < type) | |
4843 | break; | |
0b86a832 CM |
4844 | } |
4845 | return 1; | |
4846 | } |