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); | |
afe5fea7 TI |
40 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
41 | int level, int slot); | |
5de865ee | 42 | static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
f230475e | 43 | struct extent_buffer *eb); |
d97e63b6 | 44 | |
df24a2b9 | 45 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 46 | { |
df24a2b9 | 47 | struct btrfs_path *path; |
e00f7308 | 48 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 49 | return path; |
2c90e5d6 CM |
50 | } |
51 | ||
b4ce94de CM |
52 | /* |
53 | * set all locked nodes in the path to blocking locks. This should | |
54 | * be done before scheduling | |
55 | */ | |
56 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
57 | { | |
58 | int i; | |
59 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
60 | if (!p->nodes[i] || !p->locks[i]) |
61 | continue; | |
62 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
63 | if (p->locks[i] == BTRFS_READ_LOCK) | |
64 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
65 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
66 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
67 | } |
68 | } | |
69 | ||
70 | /* | |
71 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
72 | * |
73 | * held is used to keep lockdep happy, when lockdep is enabled | |
74 | * we set held to a blocking lock before we go around and | |
75 | * retake all the spinlocks in the path. You can safely use NULL | |
76 | * for held | |
b4ce94de | 77 | */ |
4008c04a | 78 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 79 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
80 | { |
81 | int i; | |
4008c04a CM |
82 | |
83 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
84 | /* lockdep really cares that we take all of these spinlocks | |
85 | * in the right order. If any of the locks in the path are not | |
86 | * currently blocking, it is going to complain. So, make really | |
87 | * really sure by forcing the path to blocking before we clear | |
88 | * the path blocking. | |
89 | */ | |
bd681513 CM |
90 | if (held) { |
91 | btrfs_set_lock_blocking_rw(held, held_rw); | |
92 | if (held_rw == BTRFS_WRITE_LOCK) | |
93 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
94 | else if (held_rw == BTRFS_READ_LOCK) | |
95 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
96 | } | |
4008c04a CM |
97 | btrfs_set_path_blocking(p); |
98 | #endif | |
99 | ||
100 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
101 | if (p->nodes[i] && p->locks[i]) { |
102 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
103 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
104 | p->locks[i] = BTRFS_WRITE_LOCK; | |
105 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
106 | p->locks[i] = BTRFS_READ_LOCK; | |
107 | } | |
b4ce94de | 108 | } |
4008c04a CM |
109 | |
110 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
111 | if (held) | |
bd681513 | 112 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 113 | #endif |
b4ce94de CM |
114 | } |
115 | ||
d352ac68 | 116 | /* this also releases the path */ |
df24a2b9 | 117 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 118 | { |
ff175d57 JJ |
119 | if (!p) |
120 | return; | |
b3b4aa74 | 121 | btrfs_release_path(p); |
df24a2b9 | 122 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
123 | } |
124 | ||
d352ac68 CM |
125 | /* |
126 | * path release drops references on the extent buffers in the path | |
127 | * and it drops any locks held by this path | |
128 | * | |
129 | * It is safe to call this on paths that no locks or extent buffers held. | |
130 | */ | |
b3b4aa74 | 131 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
132 | { |
133 | int i; | |
a2135011 | 134 | |
234b63a0 | 135 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 136 | p->slots[i] = 0; |
eb60ceac | 137 | if (!p->nodes[i]) |
925baedd CM |
138 | continue; |
139 | if (p->locks[i]) { | |
bd681513 | 140 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
141 | p->locks[i] = 0; |
142 | } | |
5f39d397 | 143 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 144 | p->nodes[i] = NULL; |
eb60ceac CM |
145 | } |
146 | } | |
147 | ||
d352ac68 CM |
148 | /* |
149 | * safely gets a reference on the root node of a tree. A lock | |
150 | * is not taken, so a concurrent writer may put a different node | |
151 | * at the root of the tree. See btrfs_lock_root_node for the | |
152 | * looping required. | |
153 | * | |
154 | * The extent buffer returned by this has a reference taken, so | |
155 | * it won't disappear. It may stop being the root of the tree | |
156 | * at any time because there are no locks held. | |
157 | */ | |
925baedd CM |
158 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
159 | { | |
160 | struct extent_buffer *eb; | |
240f62c8 | 161 | |
3083ee2e JB |
162 | while (1) { |
163 | rcu_read_lock(); | |
164 | eb = rcu_dereference(root->node); | |
165 | ||
166 | /* | |
167 | * RCU really hurts here, we could free up the root node because | |
168 | * it was cow'ed but we may not get the new root node yet so do | |
169 | * the inc_not_zero dance and if it doesn't work then | |
170 | * synchronize_rcu and try again. | |
171 | */ | |
172 | if (atomic_inc_not_zero(&eb->refs)) { | |
173 | rcu_read_unlock(); | |
174 | break; | |
175 | } | |
176 | rcu_read_unlock(); | |
177 | synchronize_rcu(); | |
178 | } | |
925baedd CM |
179 | return eb; |
180 | } | |
181 | ||
d352ac68 CM |
182 | /* loop around taking references on and locking the root node of the |
183 | * tree until you end up with a lock on the root. A locked buffer | |
184 | * is returned, with a reference held. | |
185 | */ | |
925baedd CM |
186 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
187 | { | |
188 | struct extent_buffer *eb; | |
189 | ||
d397712b | 190 | while (1) { |
925baedd CM |
191 | eb = btrfs_root_node(root); |
192 | btrfs_tree_lock(eb); | |
240f62c8 | 193 | if (eb == root->node) |
925baedd | 194 | break; |
925baedd CM |
195 | btrfs_tree_unlock(eb); |
196 | free_extent_buffer(eb); | |
197 | } | |
198 | return eb; | |
199 | } | |
200 | ||
bd681513 CM |
201 | /* loop around taking references on and locking the root node of the |
202 | * tree until you end up with a lock on the root. A locked buffer | |
203 | * is returned, with a reference held. | |
204 | */ | |
48a3b636 | 205 | static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
206 | { |
207 | struct extent_buffer *eb; | |
208 | ||
209 | while (1) { | |
210 | eb = btrfs_root_node(root); | |
211 | btrfs_tree_read_lock(eb); | |
212 | if (eb == root->node) | |
213 | break; | |
214 | btrfs_tree_read_unlock(eb); | |
215 | free_extent_buffer(eb); | |
216 | } | |
217 | return eb; | |
218 | } | |
219 | ||
d352ac68 CM |
220 | /* cowonly root (everything not a reference counted cow subvolume), just get |
221 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
222 | * get properly updated on disk. | |
223 | */ | |
0b86a832 CM |
224 | static void add_root_to_dirty_list(struct btrfs_root *root) |
225 | { | |
e5846fc6 | 226 | spin_lock(&root->fs_info->trans_lock); |
27cdeb70 MX |
227 | if (test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state) && |
228 | list_empty(&root->dirty_list)) { | |
0b86a832 CM |
229 | list_add(&root->dirty_list, |
230 | &root->fs_info->dirty_cowonly_roots); | |
231 | } | |
e5846fc6 | 232 | spin_unlock(&root->fs_info->trans_lock); |
0b86a832 CM |
233 | } |
234 | ||
d352ac68 CM |
235 | /* |
236 | * used by snapshot creation to make a copy of a root for a tree with | |
237 | * a given objectid. The buffer with the new root node is returned in | |
238 | * cow_ret, and this func returns zero on success or a negative error code. | |
239 | */ | |
be20aa9d CM |
240 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
241 | struct btrfs_root *root, | |
242 | struct extent_buffer *buf, | |
243 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
244 | { | |
245 | struct extent_buffer *cow; | |
be20aa9d CM |
246 | int ret = 0; |
247 | int level; | |
5d4f98a2 | 248 | struct btrfs_disk_key disk_key; |
be20aa9d | 249 | |
27cdeb70 MX |
250 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
251 | trans->transid != root->fs_info->running_transaction->transid); | |
252 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && | |
253 | trans->transid != root->last_trans); | |
be20aa9d CM |
254 | |
255 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
256 | if (level == 0) |
257 | btrfs_item_key(buf, &disk_key, 0); | |
258 | else | |
259 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 260 | |
5d4f98a2 YZ |
261 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
262 | new_root_objectid, &disk_key, level, | |
5581a51a | 263 | buf->start, 0); |
5d4f98a2 | 264 | if (IS_ERR(cow)) |
be20aa9d CM |
265 | return PTR_ERR(cow); |
266 | ||
267 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
268 | btrfs_set_header_bytenr(cow, cow->start); | |
269 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
270 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
271 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
272 | BTRFS_HEADER_FLAG_RELOC); | |
273 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
274 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
275 | else | |
276 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 277 | |
0a4e5586 | 278 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(), |
2b82032c YZ |
279 | BTRFS_FSID_SIZE); |
280 | ||
be20aa9d | 281 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 282 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 283 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 284 | else |
66d7e7f0 | 285 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 286 | |
be20aa9d CM |
287 | if (ret) |
288 | return ret; | |
289 | ||
290 | btrfs_mark_buffer_dirty(cow); | |
291 | *cow_ret = cow; | |
292 | return 0; | |
293 | } | |
294 | ||
bd989ba3 JS |
295 | enum mod_log_op { |
296 | MOD_LOG_KEY_REPLACE, | |
297 | MOD_LOG_KEY_ADD, | |
298 | MOD_LOG_KEY_REMOVE, | |
299 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
300 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
301 | MOD_LOG_MOVE_KEYS, | |
302 | MOD_LOG_ROOT_REPLACE, | |
303 | }; | |
304 | ||
305 | struct tree_mod_move { | |
306 | int dst_slot; | |
307 | int nr_items; | |
308 | }; | |
309 | ||
310 | struct tree_mod_root { | |
311 | u64 logical; | |
312 | u8 level; | |
313 | }; | |
314 | ||
315 | struct tree_mod_elem { | |
316 | struct rb_node node; | |
317 | u64 index; /* shifted logical */ | |
097b8a7c | 318 | u64 seq; |
bd989ba3 JS |
319 | enum mod_log_op op; |
320 | ||
321 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
322 | int slot; | |
323 | ||
324 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
325 | u64 generation; | |
326 | ||
327 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
328 | struct btrfs_disk_key key; | |
329 | u64 blockptr; | |
330 | ||
331 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
332 | struct tree_mod_move move; | |
333 | ||
334 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
335 | struct tree_mod_root old_root; | |
336 | }; | |
337 | ||
097b8a7c | 338 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 339 | { |
097b8a7c | 340 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
341 | } |
342 | ||
097b8a7c JS |
343 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
344 | { | |
345 | read_unlock(&fs_info->tree_mod_log_lock); | |
346 | } | |
347 | ||
348 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
349 | { | |
350 | write_lock(&fs_info->tree_mod_log_lock); | |
351 | } | |
352 | ||
353 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
354 | { | |
355 | write_unlock(&fs_info->tree_mod_log_lock); | |
356 | } | |
357 | ||
fc36ed7e JS |
358 | /* |
359 | * Increment the upper half of tree_mod_seq, set lower half zero. | |
360 | * | |
361 | * Must be called with fs_info->tree_mod_seq_lock held. | |
362 | */ | |
363 | static inline u64 btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info *fs_info) | |
364 | { | |
365 | u64 seq = atomic64_read(&fs_info->tree_mod_seq); | |
366 | seq &= 0xffffffff00000000ull; | |
367 | seq += 1ull << 32; | |
368 | atomic64_set(&fs_info->tree_mod_seq, seq); | |
369 | return seq; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Increment the lower half of tree_mod_seq. | |
374 | * | |
375 | * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers | |
376 | * are generated should not technically require a spin lock here. (Rationale: | |
377 | * incrementing the minor while incrementing the major seq number is between its | |
378 | * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it | |
379 | * just returns a unique sequence number as usual.) We have decided to leave | |
380 | * that requirement in here and rethink it once we notice it really imposes a | |
381 | * problem on some workload. | |
382 | */ | |
383 | static inline u64 btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info *fs_info) | |
384 | { | |
385 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
386 | } | |
387 | ||
388 | /* | |
389 | * return the last minor in the previous major tree_mod_seq number | |
390 | */ | |
391 | u64 btrfs_tree_mod_seq_prev(u64 seq) | |
392 | { | |
393 | return (seq & 0xffffffff00000000ull) - 1ull; | |
394 | } | |
395 | ||
097b8a7c JS |
396 | /* |
397 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
398 | * passed does not already have a sequence number set. So when a caller expects | |
399 | * to record tree modifications, it should ensure to set elem->seq to zero | |
400 | * before calling btrfs_get_tree_mod_seq. | |
401 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
402 | * blocker was added. | |
403 | */ | |
404 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
405 | struct seq_list *elem) | |
bd989ba3 | 406 | { |
097b8a7c JS |
407 | u64 seq; |
408 | ||
409 | tree_mod_log_write_lock(fs_info); | |
bd989ba3 | 410 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 411 | if (!elem->seq) { |
fc36ed7e | 412 | elem->seq = btrfs_inc_tree_mod_seq_major(fs_info); |
097b8a7c JS |
413 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
414 | } | |
fc36ed7e | 415 | seq = btrfs_inc_tree_mod_seq_minor(fs_info); |
bd989ba3 | 416 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
417 | tree_mod_log_write_unlock(fs_info); |
418 | ||
419 | return seq; | |
bd989ba3 JS |
420 | } |
421 | ||
422 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
423 | struct seq_list *elem) | |
424 | { | |
425 | struct rb_root *tm_root; | |
426 | struct rb_node *node; | |
427 | struct rb_node *next; | |
428 | struct seq_list *cur_elem; | |
429 | struct tree_mod_elem *tm; | |
430 | u64 min_seq = (u64)-1; | |
431 | u64 seq_putting = elem->seq; | |
432 | ||
433 | if (!seq_putting) | |
434 | return; | |
435 | ||
bd989ba3 JS |
436 | spin_lock(&fs_info->tree_mod_seq_lock); |
437 | list_del(&elem->list); | |
097b8a7c | 438 | elem->seq = 0; |
bd989ba3 JS |
439 | |
440 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 441 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
442 | if (seq_putting > cur_elem->seq) { |
443 | /* | |
444 | * blocker with lower sequence number exists, we | |
445 | * cannot remove anything from the log | |
446 | */ | |
097b8a7c JS |
447 | spin_unlock(&fs_info->tree_mod_seq_lock); |
448 | return; | |
bd989ba3 JS |
449 | } |
450 | min_seq = cur_elem->seq; | |
451 | } | |
452 | } | |
097b8a7c JS |
453 | spin_unlock(&fs_info->tree_mod_seq_lock); |
454 | ||
bd989ba3 JS |
455 | /* |
456 | * anything that's lower than the lowest existing (read: blocked) | |
457 | * sequence number can be removed from the tree. | |
458 | */ | |
097b8a7c | 459 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
460 | tm_root = &fs_info->tree_mod_log; |
461 | for (node = rb_first(tm_root); node; node = next) { | |
462 | next = rb_next(node); | |
463 | tm = container_of(node, struct tree_mod_elem, node); | |
097b8a7c | 464 | if (tm->seq > min_seq) |
bd989ba3 JS |
465 | continue; |
466 | rb_erase(node, tm_root); | |
bd989ba3 JS |
467 | kfree(tm); |
468 | } | |
097b8a7c | 469 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
470 | } |
471 | ||
472 | /* | |
473 | * key order of the log: | |
474 | * index -> sequence | |
475 | * | |
476 | * the index is the shifted logical of the *new* root node for root replace | |
477 | * operations, or the shifted logical of the affected block for all other | |
478 | * operations. | |
5de865ee FDBM |
479 | * |
480 | * Note: must be called with write lock (tree_mod_log_write_lock). | |
bd989ba3 JS |
481 | */ |
482 | static noinline int | |
483 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
484 | { | |
485 | struct rb_root *tm_root; | |
486 | struct rb_node **new; | |
487 | struct rb_node *parent = NULL; | |
488 | struct tree_mod_elem *cur; | |
c8cc6341 JB |
489 | |
490 | BUG_ON(!tm); | |
491 | ||
c8cc6341 JB |
492 | spin_lock(&fs_info->tree_mod_seq_lock); |
493 | tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info); | |
494 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
bd989ba3 | 495 | |
bd989ba3 JS |
496 | tm_root = &fs_info->tree_mod_log; |
497 | new = &tm_root->rb_node; | |
498 | while (*new) { | |
499 | cur = container_of(*new, struct tree_mod_elem, node); | |
500 | parent = *new; | |
501 | if (cur->index < tm->index) | |
502 | new = &((*new)->rb_left); | |
503 | else if (cur->index > tm->index) | |
504 | new = &((*new)->rb_right); | |
097b8a7c | 505 | else if (cur->seq < tm->seq) |
bd989ba3 | 506 | new = &((*new)->rb_left); |
097b8a7c | 507 | else if (cur->seq > tm->seq) |
bd989ba3 | 508 | new = &((*new)->rb_right); |
5de865ee FDBM |
509 | else |
510 | return -EEXIST; | |
bd989ba3 JS |
511 | } |
512 | ||
513 | rb_link_node(&tm->node, parent, new); | |
514 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 515 | return 0; |
bd989ba3 JS |
516 | } |
517 | ||
097b8a7c JS |
518 | /* |
519 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
520 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
521 | * this until all tree mod log insertions are recorded in the rb tree and then | |
522 | * call tree_mod_log_write_unlock() to release. | |
523 | */ | |
e9b7fd4d JS |
524 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
525 | struct extent_buffer *eb) { | |
526 | smp_mb(); | |
527 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
528 | return 1; | |
097b8a7c JS |
529 | if (eb && btrfs_header_level(eb) == 0) |
530 | return 1; | |
5de865ee FDBM |
531 | |
532 | tree_mod_log_write_lock(fs_info); | |
533 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { | |
534 | tree_mod_log_write_unlock(fs_info); | |
535 | return 1; | |
536 | } | |
537 | ||
e9b7fd4d JS |
538 | return 0; |
539 | } | |
540 | ||
5de865ee FDBM |
541 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
542 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
543 | struct extent_buffer *eb) | |
544 | { | |
545 | smp_mb(); | |
546 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
547 | return 0; | |
548 | if (eb && btrfs_header_level(eb) == 0) | |
549 | return 0; | |
550 | ||
551 | return 1; | |
552 | } | |
553 | ||
554 | static struct tree_mod_elem * | |
555 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
556 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 557 | { |
097b8a7c | 558 | struct tree_mod_elem *tm; |
bd989ba3 | 559 | |
c8cc6341 JB |
560 | tm = kzalloc(sizeof(*tm), flags); |
561 | if (!tm) | |
5de865ee | 562 | return NULL; |
bd989ba3 JS |
563 | |
564 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
565 | if (op != MOD_LOG_KEY_ADD) { | |
566 | btrfs_node_key(eb, &tm->key, slot); | |
567 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
568 | } | |
569 | tm->op = op; | |
570 | tm->slot = slot; | |
571 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 572 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 573 | |
5de865ee | 574 | return tm; |
097b8a7c JS |
575 | } |
576 | ||
577 | static noinline int | |
c8cc6341 JB |
578 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
579 | struct extent_buffer *eb, int slot, | |
580 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 581 | { |
5de865ee FDBM |
582 | struct tree_mod_elem *tm; |
583 | int ret; | |
584 | ||
585 | if (!tree_mod_need_log(fs_info, eb)) | |
586 | return 0; | |
587 | ||
588 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
589 | if (!tm) | |
590 | return -ENOMEM; | |
591 | ||
592 | if (tree_mod_dont_log(fs_info, eb)) { | |
593 | kfree(tm); | |
097b8a7c | 594 | return 0; |
5de865ee FDBM |
595 | } |
596 | ||
597 | ret = __tree_mod_log_insert(fs_info, tm); | |
598 | tree_mod_log_write_unlock(fs_info); | |
599 | if (ret) | |
600 | kfree(tm); | |
097b8a7c | 601 | |
5de865ee | 602 | return ret; |
097b8a7c JS |
603 | } |
604 | ||
bd989ba3 JS |
605 | static noinline int |
606 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
607 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
608 | int nr_items, gfp_t flags) | |
609 | { | |
5de865ee FDBM |
610 | struct tree_mod_elem *tm = NULL; |
611 | struct tree_mod_elem **tm_list = NULL; | |
612 | int ret = 0; | |
bd989ba3 | 613 | int i; |
5de865ee | 614 | int locked = 0; |
bd989ba3 | 615 | |
5de865ee | 616 | if (!tree_mod_need_log(fs_info, eb)) |
f395694c | 617 | return 0; |
bd989ba3 | 618 | |
5de865ee FDBM |
619 | tm_list = kzalloc(nr_items * sizeof(struct tree_mod_elem *), flags); |
620 | if (!tm_list) | |
621 | return -ENOMEM; | |
622 | ||
623 | tm = kzalloc(sizeof(*tm), flags); | |
624 | if (!tm) { | |
625 | ret = -ENOMEM; | |
626 | goto free_tms; | |
627 | } | |
628 | ||
629 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
630 | tm->slot = src_slot; | |
631 | tm->move.dst_slot = dst_slot; | |
632 | tm->move.nr_items = nr_items; | |
633 | tm->op = MOD_LOG_MOVE_KEYS; | |
634 | ||
635 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
636 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
637 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags); | |
638 | if (!tm_list[i]) { | |
639 | ret = -ENOMEM; | |
640 | goto free_tms; | |
641 | } | |
642 | } | |
643 | ||
644 | if (tree_mod_dont_log(fs_info, eb)) | |
645 | goto free_tms; | |
646 | locked = 1; | |
647 | ||
01763a2e JS |
648 | /* |
649 | * When we override something during the move, we log these removals. | |
650 | * This can only happen when we move towards the beginning of the | |
651 | * buffer, i.e. dst_slot < src_slot. | |
652 | */ | |
bd989ba3 | 653 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
5de865ee FDBM |
654 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
655 | if (ret) | |
656 | goto free_tms; | |
bd989ba3 JS |
657 | } |
658 | ||
5de865ee FDBM |
659 | ret = __tree_mod_log_insert(fs_info, tm); |
660 | if (ret) | |
661 | goto free_tms; | |
662 | tree_mod_log_write_unlock(fs_info); | |
663 | kfree(tm_list); | |
f395694c | 664 | |
5de865ee FDBM |
665 | return 0; |
666 | free_tms: | |
667 | for (i = 0; i < nr_items; i++) { | |
668 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
669 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
670 | kfree(tm_list[i]); | |
671 | } | |
672 | if (locked) | |
673 | tree_mod_log_write_unlock(fs_info); | |
674 | kfree(tm_list); | |
675 | kfree(tm); | |
bd989ba3 | 676 | |
5de865ee | 677 | return ret; |
bd989ba3 JS |
678 | } |
679 | ||
5de865ee FDBM |
680 | static inline int |
681 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
682 | struct tree_mod_elem **tm_list, | |
683 | int nritems) | |
097b8a7c | 684 | { |
5de865ee | 685 | int i, j; |
097b8a7c JS |
686 | int ret; |
687 | ||
097b8a7c | 688 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
689 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
690 | if (ret) { | |
691 | for (j = nritems - 1; j > i; j--) | |
692 | rb_erase(&tm_list[j]->node, | |
693 | &fs_info->tree_mod_log); | |
694 | return ret; | |
695 | } | |
097b8a7c | 696 | } |
5de865ee FDBM |
697 | |
698 | return 0; | |
097b8a7c JS |
699 | } |
700 | ||
bd989ba3 JS |
701 | static noinline int |
702 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
703 | struct extent_buffer *old_root, | |
90f8d62e JS |
704 | struct extent_buffer *new_root, gfp_t flags, |
705 | int log_removal) | |
bd989ba3 | 706 | { |
5de865ee FDBM |
707 | struct tree_mod_elem *tm = NULL; |
708 | struct tree_mod_elem **tm_list = NULL; | |
709 | int nritems = 0; | |
710 | int ret = 0; | |
711 | int i; | |
bd989ba3 | 712 | |
5de865ee | 713 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
714 | return 0; |
715 | ||
5de865ee FDBM |
716 | if (log_removal && btrfs_header_level(old_root) > 0) { |
717 | nritems = btrfs_header_nritems(old_root); | |
718 | tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *), | |
719 | flags); | |
720 | if (!tm_list) { | |
721 | ret = -ENOMEM; | |
722 | goto free_tms; | |
723 | } | |
724 | for (i = 0; i < nritems; i++) { | |
725 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
726 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags); | |
727 | if (!tm_list[i]) { | |
728 | ret = -ENOMEM; | |
729 | goto free_tms; | |
730 | } | |
731 | } | |
732 | } | |
d9abbf1c | 733 | |
c8cc6341 | 734 | tm = kzalloc(sizeof(*tm), flags); |
5de865ee FDBM |
735 | if (!tm) { |
736 | ret = -ENOMEM; | |
737 | goto free_tms; | |
738 | } | |
bd989ba3 JS |
739 | |
740 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
741 | tm->old_root.logical = old_root->start; | |
742 | tm->old_root.level = btrfs_header_level(old_root); | |
743 | tm->generation = btrfs_header_generation(old_root); | |
744 | tm->op = MOD_LOG_ROOT_REPLACE; | |
745 | ||
5de865ee FDBM |
746 | if (tree_mod_dont_log(fs_info, NULL)) |
747 | goto free_tms; | |
748 | ||
749 | if (tm_list) | |
750 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
751 | if (!ret) | |
752 | ret = __tree_mod_log_insert(fs_info, tm); | |
753 | ||
754 | tree_mod_log_write_unlock(fs_info); | |
755 | if (ret) | |
756 | goto free_tms; | |
757 | kfree(tm_list); | |
758 | ||
759 | return ret; | |
760 | ||
761 | free_tms: | |
762 | if (tm_list) { | |
763 | for (i = 0; i < nritems; i++) | |
764 | kfree(tm_list[i]); | |
765 | kfree(tm_list); | |
766 | } | |
767 | kfree(tm); | |
768 | ||
769 | return ret; | |
bd989ba3 JS |
770 | } |
771 | ||
772 | static struct tree_mod_elem * | |
773 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
774 | int smallest) | |
775 | { | |
776 | struct rb_root *tm_root; | |
777 | struct rb_node *node; | |
778 | struct tree_mod_elem *cur = NULL; | |
779 | struct tree_mod_elem *found = NULL; | |
780 | u64 index = start >> PAGE_CACHE_SHIFT; | |
781 | ||
097b8a7c | 782 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
783 | tm_root = &fs_info->tree_mod_log; |
784 | node = tm_root->rb_node; | |
785 | while (node) { | |
786 | cur = container_of(node, struct tree_mod_elem, node); | |
787 | if (cur->index < index) { | |
788 | node = node->rb_left; | |
789 | } else if (cur->index > index) { | |
790 | node = node->rb_right; | |
097b8a7c | 791 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
792 | node = node->rb_left; |
793 | } else if (!smallest) { | |
794 | /* we want the node with the highest seq */ | |
795 | if (found) | |
097b8a7c | 796 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
797 | found = cur; |
798 | node = node->rb_left; | |
097b8a7c | 799 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
800 | /* we want the node with the smallest seq */ |
801 | if (found) | |
097b8a7c | 802 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
803 | found = cur; |
804 | node = node->rb_right; | |
