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> |
8f282f71 | 22 | #include <linux/vmalloc.h> |
eb60ceac CM |
23 | #include "ctree.h" |
24 | #include "disk-io.h" | |
7f5c1516 | 25 | #include "transaction.h" |
5f39d397 | 26 | #include "print-tree.h" |
925baedd | 27 | #include "locking.h" |
9a8dd150 | 28 | |
e089f05c CM |
29 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
30 | *root, struct btrfs_path *path, int level); | |
31 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root | |
d4dbff95 | 32 | *root, struct btrfs_key *ins_key, |
cc0c5538 | 33 | struct btrfs_path *path, int data_size, int extend); |
5f39d397 CM |
34 | static int push_node_left(struct btrfs_trans_handle *trans, |
35 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 36 | struct extent_buffer *src, int empty); |
5f39d397 CM |
37 | static int balance_node_right(struct btrfs_trans_handle *trans, |
38 | struct btrfs_root *root, | |
39 | struct extent_buffer *dst_buf, | |
40 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
41 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
42 | int level, int slot); | |
5de865ee | 43 | static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
f230475e | 44 | struct extent_buffer *eb); |
d97e63b6 | 45 | |
df24a2b9 | 46 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 47 | { |
df24a2b9 | 48 | struct btrfs_path *path; |
e00f7308 | 49 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 50 | return path; |
2c90e5d6 CM |
51 | } |
52 | ||
b4ce94de CM |
53 | /* |
54 | * set all locked nodes in the path to blocking locks. This should | |
55 | * be done before scheduling | |
56 | */ | |
57 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
58 | { | |
59 | int i; | |
60 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
61 | if (!p->nodes[i] || !p->locks[i]) |
62 | continue; | |
63 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
64 | if (p->locks[i] == BTRFS_READ_LOCK) | |
65 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
66 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
67 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
68 | } |
69 | } | |
70 | ||
71 | /* | |
72 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
73 | * |
74 | * held is used to keep lockdep happy, when lockdep is enabled | |
75 | * we set held to a blocking lock before we go around and | |
76 | * retake all the spinlocks in the path. You can safely use NULL | |
77 | * for held | |
b4ce94de | 78 | */ |
4008c04a | 79 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 80 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
81 | { |
82 | int i; | |
4008c04a | 83 | |
bd681513 CM |
84 | if (held) { |
85 | btrfs_set_lock_blocking_rw(held, held_rw); | |
86 | if (held_rw == BTRFS_WRITE_LOCK) | |
87 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
88 | else if (held_rw == BTRFS_READ_LOCK) | |
89 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
90 | } | |
4008c04a | 91 | btrfs_set_path_blocking(p); |
4008c04a CM |
92 | |
93 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
94 | if (p->nodes[i] && p->locks[i]) { |
95 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
96 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
97 | p->locks[i] = BTRFS_WRITE_LOCK; | |
98 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
99 | p->locks[i] = BTRFS_READ_LOCK; | |
100 | } | |
b4ce94de | 101 | } |
4008c04a | 102 | |
4008c04a | 103 | if (held) |
bd681513 | 104 | btrfs_clear_lock_blocking_rw(held, held_rw); |
b4ce94de CM |
105 | } |
106 | ||
d352ac68 | 107 | /* this also releases the path */ |
df24a2b9 | 108 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 109 | { |
ff175d57 JJ |
110 | if (!p) |
111 | return; | |
b3b4aa74 | 112 | btrfs_release_path(p); |
df24a2b9 | 113 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
114 | } |
115 | ||
d352ac68 CM |
116 | /* |
117 | * path release drops references on the extent buffers in the path | |
118 | * and it drops any locks held by this path | |
119 | * | |
120 | * It is safe to call this on paths that no locks or extent buffers held. | |
121 | */ | |
b3b4aa74 | 122 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
123 | { |
124 | int i; | |
a2135011 | 125 | |
234b63a0 | 126 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 127 | p->slots[i] = 0; |
eb60ceac | 128 | if (!p->nodes[i]) |
925baedd CM |
129 | continue; |
130 | if (p->locks[i]) { | |
bd681513 | 131 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
132 | p->locks[i] = 0; |
133 | } | |
5f39d397 | 134 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 135 | p->nodes[i] = NULL; |
eb60ceac CM |
136 | } |
