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