805 | } else { | |
806 | found = cur; | |
807 | break; | |
808 | } | |
809 | } | |
097b8a7c | 810 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
811 | |
812 | return found; | |
813 | } | |
814 | ||
815 | /* | |
816 | * this returns the element from the log with the smallest time sequence | |
817 | * value that's in the log (the oldest log item). any element with a time | |
818 | * sequence lower than min_seq will be ignored. | |
819 | */ | |
820 | static struct tree_mod_elem * | |
821 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
822 | u64 min_seq) | |
823 | { | |
824 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
825 | } | |
826 | ||
827 | /* | |
828 | * this returns the element from the log with the largest time sequence | |
829 | * value that's in the log (the most recent log item). any element with | |
830 | * a time sequence lower than min_seq will be ignored. | |
831 | */ | |
832 | static struct tree_mod_elem * | |
833 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
834 | { | |
835 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
836 | } | |
837 | ||
5de865ee | 838 | static noinline int |
bd989ba3 JS |
839 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
840 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 841 | unsigned long src_offset, int nr_items) |
bd989ba3 | 842 | { |
5de865ee FDBM |
843 | int ret = 0; |
844 | struct tree_mod_elem **tm_list = NULL; | |
845 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 846 | int i; |
5de865ee | 847 | int locked = 0; |
bd989ba3 | 848 | |
5de865ee FDBM |
849 | if (!tree_mod_need_log(fs_info, NULL)) |
850 | return 0; | |
bd989ba3 | 851 | |
c8cc6341 | 852 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
853 | return 0; |
854 | ||
855 | tm_list = kzalloc(nr_items * 2 * sizeof(struct tree_mod_elem *), | |
856 | GFP_NOFS); | |
857 | if (!tm_list) | |
858 | return -ENOMEM; | |
bd989ba3 | 859 | |
5de865ee FDBM |
860 | tm_list_add = tm_list; |
861 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 862 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
863 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
864 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
865 | if (!tm_list_rem[i]) { | |
866 | ret = -ENOMEM; | |
867 | goto free_tms; | |
868 | } | |
869 | ||
870 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
871 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
872 | if (!tm_list_add[i]) { | |
873 | ret = -ENOMEM; | |
874 | goto free_tms; | |
875 | } | |
876 | } | |
877 | ||
878 | if (tree_mod_dont_log(fs_info, NULL)) | |
879 | goto free_tms; | |
880 | locked = 1; | |
881 | ||
882 | for (i = 0; i < nr_items; i++) { | |
883 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
884 | if (ret) | |
885 | goto free_tms; | |
886 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
887 | if (ret) | |
888 | goto free_tms; | |
bd989ba3 | 889 | } |
5de865ee FDBM |
890 | |
891 | tree_mod_log_write_unlock(fs_info); | |
892 | kfree(tm_list); | |
893 | ||
894 | return 0; | |
895 | ||
896 | free_tms: | |
897 | for (i = 0; i < nr_items * 2; i++) { | |
898 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
899 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
900 | kfree(tm_list[i]); | |
901 | } | |
902 | if (locked) | |
903 | tree_mod_log_write_unlock(fs_info); | |
904 | kfree(tm_list); | |
905 | ||
906 | return ret; | |
bd989ba3 JS |
907 | } |
908 | ||
909 | static inline void | |
910 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
911 | int dst_offset, int src_offset, int nr_items) | |
912 | { | |
913 | int ret; | |
914 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
915 | nr_items, GFP_NOFS); | |
916 | BUG_ON(ret < 0); | |
917 | } | |
918 | ||
097b8a7c | 919 | static noinline void |
bd989ba3 | 920 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 921 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
922 | { |
923 | int ret; | |
924 | ||
78357766 | 925 | ret = tree_mod_log_insert_key(fs_info, eb, slot, |
c8cc6341 JB |
926 | MOD_LOG_KEY_REPLACE, |
927 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
928 | BUG_ON(ret < 0); |
929 | } | |
930 | ||
5de865ee | 931 | static noinline int |
097b8a7c | 932 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) |
bd989ba3 | 933 | { |
5de865ee FDBM |
934 | struct tree_mod_elem **tm_list = NULL; |
935 | int nritems = 0; | |
936 | int i; | |
937 | int ret = 0; | |
938 | ||
939 | if (btrfs_header_level(eb) == 0) | |
940 | return 0; | |
941 | ||
942 | if (!tree_mod_need_log(fs_info, NULL)) | |
943 | return 0; | |
944 | ||
945 | nritems = btrfs_header_nritems(eb); | |
946 | tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *), | |
947 | GFP_NOFS); | |
948 | if (!tm_list) | |
949 | return -ENOMEM; | |
950 | ||
951 | for (i = 0; i < nritems; i++) { | |
952 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
953 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
954 | if (!tm_list[i]) { | |
955 | ret = -ENOMEM; | |
956 | goto free_tms; | |
957 | } | |
958 | } | |
959 | ||
e9b7fd4d | 960 | if (tree_mod_dont_log(fs_info, eb)) |
5de865ee FDBM |
961 | goto free_tms; |
962 | ||
963 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
964 | tree_mod_log_write_unlock(fs_info); | |
965 | if (ret) | |
966 | goto free_tms; | |
967 | kfree(tm_list); | |
968 | ||
969 | return 0; | |
970 | ||
971 | free_tms: | |
972 | for (i = 0; i < nritems; i++) | |
973 | kfree(tm_list[i]); | |
974 | kfree(tm_list); | |
975 | ||
976 | return ret; | |
bd989ba3 JS |
977 | } |
978 | ||
097b8a7c | 979 | static noinline void |
bd989ba3 | 980 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
981 | struct extent_buffer *new_root_node, |
982 | int log_removal) | |
bd989ba3 JS |
983 | { |
984 | int ret; | |
bd989ba3 | 985 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 986 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
987 | BUG_ON(ret < 0); |
988 | } | |
989 | ||
5d4f98a2 YZ |
990 | /* |
991 | * check if the tree block can be shared by multiple trees | |
992 | */ | |
993 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
994 | struct extent_buffer *buf) | |
995 | { | |
996 | /* | |
997 | * Tree blocks not in refernece counted trees and tree roots | |
998 | * are never shared. If a block was allocated after the last | |
999 | * snapshot and the block was not allocated by tree relocation, | |
1000 | * we know the block is not shared. | |
1001 | */ | |
27cdeb70 | 1002 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
1003 | buf != root->node && buf != root->commit_root && |
1004 | (btrfs_header_generation(buf) <= | |
1005 | btrfs_root_last_snapshot(&root->root_item) || | |
1006 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
1007 | return 1; | |
1008 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 1009 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
1010 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
1011 | return 1; | |
1012 | #endif | |
1013 | return 0; | |
1014 | } | |
1015 | ||
1016 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
1017 | struct btrfs_root *root, | |
1018 | struct extent_buffer *buf, | |
f0486c68 YZ |
1019 | struct extent_buffer *cow, |
1020 | int *last_ref) | |
5d4f98a2 YZ |
1021 | { |
1022 | u64 refs; | |
1023 | u64 owner; | |
1024 | u64 flags; | |
1025 | u64 new_flags = 0; | |
1026 | int ret; | |
1027 | ||
1028 | /* | |
1029 | * Backrefs update rules: | |
1030 | * | |
1031 | * Always use full backrefs for extent pointers in tree block | |
1032 | * allocated by tree relocation. | |
1033 | * | |
1034 | * If a shared tree block is no longer referenced by its owner | |
1035 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
1036 | * use full backrefs for extent pointers in tree block. | |
1037 | * | |
1038 | * If a tree block is been relocating | |
1039 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
1040 | * use full backrefs for extent pointers in tree block. | |
1041 | * The reason for this is some operations (such as drop tree) | |
1042 | * are only allowed for blocks use full backrefs. | |
1043 | */ | |
1044 | ||
1045 | if (btrfs_block_can_be_shared(root, buf)) { | |
1046 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
1047 | btrfs_header_level(buf), 1, |
1048 | &refs, &flags); | |
be1a5564 MF |
1049 | if (ret) |
1050 | return ret; | |
e5df9573 MF |
1051 | if (refs == 0) { |
1052 | ret = -EROFS; | |
1053 | btrfs_std_error(root->fs_info, ret); | |
1054 | return ret; | |
1055 | } | |
5d4f98a2 YZ |
1056 | } else { |
1057 | refs = 1; | |
1058 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
1059 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1060 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1061 | else | |
1062 | flags = 0; | |
1063 | } | |
1064 | ||
1065 | owner = btrfs_header_owner(buf); | |
1066 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
1067 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
1068 | ||
1069 | if (refs > 1) { | |
1070 | if ((owner == root->root_key.objectid || | |
1071 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
1072 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 1073 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 1074 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1075 | |
1076 | if (root->root_key.objectid == | |
1077 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 1078 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 1079 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 1080 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 1081 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1082 | } |
1083 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1084 | } else { | |
1085 | ||
1086 | if (root->root_key.objectid == | |
1087 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 1088 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 1089 | else |
66d7e7f0 | 1090 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 1091 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1092 | } |
1093 | if (new_flags != 0) { | |
b1c79e09 JB |
1094 | int level = btrfs_header_level(buf); |
1095 | ||
5d4f98a2 YZ |
1096 | ret = btrfs_set_disk_extent_flags(trans, root, |
1097 | buf->start, | |
1098 | buf->len, | |
b1c79e09 | 1099 | new_flags, level, 0); |
be1a5564 MF |
1100 | if (ret) |
1101 | return ret; | |
5d4f98a2 YZ |
1102 | } |
1103 | } else { | |
1104 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
1105 | if (root->root_key.objectid == | |
1106 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 1107 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 1108 | else |
66d7e7f0 | 1109 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 1110 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 1111 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 1112 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1113 | } |
1114 | clean_tree_block(trans, root, buf); | |
f0486c68 | 1115 | *last_ref = 1; |
5d4f98a2 YZ |
1116 | } |
1117 | return 0; | |
1118 | } | |
1119 | ||
d352ac68 | 1120 | /* |
d397712b CM |
1121 | * does the dirty work in cow of a single block. The parent block (if |
1122 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1123 | * dirty and returned locked. If you modify the block it needs to be marked | |
1124 | * dirty again. | |
d352ac68 CM |
1125 | * |
1126 | * search_start -- an allocation hint for the new block | |
1127 | * | |
d397712b CM |
1128 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1129 | * bytes the allocator should try to find free next to the block it returns. | |
1130 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1131 | */ |
d397712b | 1132 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1133 | struct btrfs_root *root, |
1134 | struct extent_buffer *buf, | |
1135 | struct extent_buffer *parent, int parent_slot, | |
1136 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1137 | u64 search_start, u64 empty_size) |
02217ed2 | 1138 | { |
5d4f98a2 | 1139 | struct btrfs_disk_key disk_key; |
5f39d397 | 1140 | struct extent_buffer *cow; |
be1a5564 | 1141 | int level, ret; |
f0486c68 | 1142 | int last_ref = 0; |
925baedd | 1143 | int unlock_orig = 0; |
5d4f98a2 | 1144 | u64 parent_start; |
7bb86316 | 1145 | |
925baedd CM |
1146 | if (*cow_ret == buf) |
1147 | unlock_orig = 1; | |
1148 | ||
b9447ef8 | 1149 | btrfs_assert_tree_locked(buf); |
925baedd | 1150 | |
27cdeb70 MX |
1151 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1152 | trans->transid != root->fs_info->running_transaction->transid); | |
1153 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && | |
1154 | trans->transid != root->last_trans); | |
5f39d397 | 1155 | |
7bb86316 | 1156 | level = btrfs_header_level(buf); |
31840ae1 | 1157 | |
5d4f98a2 YZ |
1158 | if (level == 0) |
1159 | btrfs_item_key(buf, &disk_key, 0); | |
1160 | else | |
1161 | btrfs_node_key(buf, &disk_key, 0); | |
1162 | ||
1163 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
1164 | if (parent) | |
1165 | parent_start = parent->start; | |
1166 | else | |
1167 | parent_start = 0; | |
1168 | } else | |
1169 | parent_start = 0; | |
1170 | ||
1171 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
1172 | root->root_key.objectid, &disk_key, | |
5581a51a | 1173 | level, search_start, empty_size); |
54aa1f4d CM |
1174 | if (IS_ERR(cow)) |
1175 | return PTR_ERR(cow); | |
6702ed49 | 1176 | |
b4ce94de CM |
1177 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1178 | ||
5f39d397 | 1179 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 1180 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1181 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1182 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1183 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1184 | BTRFS_HEADER_FLAG_RELOC); | |
1185 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1186 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1187 | else | |
1188 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1189 | |
0a4e5586 | 1190 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(), |
2b82032c YZ |
1191 | BTRFS_FSID_SIZE); |
1192 | ||
be1a5564 | 1193 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1194 | if (ret) { |
79787eaa | 1195 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1196 | return ret; |
1197 | } | |
1a40e23b | 1198 | |
27cdeb70 | 1199 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 JB |
1200 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
1201 | if (ret) | |
1202 | return ret; | |
1203 | } | |
3fd0a558 | 1204 | |
02217ed2 | 1205 | if (buf == root->node) { |
925baedd | 1206 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1207 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1208 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1209 | parent_start = buf->start; | |
1210 | else | |
1211 | parent_start = 0; | |
925baedd | 1212 | |
5f39d397 | 1213 | extent_buffer_get(cow); |
90f8d62e | 1214 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1215 | rcu_assign_pointer(root->node, cow); |
925baedd | 1216 | |
f0486c68 | 1217 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1218 | last_ref); |
5f39d397 | 1219 | free_extent_buffer(buf); |
0b86a832 | 1220 | add_root_to_dirty_list(root); |
02217ed2 | 1221 | } else { |
5d4f98a2 YZ |
1222 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1223 | parent_start = parent->start; | |
1224 | else | |
1225 | parent_start = 0; | |
1226 | ||
1227 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e | 1228 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
c8cc6341 | 1229 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1230 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1231 | cow->start); |
74493f7a CM |
1232 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1233 | trans->transid); | |
d6025579 | 1234 | btrfs_mark_buffer_dirty(parent); |
5de865ee FDBM |
1235 | if (last_ref) { |
1236 | ret = tree_mod_log_free_eb(root->fs_info, buf); | |
1237 | if (ret) { | |
1238 | btrfs_abort_transaction(trans, root, ret); | |
1239 | return ret; | |
1240 | } | |
1241 | } | |
f0486c68 | 1242 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1243 | last_ref); |
02217ed2 | 1244 | } |
925baedd CM |
1245 | if (unlock_orig) |
1246 | btrfs_tree_unlock(buf); | |
3083ee2e | 1247 | free_extent_buffer_stale(buf); |
ccd467d6 | 1248 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1249 | *cow_ret = cow; |
02217ed2 CM |
1250 | return 0; |
1251 | } | |
1252 | ||
5d9e75c4 JS |
1253 | /* |
1254 | * returns the logical address of the oldest predecessor of the given root. | |
1255 | * entries older than time_seq are ignored. | |
1256 | */ | |
1257 | static struct tree_mod_elem * | |
1258 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1259 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1260 | { |
1261 | struct tree_mod_elem *tm; | |
1262 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1263 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1264 | int looped = 0; |
1265 | ||
1266 | if (!time_seq) | |
35a3621b | 1267 | return NULL; |
5d9e75c4 JS |
1268 | |
1269 | /* | |
1270 | * the very last operation that's logged for a root is the replacement | |
1271 | * operation (if it is replaced at all). this has the index of the *new* | |
1272 | * root, making it the very first operation that's logged for this root. | |
1273 | */ | |
1274 | while (1) { | |
1275 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1276 | time_seq); | |
1277 | if (!looped && !tm) | |
35a3621b | 1278 | return NULL; |
5d9e75c4 | 1279 | /* |
28da9fb4 JS |
1280 | * if there are no tree operation for the oldest root, we simply |
1281 | * return it. this should only happen if that (old) root is at | |
1282 | * level 0. | |
5d9e75c4 | 1283 | */ |
28da9fb4 JS |
1284 | if (!tm) |
1285 | break; | |
5d9e75c4 | 1286 | |
28da9fb4 JS |
1287 | /* |
1288 | * if there's an operation that's not a root replacement, we | |
1289 | * found the oldest version of our root. normally, we'll find a | |
1290 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1291 | */ | |
5d9e75c4 JS |
1292 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1293 | break; | |
1294 | ||
1295 | found = tm; | |
1296 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1297 | looped = 1; |
1298 | } | |
1299 | ||
a95236d9 JS |
1300 | /* if there's no old root to return, return what we found instead */ |
1301 | if (!found) | |
1302 | found = tm; | |
1303 | ||
5d9e75c4 JS |
1304 | return found; |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * tm is a pointer to the first operation to rewind within eb. then, all | |
1309 | * previous operations will be rewinded (until we reach something older than | |
1310 | * time_seq). | |
1311 | */ | |
1312 | static void | |
f1ca7e98 JB |
1313 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1314 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1315 | { |
1316 | u32 n; | |
1317 | struct rb_node *next; | |
1318 | struct tree_mod_elem *tm = first_tm; | |
1319 | unsigned long o_dst; | |
1320 | unsigned long o_src; | |
1321 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1322 | ||
1323 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1324 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1325 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1326 | /* |
1327 | * all the operations are recorded with the operator used for | |
1328 | * the modification. as we're going backwards, we do the | |
1329 | * opposite of each operation here. | |
1330 | */ | |
1331 | switch (tm->op) { | |
1332 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1333 | BUG_ON(tm->slot < n); | |
1c697d4a | 1334 | /* Fallthrough */ |
95c80bb1 | 1335 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1336 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1337 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1338 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1339 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1340 | tm->generation); | |
4c3e6969 | 1341 | n++; |
5d9e75c4 JS |
1342 | break; |
1343 | case MOD_LOG_KEY_REPLACE: | |
1344 | BUG_ON(tm->slot >= n); | |
1345 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1346 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1347 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1348 | tm->generation); | |
1349 | break; | |
1350 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1351 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1352 | n--; |
1353 | break; | |
1354 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1355 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1356 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1357 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1358 | tm->move.nr_items * p_size); |
1359 | break; | |
1360 | case MOD_LOG_ROOT_REPLACE: | |
1361 | /* | |
1362 | * this operation is special. for roots, this must be | |
1363 | * handled explicitly before rewinding. | |
1364 | * for non-roots, this operation may exist if the node | |
1365 | * was a root: root A -> child B; then A gets empty and | |
1366 | * B is promoted to the new root. in the mod log, we'll | |
1367 | * have a root-replace operation for B, a tree block | |
1368 | * that is no root. we simply ignore that operation. | |
1369 | */ | |
1370 | break; | |
1371 | } | |
1372 | next = rb_next(&tm->node); | |
1373 | if (!next) | |
1374 | break; | |
1375 | tm = container_of(next, struct tree_mod_elem, node); | |
1376 | if (tm->index != first_tm->index) | |
1377 | break; | |
1378 | } | |
f1ca7e98 | 1379 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1380 | btrfs_set_header_nritems(eb, n); |
1381 | } | |
1382 | ||
47fb091f JS |
1383 | /* |
1384 | * Called with eb read locked. If the buffer cannot be rewinded, the same buffer | |
1385 | * is returned. If rewind operations happen, a fresh buffer is returned. The | |
1386 | * returned buffer is always read-locked. If the returned buffer is not the | |
1387 | * input buffer, the lock on the input buffer is released and the input buffer | |
1388 | * is freed (its refcount is decremented). | |
1389 | */ | |
5d9e75c4 | 1390 | static struct extent_buffer * |
9ec72677 JB |
1391 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1392 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1393 | { |
1394 | struct extent_buffer *eb_rewin; | |
1395 | struct tree_mod_elem *tm; | |
1396 | ||
1397 | if (!time_seq) | |
1398 | return eb; | |
1399 | ||
1400 | if (btrfs_header_level(eb) == 0) | |
1401 | return eb; | |
1402 | ||
1403 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1404 | if (!tm) | |
1405 | return eb; | |
1406 | ||
9ec72677 JB |
1407 | btrfs_set_path_blocking(path); |
1408 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1409 | ||
5d9e75c4 JS |
1410 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1411 | BUG_ON(tm->slot != 0); | |
1412 | eb_rewin = alloc_dummy_extent_buffer(eb->start, | |
1413 | fs_info->tree_root->nodesize); | |
db7f3436 | 1414 | if (!eb_rewin) { |
9ec72677 | 1415 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1416 | free_extent_buffer(eb); |
1417 | return NULL; | |
1418 | } | |
5d9e75c4 JS |
1419 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1420 | btrfs_set_header_backref_rev(eb_rewin, | |
1421 | btrfs_header_backref_rev(eb)); | |
1422 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1423 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1424 | } else { |
1425 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1426 | if (!eb_rewin) { |
9ec72677 | 1427 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1428 | free_extent_buffer(eb); |
1429 | return NULL; | |
1430 | } | |
5d9e75c4 JS |
1431 | } |
1432 | ||
9ec72677 JB |
1433 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1434 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1435 | free_extent_buffer(eb); |
1436 | ||
47fb091f JS |
1437 | extent_buffer_get(eb_rewin); |
1438 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1439 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1440 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1441 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1442 | |
1443 | return eb_rewin; | |
1444 | } | |
1445 | ||
8ba97a15 JS |
1446 | /* |
1447 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1448 | * value. If there are no changes, the current root->root_node is returned. If | |
1449 | * anything changed in between, there's a fresh buffer allocated on which the | |
1450 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1451 | * Returns NULL on error (with no locks held). | |
1452 | */ | |
5d9e75c4 JS |
1453 | static inline struct extent_buffer * |
1454 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1455 | { | |
1456 | struct tree_mod_elem *tm; | |
30b0463a JS |
1457 | struct extent_buffer *eb = NULL; |
1458 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1459 | struct extent_buffer *old; |
a95236d9 | 1460 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1461 | u64 old_generation = 0; |
a95236d9 | 1462 | u64 logical; |
834328a8 | 1463 | u32 blocksize; |
5d9e75c4 | 1464 | |
30b0463a JS |
1465 | eb_root = btrfs_read_lock_root_node(root); |
1466 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); | |
5d9e75c4 | 1467 | if (!tm) |
30b0463a | 1468 | return eb_root; |
5d9e75c4 | 1469 | |
a95236d9 JS |
1470 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1471 | old_root = &tm->old_root; | |
1472 | old_generation = tm->generation; | |
1473 | logical = old_root->logical; | |
1474 | } else { | |
30b0463a | 1475 | logical = eb_root->start; |
a95236d9 | 1476 | } |
5d9e75c4 | 1477 | |
a95236d9 | 1478 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 | 1479 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1480 | btrfs_tree_read_unlock(eb_root); |
1481 | free_extent_buffer(eb_root); | |
834328a8 | 1482 | blocksize = btrfs_level_size(root, old_root->level); |
7bfdcf7f | 1483 | old = read_tree_block(root, logical, blocksize, 0); |
fae7f21c | 1484 | if (WARN_ON(!old || !extent_buffer_uptodate(old))) { |
416bc658 | 1485 | free_extent_buffer(old); |
efe120a0 FH |
1486 | btrfs_warn(root->fs_info, |
1487 | "failed to read tree block %llu from get_old_root", logical); | |
834328a8 | 1488 | } else { |
7bfdcf7f LB |
1489 | eb = btrfs_clone_extent_buffer(old); |
1490 | free_extent_buffer(old); | |
834328a8 JS |
1491 | } |
1492 | } else if (old_root) { | |
30b0463a JS |
1493 | btrfs_tree_read_unlock(eb_root); |
1494 | free_extent_buffer(eb_root); | |
28da9fb4 | 1495 | eb = alloc_dummy_extent_buffer(logical, root->nodesize); |
834328a8 | 1496 | } else { |
9ec72677 | 1497 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1498 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1499 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1500 | free_extent_buffer(eb_root); |
834328a8 JS |
1501 | } |
1502 | ||
8ba97a15 JS |
1503 | if (!eb) |
1504 | return NULL; | |
d6381084 | 1505 | extent_buffer_get(eb); |
8ba97a15 | 1506 | btrfs_tree_read_lock(eb); |
a95236d9 | 1507 | if (old_root) { |
5d9e75c4 JS |
1508 | btrfs_set_header_bytenr(eb, eb->start); |
1509 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1510 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1511 | btrfs_set_header_level(eb, old_root->level); |
1512 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1513 | } |
28da9fb4 | 1514 | if (tm) |
f1ca7e98 | 1515 | __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1516 | else |
1517 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1518 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1519 | |
1520 | return eb; | |