137 | } | |
138 | ||
d352ac68 CM |
139 | /* |
140 | * safely gets a reference on the root node of a tree. A lock | |
141 | * is not taken, so a concurrent writer may put a different node | |
142 | * at the root of the tree. See btrfs_lock_root_node for the | |
143 | * looping required. | |
144 | * | |
145 | * The extent buffer returned by this has a reference taken, so | |
146 | * it won't disappear. It may stop being the root of the tree | |
147 | * at any time because there are no locks held. | |
148 | */ | |
925baedd CM |
149 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
150 | { | |
151 | struct extent_buffer *eb; | |
240f62c8 | 152 | |
3083ee2e JB |
153 | while (1) { |
154 | rcu_read_lock(); | |
155 | eb = rcu_dereference(root->node); | |
156 | ||
157 | /* | |
158 | * RCU really hurts here, we could free up the root node because | |
01327610 | 159 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
160 | * the inc_not_zero dance and if it doesn't work then |
161 | * synchronize_rcu and try again. | |
162 | */ | |
163 | if (atomic_inc_not_zero(&eb->refs)) { | |
164 | rcu_read_unlock(); | |
165 | break; | |
166 | } | |
167 | rcu_read_unlock(); | |
168 | synchronize_rcu(); | |
169 | } | |
925baedd CM |
170 | return eb; |
171 | } | |
172 | ||
d352ac68 CM |
173 | /* loop around taking references on and locking the root node of the |
174 | * tree until you end up with a lock on the root. A locked buffer | |
175 | * is returned, with a reference held. | |
176 | */ | |
925baedd CM |
177 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
178 | { | |
179 | struct extent_buffer *eb; | |
180 | ||
d397712b | 181 | while (1) { |
925baedd CM |
182 | eb = btrfs_root_node(root); |
183 | btrfs_tree_lock(eb); | |
240f62c8 | 184 | if (eb == root->node) |
925baedd | 185 | break; |
925baedd CM |
186 | btrfs_tree_unlock(eb); |
187 | free_extent_buffer(eb); | |
188 | } | |
189 | return eb; | |
190 | } | |
191 | ||
bd681513 CM |
192 | /* loop around taking references on and locking the root node of the |
193 | * tree until you end up with a lock on the root. A locked buffer | |
194 | * is returned, with a reference held. | |
195 | */ | |
48a3b636 | 196 | static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
197 | { |
198 | struct extent_buffer *eb; | |
199 | ||
200 | while (1) { | |
201 | eb = btrfs_root_node(root); | |
202 | btrfs_tree_read_lock(eb); | |
203 | if (eb == root->node) | |
204 | break; | |
205 | btrfs_tree_read_unlock(eb); | |
206 | free_extent_buffer(eb); | |
207 | } | |
208 | return eb; | |
209 | } | |
210 | ||
d352ac68 CM |
211 | /* cowonly root (everything not a reference counted cow subvolume), just get |
212 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
213 | * get properly updated on disk. | |
214 | */ | |
0b86a832 CM |
215 | static void add_root_to_dirty_list(struct btrfs_root *root) |
216 | { | |
e7070be1 JB |
217 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
218 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
219 | return; | |
220 | ||
e5846fc6 | 221 | spin_lock(&root->fs_info->trans_lock); |
e7070be1 JB |
222 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
223 | /* Want the extent tree to be the last on the list */ | |
224 | if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
225 | list_move_tail(&root->dirty_list, | |
226 | &root->fs_info->dirty_cowonly_roots); | |
227 | else | |
228 | list_move(&root->dirty_list, | |
229 | &root->fs_info->dirty_cowonly_roots); | |
0b86a832 | 230 | } |
e5846fc6 | 231 | spin_unlock(&root->fs_info->trans_lock); |
0b86a832 CM |
232 | } |
233 | ||
d352ac68 CM |
234 | /* |
235 | * used by snapshot creation to make a copy of a root for a tree with | |
236 | * a given objectid. The buffer with the new root node is returned in | |
237 | * cow_ret, and this func returns zero on success or a negative error code. | |
238 | */ | |
be20aa9d CM |
239 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
240 | struct btrfs_root *root, | |
241 | struct extent_buffer *buf, | |
242 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
243 | { | |
244 | struct extent_buffer *cow; | |
be20aa9d CM |
245 | int ret = 0; |
246 | int level; | |
5d4f98a2 | 247 | struct btrfs_disk_key disk_key; |
be20aa9d | 248 | |
27cdeb70 MX |
249 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
250 | trans->transid != root->fs_info->running_transaction->transid); | |
251 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && | |
252 | trans->transid != root->last_trans); | |
be20aa9d CM |
253 | |
254 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
255 | if (level == 0) |
