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