1521 | } | |
1522 | ||
5b6602e7 JS |
1523 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1524 | { | |
1525 | struct tree_mod_elem *tm; | |
1526 | int level; | |
30b0463a | 1527 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1528 | |
30b0463a | 1529 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1530 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1531 | level = tm->old_root.level; | |
1532 | } else { | |
30b0463a | 1533 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1534 | } |
30b0463a | 1535 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1536 | |
1537 | return level; | |
1538 | } | |
1539 | ||
5d4f98a2 YZ |
1540 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1541 | struct btrfs_root *root, | |
1542 | struct extent_buffer *buf) | |
1543 | { | |
f1ebcc74 LB |
1544 | /* ensure we can see the force_cow */ |
1545 | smp_rmb(); | |
1546 | ||
1547 | /* | |
1548 | * We do not need to cow a block if | |
1549 | * 1) this block is not created or changed in this transaction; | |
1550 | * 2) this block does not belong to TREE_RELOC tree; | |
1551 | * 3) the root is not forced COW. | |
1552 | * | |
1553 | * What is forced COW: | |
1554 | * when we create snapshot during commiting the transaction, | |
1555 | * after we've finished coping src root, we must COW the shared | |
1556 | * block to ensure the metadata consistency. | |
1557 | */ | |
5d4f98a2 YZ |
1558 | if (btrfs_header_generation(buf) == trans->transid && |
1559 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1560 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1561 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1562 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1563 | return 0; |
1564 | return 1; | |
1565 | } | |
1566 | ||
d352ac68 CM |
1567 | /* |
1568 | * cows a single block, see __btrfs_cow_block for the real work. | |
1569 | * This version of it has extra checks so that a block isn't cow'd more than | |
1570 | * once per transaction, as long as it hasn't been written yet | |
1571 | */ | |
d397712b | 1572 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1573 | struct btrfs_root *root, struct extent_buffer *buf, |
1574 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1575 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1576 | { |
1577 | u64 search_start; | |
f510cfec | 1578 | int ret; |
dc17ff8f | 1579 | |
31b1a2bd JL |
1580 | if (trans->transaction != root->fs_info->running_transaction) |
1581 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1582 | trans->transid, |
6702ed49 | 1583 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1584 | |
1585 | if (trans->transid != root->fs_info->generation) | |
1586 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1587 | trans->transid, root->fs_info->generation); |
dc17ff8f | 1588 | |
5d4f98a2 | 1589 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1590 | *cow_ret = buf; |
1591 | return 0; | |
1592 | } | |
c487685d | 1593 | |
0b86a832 | 1594 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1595 | |
1596 | if (parent) | |
1597 | btrfs_set_lock_blocking(parent); | |
1598 | btrfs_set_lock_blocking(buf); | |
1599 | ||
f510cfec | 1600 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1601 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1602 | |
1603 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1604 | ||
f510cfec | 1605 | return ret; |
6702ed49 CM |
1606 | } |
1607 | ||
d352ac68 CM |
1608 | /* |
1609 | * helper function for defrag to decide if two blocks pointed to by a | |
1610 | * node are actually close by | |
1611 | */ | |
6b80053d | 1612 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1613 | { |
6b80053d | 1614 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1615 | return 1; |
6b80053d | 1616 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1617 | return 1; |
1618 | return 0; | |
1619 | } | |
1620 | ||
081e9573 CM |
1621 | /* |
1622 | * compare two keys in a memcmp fashion | |
1623 | */ | |
1624 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1625 | { | |
1626 | struct btrfs_key k1; | |
1627 | ||
1628 | btrfs_disk_key_to_cpu(&k1, disk); | |
1629 | ||
20736aba | 1630 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1631 | } |
1632 | ||
f3465ca4 JB |
1633 | /* |
1634 | * same as comp_keys only with two btrfs_key's | |
1635 | */ | |
5d4f98a2 | 1636 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1637 | { |
1638 | if (k1->objectid > k2->objectid) | |
1639 | return 1; | |
1640 | if (k1->objectid < k2->objectid) | |
1641 | return -1; | |
1642 | if (k1->type > k2->type) | |
1643 | return 1; | |
1644 | if (k1->type < k2->type) | |
1645 | return -1; | |
1646 | if (k1->offset > k2->offset) | |
1647 | return 1; | |
1648 | if (k1->offset < k2->offset) | |
1649 | return -1; | |
1650 | return 0; | |
1651 | } | |
081e9573 | 1652 | |
d352ac68 CM |
1653 | /* |
1654 | * this is used by the defrag code to go through all the | |
1655 | * leaves pointed to by a node and reallocate them so that | |
1656 | * disk order is close to key order | |
1657 | */ | |
6702ed49 | 1658 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1659 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1660 | int start_slot, u64 *last_ret, |
a6b6e75e | 1661 | struct btrfs_key *progress) |
6702ed49 | 1662 | { |
6b80053d | 1663 | struct extent_buffer *cur; |
6702ed49 | 1664 | u64 blocknr; |
ca7a79ad | 1665 | u64 gen; |
e9d0b13b CM |
1666 | u64 search_start = *last_ret; |
1667 | u64 last_block = 0; | |
6702ed49 CM |
1668 | u64 other; |
1669 | u32 parent_nritems; | |
6702ed49 CM |
1670 | int end_slot; |
1671 | int i; | |
1672 | int err = 0; | |
f2183bde | 1673 | int parent_level; |
6b80053d CM |
1674 | int uptodate; |
1675 | u32 blocksize; | |
081e9573 CM |
1676 | int progress_passed = 0; |
1677 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1678 | |
5708b959 | 1679 | parent_level = btrfs_header_level(parent); |
5708b959 | 1680 | |
6c1500f2 JL |
1681 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1682 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1683 | |
6b80053d | 1684 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1685 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1686 | end_slot = parent_nritems; |
1687 | ||
1688 | if (parent_nritems == 1) | |
1689 | return 0; | |
1690 | ||
b4ce94de CM |
1691 | btrfs_set_lock_blocking(parent); |
1692 | ||
6702ed49 CM |
1693 | for (i = start_slot; i < end_slot; i++) { |
1694 | int close = 1; | |
a6b6e75e | 1695 | |
081e9573 CM |
1696 | btrfs_node_key(parent, &disk_key, i); |
1697 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1698 | continue; | |
1699 | ||
1700 | progress_passed = 1; | |
6b80053d | 1701 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1702 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1703 | if (last_block == 0) |
1704 | last_block = blocknr; | |
5708b959 | 1705 | |
6702ed49 | 1706 | if (i > 0) { |
6b80053d CM |
1707 | other = btrfs_node_blockptr(parent, i - 1); |
1708 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1709 | } |
0ef3e66b | 1710 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1711 | other = btrfs_node_blockptr(parent, i + 1); |
1712 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1713 | } |
e9d0b13b CM |
1714 | if (close) { |
1715 | last_block = blocknr; | |
6702ed49 | 1716 | continue; |
e9d0b13b | 1717 | } |
6702ed49 | 1718 | |
6b80053d CM |
1719 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1720 | if (cur) | |
b9fab919 | 1721 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1722 | else |
1723 | uptodate = 0; | |
5708b959 | 1724 | if (!cur || !uptodate) { |
6b80053d CM |
1725 | if (!cur) { |
1726 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1727 | blocksize, gen); |
416bc658 JB |
1728 | if (!cur || !extent_buffer_uptodate(cur)) { |
1729 | free_extent_buffer(cur); | |
97d9a8a4 | 1730 | return -EIO; |
416bc658 | 1731 | } |
6b80053d | 1732 | } else if (!uptodate) { |
018642a1 TI |
1733 | err = btrfs_read_buffer(cur, gen); |
1734 | if (err) { | |
1735 | free_extent_buffer(cur); | |
1736 | return err; | |
1737 | } | |
f2183bde | 1738 | } |
6702ed49 | 1739 | } |
e9d0b13b | 1740 | if (search_start == 0) |
6b80053d | 1741 | search_start = last_block; |
e9d0b13b | 1742 | |
e7a84565 | 1743 | btrfs_tree_lock(cur); |
b4ce94de | 1744 | btrfs_set_lock_blocking(cur); |
6b80053d | 1745 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1746 | &cur, search_start, |
6b80053d | 1747 | min(16 * blocksize, |
9fa8cfe7 | 1748 | (end_slot - i) * blocksize)); |
252c38f0 | 1749 | if (err) { |
e7a84565 | 1750 | btrfs_tree_unlock(cur); |
6b80053d | 1751 | free_extent_buffer(cur); |
6702ed49 | 1752 | break; |
252c38f0 | 1753 | } |
e7a84565 CM |
1754 | search_start = cur->start; |
1755 | last_block = cur->start; | |
f2183bde | 1756 | *last_ret = search_start; |
e7a84565 CM |
1757 | btrfs_tree_unlock(cur); |
1758 | free_extent_buffer(cur); | |
6702ed49 CM |
1759 | } |
1760 | return err; | |
1761 | } | |
1762 | ||
74123bd7 CM |
1763 | /* |
1764 | * The leaf data grows from end-to-front in the node. | |
1765 | * this returns the address of the start of the last item, | |
1766 | * which is the stop of the leaf data stack | |
1767 | */ | |
123abc88 | 1768 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1769 | struct extent_buffer *leaf) |
be0e5c09 | 1770 | { |
5f39d397 | 1771 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1772 | if (nr == 0) |
123abc88 | 1773 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1774 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1775 | } |
1776 | ||
aa5d6bed | 1777 | |
74123bd7 | 1778 | /* |
5f39d397 CM |
1779 | * search for key in the extent_buffer. The items start at offset p, |
1780 | * and they are item_size apart. There are 'max' items in p. | |
1781 | * | |
74123bd7 CM |
1782 | * the slot in the array is returned via slot, and it points to |
1783 | * the place where you would insert key if it is not found in | |
1784 | * the array. | |
1785 | * | |
1786 | * slot may point to max if the key is bigger than all of the keys | |
1787 | */ | |
e02119d5 CM |
1788 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1789 | unsigned long p, | |
1790 | int item_size, struct btrfs_key *key, | |
1791 | int max, int *slot) | |
be0e5c09 CM |
1792 | { |
1793 | int low = 0; | |
1794 | int high = max; | |
1795 | int mid; | |
1796 | int ret; | |
479965d6 | 1797 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1798 | struct btrfs_disk_key unaligned; |
1799 | unsigned long offset; | |
5f39d397 CM |
1800 | char *kaddr = NULL; |
1801 | unsigned long map_start = 0; | |
1802 | unsigned long map_len = 0; | |
479965d6 | 1803 | int err; |
be0e5c09 | 1804 | |
d397712b | 1805 | while (low < high) { |
be0e5c09 | 1806 | mid = (low + high) / 2; |
5f39d397 CM |
1807 | offset = p + mid * item_size; |
1808 | ||
a6591715 | 1809 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1810 | (offset + sizeof(struct btrfs_disk_key)) > |
1811 | map_start + map_len) { | |
934d375b CM |
1812 | |
1813 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1814 | sizeof(struct btrfs_disk_key), |
a6591715 | 1815 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1816 | |
1817 | if (!err) { | |
1818 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1819 | map_start); | |
1820 | } else { | |
1821 | read_extent_buffer(eb, &unaligned, | |
1822 | offset, sizeof(unaligned)); | |
1823 | tmp = &unaligned; | |
1824 | } | |
5f39d397 | 1825 | |
5f39d397 CM |
1826 | } else { |
1827 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1828 | map_start); | |
1829 | } | |
be0e5c09 CM |
1830 | ret = comp_keys(tmp, key); |
1831 | ||
1832 | if (ret < 0) | |
1833 | low = mid + 1; | |
1834 | else if (ret > 0) | |
1835 | high = mid; | |
1836 | else { | |
1837 | *slot = mid; | |
1838 | return 0; | |
1839 | } | |
1840 | } | |
1841 | *slot = low; | |
1842 | return 1; | |
1843 | } | |
1844 | ||
97571fd0 CM |
1845 | /* |
1846 | * simple bin_search frontend that does the right thing for | |
1847 | * leaves vs nodes | |
1848 | */ | |
5f39d397 CM |
1849 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1850 | int level, int *slot) | |
be0e5c09 | 1851 | { |
f775738f | 1852 | if (level == 0) |
5f39d397 CM |
1853 | return generic_bin_search(eb, |
1854 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1855 | sizeof(struct btrfs_item), |
5f39d397 | 1856 | key, btrfs_header_nritems(eb), |
7518a238 | 1857 | slot); |
f775738f | 1858 | else |
5f39d397 CM |
1859 | return generic_bin_search(eb, |
1860 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1861 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1862 | key, btrfs_header_nritems(eb), |
7518a238 | 1863 | slot); |
be0e5c09 CM |
1864 | } |
1865 | ||
5d4f98a2 YZ |
1866 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1867 | int level, int *slot) | |
1868 | { | |
1869 | return bin_search(eb, key, level, slot); | |
1870 | } | |
1871 | ||
f0486c68 YZ |
1872 | static void root_add_used(struct btrfs_root *root, u32 size) |
1873 | { | |
1874 | spin_lock(&root->accounting_lock); | |
1875 | btrfs_set_root_used(&root->root_item, | |
1876 | btrfs_root_used(&root->root_item) + size); | |
1877 | spin_unlock(&root->accounting_lock); | |
1878 | } | |
1879 | ||
1880 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1881 | { | |
1882 | spin_lock(&root->accounting_lock); | |
1883 | btrfs_set_root_used(&root->root_item, | |
1884 | btrfs_root_used(&root->root_item) - size); | |
1885 | spin_unlock(&root->accounting_lock); | |
1886 | } | |
1887 | ||
d352ac68 CM |
1888 | /* given a node and slot number, this reads the blocks it points to. The |
1889 | * extent buffer is returned with a reference taken (but unlocked). | |
1890 | * NULL is returned on error. | |
1891 | */ | |
e02119d5 | 1892 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1893 | struct extent_buffer *parent, int slot) |
bb803951 | 1894 | { |
ca7a79ad | 1895 | int level = btrfs_header_level(parent); |
416bc658 JB |
1896 | struct extent_buffer *eb; |
1897 | ||
bb803951 CM |
1898 | if (slot < 0) |
1899 | return NULL; | |
5f39d397 | 1900 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1901 | return NULL; |
ca7a79ad CM |
1902 | |
1903 | BUG_ON(level == 0); | |
1904 | ||
416bc658 JB |
1905 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
1906 | btrfs_level_size(root, level - 1), | |
1907 | btrfs_node_ptr_generation(parent, slot)); | |
1908 | if (eb && !extent_buffer_uptodate(eb)) { | |
1909 | free_extent_buffer(eb); | |
1910 | eb = NULL; | |
1911 | } | |
1912 | ||
1913 | return eb; | |
bb803951 CM |
1914 | } |
1915 | ||
d352ac68 CM |
1916 | /* |
1917 | * node level balancing, used to make sure nodes are in proper order for | |
1918 | * item deletion. We balance from the top down, so we have to make sure | |
1919 | * that a deletion won't leave an node completely empty later on. | |
1920 | */ | |
e02119d5 | 1921 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1922 | struct btrfs_root *root, |
1923 | struct btrfs_path *path, int level) | |
bb803951 | 1924 | { |
5f39d397 CM |
1925 | struct extent_buffer *right = NULL; |
1926 | struct extent_buffer *mid; | |
1927 | struct extent_buffer *left = NULL; | |
1928 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1929 | int ret = 0; |
1930 | int wret; | |
1931 | int pslot; | |
bb803951 | 1932 | int orig_slot = path->slots[level]; |
79f95c82 | 1933 | u64 orig_ptr; |
bb803951 CM |
1934 | |
1935 | if (level == 0) | |
1936 | return 0; | |
1937 | ||
5f39d397 | 1938 | mid = path->nodes[level]; |
b4ce94de | 1939 | |
bd681513 CM |
1940 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1941 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1942 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1943 | ||
1d4f8a0c | 1944 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1945 | |
a05a9bb1 | 1946 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1947 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1948 | pslot = path->slots[level + 1]; |
1949 | } | |
bb803951 | 1950 | |
40689478 CM |
1951 | /* |
1952 | * deal with the case where there is only one pointer in the root | |
1953 | * by promoting the node below to a root | |
1954 | */ | |
5f39d397 CM |
1955 | if (!parent) { |
1956 | struct extent_buffer *child; | |
bb803951 | 1957 | |
5f39d397 | 1958 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1959 | return 0; |
1960 | ||
1961 | /* promote the child to a root */ | |
5f39d397 | 1962 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1963 | if (!child) { |
1964 | ret = -EROFS; | |
1965 | btrfs_std_error(root->fs_info, ret); | |
1966 | goto enospc; | |
1967 | } | |
1968 | ||
925baedd | 1969 | btrfs_tree_lock(child); |
b4ce94de | 1970 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1971 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1972 | if (ret) { |
1973 | btrfs_tree_unlock(child); | |
1974 | free_extent_buffer(child); | |
1975 | goto enospc; | |
1976 | } | |
2f375ab9 | 1977 | |
90f8d62e | 1978 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1979 | rcu_assign_pointer(root->node, child); |
925baedd | 1980 | |
0b86a832 | 1981 | add_root_to_dirty_list(root); |
925baedd | 1982 | btrfs_tree_unlock(child); |
b4ce94de | 1983 | |
925baedd | 1984 | path->locks[level] = 0; |
bb803951 | 1985 | path->nodes[level] = NULL; |
5f39d397 | 1986 | clean_tree_block(trans, root, mid); |
925baedd | 1987 | btrfs_tree_unlock(mid); |
bb803951 | 1988 | /* once for the path */ |
5f39d397 | 1989 | free_extent_buffer(mid); |
f0486c68 YZ |
1990 | |
1991 | root_sub_used(root, mid->len); | |
5581a51a | 1992 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1993 | /* once for the root ptr */ |
3083ee2e | 1994 | free_extent_buffer_stale(mid); |
f0486c68 | 1995 | return 0; |
bb803951 | 1996 | } |
5f39d397 | 1997 | if (btrfs_header_nritems(mid) > |
123abc88 | 1998 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1999 | return 0; |
2000 | ||
5f39d397 CM |
2001 | left = read_node_slot(root, parent, pslot - 1); |
2002 | if (left) { | |
925baedd | 2003 | btrfs_tree_lock(left); |
b4ce94de | 2004 | btrfs_set_lock_blocking(left); |
5f39d397 | 2005 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 2006 | parent, pslot - 1, &left); |
54aa1f4d CM |
2007 | if (wret) { |
2008 | ret = wret; | |
2009 | goto enospc; | |
2010 | } | |
2cc58cf2 | 2011 | } |
5f39d397 CM |
2012 | right = read_node_slot(root, parent, pslot + 1); |
2013 | if (right) { | |
925baedd | 2014 | btrfs_tree_lock(right); |
b4ce94de | 2015 | btrfs_set_lock_blocking(right); |
5f39d397 | 2016 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 2017 | parent, pslot + 1, &right); |
2cc58cf2 CM |
2018 | if (wret) { |
2019 | ret = wret; | |
2020 | goto enospc; | |
2021 | } | |
2022 | } | |
2023 | ||
2024 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
2025 | if (left) { |
2026 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 2027 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
2028 | if (wret < 0) |
2029 | ret = wret; | |
bb803951 | 2030 | } |
79f95c82 CM |
2031 | |
2032 | /* | |
2033 | * then try to empty the right most buffer into the middle | |
2034 | */ | |
5f39d397 | 2035 | if (right) { |
971a1f66 | 2036 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 2037 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 2038 | ret = wret; |
5f39d397 | 2039 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 2040 | clean_tree_block(trans, root, right); |
925baedd | 2041 | btrfs_tree_unlock(right); |
afe5fea7 | 2042 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 2043 | root_sub_used(root, right->len); |
5581a51a | 2044 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 2045 | free_extent_buffer_stale(right); |
f0486c68 | 2046 | right = NULL; |
bb803951 | 2047 | } else { |
5f39d397 CM |
2048 | struct btrfs_disk_key right_key; |
2049 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 2050 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2051 | pslot + 1, 0); |
5f39d397 CM |
2052 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
2053 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
2054 | } |
2055 | } | |
5f39d397 | 2056 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
2057 | /* |
2058 | * we're not allowed to leave a node with one item in the | |
2059 | * tree during a delete. A deletion from lower in the tree | |
2060 | * could try to delete the only pointer in this node. | |
2061 | * So, pull some keys from the left. | |
2062 | * There has to be a left pointer at this point because | |
2063 | * otherwise we would have pulled some pointers from the | |
2064 | * right | |
2065 | */ | |
305a26af MF |
2066 | if (!left) { |
2067 | ret = -EROFS; | |
2068 | btrfs_std_error(root->fs_info, ret); | |
2069 | goto enospc; | |
2070 | } | |
5f39d397 | 2071 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 2072 | if (wret < 0) { |
79f95c82 | 2073 | ret = wret; |
54aa1f4d CM |
2074 | goto enospc; |
2075 | } | |
bce4eae9 CM |
2076 | if (wret == 1) { |
2077 | wret = push_node_left(trans, root, left, mid, 1); | |
2078 | if (wret < 0) | |
2079 | ret = wret; | |
2080 | } | |
79f95c82 CM |
2081 | BUG_ON(wret == 1); |
2082 | } | |
5f39d397 | 2083 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 2084 | clean_tree_block(trans, root, mid); |
925baedd | 2085 | btrfs_tree_unlock(mid); |
afe5fea7 | 2086 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2087 | root_sub_used(root, mid->len); |
5581a51a | 2088 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2089 | free_extent_buffer_stale(mid); |
f0486c68 | 2090 | mid = NULL; |
79f95c82 CM |
2091 | } else { |
2092 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2093 | struct btrfs_disk_key mid_key; |
2094 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 2095 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 2096 | pslot, 0); |
5f39d397 CM |
2097 | btrfs_set_node_key(parent, &mid_key, pslot); |
2098 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2099 | } |
bb803951 | 2100 | |
79f95c82 | 2101 | /* update the path */ |
5f39d397 CM |
2102 | if (left) { |
2103 | if (btrfs_header_nritems(left) > orig_slot) { | |
2104 | extent_buffer_get(left); | |
925baedd | 2105 | /* left was locked after cow */ |
5f39d397 | 2106 | path->nodes[level] = left; |
bb803951 CM |
2107 | path->slots[level + 1] -= 1; |
2108 | path->slots[level] = orig_slot; | |
925baedd CM |
2109 | if (mid) { |
2110 | btrfs_tree_unlock(mid); | |
5f39d397 | 2111 | free_extent_buffer(mid); |
925baedd | 2112 | } |
bb803951 | 2113 | } else { |
5f39d397 | 2114 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2115 | path->slots[level] = orig_slot; |
2116 | } | |
2117 | } | |
79f95c82 | 2118 | /* double check we haven't messed things up */ |
e20d96d6 | 2119 | if (orig_ptr != |
5f39d397 | 2120 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2121 | BUG(); |
54aa1f4d | 2122 | enospc: |
925baedd CM |
2123 | if (right) { |
2124 | btrfs_tree_unlock(right); | |
5f39d397 | 2125 | free_extent_buffer(right); |
925baedd CM |
2126 | } |
2127 | if (left) { | |
2128 | if (path->nodes[level] != left) | |
2129 | btrfs_tree_unlock(left); | |
5f39d397 | 2130 | free_extent_buffer(left); |
925baedd | 2131 | } |
bb803951 CM |
2132 | return ret; |
2133 | } | |
2134 | ||
d352ac68 CM |
2135 | /* Node balancing for insertion. Here we only split or push nodes around |
2136 | * when they are completely full. This is also done top down, so we | |
2137 | * have to be pessimistic. | |
2138 | */ | |
d397712b | 2139 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2140 | struct btrfs_root *root, |
2141 | struct btrfs_path *path, int level) | |
e66f709b | 2142 | { |
5f39d397 CM |
2143 | struct extent_buffer *right = NULL; |
2144 | struct extent_buffer *mid; | |
2145 | struct extent_buffer *left = NULL; | |
2146 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2147 | int ret = 0; |
2148 | int wret; | |
2149 | int pslot; | |
2150 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2151 | |
2152 | if (level == 0) | |
2153 | return 1; | |
2154 | ||
5f39d397 | 2155 | mid = path->nodes[level]; |
7bb86316 | 2156 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2157 | |
a05a9bb1 | 2158 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2159 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2160 | pslot = path->slots[level + 1]; |
2161 | } | |
e66f709b | 2162 | |
5f39d397 | 2163 | if (!parent) |
e66f709b | 2164 | return 1; |
e66f709b | 2165 | |
5f39d397 | 2166 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
2167 | |
2168 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2169 | if (left) { |
e66f709b | 2170 | u32 left_nr; |
925baedd CM |
2171 | |
2172 | btrfs_tree_lock(left); | |
b4ce94de CM |
2173 | btrfs_set_lock_blocking(left); |
2174 | ||
5f39d397 | 2175 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
2176 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2177 | wret = 1; | |
2178 | } else { | |
5f39d397 | 2179 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2180 | pslot - 1, &left); |
54aa1f4d CM |
2181 | if (ret) |
2182 | wret = 1; | |
2183 | else { | |
54aa1f4d | 2184 | wret = push_node_left(trans, root, |
971a1f66 | 2185 | left, mid, 0); |
54aa1f4d | 2186 | } |
33ade1f8 | 2187 | } |
e66f709b CM |
2188 | if (wret < 0) |
2189 | ret = wret; | |
2190 | if (wret == 0) { | |
5f39d397 | 2191 | struct btrfs_disk_key disk_key; |
e66f709b | 2192 | orig_slot += left_nr; |
5f39d397 | 2193 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 2194 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2195 | pslot, 0); |
5f39d397 CM |
2196 | btrfs_set_node_key(parent, &disk_key, pslot); |
2197 | btrfs_mark_buffer_dirty(parent); | |
2198 | if (btrfs_header_nritems(left) > orig_slot) { | |
2199 | path->nodes[level] = left; | |
e66f709b CM |
2200 | path->slots[level + 1] -= 1; |
2201 | path->slots[level] = orig_slot; | |
925baedd | 2202 | btrfs_tree_unlock(mid); |
5f39d397 | 2203 | free_extent_buffer(mid); |
e66f709b CM |
2204 | } else { |
2205 | orig_slot -= | |
5f39d397 | 2206 | btrfs_header_nritems(left); |
e66f709b | 2207 | path->slots[level] = orig_slot; |
925baedd | 2208 | btrfs_tree_unlock(left); |
5f39d397 | 2209 | free_extent_buffer(left); |
e66f709b | 2210 | } |
e66f709b CM |
2211 | return 0; |
2212 | } | |
925baedd | 2213 | btrfs_tree_unlock(left); |
5f39d397 | 2214 | free_extent_buffer(left); |
e66f709b | 2215 | } |
925baedd | 2216 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
2217 | |
2218 | /* | |
2219 | * then try to empty the right most buffer into the middle | |
2220 | */ | |
5f39d397 | 2221 | if (right) { |
33ade1f8 | 2222 | u32 right_nr; |
b4ce94de | 2223 | |
925baedd | 2224 | btrfs_tree_lock(right); |
b4ce94de CM |
2225 | btrfs_set_lock_blocking(right); |
2226 | ||
5f39d397 | 2227 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2228 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2229 | wret = 1; | |
2230 | } else { | |
5f39d397 CM |
2231 | ret = btrfs_cow_block(trans, root, right, |
2232 | parent, pslot + 1, | |
9fa8cfe7 | 2233 | &right); |
54aa1f4d CM |
2234 | if (ret) |
2235 | wret = 1; | |
2236 | else { | |
54aa1f4d | 2237 | wret = balance_node_right(trans, root, |
5f39d397 | 2238 | right, mid); |
54aa1f4d | 2239 | } |
33ade1f8 | 2240 | } |
e66f709b CM |
2241 | if (wret < 0) |
2242 | ret = wret; | |
2243 | if (wret == 0) { | |
5f39d397 CM |
2244 | struct btrfs_disk_key disk_key; |
2245 | ||
2246 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2247 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2248 | pslot + 1, 0); |
5f39d397 CM |