256 | btrfs_item_key(buf, &disk_key, 0); | |
257 | else | |
258 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 259 | |
4d75f8a9 DS |
260 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
261 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 262 | if (IS_ERR(cow)) |
be20aa9d CM |
263 | return PTR_ERR(cow); |
264 | ||
265 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
266 | btrfs_set_header_bytenr(cow, cow->start); | |
267 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
268 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
269 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
270 | BTRFS_HEADER_FLAG_RELOC); | |
271 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
272 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
273 | else | |
274 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 275 | |
0a4e5586 | 276 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(), |
2b82032c YZ |
277 | BTRFS_FSID_SIZE); |
278 | ||
be20aa9d | 279 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 280 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 281 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 282 | else |
e339a6b0 | 283 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 284 | |
be20aa9d CM |
285 | if (ret) |
286 | return ret; | |
287 | ||
288 | btrfs_mark_buffer_dirty(cow); | |
289 | *cow_ret = cow; | |
290 | return 0; | |
291 | } | |
292 | ||
bd989ba3 JS |
293 | enum mod_log_op { |
294 | MOD_LOG_KEY_REPLACE, | |
295 | MOD_LOG_KEY_ADD, | |
296 | MOD_LOG_KEY_REMOVE, | |
297 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
298 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
299 | MOD_LOG_MOVE_KEYS, | |
300 | MOD_LOG_ROOT_REPLACE, | |
301 | }; | |
302 | ||
303 | struct tree_mod_move { | |
304 | int dst_slot; | |
305 | int nr_items; | |
306 | }; | |
307 | ||
308 | struct tree_mod_root { | |
309 | u64 logical; | |
310 | u8 level; | |
311 | }; | |
312 | ||
313 | struct tree_mod_elem { | |
314 | struct rb_node node; | |
298cfd36 | 315 | u64 logical; |
097b8a7c | 316 | u64 seq; |
bd989ba3 JS |
317 | enum mod_log_op op; |
318 | ||
319 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
320 | int slot; | |
321 | ||
322 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
323 | u64 generation; | |
324 | ||
325 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
326 | struct btrfs_disk_key key; | |
327 | u64 blockptr; | |
328 | ||
329 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
330 | struct tree_mod_move move; | |
331 | ||
332 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
333 | struct tree_mod_root old_root; | |
334 | }; | |
335 | ||
097b8a7c | 336 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 337 | { |
097b8a7c | 338 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
339 | } |
340 | ||
097b8a7c JS |
341 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
342 | { | |
343 | read_unlock(&fs_info->tree_mod_log_lock); | |
344 | } | |
345 | ||
346 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
347 | { | |
348 | write_lock(&fs_info->tree_mod_log_lock); | |
349 | } | |
350 | ||
351 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
352 | { | |
353 | write_unlock(&fs_info->tree_mod_log_lock); | |
354 | } | |
355 | ||
fc36ed7e | 356 | /* |
fcebe456 | 357 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 358 | */ |
fcebe456 | 359 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
360 | { |
361 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
362 | } | |
363 | ||
097b8a7c JS |
364 | /* |
365 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
366 | * passed does not already have a sequence number set. So when a caller expects | |
367 | * to record tree modifications, it should ensure to set elem->seq to zero | |
368 | * before calling btrfs_get_tree_mod_seq. | |
369 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
370 | * blocker was added. | |
371 | */ | |
372 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
373 | struct seq_list *elem) | |
bd989ba3 | 374 | { |
097b8a7c | 375 | tree_mod_log_write_lock(fs_info); |
bd989ba3 | 376 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 377 | if (!elem->seq) { |
fcebe456 | 378 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
379 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
380 | } | |
bd989ba3 | 381 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
382 | tree_mod_log_write_unlock(fs_info); |
383 | ||
fcebe456 | 384 | return elem->seq; |
bd989ba3 JS |
385 | } |
386 | ||
387 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
388 | struct seq_list *elem) | |
389 | { | |
390 | struct rb_root *tm_root; | |
