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