2249 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2250 | btrfs_mark_buffer_dirty(parent); | |
2251 | ||
2252 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2253 | path->nodes[level] = right; | |
e66f709b CM |
2254 | path->slots[level + 1] += 1; |
2255 | path->slots[level] = orig_slot - | |
5f39d397 | 2256 | btrfs_header_nritems(mid); |
925baedd | 2257 | btrfs_tree_unlock(mid); |
5f39d397 | 2258 | free_extent_buffer(mid); |
e66f709b | 2259 | } else { |
925baedd | 2260 | btrfs_tree_unlock(right); |
5f39d397 | 2261 | free_extent_buffer(right); |
e66f709b | 2262 | } |
e66f709b CM |
2263 | return 0; |
2264 | } | |
925baedd | 2265 | btrfs_tree_unlock(right); |
5f39d397 | 2266 | free_extent_buffer(right); |
e66f709b | 2267 | } |
e66f709b CM |
2268 | return 1; |
2269 | } | |
2270 | ||
3c69faec | 2271 | /* |
d352ac68 CM |
2272 | * readahead one full node of leaves, finding things that are close |
2273 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2274 | */ |
c8c42864 CM |
2275 | static void reada_for_search(struct btrfs_root *root, |
2276 | struct btrfs_path *path, | |
2277 | int level, int slot, u64 objectid) | |
3c69faec | 2278 | { |
5f39d397 | 2279 | struct extent_buffer *node; |
01f46658 | 2280 | struct btrfs_disk_key disk_key; |
3c69faec | 2281 | u32 nritems; |
3c69faec | 2282 | u64 search; |
a7175319 | 2283 | u64 target; |
6b80053d | 2284 | u64 nread = 0; |
cb25c2ea | 2285 | u64 gen; |
3c69faec | 2286 | int direction = path->reada; |
5f39d397 | 2287 | struct extent_buffer *eb; |
6b80053d CM |
2288 | u32 nr; |
2289 | u32 blocksize; | |
2290 | u32 nscan = 0; | |
db94535d | 2291 | |
a6b6e75e | 2292 | if (level != 1) |
6702ed49 CM |
2293 | return; |
2294 | ||
2295 | if (!path->nodes[level]) | |
3c69faec CM |
2296 | return; |
2297 | ||
5f39d397 | 2298 | node = path->nodes[level]; |
925baedd | 2299 | |
3c69faec | 2300 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
2301 | blocksize = btrfs_level_size(root, level - 1); |
2302 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
2303 | if (eb) { |
2304 | free_extent_buffer(eb); | |
3c69faec CM |
2305 | return; |
2306 | } | |
2307 | ||
a7175319 | 2308 | target = search; |
6b80053d | 2309 | |
5f39d397 | 2310 | nritems = btrfs_header_nritems(node); |
6b80053d | 2311 | nr = slot; |
25b8b936 | 2312 | |
d397712b | 2313 | while (1) { |
6b80053d CM |
2314 | if (direction < 0) { |
2315 | if (nr == 0) | |
2316 | break; | |
2317 | nr--; | |
2318 | } else if (direction > 0) { | |
2319 | nr++; | |
2320 | if (nr >= nritems) | |
2321 | break; | |
3c69faec | 2322 | } |
01f46658 CM |
2323 | if (path->reada < 0 && objectid) { |
2324 | btrfs_node_key(node, &disk_key, nr); | |
2325 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2326 | break; | |
2327 | } | |
6b80053d | 2328 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2329 | if ((search <= target && target - search <= 65536) || |
2330 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2331 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 2332 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
2333 | nread += blocksize; |
2334 | } | |
2335 | nscan++; | |
a7175319 | 2336 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2337 | break; |
3c69faec CM |
2338 | } |
2339 | } | |
925baedd | 2340 | |
0b08851f JB |
2341 | static noinline void reada_for_balance(struct btrfs_root *root, |
2342 | struct btrfs_path *path, int level) | |
b4ce94de CM |
2343 | { |
2344 | int slot; | |
2345 | int nritems; | |
2346 | struct extent_buffer *parent; | |
2347 | struct extent_buffer *eb; | |
2348 | u64 gen; | |
2349 | u64 block1 = 0; | |
2350 | u64 block2 = 0; | |
b4ce94de CM |
2351 | int blocksize; |
2352 | ||
8c594ea8 | 2353 | parent = path->nodes[level + 1]; |
b4ce94de | 2354 | if (!parent) |
0b08851f | 2355 | return; |
b4ce94de CM |
2356 | |
2357 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2358 | slot = path->slots[level + 1]; |
b4ce94de CM |
2359 | blocksize = btrfs_level_size(root, level); |
2360 | ||
2361 | if (slot > 0) { | |
2362 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2363 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
2364 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
2365 | /* |
2366 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2367 | * don't want to return eagain here. That will loop | |
2368 | * forever | |
2369 | */ | |
2370 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2371 | block1 = 0; |
2372 | free_extent_buffer(eb); | |
2373 | } | |
8c594ea8 | 2374 | if (slot + 1 < nritems) { |
b4ce94de CM |
2375 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2376 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
2377 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 2378 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2379 | block2 = 0; |
2380 | free_extent_buffer(eb); | |
2381 | } | |
8c594ea8 | 2382 | |
0b08851f JB |
2383 | if (block1) |
2384 | readahead_tree_block(root, block1, blocksize, 0); | |
2385 | if (block2) | |
2386 | readahead_tree_block(root, block2, blocksize, 0); | |
b4ce94de CM |
2387 | } |
2388 | ||
2389 | ||
d352ac68 | 2390 | /* |
d397712b CM |
2391 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2392 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2393 | * operations on the tree might require changing key pointers higher up in the | |
2394 | * tree. | |
d352ac68 | 2395 | * |
d397712b CM |
2396 | * callers might also have set path->keep_locks, which tells this code to keep |
2397 | * the lock if the path points to the last slot in the block. This is part of | |
2398 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2399 | * |
d397712b CM |
2400 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2401 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2402 | */ |
e02119d5 | 2403 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2404 | int lowest_unlock, int min_write_lock_level, |
2405 | int *write_lock_level) | |
925baedd CM |
2406 | { |
2407 | int i; | |
2408 | int skip_level = level; | |
051e1b9f | 2409 | int no_skips = 0; |
925baedd CM |
2410 | struct extent_buffer *t; |
2411 | ||
2412 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2413 | if (!path->nodes[i]) | |
2414 | break; | |
2415 | if (!path->locks[i]) | |
2416 | break; | |
051e1b9f | 2417 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2418 | skip_level = i + 1; |
2419 | continue; | |
2420 | } | |
051e1b9f | 2421 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2422 | u32 nritems; |
2423 | t = path->nodes[i]; | |
2424 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2425 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2426 | skip_level = i + 1; |
2427 | continue; | |
2428 | } | |
2429 | } | |
051e1b9f CM |
2430 | if (skip_level < i && i >= lowest_unlock) |
2431 | no_skips = 1; | |
2432 | ||
925baedd CM |
2433 | t = path->nodes[i]; |
2434 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2435 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2436 | path->locks[i] = 0; |
f7c79f30 CM |
2437 | if (write_lock_level && |
2438 | i > min_write_lock_level && | |
2439 | i <= *write_lock_level) { | |
2440 | *write_lock_level = i - 1; | |
2441 | } | |
925baedd CM |
2442 | } |
2443 | } | |
2444 | } | |
2445 | ||
b4ce94de CM |
2446 | /* |
2447 | * This releases any locks held in the path starting at level and | |
2448 | * going all the way up to the root. | |
2449 | * | |
2450 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2451 | * corner cases, such as COW of the block at slot zero in the node. This | |
2452 | * ignores those rules, and it should only be called when there are no | |
2453 | * more updates to be done higher up in the tree. | |
2454 | */ | |
2455 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2456 | { | |
2457 | int i; | |
2458 | ||
09a2a8f9 | 2459 | if (path->keep_locks) |
b4ce94de CM |
2460 | return; |
2461 | ||
2462 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2463 | if (!path->nodes[i]) | |
12f4dacc | 2464 | continue; |
b4ce94de | 2465 | if (!path->locks[i]) |
12f4dacc | 2466 | continue; |
bd681513 | 2467 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2468 | path->locks[i] = 0; |
2469 | } | |
2470 | } | |
2471 | ||
c8c42864 CM |
2472 | /* |
2473 | * helper function for btrfs_search_slot. The goal is to find a block | |
2474 | * in cache without setting the path to blocking. If we find the block | |
2475 | * we return zero and the path is unchanged. | |
2476 | * | |
2477 | * If we can't find the block, we set the path blocking and do some | |
2478 | * reada. -EAGAIN is returned and the search must be repeated. | |
2479 | */ | |
2480 | static int | |
2481 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2482 | struct btrfs_root *root, struct btrfs_path *p, | |
2483 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2484 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2485 | { |
2486 | u64 blocknr; | |
2487 | u64 gen; | |
2488 | u32 blocksize; | |
2489 | struct extent_buffer *b = *eb_ret; | |
2490 | struct extent_buffer *tmp; | |
76a05b35 | 2491 | int ret; |
c8c42864 CM |
2492 | |
2493 | blocknr = btrfs_node_blockptr(b, slot); | |
2494 | gen = btrfs_node_ptr_generation(b, slot); | |
2495 | blocksize = btrfs_level_size(root, level - 1); | |
2496 | ||
2497 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 2498 | if (tmp) { |
b9fab919 | 2499 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2500 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2501 | *eb_ret = tmp; | |
2502 | return 0; | |
2503 | } | |
2504 | ||
2505 | /* the pages were up to date, but we failed | |
2506 | * the generation number check. Do a full | |
2507 | * read for the generation number that is correct. | |
2508 | * We must do this without dropping locks so | |
2509 | * we can trust our generation number | |
2510 | */ | |
2511 | btrfs_set_path_blocking(p); | |
2512 | ||
2513 | /* now we're allowed to do a blocking uptodate check */ | |
2514 | ret = btrfs_read_buffer(tmp, gen); | |
2515 | if (!ret) { | |
2516 | *eb_ret = tmp; | |
2517 | return 0; | |
cb44921a | 2518 | } |
bdf7c00e JB |
2519 | free_extent_buffer(tmp); |
2520 | btrfs_release_path(p); | |
2521 | return -EIO; | |
c8c42864 CM |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * reduce lock contention at high levels | |
2526 | * of the btree by dropping locks before | |
76a05b35 CM |
2527 | * we read. Don't release the lock on the current |
2528 | * level because we need to walk this node to figure | |
2529 | * out which blocks to read. | |
c8c42864 | 2530 | */ |
8c594ea8 CM |
2531 | btrfs_unlock_up_safe(p, level + 1); |
2532 | btrfs_set_path_blocking(p); | |
2533 | ||
cb44921a | 2534 | free_extent_buffer(tmp); |
c8c42864 CM |
2535 | if (p->reada) |
2536 | reada_for_search(root, p, level, slot, key->objectid); | |
2537 | ||
b3b4aa74 | 2538 | btrfs_release_path(p); |
76a05b35 CM |
2539 | |
2540 | ret = -EAGAIN; | |
5bdd3536 | 2541 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
2542 | if (tmp) { |
2543 | /* | |
2544 | * If the read above didn't mark this buffer up to date, | |
2545 | * it will never end up being up to date. Set ret to EIO now | |
2546 | * and give up so that our caller doesn't loop forever | |
2547 | * on our EAGAINs. | |
2548 | */ | |
b9fab919 | 2549 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2550 | ret = -EIO; |
c8c42864 | 2551 | free_extent_buffer(tmp); |
76a05b35 CM |
2552 | } |
2553 | return ret; | |
c8c42864 CM |
2554 | } |
2555 | ||
2556 | /* | |
2557 | * helper function for btrfs_search_slot. This does all of the checks | |
2558 | * for node-level blocks and does any balancing required based on | |
2559 | * the ins_len. | |
2560 | * | |
2561 | * If no extra work was required, zero is returned. If we had to | |
2562 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2563 | * start over | |
2564 | */ | |
2565 | static int | |
2566 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2567 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2568 | struct extent_buffer *b, int level, int ins_len, |
2569 | int *write_lock_level) | |
c8c42864 CM |
2570 | { |
2571 | int ret; | |
2572 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2573 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2574 | int sret; | |
2575 | ||
bd681513 CM |
2576 | if (*write_lock_level < level + 1) { |
2577 | *write_lock_level = level + 1; | |
2578 | btrfs_release_path(p); | |
2579 | goto again; | |
2580 | } | |
2581 | ||
c8c42864 | 2582 | btrfs_set_path_blocking(p); |
0b08851f | 2583 | reada_for_balance(root, p, level); |
c8c42864 | 2584 | sret = split_node(trans, root, p, level); |
bd681513 | 2585 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2586 | |
2587 | BUG_ON(sret > 0); | |
2588 | if (sret) { | |
2589 | ret = sret; | |
2590 | goto done; | |
2591 | } | |
2592 | b = p->nodes[level]; | |
2593 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2594 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2595 | int sret; |
2596 | ||
bd681513 CM |
2597 | if (*write_lock_level < level + 1) { |
2598 | *write_lock_level = level + 1; | |
2599 | btrfs_release_path(p); | |
2600 | goto again; | |
2601 | } | |
2602 | ||
c8c42864 | 2603 | btrfs_set_path_blocking(p); |
0b08851f | 2604 | reada_for_balance(root, p, level); |
c8c42864 | 2605 | sret = balance_level(trans, root, p, level); |
bd681513 | 2606 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2607 | |
2608 | if (sret) { | |
2609 | ret = sret; | |
2610 | goto done; | |
2611 | } | |
2612 | b = p->nodes[level]; | |
2613 | if (!b) { | |
b3b4aa74 | 2614 | btrfs_release_path(p); |
c8c42864 CM |
2615 | goto again; |
2616 | } | |
2617 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2618 | } | |
2619 | return 0; | |
2620 | ||
2621 | again: | |
2622 | ret = -EAGAIN; | |
2623 | done: | |
2624 | return ret; | |
2625 | } | |
2626 | ||
d7396f07 FDBM |
2627 | static void key_search_validate(struct extent_buffer *b, |
2628 | struct btrfs_key *key, | |
2629 | int level) | |
2630 | { | |
2631 | #ifdef CONFIG_BTRFS_ASSERT | |
2632 | struct btrfs_disk_key disk_key; | |
2633 | ||
2634 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2635 | ||
2636 | if (level == 0) | |
2637 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2638 | offsetof(struct btrfs_leaf, items[0].key), | |
2639 | sizeof(disk_key))); | |
2640 | else | |
2641 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2642 | offsetof(struct btrfs_node, ptrs[0].key), | |
2643 | sizeof(disk_key))); | |
2644 | #endif | |
2645 | } | |
2646 | ||
2647 | static int key_search(struct extent_buffer *b, struct btrfs_key *key, | |
2648 | int level, int *prev_cmp, int *slot) | |
2649 | { | |
2650 | if (*prev_cmp != 0) { | |
2651 | *prev_cmp = bin_search(b, key, level, slot); | |
2652 | return *prev_cmp; | |
2653 | } | |
2654 | ||
2655 | key_search_validate(b, key, level); | |
2656 | *slot = 0; | |
2657 | ||
2658 | return 0; | |
2659 | } | |
2660 | ||
3f870c28 | 2661 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path, |
e33d5c3d KN |
2662 | u64 iobjectid, u64 ioff, u8 key_type, |
2663 | struct btrfs_key *found_key) | |
2664 | { | |
2665 | int ret; | |
2666 | struct btrfs_key key; | |
2667 | struct extent_buffer *eb; | |
3f870c28 | 2668 | struct btrfs_path *path; |
e33d5c3d KN |
2669 | |
2670 | key.type = key_type; | |
2671 | key.objectid = iobjectid; | |
2672 | key.offset = ioff; | |
2673 | ||
3f870c28 KN |
2674 | if (found_path == NULL) { |
2675 | path = btrfs_alloc_path(); | |
2676 | if (!path) | |
2677 | return -ENOMEM; | |
2678 | } else | |
2679 | path = found_path; | |
2680 | ||
e33d5c3d | 2681 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); |
3f870c28 KN |
2682 | if ((ret < 0) || (found_key == NULL)) { |
2683 | if (path != found_path) | |
2684 | btrfs_free_path(path); | |
e33d5c3d | 2685 | return ret; |
3f870c28 | 2686 | } |
e33d5c3d KN |
2687 | |
2688 | eb = path->nodes[0]; | |
2689 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2690 | ret = btrfs_next_leaf(fs_root, path); | |
2691 | if (ret) | |
2692 | return ret; | |
2693 | eb = path->nodes[0]; | |
2694 | } | |
2695 | ||
2696 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2697 | if (found_key->type != key.type || | |
2698 | found_key->objectid != key.objectid) | |
2699 | return 1; | |
2700 | ||
2701 | return 0; | |
2702 | } | |
2703 | ||
74123bd7 CM |
2704 | /* |
2705 | * look for key in the tree. path is filled in with nodes along the way | |
2706 | * if key is found, we return zero and you can find the item in the leaf | |
2707 | * level of the path (level 0) | |
2708 | * | |
2709 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2710 | * be inserted, and 1 is returned. If there are other errors during the |
2711 | * search a negative error number is returned. | |
97571fd0 CM |
2712 | * |
2713 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2714 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2715 | * possible) | |
74123bd7 | 2716 | */ |
e089f05c CM |
2717 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2718 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2719 | ins_len, int cow) | |
be0e5c09 | 2720 | { |
5f39d397 | 2721 | struct extent_buffer *b; |
be0e5c09 CM |
2722 | int slot; |
2723 | int ret; | |
33c66f43 | 2724 | int err; |
be0e5c09 | 2725 | int level; |
925baedd | 2726 | int lowest_unlock = 1; |
bd681513 CM |
2727 | int root_lock; |
2728 | /* everything at write_lock_level or lower must be write locked */ | |
2729 | int write_lock_level = 0; | |
9f3a7427 | 2730 | u8 lowest_level = 0; |
f7c79f30 | 2731 | int min_write_lock_level; |
d7396f07 | 2732 | int prev_cmp; |
9f3a7427 | 2733 | |
6702ed49 | 2734 | lowest_level = p->lowest_level; |
323ac95b | 2735 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2736 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2737 | BUG_ON(!cow && ins_len); |
25179201 | 2738 | |
bd681513 | 2739 | if (ins_len < 0) { |
925baedd | 2740 | lowest_unlock = 2; |
65b51a00 | 2741 | |
bd681513 CM |
2742 | /* when we are removing items, we might have to go up to level |
2743 | * two as we update tree pointers Make sure we keep write | |
2744 | * for those levels as well | |
2745 | */ | |
2746 | write_lock_level = 2; | |
2747 | } else if (ins_len > 0) { | |
2748 | /* | |
2749 | * for inserting items, make sure we have a write lock on | |
2750 | * level 1 so we can update keys | |
2751 | */ | |
2752 | write_lock_level = 1; | |
2753 | } | |
2754 | ||
2755 | if (!cow) | |
2756 | write_lock_level = -1; | |
2757 | ||
09a2a8f9 | 2758 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2759 | write_lock_level = BTRFS_MAX_LEVEL; |
2760 | ||
f7c79f30 CM |
2761 | min_write_lock_level = write_lock_level; |
2762 | ||
bb803951 | 2763 | again: |
d7396f07 | 2764 | prev_cmp = -1; |
bd681513 CM |
2765 | /* |
2766 | * we try very hard to do read locks on the root | |
2767 | */ | |
2768 | root_lock = BTRFS_READ_LOCK; | |
2769 | level = 0; | |
5d4f98a2 | 2770 | if (p->search_commit_root) { |
bd681513 CM |
2771 | /* |
2772 | * the commit roots are read only | |
2773 | * so we always do read locks | |
2774 | */ | |
3f8a18cc JB |
2775 | if (p->need_commit_sem) |
2776 | down_read(&root->fs_info->commit_root_sem); | |
5d4f98a2 YZ |
2777 | b = root->commit_root; |
2778 | extent_buffer_get(b); | |
bd681513 | 2779 | level = btrfs_header_level(b); |
3f8a18cc JB |
2780 | if (p->need_commit_sem) |
2781 | up_read(&root->fs_info->commit_root_sem); | |
5d4f98a2 | 2782 | if (!p->skip_locking) |
bd681513 | 2783 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2784 | } else { |
bd681513 | 2785 | if (p->skip_locking) { |
5d4f98a2 | 2786 | b = btrfs_root_node(root); |
bd681513 CM |
2787 | level = btrfs_header_level(b); |
2788 | } else { | |
2789 | /* we don't know the level of the root node | |
2790 | * until we actually have it read locked | |
2791 | */ | |
2792 | b = btrfs_read_lock_root_node(root); | |
2793 | level = btrfs_header_level(b); | |
2794 | if (level <= write_lock_level) { | |
2795 | /* whoops, must trade for write lock */ | |
2796 | btrfs_tree_read_unlock(b); | |
2797 | free_extent_buffer(b); | |
2798 | b = btrfs_lock_root_node(root); | |
2799 | root_lock = BTRFS_WRITE_LOCK; | |
2800 | ||
2801 | /* the level might have changed, check again */ | |
2802 | level = btrfs_header_level(b); | |
2803 | } | |
2804 | } | |
5d4f98a2 | 2805 | } |
bd681513 CM |
2806 | p->nodes[level] = b; |
2807 | if (!p->skip_locking) | |
2808 | p->locks[level] = root_lock; | |
925baedd | 2809 | |
eb60ceac | 2810 | while (b) { |
5f39d397 | 2811 | level = btrfs_header_level(b); |
65b51a00 CM |
2812 | |
2813 | /* | |
2814 | * setup the path here so we can release it under lock | |
2815 | * contention with the cow code | |
2816 | */ | |
02217ed2 | 2817 | if (cow) { |
c8c42864 CM |
2818 | /* |
2819 | * if we don't really need to cow this block | |
2820 | * then we don't want to set the path blocking, | |
2821 | * so we test it here | |
2822 | */ | |
5d4f98a2 | 2823 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2824 | goto cow_done; |
5d4f98a2 | 2825 | |
b4ce94de CM |
2826 | btrfs_set_path_blocking(p); |
2827 | ||
bd681513 CM |
2828 | /* |
2829 | * must have write locks on this node and the | |
2830 | * parent | |
2831 | */ | |
5124e00e JB |
2832 | if (level > write_lock_level || |
2833 | (level + 1 > write_lock_level && | |
2834 | level + 1 < BTRFS_MAX_LEVEL && | |
2835 | p->nodes[level + 1])) { | |
bd681513 CM |
2836 | write_lock_level = level + 1; |
2837 | btrfs_release_path(p); | |
2838 | goto again; | |
2839 | } | |
2840 | ||
33c66f43 YZ |
2841 | err = btrfs_cow_block(trans, root, b, |
2842 | p->nodes[level + 1], | |
2843 | p->slots[level + 1], &b); | |
2844 | if (err) { | |
33c66f43 | 2845 | ret = err; |
65b51a00 | 2846 | goto done; |
54aa1f4d | 2847 | } |
02217ed2 | 2848 | } |
65b51a00 | 2849 | cow_done: |
eb60ceac | 2850 | p->nodes[level] = b; |
bd681513 | 2851 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2852 | |
2853 | /* | |
2854 | * we have a lock on b and as long as we aren't changing | |
2855 | * the tree, there is no way to for the items in b to change. | |
2856 | * It is safe to drop the lock on our parent before we | |
2857 | * go through the expensive btree search on b. | |
2858 | * | |
eb653de1 FDBM |
2859 | * If we're inserting or deleting (ins_len != 0), then we might |
2860 | * be changing slot zero, which may require changing the parent. | |
2861 | * So, we can't drop the lock until after we know which slot | |
2862 | * we're operating on. | |
b4ce94de | 2863 | */ |
eb653de1 FDBM |
2864 | if (!ins_len && !p->keep_locks) { |
2865 | int u = level + 1; | |
2866 | ||
2867 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2868 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2869 | p->locks[u] = 0; | |
2870 | } | |
2871 | } | |
b4ce94de | 2872 | |
d7396f07 | 2873 | ret = key_search(b, key, level, &prev_cmp, &slot); |
b4ce94de | 2874 | |
5f39d397 | 2875 | if (level != 0) { |
33c66f43 YZ |
2876 | int dec = 0; |
2877 | if (ret && slot > 0) { | |
2878 | dec = 1; | |
be0e5c09 | 2879 | slot -= 1; |
33c66f43 | 2880 | } |
be0e5c09 | 2881 | p->slots[level] = slot; |
33c66f43 | 2882 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2883 | ins_len, &write_lock_level); |
33c66f43 | 2884 | if (err == -EAGAIN) |
c8c42864 | 2885 | goto again; |
33c66f43 YZ |
2886 | if (err) { |
2887 | ret = err; | |
c8c42864 | 2888 | goto done; |
33c66f43 | 2889 | } |
c8c42864 CM |
2890 | b = p->nodes[level]; |
2891 | slot = p->slots[level]; | |
b4ce94de | 2892 | |
bd681513 CM |
2893 | /* |
2894 | * slot 0 is special, if we change the key | |
2895 | * we have to update the parent pointer | |
2896 | * which means we must have a write lock | |
2897 | * on the parent | |
2898 | */ | |
eb653de1 | 2899 | if (slot == 0 && ins_len && |
bd681513 CM |
2900 | write_lock_level < level + 1) { |
2901 | write_lock_level = level + 1; | |
2902 | btrfs_release_path(p); | |
2903 | goto again; | |
2904 | } | |
2905 | ||
f7c79f30 CM |
2906 | unlock_up(p, level, lowest_unlock, |
2907 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2908 | |
925baedd | 2909 | if (level == lowest_level) { |
33c66f43 YZ |
2910 | if (dec) |
2911 | p->slots[level]++; | |
5b21f2ed | 2912 | goto done; |
925baedd | 2913 | } |
ca7a79ad | 2914 | |
33c66f43 | 2915 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2916 | &b, level, slot, key, 0); |
33c66f43 | 2917 | if (err == -EAGAIN) |
c8c42864 | 2918 | goto again; |
33c66f43 YZ |
2919 | if (err) { |
2920 | ret = err; | |
76a05b35 | 2921 | goto done; |
33c66f43 | 2922 | } |
76a05b35 | 2923 | |
b4ce94de | 2924 | if (!p->skip_locking) { |
bd681513 CM |
2925 | level = btrfs_header_level(b); |
2926 | if (level <= write_lock_level) { | |
2927 | err = btrfs_try_tree_write_lock(b); | |
2928 | if (!err) { | |
2929 | btrfs_set_path_blocking(p); | |
2930 | btrfs_tree_lock(b); | |
2931 | btrfs_clear_path_blocking(p, b, | |
2932 | BTRFS_WRITE_LOCK); | |
2933 | } | |
2934 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2935 | } else { | |
2936 | err = btrfs_try_tree_read_lock(b); | |
2937 | if (!err) { | |
2938 | btrfs_set_path_blocking(p); | |
2939 | btrfs_tree_read_lock(b); | |
2940 | btrfs_clear_path_blocking(p, b, | |
2941 | BTRFS_READ_LOCK); | |
2942 | } | |
2943 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2944 | } |
bd681513 | 2945 | p->nodes[level] = b; |
b4ce94de | 2946 | } |
be0e5c09 CM |
2947 | } else { |
2948 | p->slots[level] = slot; | |
87b29b20 YZ |
2949 | if (ins_len > 0 && |
2950 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2951 | if (write_lock_level < 1) { |
2952 | write_lock_level = 1; | |
2953 | btrfs_release_path(p); | |
2954 | goto again; | |
2955 | } | |
2956 | ||
b4ce94de | 2957 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2958 | err = split_leaf(trans, root, key, |
2959 | p, ins_len, ret == 0); | |
bd681513 | 2960 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2961 | |
33c66f43 YZ |
2962 | BUG_ON(err > 0); |
2963 | if (err) { | |
2964 | ret = err; | |
65b51a00 CM |
2965 | goto done; |
2966 | } | |
5c680ed6 | 2967 | } |
459931ec | 2968 | if (!p->search_for_split) |
f7c79f30 CM |
2969 | unlock_up(p, level, lowest_unlock, |
2970 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2971 | goto done; |
be0e5c09 CM |
2972 | } |
2973 | } | |
65b51a00 CM |
2974 | ret = 1; |
2975 | done: | |
b4ce94de CM |
2976 | /* |
2977 | * we don't really know what they plan on doing with the path | |
2978 | * from here on, so for now just mark it as blocking | |
2979 | */ | |
b9473439 CM |
2980 | if (!p->leave_spinning) |
2981 | btrfs_set_path_blocking(p); | |
76a05b35 | 2982 | if (ret < 0) |
b3b4aa74 | 2983 | btrfs_release_path(p); |
65b51a00 | 2984 | return ret; |
be0e5c09 CM |
2985 | } |
2986 | ||
5d9e75c4 JS |
2987 | /* |
2988 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2989 | * current state of the tree together with the operations recorded in the tree | |
2990 | * modification log to search for the key in a previous version of this tree, as | |
2991 | * denoted by the time_seq parameter. | |
2992 | * | |
2993 | * Naturally, there is no support for insert, delete or cow operations. | |
2994 | * | |
2995 | * The resulting path and return value will be set up as if we called | |
2996 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2997 | */ | |
2998 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2999 | struct btrfs_path *p, u64 time_seq) | |
3000 | { | |
3001 | struct extent_buffer *b; | |
3002 | int slot; | |
3003 | int ret; | |
3004 | int err; | |
3005 | int level; | |
3006 | int lowest_unlock = 1; | |