391 | struct rb_node *node; | |
392 | struct rb_node *next; | |
393 | struct seq_list *cur_elem; | |
394 | struct tree_mod_elem *tm; | |
395 | u64 min_seq = (u64)-1; | |
396 | u64 seq_putting = elem->seq; | |
397 | ||
398 | if (!seq_putting) | |
399 | return; | |
400 | ||
bd989ba3 JS |
401 | spin_lock(&fs_info->tree_mod_seq_lock); |
402 | list_del(&elem->list); | |
097b8a7c | 403 | elem->seq = 0; |
bd989ba3 JS |
404 | |
405 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 406 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
407 | if (seq_putting > cur_elem->seq) { |
408 | /* | |
409 | * blocker with lower sequence number exists, we | |
410 | * cannot remove anything from the log | |
411 | */ | |
097b8a7c JS |
412 | spin_unlock(&fs_info->tree_mod_seq_lock); |
413 | return; | |
bd989ba3 JS |
414 | } |
415 | min_seq = cur_elem->seq; | |
416 | } | |
417 | } | |
097b8a7c JS |
418 | spin_unlock(&fs_info->tree_mod_seq_lock); |
419 | ||
bd989ba3 JS |
420 | /* |
421 | * anything that's lower than the lowest existing (read: blocked) | |
422 | * sequence number can be removed from the tree. | |
423 | */ | |
097b8a7c | 424 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
425 | tm_root = &fs_info->tree_mod_log; |
426 | for (node = rb_first(tm_root); node; node = next) { | |
427 | next = rb_next(node); | |
428 | tm = container_of(node, struct tree_mod_elem, node); | |
097b8a7c | 429 | if (tm->seq > min_seq) |
bd989ba3 JS |
430 | continue; |
431 | rb_erase(node, tm_root); | |
bd989ba3 JS |
432 | kfree(tm); |
433 | } | |
097b8a7c | 434 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
435 | } |
436 | ||
437 | /* | |
438 | * key order of the log: | |
298cfd36 | 439 | * node/leaf start address -> sequence |
bd989ba3 | 440 | * |
298cfd36 CR |
441 | * The 'start address' is the logical address of the *new* root node |
442 | * for root replace operations, or the logical address of the affected | |
443 | * block for all other operations. | |
5de865ee FDBM |
444 | * |
445 | * Note: must be called with write lock (tree_mod_log_write_lock). | |
bd989ba3 JS |
446 | */ |
447 | static noinline int | |
448 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
449 | { | |
450 | struct rb_root *tm_root; | |
451 | struct rb_node **new; | |
452 | struct rb_node *parent = NULL; | |
453 | struct tree_mod_elem *cur; | |
c8cc6341 JB |
454 | |
455 | BUG_ON(!tm); | |
456 | ||
fcebe456 | 457 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 458 | |
bd989ba3 JS |
459 | tm_root = &fs_info->tree_mod_log; |
460 | new = &tm_root->rb_node; | |
461 | while (*new) { | |
462 | cur = container_of(*new, struct tree_mod_elem, node); | |
463 | parent = *new; | |
298cfd36 | 464 | if (cur->logical < tm->logical) |
bd989ba3 | 465 | new = &((*new)->rb_left); |
298cfd36 | 466 | else if (cur->logical > tm->logical) |
bd989ba3 | 467 | new = &((*new)->rb_right); |
097b8a7c | 468 | else if (cur->seq < tm->seq) |
bd989ba3 | 469 | new = &((*new)->rb_left); |
097b8a7c | 470 | else if (cur->seq > tm->seq) |
bd989ba3 | 471 | new = &((*new)->rb_right); |
5de865ee FDBM |
472 | else |
473 | return -EEXIST; | |
bd989ba3 JS |
474 | } |
475 | ||
476 | rb_link_node(&tm->node, parent, new); | |
477 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 478 | return 0; |
bd989ba3 JS |
479 | } |
480 | ||
097b8a7c JS |
481 | /* |
482 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
483 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
484 | * this until all tree mod log insertions are recorded in the rb tree and then | |
485 | * call tree_mod_log_write_unlock() to release. | |
486 | */ | |
e9b7fd4d JS |
487 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
488 | struct extent_buffer *eb) { | |
489 | smp_mb(); | |
490 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
491 | return 1; | |
097b8a7c JS |
492 | if (eb && btrfs_header_level(eb) == 0) |
493 | return 1; | |
5de865ee FDBM |
494 | |
495 | tree_mod_log_write_lock(fs_info); | |
496 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { | |
497 | tree_mod_log_write_unlock(fs_info); | |
498 | return 1; | |
499 | } | |
500 | ||
e9b7fd4d JS |
501 | return 0; |
502 | } | |
503 | ||
5de865ee FDBM |
504 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
505 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
506 | struct extent_buffer *eb) | |
507 | { | |
508 | smp_mb(); | |
509 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
510 | return 0; | |
511 | if (eb && btrfs_header_level(eb) == 0) | |
512 | return 0; | |
513 | ||
514 | return 1; | |
515 | } | |
516 | ||
517 | static struct tree_mod_elem * | |