3007 | u8 lowest_level = 0; | |
d4b4087c | 3008 | int prev_cmp = -1; |
5d9e75c4 JS |
3009 | |
3010 | lowest_level = p->lowest_level; | |
3011 | WARN_ON(p->nodes[0] != NULL); | |
3012 | ||
3013 | if (p->search_commit_root) { | |
3014 | BUG_ON(time_seq); | |
3015 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3016 | } | |
3017 | ||
3018 | again: | |
5d9e75c4 | 3019 | b = get_old_root(root, time_seq); |
5d9e75c4 | 3020 | level = btrfs_header_level(b); |
5d9e75c4 JS |
3021 | p->locks[level] = BTRFS_READ_LOCK; |
3022 | ||
3023 | while (b) { | |
3024 | level = btrfs_header_level(b); | |
3025 | p->nodes[level] = b; | |
3026 | btrfs_clear_path_blocking(p, NULL, 0); | |
3027 | ||
3028 | /* | |
3029 | * we have a lock on b and as long as we aren't changing | |
3030 | * the tree, there is no way to for the items in b to change. | |
3031 | * It is safe to drop the lock on our parent before we | |
3032 | * go through the expensive btree search on b. | |
3033 | */ | |
3034 | btrfs_unlock_up_safe(p, level + 1); | |
3035 | ||
d4b4087c JB |
3036 | /* |
3037 | * Since we can unwind eb's we want to do a real search every | |
3038 | * time. | |
3039 | */ | |
3040 | prev_cmp = -1; | |
d7396f07 | 3041 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
3042 | |
3043 | if (level != 0) { | |
3044 | int dec = 0; | |
3045 | if (ret && slot > 0) { | |
3046 | dec = 1; | |
3047 | slot -= 1; | |
3048 | } | |
3049 | p->slots[level] = slot; | |
3050 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3051 | ||
3052 | if (level == lowest_level) { | |
3053 | if (dec) | |
3054 | p->slots[level]++; | |
3055 | goto done; | |
3056 | } | |
3057 | ||
3058 | err = read_block_for_search(NULL, root, p, &b, level, | |
3059 | slot, key, time_seq); | |
3060 | if (err == -EAGAIN) | |
3061 | goto again; | |
3062 | if (err) { | |
3063 | ret = err; | |
3064 | goto done; | |
3065 | } | |
3066 | ||
3067 | level = btrfs_header_level(b); | |
3068 | err = btrfs_try_tree_read_lock(b); | |
3069 | if (!err) { | |
3070 | btrfs_set_path_blocking(p); | |
3071 | btrfs_tree_read_lock(b); | |
3072 | btrfs_clear_path_blocking(p, b, | |
3073 | BTRFS_READ_LOCK); | |
3074 | } | |
9ec72677 | 3075 | b = tree_mod_log_rewind(root->fs_info, p, b, time_seq); |
db7f3436 JB |
3076 | if (!b) { |
3077 | ret = -ENOMEM; | |
3078 | goto done; | |
3079 | } | |
5d9e75c4 JS |
3080 | p->locks[level] = BTRFS_READ_LOCK; |
3081 | p->nodes[level] = b; | |
5d9e75c4 JS |
3082 | } else { |
3083 | p->slots[level] = slot; | |
3084 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3085 | goto done; | |
3086 | } | |
3087 | } | |
3088 | ret = 1; | |
3089 | done: | |
3090 | if (!p->leave_spinning) | |
3091 | btrfs_set_path_blocking(p); | |
3092 | if (ret < 0) | |
3093 | btrfs_release_path(p); | |
3094 | ||
3095 | return ret; | |
3096 | } | |
3097 | ||
2f38b3e1 AJ |
3098 | /* |
3099 | * helper to use instead of search slot if no exact match is needed but | |
3100 | * instead the next or previous item should be returned. | |
3101 | * When find_higher is true, the next higher item is returned, the next lower | |
3102 | * otherwise. | |
3103 | * When return_any and find_higher are both true, and no higher item is found, | |
3104 | * return the next lower instead. | |
3105 | * When return_any is true and find_higher is false, and no lower item is found, | |
3106 | * return the next higher instead. | |
3107 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3108 | * < 0 on error | |
3109 | */ | |
3110 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
3111 | struct btrfs_key *key, struct btrfs_path *p, | |
3112 | int find_higher, int return_any) | |
3113 | { | |
3114 | int ret; | |
3115 | struct extent_buffer *leaf; | |
3116 | ||
3117 | again: | |
3118 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3119 | if (ret <= 0) | |
3120 | return ret; | |
3121 | /* | |
3122 | * a return value of 1 means the path is at the position where the | |
3123 | * item should be inserted. Normally this is the next bigger item, | |
3124 | * but in case the previous item is the last in a leaf, path points | |
3125 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3126 | * item. | |
3127 | */ | |
3128 | leaf = p->nodes[0]; | |
3129 | ||
3130 | if (find_higher) { | |
3131 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3132 | ret = btrfs_next_leaf(root, p); | |
3133 | if (ret <= 0) | |
3134 | return ret; | |
3135 | if (!return_any) | |
3136 | return 1; | |
3137 | /* | |
3138 | * no higher item found, return the next | |
3139 | * lower instead | |
3140 | */ | |
3141 | return_any = 0; | |
3142 | find_higher = 0; | |
3143 | btrfs_release_path(p); | |
3144 | goto again; | |
3145 | } | |
3146 | } else { | |
e6793769 AJ |
3147 | if (p->slots[0] == 0) { |
3148 | ret = btrfs_prev_leaf(root, p); | |
3149 | if (ret < 0) | |
3150 | return ret; | |
3151 | if (!ret) { | |
23c6bf6a FDBM |
3152 | leaf = p->nodes[0]; |
3153 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3154 | p->slots[0]--; | |
e6793769 | 3155 | return 0; |
2f38b3e1 | 3156 | } |
e6793769 AJ |
3157 | if (!return_any) |
3158 | return 1; | |
3159 | /* | |
3160 | * no lower item found, return the next | |
3161 | * higher instead | |
3162 | */ | |
3163 | return_any = 0; | |
3164 | find_higher = 1; | |
3165 | btrfs_release_path(p); | |
3166 | goto again; | |
3167 | } else { | |
2f38b3e1 AJ |
3168 | --p->slots[0]; |
3169 | } | |
3170 | } | |
3171 | return 0; | |
3172 | } | |
3173 | ||
74123bd7 CM |
3174 | /* |
3175 | * adjust the pointers going up the tree, starting at level | |
3176 | * making sure the right key of each node is points to 'key'. | |
3177 | * This is used after shifting pointers to the left, so it stops | |
3178 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3179 | * higher levels | |
aa5d6bed | 3180 | * |
74123bd7 | 3181 | */ |
d6a0a126 | 3182 | static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 3183 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3184 | { |
3185 | int i; | |
5f39d397 CM |
3186 | struct extent_buffer *t; |
3187 | ||
234b63a0 | 3188 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3189 | int tslot = path->slots[i]; |
eb60ceac | 3190 | if (!path->nodes[i]) |
be0e5c09 | 3191 | break; |
5f39d397 | 3192 | t = path->nodes[i]; |
32adf090 | 3193 | tree_mod_log_set_node_key(root->fs_info, t, tslot, 1); |
5f39d397 | 3194 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3195 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3196 | if (tslot != 0) |
3197 | break; | |
3198 | } | |
3199 | } | |
3200 | ||
31840ae1 ZY |
3201 | /* |
3202 | * update item key. | |
3203 | * | |
3204 | * This function isn't completely safe. It's the caller's responsibility | |
3205 | * that the new key won't break the order | |
3206 | */ | |
afe5fea7 | 3207 | void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 3208 | struct btrfs_key *new_key) |
31840ae1 ZY |
3209 | { |
3210 | struct btrfs_disk_key disk_key; | |
3211 | struct extent_buffer *eb; | |
3212 | int slot; | |
3213 | ||
3214 | eb = path->nodes[0]; | |
3215 | slot = path->slots[0]; | |
3216 | if (slot > 0) { | |
3217 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3218 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3219 | } |
3220 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3221 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3222 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3223 | } |
3224 | ||
3225 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3226 | btrfs_set_item_key(eb, &disk_key, slot); | |
3227 | btrfs_mark_buffer_dirty(eb); | |
3228 | if (slot == 0) | |
d6a0a126 | 3229 | fixup_low_keys(root, path, &disk_key, 1); |
31840ae1 ZY |
3230 | } |
3231 | ||
74123bd7 CM |
3232 | /* |
3233 | * try to push data from one node into the next node left in the | |
79f95c82 | 3234 | * tree. |
aa5d6bed CM |
3235 | * |
3236 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3237 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3238 | */ |
98ed5174 CM |
3239 | static int push_node_left(struct btrfs_trans_handle *trans, |
3240 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 3241 | struct extent_buffer *src, int empty) |
be0e5c09 | 3242 | { |
be0e5c09 | 3243 | int push_items = 0; |
bb803951 CM |
3244 | int src_nritems; |
3245 | int dst_nritems; | |
aa5d6bed | 3246 | int ret = 0; |
be0e5c09 | 3247 | |
5f39d397 CM |
3248 | src_nritems = btrfs_header_nritems(src); |
3249 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3250 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
3251 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3252 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3253 | |
bce4eae9 | 3254 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3255 | return 1; |
3256 | ||
d397712b | 3257 | if (push_items <= 0) |
be0e5c09 CM |
3258 | return 1; |
3259 | ||
bce4eae9 | 3260 | if (empty) { |
971a1f66 | 3261 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3262 | if (push_items < src_nritems) { |
3263 | /* leave at least 8 pointers in the node if | |
3264 | * we aren't going to empty it | |
3265 | */ | |
3266 | if (src_nritems - push_items < 8) { | |
3267 | if (push_items <= 8) | |
3268 | return 1; | |
3269 | push_items -= 8; | |
3270 | } | |
3271 | } | |
3272 | } else | |
3273 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3274 | |
5de865ee FDBM |
3275 | ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
3276 | push_items); | |
3277 | if (ret) { | |
3278 | btrfs_abort_transaction(trans, root, ret); | |
3279 | return ret; | |
3280 | } | |
5f39d397 CM |
3281 | copy_extent_buffer(dst, src, |
3282 | btrfs_node_key_ptr_offset(dst_nritems), | |
3283 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3284 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3285 | |
bb803951 | 3286 | if (push_items < src_nritems) { |
57911b8b JS |
3287 | /* |
3288 | * don't call tree_mod_log_eb_move here, key removal was already | |
3289 | * fully logged by tree_mod_log_eb_copy above. | |
3290 | */ | |
5f39d397 CM |
3291 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3292 | btrfs_node_key_ptr_offset(push_items), | |
3293 | (src_nritems - push_items) * | |
3294 | sizeof(struct btrfs_key_ptr)); | |
3295 | } | |
3296 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3297 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3298 | btrfs_mark_buffer_dirty(src); | |
3299 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3300 | |
79f95c82 CM |
3301 | return ret; |
3302 | } | |
3303 | ||
3304 | /* | |
3305 | * try to push data from one node into the next node right in the | |
3306 | * tree. | |
3307 | * | |
3308 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3309 | * error, and > 0 if there was no room in the right hand block. | |
3310 | * | |
3311 | * this will only push up to 1/2 the contents of the left node over | |
3312 | */ | |
5f39d397 CM |
3313 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3314 | struct btrfs_root *root, | |
3315 | struct extent_buffer *dst, | |
3316 | struct extent_buffer *src) | |
79f95c82 | 3317 | { |
79f95c82 CM |
3318 | int push_items = 0; |
3319 | int max_push; | |
3320 | int src_nritems; | |
3321 | int dst_nritems; | |
3322 | int ret = 0; | |
79f95c82 | 3323 | |
7bb86316 CM |
3324 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3325 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3326 | ||
5f39d397 CM |
3327 | src_nritems = btrfs_header_nritems(src); |
3328 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3329 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3330 | if (push_items <= 0) |
79f95c82 | 3331 | return 1; |
bce4eae9 | 3332 | |
d397712b | 3333 | if (src_nritems < 4) |
bce4eae9 | 3334 | return 1; |
79f95c82 CM |
3335 | |
3336 | max_push = src_nritems / 2 + 1; | |
3337 | /* don't try to empty the node */ | |
d397712b | 3338 | if (max_push >= src_nritems) |
79f95c82 | 3339 | return 1; |
252c38f0 | 3340 | |
79f95c82 CM |
3341 | if (max_push < push_items) |
3342 | push_items = max_push; | |
3343 | ||
f230475e | 3344 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3345 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3346 | btrfs_node_key_ptr_offset(0), | |
3347 | (dst_nritems) * | |
3348 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3349 | |
5de865ee FDBM |
3350 | ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
3351 | src_nritems - push_items, push_items); | |
3352 | if (ret) { | |
3353 | btrfs_abort_transaction(trans, root, ret); | |
3354 | return ret; | |
3355 | } | |
5f39d397 CM |
3356 | copy_extent_buffer(dst, src, |
3357 | btrfs_node_key_ptr_offset(0), | |
3358 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3359 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3360 | |
5f39d397 CM |
3361 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3362 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3363 | |
5f39d397 CM |
3364 | btrfs_mark_buffer_dirty(src); |
3365 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3366 | |
aa5d6bed | 3367 | return ret; |
be0e5c09 CM |
3368 | } |
3369 | ||
97571fd0 CM |
3370 | /* |
3371 | * helper function to insert a new root level in the tree. | |
3372 | * A new node is allocated, and a single item is inserted to | |
3373 | * point to the existing root | |
aa5d6bed CM |
3374 | * |
3375 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3376 | */ |
d397712b | 3377 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3378 | struct btrfs_root *root, |
fdd99c72 | 3379 | struct btrfs_path *path, int level) |
5c680ed6 | 3380 | { |
7bb86316 | 3381 | u64 lower_gen; |
5f39d397 CM |
3382 | struct extent_buffer *lower; |
3383 | struct extent_buffer *c; | |
925baedd | 3384 | struct extent_buffer *old; |
5f39d397 | 3385 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3386 | |
3387 | BUG_ON(path->nodes[level]); | |
3388 | BUG_ON(path->nodes[level-1] != root->node); | |
3389 | ||
7bb86316 CM |
3390 | lower = path->nodes[level-1]; |
3391 | if (level == 1) | |
3392 | btrfs_item_key(lower, &lower_key, 0); | |
3393 | else | |
3394 | btrfs_node_key(lower, &lower_key, 0); | |
3395 | ||
31840ae1 | 3396 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 3397 | root->root_key.objectid, &lower_key, |
5581a51a | 3398 | level, root->node->start, 0); |
5f39d397 CM |
3399 | if (IS_ERR(c)) |
3400 | return PTR_ERR(c); | |
925baedd | 3401 | |
f0486c68 YZ |
3402 | root_add_used(root, root->nodesize); |
3403 | ||
5d4f98a2 | 3404 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3405 | btrfs_set_header_nritems(c, 1); |
3406 | btrfs_set_header_level(c, level); | |
db94535d | 3407 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3408 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3409 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3410 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3411 | |
0a4e5586 | 3412 | write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(), |
5f39d397 | 3413 | BTRFS_FSID_SIZE); |
e17cade2 CM |
3414 | |
3415 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 3416 | btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE); |
e17cade2 | 3417 | |
5f39d397 | 3418 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3419 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3420 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3421 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3422 | |
3423 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3424 | |
5f39d397 | 3425 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3426 | |
925baedd | 3427 | old = root->node; |
fdd99c72 | 3428 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3429 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3430 | |
3431 | /* the super has an extra ref to root->node */ | |
3432 | free_extent_buffer(old); | |
3433 | ||
0b86a832 | 3434 | add_root_to_dirty_list(root); |
5f39d397 CM |
3435 | extent_buffer_get(c); |
3436 | path->nodes[level] = c; | |
bd681513 | 3437 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3438 | path->slots[level] = 0; |
3439 | return 0; | |
3440 | } | |
3441 | ||
74123bd7 CM |
3442 | /* |
3443 | * worker function to insert a single pointer in a node. | |
3444 | * the node should have enough room for the pointer already | |
97571fd0 | 3445 | * |
74123bd7 CM |
3446 | * slot and level indicate where you want the key to go, and |
3447 | * blocknr is the block the key points to. | |
3448 | */ | |
143bede5 JM |
3449 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3450 | struct btrfs_root *root, struct btrfs_path *path, | |
3451 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3452 | int slot, int level) |
74123bd7 | 3453 | { |
5f39d397 | 3454 | struct extent_buffer *lower; |
74123bd7 | 3455 | int nritems; |
f3ea38da | 3456 | int ret; |
5c680ed6 CM |
3457 | |
3458 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3459 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3460 | lower = path->nodes[level]; |
3461 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3462 | BUG_ON(slot > nritems); |
143bede5 | 3463 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3464 | if (slot != nritems) { |
c3e06965 | 3465 | if (level) |
f3ea38da JS |
3466 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3467 | slot, nritems - slot); | |
5f39d397 CM |
3468 | memmove_extent_buffer(lower, |
3469 | btrfs_node_key_ptr_offset(slot + 1), | |
3470 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3471 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3472 | } |
c3e06965 | 3473 | if (level) { |
f3ea38da | 3474 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
c8cc6341 | 3475 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3476 | BUG_ON(ret < 0); |
3477 | } | |
5f39d397 | 3478 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3479 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3480 | WARN_ON(trans->transid == 0); |
3481 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3482 | btrfs_set_header_nritems(lower, nritems + 1); |
3483 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3484 | } |
3485 | ||
97571fd0 CM |
3486 | /* |
3487 | * split the node at the specified level in path in two. | |
3488 | * The path is corrected to point to the appropriate node after the split | |
3489 | * | |
3490 | * Before splitting this tries to make some room in the node by pushing | |
3491 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3492 | * |
3493 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3494 | */ |
e02119d5 CM |
3495 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3496 | struct btrfs_root *root, | |
3497 | struct btrfs_path *path, int level) | |
be0e5c09 | 3498 | { |
5f39d397 CM |
3499 | struct extent_buffer *c; |
3500 | struct extent_buffer *split; | |
3501 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3502 | int mid; |
5c680ed6 | 3503 | int ret; |
7518a238 | 3504 | u32 c_nritems; |
eb60ceac | 3505 | |
5f39d397 | 3506 | c = path->nodes[level]; |
7bb86316 | 3507 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3508 | if (c == root->node) { |
d9abbf1c | 3509 | /* |
90f8d62e JS |
3510 | * trying to split the root, lets make a new one |
3511 | * | |
fdd99c72 | 3512 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3513 | * insert_new_root, because that root buffer will be kept as a |
3514 | * normal node. We are going to log removal of half of the | |
3515 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3516 | * tree lock on the buffer, which is why we cannot race with | |
3517 | * other tree_mod_log users. | |
d9abbf1c | 3518 | */ |
fdd99c72 | 3519 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3520 | if (ret) |
3521 | return ret; | |
b3612421 | 3522 | } else { |
e66f709b | 3523 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3524 | c = path->nodes[level]; |
3525 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3526 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3527 | return 0; |
54aa1f4d CM |
3528 | if (ret < 0) |
3529 | return ret; | |
be0e5c09 | 3530 | } |
e66f709b | 3531 | |
5f39d397 | 3532 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3533 | mid = (c_nritems + 1) / 2; |
3534 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3535 | |
5d4f98a2 | 3536 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 3537 | root->root_key.objectid, |
5581a51a | 3538 | &disk_key, level, c->start, 0); |
5f39d397 CM |
3539 | if (IS_ERR(split)) |
3540 | return PTR_ERR(split); | |
3541 | ||
f0486c68 YZ |
3542 | root_add_used(root, root->nodesize); |
3543 | ||
5d4f98a2 | 3544 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3545 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3546 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3547 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3548 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3549 | btrfs_set_header_owner(split, root->root_key.objectid); |
3550 | write_extent_buffer(split, root->fs_info->fsid, | |
0a4e5586 | 3551 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 | 3552 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
b308bc2f | 3553 | btrfs_header_chunk_tree_uuid(split), |
e17cade2 | 3554 | BTRFS_UUID_SIZE); |
54aa1f4d | 3555 | |
5de865ee FDBM |
3556 | ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0, |
3557 | mid, c_nritems - mid); | |
3558 | if (ret) { | |
3559 | btrfs_abort_transaction(trans, root, ret); | |
3560 | return ret; | |
3561 | } | |
5f39d397 CM |
3562 | copy_extent_buffer(split, c, |
3563 | btrfs_node_key_ptr_offset(0), | |
3564 | btrfs_node_key_ptr_offset(mid), | |
3565 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3566 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3567 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3568 | ret = 0; |
3569 | ||
5f39d397 CM |
3570 | btrfs_mark_buffer_dirty(c); |
3571 | btrfs_mark_buffer_dirty(split); | |
3572 | ||
143bede5 | 3573 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3574 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3575 | |
5de08d7d | 3576 | if (path->slots[level] >= mid) { |
5c680ed6 | 3577 | path->slots[level] -= mid; |
925baedd | 3578 | btrfs_tree_unlock(c); |
5f39d397 CM |
3579 | free_extent_buffer(c); |
3580 | path->nodes[level] = split; | |
5c680ed6 CM |
3581 | path->slots[level + 1] += 1; |
3582 | } else { | |
925baedd | 3583 | btrfs_tree_unlock(split); |
5f39d397 | 3584 | free_extent_buffer(split); |
be0e5c09 | 3585 | } |
aa5d6bed | 3586 | return ret; |
be0e5c09 CM |
3587 | } |
3588 | ||
74123bd7 CM |
3589 | /* |
3590 | * how many bytes are required to store the items in a leaf. start | |
3591 | * and nr indicate which items in the leaf to check. This totals up the | |
3592 | * space used both by the item structs and the item data | |
3593 | */ | |
5f39d397 | 3594 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3595 | { |
41be1f3b JB |
3596 | struct btrfs_item *start_item; |
3597 | struct btrfs_item *end_item; | |
3598 | struct btrfs_map_token token; | |
be0e5c09 | 3599 | int data_len; |
5f39d397 | 3600 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3601 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3602 | |
3603 | if (!nr) | |
3604 | return 0; | |
41be1f3b | 3605 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3606 | start_item = btrfs_item_nr(start); |
3607 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3608 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3609 | btrfs_token_item_size(l, start_item, &token); | |
3610 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3611 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3612 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3613 | return data_len; |
3614 | } | |
3615 | ||
d4dbff95 CM |
3616 | /* |
3617 | * The space between the end of the leaf items and | |
3618 | * the start of the leaf data. IOW, how much room | |
3619 | * the leaf has left for both items and data | |
3620 | */ | |
d397712b | 3621 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3622 | struct extent_buffer *leaf) |
d4dbff95 | 3623 | { |
5f39d397 CM |
3624 | int nritems = btrfs_header_nritems(leaf); |
3625 | int ret; | |
3626 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3627 | if (ret < 0) { | |
efe120a0 FH |
3628 | btrfs_crit(root->fs_info, |
3629 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
ae2f5411 | 3630 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3631 | leaf_space_used(leaf, 0, nritems), nritems); |
3632 | } | |
3633 | return ret; | |
d4dbff95 CM |
3634 | } |
3635 | ||
99d8f83c CM |
3636 | /* |
3637 | * min slot controls the lowest index we're willing to push to the | |
3638 | * right. We'll push up to and including min_slot, but no lower | |
3639 | */ | |
44871b1b CM |
3640 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3641 | struct btrfs_root *root, | |
3642 | struct btrfs_path *path, | |
3643 | int data_size, int empty, | |
3644 | struct extent_buffer *right, | |
99d8f83c CM |
3645 | int free_space, u32 left_nritems, |
3646 | u32 min_slot) | |
00ec4c51 | 3647 | { |
5f39d397 | 3648 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3649 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3650 | struct btrfs_map_token token; |
5f39d397 | 3651 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3652 | int slot; |
34a38218 | 3653 | u32 i; |
00ec4c51 CM |
3654 | int push_space = 0; |
3655 | int push_items = 0; | |
0783fcfc | 3656 | struct btrfs_item *item; |
34a38218 | 3657 | u32 nr; |
7518a238 | 3658 | u32 right_nritems; |
5f39d397 | 3659 | u32 data_end; |
db94535d | 3660 | u32 this_item_size; |
00ec4c51 | 3661 | |
cfed81a0 CM |
3662 | btrfs_init_map_token(&token); |
3663 | ||
34a38218 CM |
3664 | if (empty) |
3665 | nr = 0; | |
3666 | else | |
99d8f83c | 3667 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3668 | |
31840ae1 | 3669 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3670 | push_space += data_size; |
31840ae1 | 3671 | |
44871b1b | 3672 | slot = path->slots[1]; |
34a38218 CM |
3673 | i = left_nritems - 1; |
3674 | while (i >= nr) { | |
dd3cc16b | 3675 | item = btrfs_item_nr(i); |
db94535d | 3676 | |
31840ae1 ZY |
3677 | if (!empty && push_items > 0) { |
3678 | if (path->slots[0] > i) | |
3679 | break; | |
3680 | if (path->slots[0] == i) { | |
3681 | int space = btrfs_leaf_free_space(root, left); | |
3682 | if (space + push_space * 2 > free_space) | |
3683 | break; | |
3684 | } | |
3685 | } | |
3686 | ||
00ec4c51 | 3687 | if (path->slots[0] == i) |
87b29b20 | 3688 | push_space += data_size; |
db94535d | 3689 | |
db94535d CM |
3690 | this_item_size = btrfs_item_size(left, item); |
3691 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3692 | break; |
31840ae1 | 3693 | |
00ec4c51 | 3694 | push_items++; |
db94535d | 3695 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3696 | if (i == 0) |
3697 | break; | |
3698 | i--; | |
db94535d | 3699 | } |
5f39d397 | 3700 | |
925baedd CM |
3701 | if (push_items == 0) |
3702 | goto out_unlock; | |
5f39d397 | 3703 | |
6c1500f2 | 3704 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3705 | |
00ec4c51 | 3706 | /* push left to right */ |
5f39d397 | 3707 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3708 | |
5f39d397 | 3709 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3710 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3711 | |
00ec4c51 | 3712 | /* make room in the right data area */ |
5f39d397 CM |
3713 | data_end = leaf_data_end(root, right); |
3714 | memmove_extent_buffer(right, | |
3715 | btrfs_leaf_data(right) + data_end - push_space, | |
3716 | btrfs_leaf_data(right) + data_end, | |
3717 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3718 | ||