518 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
519 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 520 | { |
097b8a7c | 521 | struct tree_mod_elem *tm; |
bd989ba3 | 522 | |
c8cc6341 JB |
523 | tm = kzalloc(sizeof(*tm), flags); |
524 | if (!tm) | |
5de865ee | 525 | return NULL; |
bd989ba3 | 526 | |
298cfd36 | 527 | tm->logical = eb->start; |
bd989ba3 JS |
528 | if (op != MOD_LOG_KEY_ADD) { |
529 | btrfs_node_key(eb, &tm->key, slot); | |
530 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
531 | } | |
532 | tm->op = op; | |
533 | tm->slot = slot; | |
534 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 535 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 536 | |
5de865ee | 537 | return tm; |
097b8a7c JS |
538 | } |
539 | ||
540 | static noinline int | |
c8cc6341 JB |
541 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
542 | struct extent_buffer *eb, int slot, | |
543 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 544 | { |
5de865ee FDBM |
545 | struct tree_mod_elem *tm; |
546 | int ret; | |
547 | ||
548 | if (!tree_mod_need_log(fs_info, eb)) | |
549 | return 0; | |
550 | ||
551 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
552 | if (!tm) | |
553 | return -ENOMEM; | |
554 | ||
555 | if (tree_mod_dont_log(fs_info, eb)) { | |
556 | kfree(tm); | |
097b8a7c | 557 | return 0; |
5de865ee FDBM |
558 | } |
559 | ||
560 | ret = __tree_mod_log_insert(fs_info, tm); | |
561 | tree_mod_log_write_unlock(fs_info); | |
562 | if (ret) | |
563 | kfree(tm); | |
097b8a7c | 564 | |
5de865ee | 565 | return ret; |
097b8a7c JS |
566 | } |
567 | ||
bd989ba3 JS |
568 | static noinline int |
569 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
570 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
571 | int nr_items, gfp_t flags) | |
572 | { | |
5de865ee FDBM |
573 | struct tree_mod_elem *tm = NULL; |
574 | struct tree_mod_elem **tm_list = NULL; | |
575 | int ret = 0; | |
bd989ba3 | 576 | int i; |
5de865ee | 577 | int locked = 0; |
bd989ba3 | 578 | |
5de865ee | 579 | if (!tree_mod_need_log(fs_info, eb)) |
f395694c | 580 | return 0; |
bd989ba3 | 581 | |
31e818fe | 582 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), flags); |
5de865ee FDBM |
583 | if (!tm_list) |
584 | return -ENOMEM; | |
585 | ||
586 | tm = kzalloc(sizeof(*tm), flags); | |
587 | if (!tm) { | |
588 | ret = -ENOMEM; | |
589 | goto free_tms; | |
590 | } | |
591 | ||
298cfd36 | 592 | tm->logical = eb->start; |
5de865ee FDBM |
593 | tm->slot = src_slot; |
594 | tm->move.dst_slot = dst_slot; | |
595 | tm->move.nr_items = nr_items; | |
596 | tm->op = MOD_LOG_MOVE_KEYS; | |
597 | ||
598 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
599 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
600 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags); | |
601 | if (!tm_list[i]) { | |
602 | ret = -ENOMEM; | |
603 | goto free_tms; | |
604 | } | |
605 | } | |
606 | ||
607 | if (tree_mod_dont_log(fs_info, eb)) | |
608 | goto free_tms; | |
609 | locked = 1; | |
610 | ||
01763a2e JS |
611 | /* |
612 | * When we override something during the move, we log these removals. | |
613 | * This can only happen when we move towards the beginning of the | |
614 | * buffer, i.e. dst_slot < src_slot. | |
615 | */ | |
bd989ba3 | 616 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
5de865ee FDBM |
617 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
618 | if (ret) | |
619 | goto free_tms; | |
bd989ba3 JS |
620 | } |
621 | ||
5de865ee FDBM |
622 | ret = __tree_mod_log_insert(fs_info, tm); |
623 | if (ret) | |
624 | goto free_tms; | |
625 | tree_mod_log_write_unlock(fs_info); | |
626 | kfree(tm_list); | |
f395694c | 627 | |
5de865ee FDBM |
628 | return 0; |
629 | free_tms: | |
630 | for (i = 0; i < nr_items; i++) { | |
631 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
632 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
633 | kfree(tm_list[i]); | |
634 | } | |
635 | if (locked) | |
636 | tree_mod_log_write_unlock(fs_info); | |
637 | kfree(tm_list); | |
638 | kfree(tm); | |
bd989ba3 | 639 | |
5de865ee | 640 | return ret; |
bd989ba3 JS |
641 | } |
642 | ||
5de865ee FDBM |
643 | static inline int |
644 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
645 | struct tree_mod_elem **tm_list, | |
646 | int nritems) | |
097b8a7c | 647 | { |
5de865ee | 648 | int i, j; |
097b8a7c JS |
649 | int ret; |
650 | ||
097b8a7c | 651 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
652 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
653 | if (ret) { | |
654 | for (j = nritems - 1; j > i; j--) | |
655 | rb_erase(&tm_list[j]->node, | |
656 | &fs_info->tree_mod_log); | |
657 | return ret; | |
658 | } | |
097b8a7c | 659 | } |
5de865ee FDBM |
660 | |
661 | return 0; | |
097b8a7c JS |
662 | } |
663 | ||
bd989ba3 JS |
664 | static noinline int |
665 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
666 | struct extent_buffer *old_root, | |
90f8d62e JS |
667 | struct extent_buffer *new_root, gfp_t flags, |
668 | int log_removal) | |
bd989ba3 | 669 | { |
5de865ee FDBM |
670 | struct tree_mod_elem *tm = NULL; |
671 | struct tree_mod_elem **tm_list = NULL; | |
672 | int nritems = 0; | |
673 | int ret = 0; | |
674 | int i; | |
bd989ba3 | 675 | |
5de865ee | 676 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
677 | return 0; |
678 | ||
5de865ee FDBM |
679 | if (log_removal && btrfs_header_level(old_root) > 0) { |
680 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 681 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
682 | flags); |
683 | if (!tm_list) { | |
684 | ret = -ENOMEM; | |
685 | goto free_tms; | |
686 | } | |
687 | for (i = 0; i < nritems; i++) { | |
688 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
689 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags); | |
690 | if (!tm_list[i]) { | |
691 | ret = -ENOMEM; | |
692 | goto free_tms; | |
693 | } | |
694 | } | |
695 | } | |
d9abbf1c | 696 | |
c8cc6341 | 697 | tm = kzalloc(sizeof(*tm), flags); |
5de865ee FDBM |
698 | if (!tm) { |
699 | ret = -ENOMEM; | |
700 | goto free_tms; | |
701 | } | |
bd989ba3 | 702 | |
298cfd36 | 703 | tm->logical = new_root->start; |
bd989ba3 JS |
704 | tm->old_root.logical = old_root->start; |
705 | tm->old_root.level = btrfs_header_level(old_root); | |
706 | tm->generation = btrfs_header_generation(old_root); | |
707 | tm->op = MOD_LOG_ROOT_REPLACE; | |
708 | ||
5de865ee FDBM |
709 | if (tree_mod_dont_log(fs_info, NULL)) |
710 | goto free_tms; | |
711 | ||
712 | if (tm_list) | |
713 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
714 | if (!ret) | |
715 | ret = __tree_mod_log_insert(fs_info, tm); | |
716 | ||
717 | tree_mod_log_write_unlock(fs_info); | |
718 | if (ret) | |
719 | goto free_tms; | |
720 | kfree(tm_list); | |
721 | ||
722 | return ret; | |
723 | ||
724 | free_tms: | |
725 | if (tm_list) { | |
726 | for (i = 0; i < nritems; i++) | |
727 | kfree(tm_list[i]); | |
728 | kfree(tm_list); | |
729 | } | |
730 | kfree(tm); | |
731 | ||
732 | return ret; | |
bd989ba3 JS |
733 | } |
734 | ||
735 | static struct tree_mod_elem * | |
736 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
737 | int smallest) | |
738 | { | |
739 | struct rb_root *tm_root; | |
740 | struct rb_node *node; | |
741 | struct tree_mod_elem *cur = NULL; | |
742 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 743 | |
097b8a7c | 744 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
745 | tm_root = &fs_info->tree_mod_log; |
746 | node = tm_root->rb_node; | |
747 | while (node) { | |
748 | cur = container_of(node, struct tree_mod_elem, node); | |
298cfd36 | 749 | if (cur->logical < start) { |
bd989ba3 | 750 | node = node->rb_left; |
298cfd36 | 751 | } else if (cur->logical > start) { |
bd989ba3 | 752 | node = node->rb_right; |
097b8a7c | 753 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
754 | node = node->rb_left; |
755 | } else if (!smallest) { | |
756 | /* we want the node with the highest seq */ | |
757 | if (found) | |
097b8a7c | 758 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
759 | found = cur; |
760 | node = node->rb_left; | |
097b8a7c | 761 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
762 | /* we want the node with the smallest seq */ |
763 | if (found) | |
097b8a7c | 764 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
765 | found = cur; |
766 | node = node->rb_right; | |
767 | } else { | |
768 | found = cur; | |
769 | break; | |
770 | } | |
771 | } | |
097b8a7c | 772 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
773 | |
774 | return found; | |
775 | } | |
776 | ||
777 | /* | |
778 | * this returns the element from the log with the smallest time sequence | |
779 | * value that's in the log (the oldest log item). any element with a time | |
780 | * sequence lower than min_seq will be ignored. | |
781 | */ | |
782 | static struct tree_mod_elem * | |
783 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
784 | u64 min_seq) | |
785 | { | |
786 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
787 | } | |
788 | ||
789 | /* | |
790 | * this returns the element from the log with the largest time sequence | |
791 | * value that's in the log (the most recent log item). any element with | |
792 | * a time sequence lower than min_seq will be ignored. | |
793 | */ | |