00ec4c51 | 3719 | /* copy from the left data area */ |
5f39d397 | 3720 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3721 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3722 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3723 | push_space); | |
5f39d397 CM |
3724 | |
3725 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3726 | btrfs_item_nr_offset(0), | |
3727 | right_nritems * sizeof(struct btrfs_item)); | |
3728 | ||
00ec4c51 | 3729 | /* copy the items from left to right */ |
5f39d397 CM |
3730 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3731 | btrfs_item_nr_offset(left_nritems - push_items), | |
3732 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3733 | |
3734 | /* update the item pointers */ | |
7518a238 | 3735 | right_nritems += push_items; |
5f39d397 | 3736 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3737 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3738 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3739 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3740 | push_space -= btrfs_token_item_size(right, item, &token); |
3741 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3742 | } |
3743 | ||
7518a238 | 3744 | left_nritems -= push_items; |
5f39d397 | 3745 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3746 | |
34a38218 CM |
3747 | if (left_nritems) |
3748 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
3749 | else |
3750 | clean_tree_block(trans, root, left); | |
3751 | ||
5f39d397 | 3752 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3753 | |
5f39d397 CM |
3754 | btrfs_item_key(right, &disk_key, 0); |
3755 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3756 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3757 | |
00ec4c51 | 3758 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3759 | if (path->slots[0] >= left_nritems) { |
3760 | path->slots[0] -= left_nritems; | |
925baedd CM |
3761 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
3762 | clean_tree_block(trans, root, path->nodes[0]); | |
3763 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
3764 | free_extent_buffer(path->nodes[0]); |
3765 | path->nodes[0] = right; | |
00ec4c51 CM |
3766 | path->slots[1] += 1; |
3767 | } else { | |
925baedd | 3768 | btrfs_tree_unlock(right); |
5f39d397 | 3769 | free_extent_buffer(right); |
00ec4c51 CM |
3770 | } |
3771 | return 0; | |
925baedd CM |
3772 | |
3773 | out_unlock: | |
3774 | btrfs_tree_unlock(right); | |
3775 | free_extent_buffer(right); | |
3776 | return 1; | |
00ec4c51 | 3777 | } |
925baedd | 3778 | |
44871b1b CM |
3779 | /* |
3780 | * push some data in the path leaf to the right, trying to free up at | |
3781 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3782 | * | |
3783 | * returns 1 if the push failed because the other node didn't have enough | |
3784 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3785 | * |
3786 | * this will push starting from min_slot to the end of the leaf. It won't | |
3787 | * push any slot lower than min_slot | |
44871b1b CM |
3788 | */ |
3789 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3790 | *root, struct btrfs_path *path, |
3791 | int min_data_size, int data_size, | |
3792 | int empty, u32 min_slot) | |
44871b1b CM |
3793 | { |
3794 | struct extent_buffer *left = path->nodes[0]; | |
3795 | struct extent_buffer *right; | |
3796 | struct extent_buffer *upper; | |
3797 | int slot; | |
3798 | int free_space; | |
3799 | u32 left_nritems; | |
3800 | int ret; | |
3801 | ||
3802 | if (!path->nodes[1]) | |
3803 | return 1; | |
3804 | ||
3805 | slot = path->slots[1]; | |
3806 | upper = path->nodes[1]; | |
3807 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3808 | return 1; | |
3809 | ||
3810 | btrfs_assert_tree_locked(path->nodes[1]); | |
3811 | ||
3812 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3813 | if (right == NULL) |
3814 | return 1; | |
3815 | ||
44871b1b CM |
3816 | btrfs_tree_lock(right); |
3817 | btrfs_set_lock_blocking(right); | |
3818 | ||
3819 | free_space = btrfs_leaf_free_space(root, right); | |
3820 | if (free_space < data_size) | |
3821 | goto out_unlock; | |
3822 | ||
3823 | /* cow and double check */ | |
3824 | ret = btrfs_cow_block(trans, root, right, upper, | |
3825 | slot + 1, &right); | |
3826 | if (ret) | |
3827 | goto out_unlock; | |
3828 | ||
3829 | free_space = btrfs_leaf_free_space(root, right); | |
3830 | if (free_space < data_size) | |
3831 | goto out_unlock; | |
3832 | ||
3833 | left_nritems = btrfs_header_nritems(left); | |
3834 | if (left_nritems == 0) | |
3835 | goto out_unlock; | |
3836 | ||
2ef1fed2 FDBM |
3837 | if (path->slots[0] == left_nritems && !empty) { |
3838 | /* Key greater than all keys in the leaf, right neighbor has | |
3839 | * enough room for it and we're not emptying our leaf to delete | |
3840 | * it, therefore use right neighbor to insert the new item and | |
3841 | * no need to touch/dirty our left leaft. */ | |
3842 | btrfs_tree_unlock(left); | |
3843 | free_extent_buffer(left); | |
3844 | path->nodes[0] = right; | |
3845 | path->slots[0] = 0; | |
3846 | path->slots[1]++; | |
3847 | return 0; | |
3848 | } | |
3849 | ||
99d8f83c CM |
3850 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3851 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3852 | out_unlock: |
3853 | btrfs_tree_unlock(right); | |
3854 | free_extent_buffer(right); | |
3855 | return 1; | |
3856 | } | |
3857 | ||
74123bd7 CM |
3858 | /* |
3859 | * push some data in the path leaf to the left, trying to free up at | |
3860 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3861 | * |
3862 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3863 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3864 | * items | |
74123bd7 | 3865 | */ |
44871b1b CM |
3866 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3867 | struct btrfs_root *root, | |
3868 | struct btrfs_path *path, int data_size, | |
3869 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3870 | int free_space, u32 right_nritems, |
3871 | u32 max_slot) | |
be0e5c09 | 3872 | { |
5f39d397 CM |
3873 | struct btrfs_disk_key disk_key; |
3874 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3875 | int i; |
be0e5c09 CM |
3876 | int push_space = 0; |
3877 | int push_items = 0; | |
0783fcfc | 3878 | struct btrfs_item *item; |
7518a238 | 3879 | u32 old_left_nritems; |
34a38218 | 3880 | u32 nr; |
aa5d6bed | 3881 | int ret = 0; |
db94535d CM |
3882 | u32 this_item_size; |
3883 | u32 old_left_item_size; | |
cfed81a0 CM |
3884 | struct btrfs_map_token token; |
3885 | ||
3886 | btrfs_init_map_token(&token); | |
be0e5c09 | 3887 | |
34a38218 | 3888 | if (empty) |
99d8f83c | 3889 | nr = min(right_nritems, max_slot); |
34a38218 | 3890 | else |
99d8f83c | 3891 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3892 | |
3893 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3894 | item = btrfs_item_nr(i); |
db94535d | 3895 | |
31840ae1 ZY |
3896 | if (!empty && push_items > 0) { |
3897 | if (path->slots[0] < i) | |
3898 | break; | |
3899 | if (path->slots[0] == i) { | |
3900 | int space = btrfs_leaf_free_space(root, right); | |
3901 | if (space + push_space * 2 > free_space) | |
3902 | break; | |
3903 | } | |
3904 | } | |
3905 | ||
be0e5c09 | 3906 | if (path->slots[0] == i) |
87b29b20 | 3907 | push_space += data_size; |
db94535d CM |
3908 | |
3909 | this_item_size = btrfs_item_size(right, item); | |
3910 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3911 | break; |
db94535d | 3912 | |
be0e5c09 | 3913 | push_items++; |
db94535d CM |
3914 | push_space += this_item_size + sizeof(*item); |
3915 | } | |
3916 | ||
be0e5c09 | 3917 | if (push_items == 0) { |
925baedd CM |
3918 | ret = 1; |
3919 | goto out; | |
be0e5c09 | 3920 | } |
fae7f21c | 3921 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3922 | |
be0e5c09 | 3923 | /* push data from right to left */ |
5f39d397 CM |
3924 | copy_extent_buffer(left, right, |
3925 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3926 | btrfs_item_nr_offset(0), | |
3927 | push_items * sizeof(struct btrfs_item)); | |
3928 | ||
123abc88 | 3929 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3930 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3931 | |
3932 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3933 | leaf_data_end(root, left) - push_space, |
3934 | btrfs_leaf_data(right) + | |
5f39d397 | 3935 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3936 | push_space); |
5f39d397 | 3937 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3938 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3939 | |
db94535d | 3940 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3941 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3942 | u32 ioff; |
db94535d | 3943 | |
dd3cc16b | 3944 | item = btrfs_item_nr(i); |
db94535d | 3945 | |
cfed81a0 CM |
3946 | ioff = btrfs_token_item_offset(left, item, &token); |
3947 | btrfs_set_token_item_offset(left, item, | |
3948 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3949 | &token); | |
be0e5c09 | 3950 | } |
5f39d397 | 3951 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3952 | |
3953 | /* fixup right node */ | |
31b1a2bd JL |
3954 | if (push_items > right_nritems) |
3955 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3956 | right_nritems); |
34a38218 CM |
3957 | |
3958 | if (push_items < right_nritems) { | |
3959 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3960 | leaf_data_end(root, right); | |
3961 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3962 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3963 | btrfs_leaf_data(right) + | |
3964 | leaf_data_end(root, right), push_space); | |
3965 | ||
3966 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3967 | btrfs_item_nr_offset(push_items), |
3968 | (btrfs_header_nritems(right) - push_items) * | |
3969 | sizeof(struct btrfs_item)); | |
34a38218 | 3970 | } |
eef1c494 Y |
3971 | right_nritems -= push_items; |
3972 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3973 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 3974 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3975 | item = btrfs_item_nr(i); |
db94535d | 3976 | |
cfed81a0 CM |
3977 | push_space = push_space - btrfs_token_item_size(right, |
3978 | item, &token); | |
3979 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3980 | } |
eb60ceac | 3981 | |
5f39d397 | 3982 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3983 | if (right_nritems) |
3984 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3985 | else |
3986 | clean_tree_block(trans, root, right); | |
098f59c2 | 3987 | |
5f39d397 | 3988 | btrfs_item_key(right, &disk_key, 0); |
d6a0a126 | 3989 | fixup_low_keys(root, path, &disk_key, 1); |
be0e5c09 CM |
3990 | |
3991 | /* then fixup the leaf pointer in the path */ | |
3992 | if (path->slots[0] < push_items) { | |
3993 | path->slots[0] += old_left_nritems; | |
925baedd | 3994 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3995 | free_extent_buffer(path->nodes[0]); |
3996 | path->nodes[0] = left; | |
be0e5c09 CM |
3997 | path->slots[1] -= 1; |
3998 | } else { | |
925baedd | 3999 | btrfs_tree_unlock(left); |
5f39d397 | 4000 | free_extent_buffer(left); |
be0e5c09 CM |
4001 | path->slots[0] -= push_items; |
4002 | } | |
eb60ceac | 4003 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 4004 | return ret; |
925baedd CM |
4005 | out: |
4006 | btrfs_tree_unlock(left); | |
4007 | free_extent_buffer(left); | |
4008 | return ret; | |
be0e5c09 CM |
4009 | } |
4010 | ||
44871b1b CM |
4011 | /* |
4012 | * push some data in the path leaf to the left, trying to free up at | |
4013 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
4014 | * |
4015 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
4016 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
4017 | * items | |
44871b1b CM |
4018 | */ |
4019 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
4020 | *root, struct btrfs_path *path, int min_data_size, |
4021 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
4022 | { |
4023 | struct extent_buffer *right = path->nodes[0]; | |
4024 | struct extent_buffer *left; | |
4025 | int slot; | |
4026 | int free_space; | |
4027 | u32 right_nritems; | |
4028 | int ret = 0; | |
4029 | ||
4030 | slot = path->slots[1]; | |
4031 | if (slot == 0) | |
4032 | return 1; | |
4033 | if (!path->nodes[1]) | |
4034 | return 1; | |
4035 | ||
4036 | right_nritems = btrfs_header_nritems(right); | |
4037 | if (right_nritems == 0) | |
4038 | return 1; | |
4039 | ||
4040 | btrfs_assert_tree_locked(path->nodes[1]); | |
4041 | ||
4042 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
4043 | if (left == NULL) |
4044 | return 1; | |
4045 | ||
44871b1b CM |
4046 | btrfs_tree_lock(left); |
4047 | btrfs_set_lock_blocking(left); | |
4048 | ||
4049 | free_space = btrfs_leaf_free_space(root, left); | |
4050 | if (free_space < data_size) { | |
4051 | ret = 1; | |
4052 | goto out; | |
4053 | } | |
4054 | ||
4055 | /* cow and double check */ | |
4056 | ret = btrfs_cow_block(trans, root, left, | |
4057 | path->nodes[1], slot - 1, &left); | |
4058 | if (ret) { | |
4059 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4060 | if (ret == -ENOSPC) |
4061 | ret = 1; | |
44871b1b CM |
4062 | goto out; |
4063 | } | |
4064 | ||
4065 | free_space = btrfs_leaf_free_space(root, left); | |
4066 | if (free_space < data_size) { | |
4067 | ret = 1; | |
4068 | goto out; | |
4069 | } | |
4070 | ||
99d8f83c CM |
4071 | return __push_leaf_left(trans, root, path, min_data_size, |
4072 | empty, left, free_space, right_nritems, | |
4073 | max_slot); | |
44871b1b CM |
4074 | out: |
4075 | btrfs_tree_unlock(left); | |
4076 | free_extent_buffer(left); | |
4077 | return ret; | |
4078 | } | |
4079 | ||
4080 | /* | |
4081 | * split the path's leaf in two, making sure there is at least data_size | |
4082 | * available for the resulting leaf level of the path. | |
44871b1b | 4083 | */ |
143bede5 JM |
4084 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
4085 | struct btrfs_root *root, | |
4086 | struct btrfs_path *path, | |
4087 | struct extent_buffer *l, | |
4088 | struct extent_buffer *right, | |
4089 | int slot, int mid, int nritems) | |
44871b1b CM |
4090 | { |
4091 | int data_copy_size; | |
4092 | int rt_data_off; | |
4093 | int i; | |
44871b1b | 4094 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4095 | struct btrfs_map_token token; |
4096 | ||
4097 | btrfs_init_map_token(&token); | |
44871b1b CM |
4098 | |
4099 | nritems = nritems - mid; | |
4100 | btrfs_set_header_nritems(right, nritems); | |
4101 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
4102 | ||
4103 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4104 | btrfs_item_nr_offset(mid), | |
4105 | nritems * sizeof(struct btrfs_item)); | |
4106 | ||
4107 | copy_extent_buffer(right, l, | |
4108 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
4109 | data_copy_size, btrfs_leaf_data(l) + | |
4110 | leaf_data_end(root, l), data_copy_size); | |
4111 | ||
4112 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
4113 | btrfs_item_end_nr(l, mid); | |
4114 | ||
4115 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4116 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4117 | u32 ioff; |
4118 | ||
cfed81a0 CM |
4119 | ioff = btrfs_token_item_offset(right, item, &token); |
4120 | btrfs_set_token_item_offset(right, item, | |
4121 | ioff + rt_data_off, &token); | |
44871b1b CM |
4122 | } |
4123 | ||
44871b1b | 4124 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4125 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 4126 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4127 | path->slots[1] + 1, 1); |
44871b1b CM |
4128 | |
4129 | btrfs_mark_buffer_dirty(right); | |
4130 | btrfs_mark_buffer_dirty(l); | |
4131 | BUG_ON(path->slots[0] != slot); | |
4132 | ||
44871b1b CM |
4133 | if (mid <= slot) { |
4134 | btrfs_tree_unlock(path->nodes[0]); | |
4135 | free_extent_buffer(path->nodes[0]); | |
4136 | path->nodes[0] = right; | |
4137 | path->slots[0] -= mid; | |
4138 | path->slots[1] += 1; | |
4139 | } else { | |
4140 | btrfs_tree_unlock(right); | |
4141 | free_extent_buffer(right); | |
4142 | } | |
4143 | ||
4144 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4145 | } |
4146 | ||
99d8f83c CM |
4147 | /* |
4148 | * double splits happen when we need to insert a big item in the middle | |
4149 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4150 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4151 | * A B C | |
4152 | * | |
4153 | * We avoid this by trying to push the items on either side of our target | |
4154 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4155 | * completely. | |
4156 | */ | |
4157 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4158 | struct btrfs_root *root, | |
4159 | struct btrfs_path *path, | |
4160 | int data_size) | |
4161 | { | |
4162 | int ret; | |
4163 | int progress = 0; | |
4164 | int slot; | |
4165 | u32 nritems; | |
5a4267ca | 4166 | int space_needed = data_size; |
99d8f83c CM |
4167 | |
4168 | slot = path->slots[0]; | |
5a4267ca FDBM |
4169 | if (slot < btrfs_header_nritems(path->nodes[0])) |
4170 | space_needed -= btrfs_leaf_free_space(root, path->nodes[0]); | |
99d8f83c CM |
4171 | |
4172 | /* | |
4173 | * try to push all the items after our slot into the | |
4174 | * right leaf | |
4175 | */ | |
5a4267ca | 4176 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4177 | if (ret < 0) |
4178 | return ret; | |
4179 | ||
4180 | if (ret == 0) | |
4181 | progress++; | |
4182 | ||
4183 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4184 | /* | |
4185 | * our goal is to get our slot at the start or end of a leaf. If | |
4186 | * we've done so we're done | |
4187 | */ | |
4188 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4189 | return 0; | |
4190 | ||
4191 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4192 | return 0; | |
4193 | ||
4194 | /* try to push all the items before our slot into the next leaf */ | |
4195 | slot = path->slots[0]; | |
5a4267ca | 4196 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4197 | if (ret < 0) |
4198 | return ret; | |
4199 | ||
4200 | if (ret == 0) | |
4201 | progress++; | |
4202 | ||
4203 | if (progress) | |
4204 | return 0; | |
4205 | return 1; | |
4206 | } | |
4207 | ||
74123bd7 CM |
4208 | /* |
4209 | * split the path's leaf in two, making sure there is at least data_size | |
4210 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4211 | * |
4212 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4213 | */ |
e02119d5 CM |
4214 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4215 | struct btrfs_root *root, | |
4216 | struct btrfs_key *ins_key, | |
4217 | struct btrfs_path *path, int data_size, | |
4218 | int extend) | |
be0e5c09 | 4219 | { |
5d4f98a2 | 4220 | struct btrfs_disk_key disk_key; |
5f39d397 | 4221 | struct extent_buffer *l; |
7518a238 | 4222 | u32 nritems; |
eb60ceac CM |
4223 | int mid; |
4224 | int slot; | |
5f39d397 | 4225 | struct extent_buffer *right; |
d4dbff95 | 4226 | int ret = 0; |
aa5d6bed | 4227 | int wret; |
5d4f98a2 | 4228 | int split; |
cc0c5538 | 4229 | int num_doubles = 0; |
99d8f83c | 4230 | int tried_avoid_double = 0; |
aa5d6bed | 4231 | |
a5719521 YZ |
4232 | l = path->nodes[0]; |
4233 | slot = path->slots[0]; | |
4234 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
4235 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
4236 | return -EOVERFLOW; | |
4237 | ||
40689478 | 4238 | /* first try to make some room by pushing left and right */ |
33157e05 | 4239 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4240 | int space_needed = data_size; |
4241 | ||
4242 | if (slot < btrfs_header_nritems(l)) | |
4243 | space_needed -= btrfs_leaf_free_space(root, l); | |
4244 | ||
4245 | wret = push_leaf_right(trans, root, path, space_needed, | |
4246 | space_needed, 0, 0); | |
d397712b | 4247 | if (wret < 0) |
eaee50e8 | 4248 | return wret; |
3685f791 | 4249 | if (wret) { |
5a4267ca FDBM |
4250 | wret = push_leaf_left(trans, root, path, space_needed, |
4251 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4252 | if (wret < 0) |
4253 | return wret; | |
4254 | } | |
4255 | l = path->nodes[0]; | |
aa5d6bed | 4256 | |
3685f791 | 4257 | /* did the pushes work? */ |
87b29b20 | 4258 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 4259 | return 0; |
3326d1b0 | 4260 | } |
aa5d6bed | 4261 | |
5c680ed6 | 4262 | if (!path->nodes[1]) { |
fdd99c72 | 4263 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4264 | if (ret) |
4265 | return ret; | |
4266 | } | |
cc0c5538 | 4267 | again: |
5d4f98a2 | 4268 | split = 1; |
cc0c5538 | 4269 | l = path->nodes[0]; |
eb60ceac | 4270 | slot = path->slots[0]; |
5f39d397 | 4271 | nritems = btrfs_header_nritems(l); |
d397712b | 4272 | mid = (nritems + 1) / 2; |
54aa1f4d | 4273 | |
5d4f98a2 YZ |
4274 | if (mid <= slot) { |
4275 | if (nritems == 1 || | |
4276 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
4277 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4278 | if (slot >= nritems) { | |
4279 | split = 0; | |
4280 | } else { | |
4281 | mid = slot; | |
4282 | if (mid != nritems && | |
4283 | leaf_space_used(l, mid, nritems - mid) + | |
4284 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4285 | if (data_size && !tried_avoid_double) |
4286 | goto push_for_double; | |
5d4f98a2 YZ |
4287 | split = 2; |
4288 | } | |
4289 | } | |
4290 | } | |
4291 | } else { | |
4292 | if (leaf_space_used(l, 0, mid) + data_size > | |
4293 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4294 | if (!extend && data_size && slot == 0) { | |
4295 | split = 0; | |
4296 | } else if ((extend || !data_size) && slot == 0) { | |
4297 | mid = 1; | |
4298 | } else { | |
4299 | mid = slot; | |
4300 | if (mid != nritems && | |
4301 | leaf_space_used(l, mid, nritems - mid) + | |
4302 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4303 | if (data_size && !tried_avoid_double) |
4304 | goto push_for_double; | |
67871254 | 4305 | split = 2; |
5d4f98a2 YZ |
4306 | } |
4307 | } | |
4308 | } | |
4309 | } | |
4310 | ||
4311 | if (split == 0) | |
4312 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4313 | else | |
4314 | btrfs_item_key(l, &disk_key, mid); | |
4315 | ||
4316 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 4317 | root->root_key.objectid, |
5581a51a | 4318 | &disk_key, 0, l->start, 0); |
f0486c68 | 4319 | if (IS_ERR(right)) |
5f39d397 | 4320 | return PTR_ERR(right); |
f0486c68 YZ |
4321 | |
4322 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
4323 | |
4324 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 4325 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4326 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4327 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4328 | btrfs_set_header_owner(right, root->root_key.objectid); |
4329 | btrfs_set_header_level(right, 0); | |
4330 | write_extent_buffer(right, root->fs_info->fsid, | |
0a4e5586 | 4331 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 CM |
4332 | |
4333 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 4334 | btrfs_header_chunk_tree_uuid(right), |
e17cade2 | 4335 | BTRFS_UUID_SIZE); |
44871b1b | 4336 | |
5d4f98a2 YZ |
4337 | if (split == 0) { |
4338 | if (mid <= slot) { | |
4339 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4340 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4341 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4342 | btrfs_tree_unlock(path->nodes[0]); |
4343 | free_extent_buffer(path->nodes[0]); | |
4344 | path->nodes[0] = right; | |
4345 | path->slots[0] = 0; | |
4346 | path->slots[1] += 1; | |
4347 | } else { | |
4348 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4349 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4350 | path->slots[1], 1); |
5d4f98a2 YZ |
4351 | btrfs_tree_unlock(path->nodes[0]); |
4352 | free_extent_buffer(path->nodes[0]); | |
4353 | path->nodes[0] = right; | |
4354 | path->slots[0] = 0; | |
143bede5 | 4355 | if (path->slots[1] == 0) |
d6a0a126 | 4356 | fixup_low_keys(root, path, &disk_key, 1); |
d4dbff95 | 4357 | } |
5d4f98a2 YZ |
4358 | btrfs_mark_buffer_dirty(right); |
4359 | return ret; | |
d4dbff95 | 4360 | } |
74123bd7 | 4361 | |
143bede5 | 4362 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4363 | |
5d4f98a2 | 4364 | if (split == 2) { |
cc0c5538 CM |
4365 | BUG_ON(num_doubles != 0); |
4366 | num_doubles++; | |
4367 | goto again; | |
a429e513 | 4368 | } |
44871b1b | 4369 | |
143bede5 | 4370 | return 0; |
99d8f83c CM |
4371 | |
4372 | push_for_double: | |
4373 | push_for_double_split(trans, root, path, data_size); | |
4374 | tried_avoid_double = 1; | |
4375 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4376 | return 0; | |
4377 | goto again; | |
be0e5c09 CM |
4378 | } |
4379 | ||
ad48fd75 YZ |
4380 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4381 | struct btrfs_root *root, | |
4382 | struct btrfs_path *path, int ins_len) | |
459931ec | 4383 | { |
ad48fd75 | 4384 | struct btrfs_key key; |
459931ec | 4385 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4386 | struct btrfs_file_extent_item *fi; |
4387 | u64 extent_len = 0; | |
4388 | u32 item_size; | |
4389 | int ret; | |
459931ec CM |
4390 | |
4391 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4392 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4393 | ||
4394 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4395 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4396 | ||
4397 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4398 | return 0; | |
459931ec CM |
4399 | |
4400 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4401 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4402 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4403 | struct btrfs_file_extent_item); | |
4404 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4405 | } | |
b3b4aa74 | 4406 | btrfs_release_path(path); |
459931ec | 4407 | |
459931ec | 4408 | path->keep_locks = 1; |
ad48fd75 YZ |
4409 | path->search_for_split = 1; |
4410 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4411 | path->search_for_split = 0; |
ad48fd75 YZ |
4412 | if (ret < 0) |
4413 | goto err; | |
459931ec | 4414 | |
ad48fd75 YZ |
4415 | ret = -EAGAIN; |
4416 | leaf = path->nodes[0]; | |
459931ec | 4417 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
4418 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
4419 | goto err; | |
4420 | ||
109f6aef CM |
4421 | /* the leaf has changed, it now has room. return now */ |
4422 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4423 | goto err; | |
4424 | ||
ad48fd75 YZ |
4425 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4426 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4427 | struct btrfs_file_extent_item); | |
4428 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4429 | goto err; | |
459931ec CM |
4430 | } |
4431 | ||
b9473439 | 4432 | btrfs_set_path_blocking(path); |
ad48fd75 | 4433 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4434 | if (ret) |
4435 | goto err; | |
459931ec | 4436 | |
ad48fd75 | 4437 | path->keep_locks = 0; |
b9473439 | 4438 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4439 | return 0; |
4440 | err: | |
4441 | path->keep_locks = 0; | |
4442 | return ret; | |
4443 | } | |
4444 | ||
4445 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4446 | struct btrfs_root *root, | |
4447 | struct btrfs_path *path, | |
4448 | struct btrfs_key *new_key, | |
4449 | unsigned long split_offset) | |
4450 | { | |
4451 | struct extent_buffer *leaf; | |
4452 | struct btrfs_item *item; | |