794 | static struct tree_mod_elem * | |
795 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
796 | { | |
797 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
798 | } | |
799 | ||
5de865ee | 800 | static noinline int |
bd989ba3 JS |
801 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
802 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 803 | unsigned long src_offset, int nr_items) |
bd989ba3 | 804 | { |
5de865ee FDBM |
805 | int ret = 0; |
806 | struct tree_mod_elem **tm_list = NULL; | |
807 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 808 | int i; |
5de865ee | 809 | int locked = 0; |
bd989ba3 | 810 | |
5de865ee FDBM |
811 | if (!tree_mod_need_log(fs_info, NULL)) |
812 | return 0; | |
bd989ba3 | 813 | |
c8cc6341 | 814 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
815 | return 0; |
816 | ||
31e818fe | 817 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
818 | GFP_NOFS); |
819 | if (!tm_list) | |
820 | return -ENOMEM; | |
bd989ba3 | 821 | |
5de865ee FDBM |
822 | tm_list_add = tm_list; |
823 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 824 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
825 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
826 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
827 | if (!tm_list_rem[i]) { | |
828 | ret = -ENOMEM; | |
829 | goto free_tms; | |
830 | } | |
831 | ||
832 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
833 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
834 | if (!tm_list_add[i]) { | |
835 | ret = -ENOMEM; | |
836 | goto free_tms; | |
837 | } | |
838 | } | |
839 | ||
840 | if (tree_mod_dont_log(fs_info, NULL)) | |
841 | goto free_tms; | |
842 | locked = 1; | |
843 | ||
844 | for (i = 0; i < nr_items; i++) { | |
845 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
846 | if (ret) | |
847 | goto free_tms; | |
848 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
849 | if (ret) | |
850 | goto free_tms; | |
bd989ba3 | 851 | } |
5de865ee FDBM |
852 | |
853 | tree_mod_log_write_unlock(fs_info); | |
854 | kfree(tm_list); | |
855 | ||
856 | return 0; | |
857 | ||
858 | free_tms: | |
859 | for (i = 0; i < nr_items * 2; i++) { | |
860 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
861 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
862 | kfree(tm_list[i]); | |
863 | } | |
864 | if (locked) | |
865 | tree_mod_log_write_unlock(fs_info); | |
866 | kfree(tm_list); | |
867 | ||
868 | return ret; | |
bd989ba3 JS |
869 | } |
870 | ||
871 | static inline void | |
872 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
873 | int dst_offset, int src_offset, int nr_items) | |
874 | { | |
875 | int ret; | |
876 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
877 | nr_items, GFP_NOFS); | |
878 | BUG_ON(ret < 0); | |
879 | } | |
880 | ||
097b8a7c | 881 | static noinline void |
bd989ba3 | 882 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 883 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
884 | { |
885 | int ret; | |
886 | ||
78357766 | 887 | ret = tree_mod_log_insert_key(fs_info, eb, slot, |
c8cc6341 JB |
888 | MOD_LOG_KEY_REPLACE, |
889 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
890 | BUG_ON(ret < 0); |
891 | } | |
892 | ||
5de865ee | 893 | static noinline int |
097b8a7c | 894 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) |
bd989ba3 | 895 | { |
5de865ee FDBM |
896 | struct tree_mod_elem **tm_list = NULL; |
897 | int nritems = 0; | |
898 | int i; | |
899 | int ret = 0; | |
900 | ||
901 | if (btrfs_header_level(eb) == 0) | |
902 | return 0; | |
903 | ||
904 | if (!tree_mod_need_log(fs_info, NULL)) | |
905 | return 0; | |
906 | ||
907 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 908 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
909 | if (!tm_list) |
910 | return -ENOMEM; | |
911 | ||
912 | for (i = 0; i < nritems; i++) { | |
913 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
914 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
915 | if (!tm_list[i]) { | |
916 | ret = -ENOMEM; | |
917 | goto free_tms; | |
918 | } | |
919 | } | |
920 | ||
e9b7fd4d | 921 | if (tree_mod_dont_log(fs_info, eb)) |
5de865ee FDBM |
922 | goto free_tms; |
923 | ||
924 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
925 | tree_mod_log_write_unlock(fs_info); | |
926 | if (ret) | |
927 | goto free_tms; | |
928 | kfree(tm_list); | |
929 | ||
930 | return 0; | |
931 | ||
932 | free_tms: | |
933 | for (i = 0; i < nritems; i++) | |
934 | kfree(tm_list[i]); | |
935 | kfree(tm_list); | |
936 | ||
937 | return ret; | |
bd989ba3 JS |
938 | } |
939 | ||
097b8a7c | 940 | static noinline void |