4453 | struct btrfs_item *new_item; | |
4454 | int slot; | |
4455 | char *buf; | |
4456 | u32 nritems; | |
4457 | u32 item_size; | |
4458 | u32 orig_offset; | |
4459 | struct btrfs_disk_key disk_key; | |
4460 | ||
b9473439 CM |
4461 | leaf = path->nodes[0]; |
4462 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4463 | ||
b4ce94de CM |
4464 | btrfs_set_path_blocking(path); |
4465 | ||
dd3cc16b | 4466 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4467 | orig_offset = btrfs_item_offset(leaf, item); |
4468 | item_size = btrfs_item_size(leaf, item); | |
4469 | ||
459931ec | 4470 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4471 | if (!buf) |
4472 | return -ENOMEM; | |
4473 | ||
459931ec CM |
4474 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4475 | path->slots[0]), item_size); | |
459931ec | 4476 | |
ad48fd75 | 4477 | slot = path->slots[0] + 1; |
459931ec | 4478 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4479 | if (slot != nritems) { |
4480 | /* shift the items */ | |
4481 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4482 | btrfs_item_nr_offset(slot), |
4483 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4484 | } |
4485 | ||
4486 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4487 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4488 | ||
dd3cc16b | 4489 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4490 | |
4491 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4492 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4493 | ||
4494 | btrfs_set_item_offset(leaf, item, | |
4495 | orig_offset + item_size - split_offset); | |
4496 | btrfs_set_item_size(leaf, item, split_offset); | |
4497 | ||
4498 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4499 | ||
4500 | /* write the data for the start of the original item */ | |
4501 | write_extent_buffer(leaf, buf, | |
4502 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4503 | split_offset); | |
4504 | ||
4505 | /* write the data for the new item */ | |
4506 | write_extent_buffer(leaf, buf + split_offset, | |
4507 | btrfs_item_ptr_offset(leaf, slot), | |
4508 | item_size - split_offset); | |
4509 | btrfs_mark_buffer_dirty(leaf); | |
4510 | ||
ad48fd75 | 4511 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4512 | kfree(buf); |
ad48fd75 YZ |
4513 | return 0; |
4514 | } | |
4515 | ||
4516 | /* | |
4517 | * This function splits a single item into two items, | |
4518 | * giving 'new_key' to the new item and splitting the | |
4519 | * old one at split_offset (from the start of the item). | |
4520 | * | |
4521 | * The path may be released by this operation. After | |
4522 | * the split, the path is pointing to the old item. The | |
4523 | * new item is going to be in the same node as the old one. | |
4524 | * | |
4525 | * Note, the item being split must be smaller enough to live alone on | |
4526 | * a tree block with room for one extra struct btrfs_item | |
4527 | * | |
4528 | * This allows us to split the item in place, keeping a lock on the | |
4529 | * leaf the entire time. | |
4530 | */ | |
4531 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4532 | struct btrfs_root *root, | |
4533 | struct btrfs_path *path, | |
4534 | struct btrfs_key *new_key, | |
4535 | unsigned long split_offset) | |
4536 | { | |
4537 | int ret; | |
4538 | ret = setup_leaf_for_split(trans, root, path, | |
4539 | sizeof(struct btrfs_item)); | |
4540 | if (ret) | |
4541 | return ret; | |
4542 | ||
4543 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4544 | return ret; |
4545 | } | |
4546 | ||
ad48fd75 YZ |
4547 | /* |
4548 | * This function duplicate a item, giving 'new_key' to the new item. | |
4549 | * It guarantees both items live in the same tree leaf and the new item | |
4550 | * is contiguous with the original item. | |
4551 | * | |
4552 | * This allows us to split file extent in place, keeping a lock on the | |
4553 | * leaf the entire time. | |
4554 | */ | |
4555 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4556 | struct btrfs_root *root, | |
4557 | struct btrfs_path *path, | |
4558 | struct btrfs_key *new_key) | |
4559 | { | |
4560 | struct extent_buffer *leaf; | |
4561 | int ret; | |
4562 | u32 item_size; | |
4563 | ||
4564 | leaf = path->nodes[0]; | |
4565 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4566 | ret = setup_leaf_for_split(trans, root, path, | |
4567 | item_size + sizeof(struct btrfs_item)); | |
4568 | if (ret) | |
4569 | return ret; | |
4570 | ||
4571 | path->slots[0]++; | |
afe5fea7 | 4572 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4573 | item_size, item_size + |
4574 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4575 | leaf = path->nodes[0]; |
4576 | memcpy_extent_buffer(leaf, | |
4577 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4578 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4579 | item_size); | |
4580 | return 0; | |
4581 | } | |
4582 | ||
d352ac68 CM |
4583 | /* |
4584 | * make the item pointed to by the path smaller. new_size indicates | |
4585 | * how small to make it, and from_end tells us if we just chop bytes | |
4586 | * off the end of the item or if we shift the item to chop bytes off | |
4587 | * the front. | |
4588 | */ | |
afe5fea7 | 4589 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4590 | u32 new_size, int from_end) |
b18c6685 | 4591 | { |
b18c6685 | 4592 | int slot; |
5f39d397 CM |
4593 | struct extent_buffer *leaf; |
4594 | struct btrfs_item *item; | |
b18c6685 CM |
4595 | u32 nritems; |
4596 | unsigned int data_end; | |
4597 | unsigned int old_data_start; | |
4598 | unsigned int old_size; | |
4599 | unsigned int size_diff; | |
4600 | int i; | |
cfed81a0 CM |
4601 | struct btrfs_map_token token; |
4602 | ||
4603 | btrfs_init_map_token(&token); | |
b18c6685 | 4604 | |
5f39d397 | 4605 | leaf = path->nodes[0]; |
179e29e4 CM |
4606 | slot = path->slots[0]; |
4607 | ||
4608 | old_size = btrfs_item_size_nr(leaf, slot); | |
4609 | if (old_size == new_size) | |
143bede5 | 4610 | return; |
b18c6685 | 4611 | |
5f39d397 | 4612 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4613 | data_end = leaf_data_end(root, leaf); |
4614 | ||
5f39d397 | 4615 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4616 | |
b18c6685 CM |
4617 | size_diff = old_size - new_size; |
4618 | ||
4619 | BUG_ON(slot < 0); | |
4620 | BUG_ON(slot >= nritems); | |
4621 | ||
4622 | /* | |
4623 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4624 | */ | |
4625 | /* first correct the data pointers */ | |
4626 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4627 | u32 ioff; |
dd3cc16b | 4628 | item = btrfs_item_nr(i); |
db94535d | 4629 | |
cfed81a0 CM |
4630 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4631 | btrfs_set_token_item_offset(leaf, item, | |
4632 | ioff + size_diff, &token); | |
b18c6685 | 4633 | } |
db94535d | 4634 | |
b18c6685 | 4635 | /* shift the data */ |
179e29e4 CM |
4636 | if (from_end) { |
4637 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4638 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4639 | data_end, old_data_start + new_size - data_end); | |
4640 | } else { | |
4641 | struct btrfs_disk_key disk_key; | |
4642 | u64 offset; | |
4643 | ||
4644 | btrfs_item_key(leaf, &disk_key, slot); | |
4645 | ||
4646 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4647 | unsigned long ptr; | |
4648 | struct btrfs_file_extent_item *fi; | |
4649 | ||
4650 | fi = btrfs_item_ptr(leaf, slot, | |
4651 | struct btrfs_file_extent_item); | |
4652 | fi = (struct btrfs_file_extent_item *)( | |
4653 | (unsigned long)fi - size_diff); | |
4654 | ||
4655 | if (btrfs_file_extent_type(leaf, fi) == | |
4656 | BTRFS_FILE_EXTENT_INLINE) { | |
4657 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4658 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
4659 | (unsigned long)fi, |
4660 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
4661 | disk_bytenr)); |
4662 | } | |
4663 | } | |
4664 | ||
4665 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4666 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4667 | data_end, old_data_start - data_end); | |
4668 | ||
4669 | offset = btrfs_disk_key_offset(&disk_key); | |
4670 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4671 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4672 | if (slot == 0) | |
d6a0a126 | 4673 | fixup_low_keys(root, path, &disk_key, 1); |
179e29e4 | 4674 | } |
5f39d397 | 4675 | |
dd3cc16b | 4676 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4677 | btrfs_set_item_size(leaf, item, new_size); |
4678 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4679 | |
5f39d397 CM |
4680 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4681 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4682 | BUG(); |
5f39d397 | 4683 | } |
b18c6685 CM |
4684 | } |
4685 | ||
d352ac68 | 4686 | /* |
8f69dbd2 | 4687 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4688 | */ |
4b90c680 | 4689 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4690 | u32 data_size) |
6567e837 | 4691 | { |
6567e837 | 4692 | int slot; |
5f39d397 CM |
4693 | struct extent_buffer *leaf; |
4694 | struct btrfs_item *item; | |
6567e837 CM |
4695 | u32 nritems; |
4696 | unsigned int data_end; | |
4697 | unsigned int old_data; | |
4698 | unsigned int old_size; | |
4699 | int i; | |
cfed81a0 CM |
4700 | struct btrfs_map_token token; |
4701 | ||
4702 | btrfs_init_map_token(&token); | |
6567e837 | 4703 | |
5f39d397 | 4704 | leaf = path->nodes[0]; |
6567e837 | 4705 | |
5f39d397 | 4706 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4707 | data_end = leaf_data_end(root, leaf); |
4708 | ||
5f39d397 CM |
4709 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4710 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4711 | BUG(); |
5f39d397 | 4712 | } |
6567e837 | 4713 | slot = path->slots[0]; |
5f39d397 | 4714 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4715 | |
4716 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4717 | if (slot >= nritems) { |
4718 | btrfs_print_leaf(root, leaf); | |
efe120a0 | 4719 | btrfs_crit(root->fs_info, "slot %d too large, nritems %d", |
d397712b | 4720 | slot, nritems); |
3326d1b0 CM |
4721 | BUG_ON(1); |
4722 | } | |
6567e837 CM |
4723 | |
4724 | /* | |
4725 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4726 | */ | |
4727 | /* first correct the data pointers */ | |
4728 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4729 | u32 ioff; |
dd3cc16b | 4730 | item = btrfs_item_nr(i); |
db94535d | 4731 | |
cfed81a0 CM |
4732 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4733 | btrfs_set_token_item_offset(leaf, item, | |
4734 | ioff - data_size, &token); | |
6567e837 | 4735 | } |
5f39d397 | 4736 | |
6567e837 | 4737 | /* shift the data */ |
5f39d397 | 4738 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4739 | data_end - data_size, btrfs_leaf_data(leaf) + |
4740 | data_end, old_data - data_end); | |
5f39d397 | 4741 | |
6567e837 | 4742 | data_end = old_data; |
5f39d397 | 4743 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4744 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4745 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4746 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4747 | |
5f39d397 CM |
4748 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4749 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4750 | BUG(); |
5f39d397 | 4751 | } |
6567e837 CM |
4752 | } |
4753 | ||
74123bd7 | 4754 | /* |
44871b1b CM |
4755 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4756 | * to save stack depth by doing the bulk of the work in a function | |
4757 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4758 | */ |
afe5fea7 | 4759 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4760 | struct btrfs_key *cpu_key, u32 *data_size, |
4761 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4762 | { |
5f39d397 | 4763 | struct btrfs_item *item; |
9c58309d | 4764 | int i; |
7518a238 | 4765 | u32 nritems; |
be0e5c09 | 4766 | unsigned int data_end; |
e2fa7227 | 4767 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4768 | struct extent_buffer *leaf; |
4769 | int slot; | |
cfed81a0 CM |
4770 | struct btrfs_map_token token; |
4771 | ||
4772 | btrfs_init_map_token(&token); | |
e2fa7227 | 4773 | |
5f39d397 | 4774 | leaf = path->nodes[0]; |
44871b1b | 4775 | slot = path->slots[0]; |
74123bd7 | 4776 | |
5f39d397 | 4777 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4778 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4779 | |
f25956cc | 4780 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4781 | btrfs_print_leaf(root, leaf); |
efe120a0 | 4782 | btrfs_crit(root->fs_info, "not enough freespace need %u have %d", |
9c58309d | 4783 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4784 | BUG(); |
d4dbff95 | 4785 | } |
5f39d397 | 4786 | |
be0e5c09 | 4787 | if (slot != nritems) { |
5f39d397 | 4788 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4789 | |
5f39d397 CM |
4790 | if (old_data < data_end) { |
4791 | btrfs_print_leaf(root, leaf); | |
efe120a0 | 4792 | btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d", |
5f39d397 CM |
4793 | slot, old_data, data_end); |
4794 | BUG_ON(1); | |
4795 | } | |
be0e5c09 CM |
4796 | /* |
4797 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4798 | */ | |
4799 | /* first correct the data pointers */ | |
0783fcfc | 4800 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4801 | u32 ioff; |
db94535d | 4802 | |
dd3cc16b | 4803 | item = btrfs_item_nr( i); |
cfed81a0 CM |
4804 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4805 | btrfs_set_token_item_offset(leaf, item, | |
4806 | ioff - total_data, &token); | |
0783fcfc | 4807 | } |
be0e5c09 | 4808 | /* shift the items */ |
9c58309d | 4809 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4810 | btrfs_item_nr_offset(slot), |
d6025579 | 4811 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4812 | |
4813 | /* shift the data */ | |
5f39d397 | 4814 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4815 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4816 | data_end, old_data - data_end); |
be0e5c09 CM |
4817 | data_end = old_data; |
4818 | } | |
5f39d397 | 4819 | |
62e2749e | 4820 | /* setup the item for the new data */ |
9c58309d CM |
4821 | for (i = 0; i < nr; i++) { |
4822 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4823 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4824 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4825 | btrfs_set_token_item_offset(leaf, item, |
4826 | data_end - data_size[i], &token); | |
9c58309d | 4827 | data_end -= data_size[i]; |
cfed81a0 | 4828 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4829 | } |
44871b1b | 4830 | |
9c58309d | 4831 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4832 | |
5a01a2e3 CM |
4833 | if (slot == 0) { |
4834 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
d6a0a126 | 4835 | fixup_low_keys(root, path, &disk_key, 1); |
5a01a2e3 | 4836 | } |
b9473439 CM |
4837 | btrfs_unlock_up_safe(path, 1); |
4838 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4839 | |
5f39d397 CM |
4840 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4841 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4842 | BUG(); |
5f39d397 | 4843 | } |
44871b1b CM |
4844 | } |
4845 | ||
4846 | /* | |
4847 | * Given a key and some data, insert items into the tree. | |
4848 | * This does all the path init required, making room in the tree if needed. | |
4849 | */ | |
4850 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4851 | struct btrfs_root *root, | |
4852 | struct btrfs_path *path, | |
4853 | struct btrfs_key *cpu_key, u32 *data_size, | |
4854 | int nr) | |
4855 | { | |
44871b1b CM |
4856 | int ret = 0; |
4857 | int slot; | |
4858 | int i; | |
4859 | u32 total_size = 0; | |
4860 | u32 total_data = 0; | |
4861 | ||
4862 | for (i = 0; i < nr; i++) | |
4863 | total_data += data_size[i]; | |
4864 | ||
4865 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4866 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4867 | if (ret == 0) | |
4868 | return -EEXIST; | |
4869 | if (ret < 0) | |
143bede5 | 4870 | return ret; |
44871b1b | 4871 | |
44871b1b CM |
4872 | slot = path->slots[0]; |
4873 | BUG_ON(slot < 0); | |
4874 | ||
afe5fea7 | 4875 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4876 | total_data, total_size, nr); |
143bede5 | 4877 | return 0; |
62e2749e CM |
4878 | } |
4879 | ||
4880 | /* | |
4881 | * Given a key and some data, insert an item into the tree. | |
4882 | * This does all the path init required, making room in the tree if needed. | |
4883 | */ | |
e089f05c CM |
4884 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4885 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4886 | data_size) | |
62e2749e CM |
4887 | { |
4888 | int ret = 0; | |
2c90e5d6 | 4889 | struct btrfs_path *path; |
5f39d397 CM |
4890 | struct extent_buffer *leaf; |
4891 | unsigned long ptr; | |
62e2749e | 4892 | |
2c90e5d6 | 4893 | path = btrfs_alloc_path(); |
db5b493a TI |
4894 | if (!path) |
4895 | return -ENOMEM; | |
2c90e5d6 | 4896 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4897 | if (!ret) { |
5f39d397 CM |
4898 | leaf = path->nodes[0]; |
4899 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4900 | write_extent_buffer(leaf, data, ptr, data_size); | |
4901 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4902 | } |
2c90e5d6 | 4903 | btrfs_free_path(path); |
aa5d6bed | 4904 | return ret; |
be0e5c09 CM |
4905 | } |
4906 | ||
74123bd7 | 4907 | /* |
5de08d7d | 4908 | * delete the pointer from a given node. |
74123bd7 | 4909 | * |
d352ac68 CM |
4910 | * the tree should have been previously balanced so the deletion does not |
4911 | * empty a node. | |
74123bd7 | 4912 | */ |
afe5fea7 TI |
4913 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4914 | int level, int slot) | |
be0e5c09 | 4915 | { |
5f39d397 | 4916 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4917 | u32 nritems; |
f3ea38da | 4918 | int ret; |
be0e5c09 | 4919 | |
5f39d397 | 4920 | nritems = btrfs_header_nritems(parent); |
d397712b | 4921 | if (slot != nritems - 1) { |
0e411ece | 4922 | if (level) |
f3ea38da JS |
4923 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4924 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4925 | memmove_extent_buffer(parent, |
4926 | btrfs_node_key_ptr_offset(slot), | |
4927 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4928 | sizeof(struct btrfs_key_ptr) * |
4929 | (nritems - slot - 1)); | |
57ba86c0 CM |
4930 | } else if (level) { |
4931 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
c8cc6341 | 4932 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4933 | BUG_ON(ret < 0); |
bb803951 | 4934 | } |
f3ea38da | 4935 | |
7518a238 | 4936 | nritems--; |
5f39d397 | 4937 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4938 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4939 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4940 | /* just turn the root into a leaf and break */ |
5f39d397 | 4941 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4942 | } else if (slot == 0) { |
5f39d397 CM |
4943 | struct btrfs_disk_key disk_key; |
4944 | ||
4945 | btrfs_node_key(parent, &disk_key, 0); | |
d6a0a126 | 4946 | fixup_low_keys(root, path, &disk_key, level + 1); |
be0e5c09 | 4947 | } |
d6025579 | 4948 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4949 | } |
4950 | ||
323ac95b CM |
4951 | /* |
4952 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4953 | * path->nodes[1]. |
323ac95b CM |
4954 | * |
4955 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4956 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4957 | * | |
4958 | * The path must have already been setup for deleting the leaf, including | |
4959 | * all the proper balancing. path->nodes[1] must be locked. | |
4960 | */ | |
143bede5 JM |
4961 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4962 | struct btrfs_root *root, | |
4963 | struct btrfs_path *path, | |
4964 | struct extent_buffer *leaf) | |
323ac95b | 4965 | { |
5d4f98a2 | 4966 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4967 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4968 | |
4d081c41 CM |
4969 | /* |
4970 | * btrfs_free_extent is expensive, we want to make sure we | |
4971 | * aren't holding any locks when we call it | |
4972 | */ | |
4973 | btrfs_unlock_up_safe(path, 0); | |
4974 | ||
f0486c68 YZ |
4975 | root_sub_used(root, leaf->len); |
4976 | ||
3083ee2e | 4977 | extent_buffer_get(leaf); |
5581a51a | 4978 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4979 | free_extent_buffer_stale(leaf); |
323ac95b | 4980 | } |
74123bd7 CM |
4981 | /* |
4982 | * delete the item at the leaf level in path. If that empties | |
4983 | * the leaf, remove it from the tree | |
4984 | */ | |
85e21bac CM |
4985 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4986 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4987 | { |
5f39d397 CM |
4988 | struct extent_buffer *leaf; |
4989 | struct btrfs_item *item; | |
85e21bac CM |
4990 | int last_off; |
4991 | int dsize = 0; | |
aa5d6bed CM |
4992 | int ret = 0; |
4993 | int wret; | |
85e21bac | 4994 | int i; |
7518a238 | 4995 | u32 nritems; |
cfed81a0 CM |
4996 | struct btrfs_map_token token; |
4997 | ||
4998 | btrfs_init_map_token(&token); | |
be0e5c09 | 4999 | |
5f39d397 | 5000 | leaf = path->nodes[0]; |
85e21bac CM |
5001 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
5002 | ||
5003 | for (i = 0; i < nr; i++) | |
5004 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
5005 | ||
5f39d397 | 5006 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 5007 | |
85e21bac | 5008 | if (slot + nr != nritems) { |
123abc88 | 5009 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
5010 | |
5011 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
5012 | data_end + dsize, |
5013 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 5014 | last_off - data_end); |
5f39d397 | 5015 | |
85e21bac | 5016 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 5017 | u32 ioff; |
db94535d | 5018 | |
dd3cc16b | 5019 | item = btrfs_item_nr(i); |
cfed81a0 CM |
5020 | ioff = btrfs_token_item_offset(leaf, item, &token); |
5021 | btrfs_set_token_item_offset(leaf, item, | |
5022 | ioff + dsize, &token); | |
0783fcfc | 5023 | } |
db94535d | 5024 | |
5f39d397 | 5025 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5026 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5027 | sizeof(struct btrfs_item) * |
85e21bac | 5028 | (nritems - slot - nr)); |
be0e5c09 | 5029 | } |
85e21bac CM |
5030 | btrfs_set_header_nritems(leaf, nritems - nr); |
5031 | nritems -= nr; | |
5f39d397 | 5032 | |
74123bd7 | 5033 | /* delete the leaf if we've emptied it */ |
7518a238 | 5034 | if (nritems == 0) { |
5f39d397 CM |
5035 | if (leaf == root->node) { |
5036 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5037 | } else { |
f0486c68 YZ |
5038 | btrfs_set_path_blocking(path); |
5039 | clean_tree_block(trans, root, leaf); | |
143bede5 | 5040 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5041 | } |
be0e5c09 | 5042 | } else { |
7518a238 | 5043 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5044 | if (slot == 0) { |
5f39d397 CM |
5045 | struct btrfs_disk_key disk_key; |
5046 | ||
5047 | btrfs_item_key(leaf, &disk_key, 0); | |
d6a0a126 | 5048 | fixup_low_keys(root, path, &disk_key, 1); |
aa5d6bed | 5049 | } |
aa5d6bed | 5050 | |
74123bd7 | 5051 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 5052 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
5053 | /* push_leaf_left fixes the path. |
5054 | * make sure the path still points to our leaf | |
5055 | * for possible call to del_ptr below | |
5056 | */ | |
4920c9ac | 5057 | slot = path->slots[1]; |
5f39d397 CM |
5058 | extent_buffer_get(leaf); |
5059 | ||
b9473439 | 5060 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5061 | wret = push_leaf_left(trans, root, path, 1, 1, |
5062 | 1, (u32)-1); | |
54aa1f4d | 5063 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5064 | ret = wret; |
5f39d397 CM |
5065 | |
5066 | if (path->nodes[0] == leaf && | |
5067 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5068 | wret = push_leaf_right(trans, root, path, 1, |
5069 | 1, 1, 0); | |
54aa1f4d | 5070 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5071 | ret = wret; |
5072 | } | |
5f39d397 CM |
5073 | |
5074 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5075 | path->slots[1] = slot; |
143bede5 | 5076 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5077 | free_extent_buffer(leaf); |
143bede5 | 5078 | ret = 0; |
5de08d7d | 5079 | } else { |
925baedd CM |
5080 | /* if we're still in the path, make sure |
5081 | * we're dirty. Otherwise, one of the | |
5082 | * push_leaf functions must have already | |
5083 | * dirtied this buffer | |
5084 | */ | |
5085 | if (path->nodes[0] == leaf) | |
5086 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5087 | free_extent_buffer(leaf); |
be0e5c09 | 5088 | } |
d5719762 | 5089 | } else { |
5f39d397 | 5090 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5091 | } |
5092 | } | |
aa5d6bed | 5093 | return ret; |
be0e5c09 CM |
5094 | } |
5095 | ||
7bb86316 | 5096 | /* |
925baedd | 5097 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5098 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5099 | * returns < 0 on io errors. | |
d352ac68 CM |
5100 | * |
5101 | * This may release the path, and so you may lose any locks held at the | |
5102 | * time you call it. | |
7bb86316 | 5103 | */ |
16e7549f | 5104 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5105 | { |
925baedd CM |
5106 | struct btrfs_key key; |
5107 | struct btrfs_disk_key found_key; | |
5108 | int ret; | |
7bb86316 | 5109 | |
925baedd | 5110 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5111 | |
e8b0d724 | 5112 | if (key.offset > 0) { |
925baedd | 5113 | key.offset--; |
e8b0d724 | 5114 | } else if (key.type > 0) { |
925baedd | 5115 | key.type--; |
e8b0d724 FDBM |
5116 | key.offset = (u64)-1; |
5117 | } else if (key.objectid > 0) { | |
925baedd | 5118 | key.objectid--; |
e8b0d724 FDBM |
5119 | key.type = (u8)-1; |
5120 | key.offset = (u64)-1; | |
5121 | } else { | |
925baedd | 5122 | return 1; |
e8b0d724 | 5123 | } |
7bb86316 | 5124 | |
b3b4aa74 | 5125 | btrfs_release_path(path); |
925baedd CM |
5126 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5127 | if (ret < 0) | |
5128 | return ret; | |
5129 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5130 | ret = comp_keys(&found_key, &key); | |
5131 | if (ret < 0) | |
5132 | return 0; | |
5133 | return 1; | |
7bb86316 CM |
5134 | } |
5135 | ||
3f157a2f CM |
5136 | /* |
5137 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5138 | * for nodes or leaves that are have a minimum transaction id. |
5139 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5140 | * |
5141 | * This does not cow, but it does stuff the starting key it finds back | |
5142 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5143 | * key and get a writable path. | |
5144 | * | |
5145 | * This does lock as it descends, and path->keep_locks should be set | |
5146 | * to 1 by the caller. | |
5147 | * | |
5148 | * This honors path->lowest_level to prevent descent past a given level | |
5149 | * of the tree. | |
5150 | * | |
d352ac68 CM |
5151 | * min_trans indicates the oldest transaction that you are interested |
5152 | * in walking through. Any nodes or leaves older than min_trans are | |
5153 | * skipped over (without reading them). | |
5154 | * | |
3f157a2f CM |
5155 | * returns zero if something useful was found, < 0 on error and 1 if there |
5156 | * was nothing in the tree that matched the search criteria. | |
5157 | */ | |
5158 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5159 | struct btrfs_path *path, |
3f157a2f CM |
5160 | u64 min_trans) |
5161 | { | |
5162 | struct extent_buffer *cur; | |
5163 | struct btrfs_key found_key; | |
5164 | int slot; | |
9652480b | 5165 | int sret; |
3f157a2f CM |
5166 | u32 nritems; |
5167 | int level; | |
5168 | int ret = 1; | |
5169 | ||
934d375b | 5170 | WARN_ON(!path->keep_locks); |
3f157a2f | 5171 | again: |
bd681513 | 5172 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5173 | level = btrfs_header_level(cur); |
e02119d5 | 5174 | WARN_ON(path->nodes[level]); |
3f157a2f | 5175 | path->nodes[level] = cur; |
bd681513 | 5176 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5177 | |
5178 | if (btrfs_header_generation(cur) < min_trans) { | |
5179 | ret = 1; | |
5180 | goto out; | |
5181 | } | |
d397712b | 5182 | while (1) { |
3f157a2f CM |
5183 | nritems = btrfs_header_nritems(cur); |
5184 | level = btrfs_header_level(cur); | |
9652480b | 5185 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 5186 | |
323ac95b CM |
5187 | /* at the lowest level, we're done, setup the path and exit */ |
5188 | if (level == path->lowest_level) { | |
e02119d5 CM |
5189 | if (slot >= nritems) |
5190 | goto find_next_key; | |
3f157a2f CM |
5191 | ret = 0; |
5192 | path->slots[level] = slot; | |
5193 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5194 | goto out; | |
5195 | } | |
9652480b Y |
5196 | if (sret && slot > 0) |
5197 | slot--; | |
3f157a2f | 5198 | /* |
de78b51a ES |
5199 | * check this node pointer against the min_trans parameters. |
5200 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5201 | */ |
d397712b | 5202 | while (slot < nritems) { |
3f157a2f | 5203 | u64 gen; |
e02119d5 | 5204 | |
3f157a2f CM |
5205 | gen = btrfs_node_ptr_generation(cur, slot); |
5206 | if (gen < min_trans) { | |
5207 | slot++; | |
5208 | continue; | |
5209 | } | |
de78b51a | 5210 | break; |
3f157a2f | 5211 | } |
e02119d5 | 5212 | find_next_key: |
3f157a2f CM |
5213 | /* |
5214 | * we didn't find a candidate key in this node, walk forward | |
5215 | * and find another one | |
5216 | */ | |
5217 | if (slot >= nritems) { | |
e02119d5 | 5218 | path->slots[level] = slot; |
b4ce94de | 5219 | btrfs_set_path_blocking(path); |
e02119d5 | 5220 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5221 | min_trans); |
e02119d5 | 5222 | if (sret == 0) { |
b3b4aa74 | 5223 | btrfs_release_path(path); |
3f157a2f CM |
5224 | goto again; |
5225 | } else { | |
5226 | goto out; | |
5227 | } | |
5228 | } | |
5229 | /* save our key for returning back */ | |
5230 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5231 | path->slots[level] = slot; | |
5232 | if (level == path->lowest_level) { | |
5233 | ret = 0; | |
f7c79f30 | 5234 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5235 | goto out; |
5236 | } | |
b4ce94de | 5237 | btrfs_set_path_blocking(path); |
3f157a2f | 5238 | cur = read_node_slot(root, cur, slot); |
79787eaa | 5239 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 5240 | |
bd681513 | 5241 | btrfs_tree_read_lock(cur); |
b4ce94de | 5242 | |
bd681513 | 5243 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5244 | path->nodes[level - 1] = cur; |
f7c79f30 | 5245 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5246 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5247 | } |
5248 | out: | |
5249 | if (ret == 0) | |
5250 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 5251 | btrfs_set_path_blocking(path); |
3f157a2f CM |
5252 | return ret; |
5253 | } | |
5254 | ||
7069830a AB |
5255 | static void tree_move_down(struct btrfs_root *root, |
5256 | struct btrfs_path *path, | |
5257 | int *level, int root_level) | |
5258 | { | |
74dd17fb | 5259 | BUG_ON(*level == 0); |
7069830a AB |
5260 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
5261 | path->slots[*level]); | |
5262 | path->slots[*level - 1] = 0; | |
5263 | (*level)--; | |
5264 | } | |
5265 | ||
5266 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
5267 | struct btrfs_path *path, | |
5268 | int *level, int root_level) | |
5269 | { | |
5270 | int ret = 0; | |
5271 | int nritems; | |
5272 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5273 | ||
5274 | path->slots[*level]++; | |
5275 | ||
74dd17fb | 5276 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5277 | if (*level == root_level) |
5278 | return -1; | |
5279 | ||
5280 | /* move upnext */ | |
5281 | path->slots[*level] = 0; | |
5282 | free_extent_buffer(path->nodes[*level]); | |
5283 | path->nodes[*level] = NULL; | |
5284 | (*level)++; | |
5285 | path->slots[*level]++; | |
5286 | ||
5287 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5288 | ret = 1; | |
5289 | } | |
5290 | return ret; | |
5291 | } | |
5292 | ||
5293 | /* | |
5294 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5295 | * or down. | |
5296 | */ | |
5297 | static int tree_advance(struct btrfs_root *root, | |
5298 | struct btrfs_path *path, | |
5299 | int *level, int root_level, | |
5300 | int allow_down, | |
5301 | struct btrfs_key *key) | |
5302 | { | |
5303 | int ret; | |
5304 | ||
5305 | if (*level == 0 || !allow_down) { | |
5306 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
5307 | } else { | |
5308 | tree_move_down(root, path, level, root_level); | |
5309 | ret = 0; | |
5310 | } | |
5311 | if (ret >= 0) { | |
5312 | if (*level == 0) | |
5313 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5314 | path->slots[*level]); | |
5315 | else | |
5316 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5317 | path->slots[*level]); | |
5318 | } | |
5319 | return ret; | |
5320 | } | |
5321 | ||
5322 | static int tree_compare_item(struct btrfs_root *left_root, | |
5323 | struct btrfs_path *left_path, | |
5324 | struct btrfs_path *right_path, | |
5325 | char *tmp_buf) | |
5326 | { | |
5327 | int cmp; | |
5328 | int len1, len2; | |
5329 | unsigned long off1, off2; | |
5330 | ||
5331 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5332 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5333 | if (len1 != len2) | |
5334 | return 1; | |
5335 | ||
5336 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5337 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5338 | right_path->slots[0]); | |
5339 | ||
5340 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5341 | ||
5342 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5343 | if (cmp) | |
5344 | return 1; | |
5345 | return 0; | |
5346 | } | |
5347 | ||
5348 | #define ADVANCE 1 | |
5349 | #define ADVANCE_ONLY_NEXT -1 | |
5350 | ||
5351 | /* | |
5352 | * This function compares two trees and calls the provided callback for | |
5353 | * every changed/new/deleted item it finds. | |
5354 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5355 | * the compare pretty fast on snapshotted subvolumes. | |
5356 | * | |
5357 | * This currently works on commit roots only. As commit roots are read only, | |
5358 | * we don't do any locking. The commit roots are protected with transactions. | |
5359 | * Transactions are ended and rejoined when a commit is tried in between. | |
5360 | * | |
5361 | * This function checks for modifications done to the trees while comparing. | |
5362 | * If it detects a change, it aborts immediately. | |
5363 | */ | |
5364 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5365 | struct btrfs_root *right_root, | |
5366 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5367 | { | |
5368 | int ret; | |
5369 | int cmp; | |
7069830a AB |
5370 | struct btrfs_path *left_path = NULL; |
5371 | struct btrfs_path *right_path = NULL; | |
5372 | struct btrfs_key left_key; | |
5373 | struct btrfs_key right_key; | |
5374 | char *tmp_buf = NULL; | |
5375 | int left_root_level; | |
5376 | int right_root_level; | |
5377 | int left_level; | |
5378 | int right_level; | |
5379 | int left_end_reached; | |
5380 | int right_end_reached; | |
5381 | int advance_left; | |
5382 | int advance_right; | |
5383 | u64 left_blockptr; | |
5384 | u64 right_blockptr; | |
6baa4293 FM |
5385 | u64 left_gen; |
5386 | u64 right_gen; | |
7069830a AB |
5387 | |
5388 | left_path = btrfs_alloc_path(); | |
5389 | if (!left_path) { | |
5390 | ret = -ENOMEM; | |
5391 | goto out; | |
5392 | } | |
5393 | right_path = btrfs_alloc_path(); | |
5394 | if (!right_path) { | |
5395 | ret = -ENOMEM; | |
5396 | goto out; | |
5397 | } | |
5398 | ||
5399 | tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS); | |
5400 | if (!tmp_buf) { | |
5401 | ret = -ENOMEM; | |
5402 | goto out; | |
5403 | } | |
5404 | ||
5405 | left_path->search_commit_root = 1; | |
5406 | left_path->skip_locking = 1; | |
5407 | right_path->search_commit_root = 1; | |
5408 | right_path->skip_locking = 1; | |
5409 | ||
7069830a AB |
5410 | /* |
5411 | * Strategy: Go to the first items of both trees. Then do | |
5412 | * | |
5413 | * If both trees are at level 0 | |
5414 | * Compare keys of current items | |
5415 | * If left < right treat left item as new, advance left tree | |
5416 | * and repeat | |
5417 | * If left > right treat right item as deleted, advance right tree | |
5418 | * and repeat | |
5419 | * If left == right do deep compare of items, treat as changed if | |
5420 | * needed, advance both trees and repeat | |
5421 | * If both trees are at the same level but not at level 0 | |
5422 | * Compare keys of current nodes/leafs | |
5423 | * If left < right advance left tree and repeat | |
5424 | * If left > right advance right tree and repeat | |
5425 | * If left == right compare blockptrs of the next nodes/leafs | |
5426 | * If they match advance both trees but stay at the same level | |
5427 | * and repeat | |
5428 | * If they don't match advance both trees while allowing to go | |
5429 | * deeper and repeat | |
5430 | * If tree levels are different | |
5431 | * Advance the tree that needs it and repeat | |
5432 | * | |
5433 | * Advancing a tree means: | |
5434 | * If we are at level 0, try to go to the next slot. If that's not | |
5435 | * possible, go one level up and repeat. Stop when we found a level | |
5436 | * where we could go to the next slot. We may at this point be on a | |
5437 | * node or a leaf. | |
5438 | * | |
5439 | * If we are not at level 0 and not on shared tree blocks, go one | |
5440 | * level deeper. | |
5441 | * | |
5442 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5443 | * the right if possible or go up and right. | |
5444 | */ | |
5445 | ||
3f8a18cc | 5446 | down_read(&left_root->fs_info->commit_root_sem); |
7069830a AB |
5447 | left_level = btrfs_header_level(left_root->commit_root); |
5448 | left_root_level = left_level; | |
5449 | left_path->nodes[left_level] = left_root->commit_root; | |
5450 | extent_buffer_get(left_path->nodes[left_level]); | |
5451 | ||
5452 | right_level = btrfs_header_level(right_root->commit_root); | |
5453 | right_root_level = right_level; | |
5454 | right_path->nodes[right_level] = right_root->commit_root; | |
5455 | extent_buffer_get(right_path->nodes[right_level]); | |
3f8a18cc | 5456 | up_read(&left_root->fs_info->commit_root_sem); |
7069830a AB |
5457 | |
5458 | if (left_level == 0) | |
5459 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5460 | &left_key, left_path->slots[left_level]); | |
5461 | else | |
5462 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5463 | &left_key, left_path->slots[left_level]); | |
5464 | if (right_level == 0) | |
5465 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5466 | &right_key, right_path->slots[right_level]); | |
5467 | else | |
5468 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5469 | &right_key, right_path->slots[right_level]); | |
5470 | ||
5471 | left_end_reached = right_end_reached = 0; | |
5472 | advance_left = advance_right = 0; | |
5473 | ||
5474 | while (1) { | |
7069830a AB |
5475 | if (advance_left && !left_end_reached) { |
5476 | ret = tree_advance(left_root, left_path, &left_level, | |
5477 | left_root_level, | |
5478 | advance_left != ADVANCE_ONLY_NEXT, | |
5479 | &left_key); | |
5480 | if (ret < 0) | |
5481 | left_end_reached = ADVANCE; | |
5482 | advance_left = 0; | |
5483 | } | |
5484 | if (advance_right && !right_end_reached) { | |
5485 | ret = tree_advance(right_root, right_path, &right_level, | |
5486 | right_root_level, | |
5487 | advance_right != ADVANCE_ONLY_NEXT, | |
5488 | &right_key); | |
5489 | if (ret < 0) | |
5490 | right_end_reached = ADVANCE; | |
5491 | advance_right = 0; | |
5492 | } | |
5493 | ||
5494 | if (left_end_reached && right_end_reached) { | |
5495 | ret = 0; | |
5496 | goto out; | |
5497 | } else if (left_end_reached) { | |
5498 | if (right_level == 0) { | |
5499 | ret = changed_cb(left_root, right_root, | |
5500 | left_path, right_path, | |
5501 | &right_key, | |
5502 | BTRFS_COMPARE_TREE_DELETED, | |
5503 | ctx); | |
5504 | if (ret < 0) | |
5505 | goto out; | |
5506 | } | |
5507 | advance_right = ADVANCE; | |
5508 | continue; | |
5509 | } else if (right_end_reached) { | |
5510 | if (left_level == 0) { | |
5511 | ret = changed_cb(left_root, right_root, | |
5512 | left_path, right_path, | |
5513 | &left_key, | |
5514 | BTRFS_COMPARE_TREE_NEW, | |
5515 | ctx); | |
5516 | if (ret < 0) | |
5517 | goto out; | |
5518 | } | |
5519 | advance_left = ADVANCE; | |
5520 | continue; | |
5521 | } | |
5522 | ||
5523 | if (left_level == 0 && right_level == 0) { | |
5524 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5525 | if (cmp < 0) { | |
5526 | ret = changed_cb(left_root, right_root, | |
5527 | left_path, right_path, | |
5528 | &left_key, | |
5529 | BTRFS_COMPARE_TREE_NEW, | |
5530 | ctx); | |
5531 | if (ret < 0) | |
5532 | goto out; | |
5533 | advance_left = ADVANCE; | |
5534 | } else if (cmp > 0) { | |
5535 | ret = changed_cb(left_root, right_root, | |
5536 | left_path, right_path, | |
5537 | &right_key, | |
5538 | BTRFS_COMPARE_TREE_DELETED, | |
5539 | ctx); | |
5540 | if (ret < 0) | |
5541 | goto out; | |
5542 | advance_right = ADVANCE; | |
5543 | } else { | |
ba5e8f2e JB |
5544 | enum btrfs_compare_tree_result cmp; |
5545 | ||
74dd17fb | 5546 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5547 | ret = tree_compare_item(left_root, left_path, |
5548 | right_path, tmp_buf); | |
ba5e8f2e JB |
5549 | if (ret) |
5550 | cmp = BTRFS_COMPARE_TREE_CHANGED; | |
5551 | else | |
5552 | cmp = BTRFS_COMPARE_TREE_SAME; | |
5553 | ret = changed_cb(left_root, right_root, | |
5554 | left_path, right_path, | |
5555 | &left_key, cmp, ctx); | |
5556 | if (ret < 0) | |
5557 | goto out; | |
7069830a AB |
5558 | advance_left = ADVANCE; |
5559 | advance_right = ADVANCE; | |
5560 | } | |
5561 | } else if (left_level == right_level) { | |
5562 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5563 | if (cmp < 0) { | |
5564 | advance_left = ADVANCE; | |
5565 | } else if (cmp > 0) { | |
5566 | advance_right = ADVANCE; | |
5567 | } else { | |
5568 | left_blockptr = btrfs_node_blockptr( | |
5569 | left_path->nodes[left_level], | |
5570 | left_path->slots[left_level]); | |
5571 | right_blockptr = btrfs_node_blockptr( | |
5572 | right_path->nodes[right_level], | |
5573 | right_path->slots[right_level]); | |
6baa4293 FM |
5574 | left_gen = btrfs_node_ptr_generation( |
5575 | left_path->nodes[left_level], | |
5576 | left_path->slots[left_level]); | |
5577 | right_gen = btrfs_node_ptr_generation( | |
5578 | right_path->nodes[right_level], | |
5579 | right_path->slots[right_level]); | |
5580 | if (left_blockptr == right_blockptr && | |
5581 | left_gen == right_gen) { | |
7069830a AB |
5582 | /* |
5583 | * As we're on a shared block, don't | |
5584 | * allow to go deeper. | |
5585 | */ | |
5586 | advance_left = ADVANCE_ONLY_NEXT; | |
5587 | advance_right = ADVANCE_ONLY_NEXT; | |
5588 | } else { | |
5589 | advance_left = ADVANCE; | |
5590 | advance_right = ADVANCE; | |
5591 | } | |
5592 | } | |
5593 | } else if (left_level < right_level) { | |
5594 | advance_right = ADVANCE; | |
5595 | } else { | |
5596 | advance_left = ADVANCE; | |
5597 | } | |
5598 | } | |
5599 | ||
5600 | out: | |
5601 | btrfs_free_path(left_path); | |
5602 | btrfs_free_path(right_path); | |
5603 | kfree(tmp_buf); | |
7069830a AB |
5604 | return ret; |
5605 | } | |
5606 | ||
3f157a2f CM |
5607 | /* |
5608 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5609 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5610 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5611 | * |
5612 | * 0 is returned if another key is found, < 0 if there are any errors | |
5613 | * and 1 is returned if there are no higher keys in the tree | |
5614 | * | |
5615 | * path->keep_locks should be set to 1 on the search made before | |
5616 | * calling this function. | |
5617 | */ | |
e7a84565 | 5618 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5619 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5620 | { |
e7a84565 CM |
5621 | int slot; |
5622 | struct extent_buffer *c; | |
5623 | ||
934d375b | 5624 | WARN_ON(!path->keep_locks); |
d397712b | 5625 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5626 | if (!path->nodes[level]) |
5627 | return 1; | |
5628 | ||
5629 | slot = path->slots[level] + 1; | |
5630 | c = path->nodes[level]; | |
3f157a2f | 5631 | next: |
e7a84565 | 5632 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5633 | int ret; |
5634 | int orig_lowest; | |
5635 | struct btrfs_key cur_key; | |
5636 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5637 | !path->nodes[level + 1]) | |
e7a84565 | 5638 | return 1; |
33c66f43 YZ |
5639 | |
5640 | if (path->locks[level + 1]) { | |
5641 | level++; | |
5642 | continue; | |
5643 | } | |
5644 | ||
5645 | slot = btrfs_header_nritems(c) - 1; | |
5646 | if (level == 0) | |
5647 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5648 | else | |
5649 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5650 | ||
5651 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5652 | btrfs_release_path(path); |
33c66f43 YZ |
5653 | path->lowest_level = level; |
5654 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5655 | 0, 0); | |
5656 | path->lowest_level = orig_lowest; | |
5657 | if (ret < 0) | |
5658 | return ret; | |
5659 | ||
5660 | c = path->nodes[level]; | |
5661 | slot = path->slots[level]; | |
5662 | if (ret == 0) | |
5663 | slot++; | |
5664 | goto next; | |
e7a84565 | 5665 | } |
33c66f43 | 5666 | |
e7a84565 CM |
5667 | if (level == 0) |
5668 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5669 | else { |
3f157a2f CM |
5670 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5671 | ||
3f157a2f CM |
5672 | if (gen < min_trans) { |
5673 | slot++; | |
5674 | goto next; | |
5675 | } | |
e7a84565 | 5676 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5677 | } |
e7a84565 CM |
5678 | return 0; |
5679 | } | |
5680 | return 1; | |
5681 | } | |
5682 | ||
97571fd0 | 5683 | /* |
925baedd | 5684 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5685 | * returns 0 if it found something or 1 if there are no greater leaves. |
5686 | * returns < 0 on io errors. | |
97571fd0 | 5687 | */ |
234b63a0 | 5688 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5689 | { |
5690 | return btrfs_next_old_leaf(root, path, 0); | |
5691 | } | |
5692 | ||
5693 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5694 | u64 time_seq) | |
d97e63b6 CM |
5695 | { |
5696 | int slot; | |
8e73f275 | 5697 | int level; |
5f39d397 | 5698 | struct extent_buffer *c; |
8e73f275 | 5699 | struct extent_buffer *next; |
925baedd CM |
5700 | struct btrfs_key key; |
5701 | u32 nritems; | |
5702 | int ret; | |
8e73f275 | 5703 | int old_spinning = path->leave_spinning; |
bd681513 | 5704 | int next_rw_lock = 0; |
925baedd CM |
5705 | |
5706 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5707 | if (nritems == 0) |
925baedd | 5708 | return 1; |
925baedd | 5709 | |
8e73f275 CM |
5710 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5711 | again: | |
5712 | level = 1; | |
5713 | next = NULL; | |
bd681513 | 5714 | next_rw_lock = 0; |
b3b4aa74 | 5715 | btrfs_release_path(path); |
8e73f275 | 5716 | |
a2135011 | 5717 | path->keep_locks = 1; |
31533fb2 | 5718 | path->leave_spinning = 1; |
8e73f275 | 5719 | |
3d7806ec JS |
5720 | if (time_seq) |
5721 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5722 | else | |
5723 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5724 | path->keep_locks = 0; |
5725 | ||
5726 | if (ret < 0) | |
5727 | return ret; | |
5728 | ||
a2135011 | 5729 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5730 | /* |
5731 | * by releasing the path above we dropped all our locks. A balance | |
5732 | * could have added more items next to the key that used to be | |
5733 | * at the very end of the block. So, check again here and | |
5734 | * advance the path if there are now more items available. | |
5735 | */ | |
a2135011 | 5736 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5737 | if (ret == 0) |
5738 | path->slots[0]++; | |
8e73f275 | 5739 | ret = 0; |
925baedd CM |
5740 | goto done; |
5741 | } | |
d97e63b6 | 5742 | |
d397712b | 5743 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5744 | if (!path->nodes[level]) { |
5745 | ret = 1; | |
5746 | goto done; | |
5747 | } | |
5f39d397 | 5748 | |
d97e63b6 CM |
5749 | slot = path->slots[level] + 1; |
5750 | c = path->nodes[level]; | |
5f39d397 | 5751 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5752 | level++; |
8e73f275 CM |
5753 | if (level == BTRFS_MAX_LEVEL) { |
5754 | ret = 1; | |
5755 | goto done; | |
5756 | } | |
d97e63b6 CM |
5757 | continue; |
5758 | } | |
5f39d397 | 5759 | |
925baedd | 5760 | if (next) { |
bd681513 | 5761 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5762 | free_extent_buffer(next); |
925baedd | 5763 | } |
5f39d397 | 5764 | |
8e73f275 | 5765 | next = c; |
bd681513 | 5766 | next_rw_lock = path->locks[level]; |
8e73f275 | 5767 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5768 | slot, &key, 0); |
8e73f275 CM |
5769 | if (ret == -EAGAIN) |
5770 | goto again; | |
5f39d397 | 5771 | |
76a05b35 | 5772 | if (ret < 0) { |
b3b4aa74 | 5773 | btrfs_release_path(path); |
76a05b35 CM |
5774 | goto done; |
5775 | } | |
5776 | ||
5cd57b2c | 5777 | if (!path->skip_locking) { |
bd681513 | 5778 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5779 | if (!ret && time_seq) { |
5780 | /* | |
5781 | * If we don't get the lock, we may be racing | |
5782 | * with push_leaf_left, holding that lock while | |
5783 | * itself waiting for the leaf we've currently | |
5784 | * locked. To solve this situation, we give up | |
5785 | * on our lock and cycle. | |
5786 | */ | |
cf538830 | 5787 | free_extent_buffer(next); |
d42244a0 JS |
5788 | btrfs_release_path(path); |
5789 | cond_resched(); | |
5790 | goto again; | |
5791 | } | |
8e73f275 CM |
5792 | if (!ret) { |
5793 | btrfs_set_path_blocking(path); | |
bd681513 | 5794 | btrfs_tree_read_lock(next); |
31533fb2 | 5795 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5796 | BTRFS_READ_LOCK); |
8e73f275 | 5797 | } |
31533fb2 | 5798 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5799 | } |
d97e63b6 CM |
5800 | break; |
5801 | } | |
5802 | path->slots[level] = slot; | |
d397712b | 5803 | while (1) { |
d97e63b6 CM |
5804 | level--; |
5805 | c = path->nodes[level]; | |
925baedd | 5806 | if (path->locks[level]) |
bd681513 | 5807 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5808 | |
5f39d397 | 5809 | free_extent_buffer(c); |
d97e63b6 CM |
5810 | path->nodes[level] = next; |
5811 | path->slots[level] = 0; | |
a74a4b97 | 5812 | if (!path->skip_locking) |
bd681513 | 5813 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5814 | if (!level) |
5815 | break; | |
b4ce94de | 5816 | |
8e73f275 | 5817 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5818 | 0, &key, 0); |
8e73f275 CM |
5819 | if (ret == -EAGAIN) |
5820 | goto again; | |
5821 | ||
76a05b35 | 5822 | if (ret < 0) { |
b3b4aa74 | 5823 | btrfs_release_path(path); |
76a05b35 CM |
5824 | goto done; |
5825 | } | |
5826 | ||
5cd57b2c | 5827 | if (!path->skip_locking) { |
bd681513 | 5828 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5829 | if (!ret) { |
5830 | btrfs_set_path_blocking(path); | |
bd681513 | 5831 | btrfs_tree_read_lock(next); |
31533fb2 | 5832 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5833 | BTRFS_READ_LOCK); |
5834 | } | |
31533fb2 | 5835 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5836 | } |
d97e63b6 | 5837 | } |
8e73f275 | 5838 | ret = 0; |
925baedd | 5839 | done: |
f7c79f30 | 5840 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5841 | path->leave_spinning = old_spinning; |
5842 | if (!old_spinning) | |
5843 | btrfs_set_path_blocking(path); | |
5844 | ||
5845 | return ret; | |
d97e63b6 | 5846 | } |
0b86a832 | 5847 | |
3f157a2f CM |
5848 | /* |
5849 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5850 | * searching until it gets past min_objectid or finds an item of 'type' | |
5851 | * | |
5852 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5853 | */ | |
0b86a832 CM |
5854 | int btrfs_previous_item(struct btrfs_root *root, |
5855 | struct btrfs_path *path, u64 min_objectid, | |
5856 | int type) | |
5857 | { | |
5858 | struct btrfs_key found_key; | |
5859 | struct extent_buffer *leaf; | |
e02119d5 | 5860 | u32 nritems; |
0b86a832 CM |
5861 | int ret; |
5862 | ||
d397712b | 5863 | while (1) { |
0b86a832 | 5864 | if (path->slots[0] == 0) { |
b4ce94de | 5865 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5866 | ret = btrfs_prev_leaf(root, path); |
5867 | if (ret != 0) | |
5868 | return ret; | |
5869 | } else { | |
5870 | path->slots[0]--; | |
5871 | } | |
5872 | leaf = path->nodes[0]; | |
e02119d5 CM |
5873 | nritems = btrfs_header_nritems(leaf); |
5874 | if (nritems == 0) | |
5875 | return 1; | |
5876 | if (path->slots[0] == nritems) | |
5877 | path->slots[0]--; | |
5878 | ||
0b86a832 | 5879 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5880 | if (found_key.objectid < min_objectid) |
5881 | break; | |
0a4eefbb YZ |
5882 | if (found_key.type == type) |
5883 | return 0; | |
e02119d5 CM |
5884 | if (found_key.objectid == min_objectid && |
5885 | found_key.type < type) | |
5886 | break; | |
0b86a832 CM |
5887 | } |
5888 | return 1; | |
5889 | } | |
ade2e0b3 WS |
5890 | |
5891 | /* | |
5892 | * search in extent tree to find a previous Metadata/Data extent item with | |
5893 | * min objecitd. | |
5894 | * | |
5895 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5896 | */ | |
5897 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5898 | struct btrfs_path *path, u64 min_objectid) | |
5899 | { | |
5900 | struct btrfs_key found_key; | |
5901 | struct extent_buffer *leaf; | |
5902 | u32 nritems; | |
5903 | int ret; | |
5904 | ||
5905 | while (1) { | |
5906 | if (path->slots[0] == 0) { | |
5907 | btrfs_set_path_blocking(path); | |
5908 | ret = btrfs_prev_leaf(root, path); | |
5909 | if (ret != 0) | |
5910 | return ret; | |
5911 | } else { | |
5912 | path->slots[0]--; | |
5913 | } | |
5914 | leaf = path->nodes[0]; | |
5915 | nritems = btrfs_header_nritems(leaf); | |
5916 | if (nritems == 0) | |
5917 | return 1; | |
5918 | if (path->slots[0] == nritems) | |
5919 | path->slots[0]--; | |
5920 | ||
5921 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5922 | if (found_key.objectid < min_objectid) | |
5923 | break; | |
5924 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5925 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5926 | return 0; | |
5927 | if (found_key.objectid == min_objectid && | |
5928 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5929 | break; | |
5930 | } | |
5931 | return 1; | |
5932 | } |