bd989ba3 | 941 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
942 | struct extent_buffer *new_root_node, |
943 | int log_removal) | |
bd989ba3 JS |
944 | { |
945 | int ret; | |
bd989ba3 | 946 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 947 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
948 | BUG_ON(ret < 0); |
949 | } | |
950 | ||
5d4f98a2 YZ |
951 | /* |
952 | * check if the tree block can be shared by multiple trees | |
953 | */ | |
954 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
955 | struct extent_buffer *buf) | |
956 | { | |
957 | /* | |
01327610 | 958 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
959 | * are never shared. If a block was allocated after the last |
960 | * snapshot and the block was not allocated by tree relocation, | |
961 | * we know the block is not shared. | |
962 | */ | |
27cdeb70 | 963 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
964 | buf != root->node && buf != root->commit_root && |
965 | (btrfs_header_generation(buf) <= | |
966 | btrfs_root_last_snapshot(&root->root_item) || | |
967 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
968 | return 1; | |
969 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 970 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
971 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
972 | return 1; | |
973 | #endif | |
974 | return 0; | |
975 | } | |
976 | ||
977 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
978 | struct btrfs_root *root, | |
979 | struct extent_buffer *buf, | |
f0486c68 YZ |
980 | struct extent_buffer *cow, |
981 | int *last_ref) | |
5d4f98a2 YZ |
982 | { |
983 | u64 refs; | |
984 | u64 owner; | |
985 | u64 flags; | |
986 | u64 new_flags = 0; | |
987 | int ret; | |
988 | ||
989 | /* | |
990 | * Backrefs update rules: | |
991 | * | |
992 | * Always use full backrefs for extent pointers in tree block | |
993 | * allocated by tree relocation. | |
994 | * | |
995 | * If a shared tree block is no longer referenced by its owner | |
996 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
997 | * use full backrefs for extent pointers in tree block. | |
998 | * | |
999 | * If a tree block is been relocating | |
1000 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
1001 | * use full backrefs for extent pointers in tree block. | |
1002 | * The reason for this is some operations (such as drop tree) | |
1003 | * are only allowed for blocks use full backrefs. | |
1004 | */ | |
1005 | ||
1006 | if (btrfs_block_can_be_shared(root, buf)) { | |
1007 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
1008 | btrfs_header_level(buf), 1, |
1009 | &refs, &flags); | |
be1a5564 MF |
1010 | if (ret) |
1011 | return ret; | |
e5df9573 MF |
1012 | if (refs == 0) { |
1013 | ret = -EROFS; | |
34d97007 | 1014 | btrfs_handle_fs_error(root->fs_info, ret, NULL); |
e5df9573 MF |
1015 | return ret; |
1016 | } | |
5d4f98a2 YZ |
1017 | } else { |
1018 | refs = 1; | |
1019 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
1020 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1021 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1022 | else | |
1023 | flags = 0; | |
1024 | } | |
1025 | ||
1026 | owner = btrfs_header_owner(buf); | |
1027 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
1028 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
1029 | ||
1030 | if (refs > 1) { | |
1031 | if ((owner == root->root_key.objectid || | |
1032 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
1033 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 1034 | ret = btrfs_inc_ref(trans, root, buf, 1); |
79787eaa | 1035 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1036 | |
1037 | if (root->root_key.objectid == | |
1038 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 1039 | ret = btrfs_dec_ref(trans, root, buf, 0); |
79787eaa | 1040 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 1041 | ret = btrfs_inc_ref(trans, root, cow, 1); |
79787eaa | 1042 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1043 | } |
1044 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1045 | } else { | |
1046 | ||
1047 | if (root->root_key.objectid == | |
1048 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1049 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1050 | else |
e339a6b0 | 1051 | ret = btrfs_inc_ref(trans, root, cow, 0); |
79787eaa | 1052 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1053 | } |
1054 | if (new_flags != 0) { | |
b1c79e09 JB |
1055 | int level = btrfs_header_level(buf); |
1056 | ||
5d4f98a2 YZ |
1057 | ret = btrfs_set_disk_extent_flags(trans, root, |
1058 | buf->start, | |
1059 | buf->len, | |
b1c79e09 | 1060 | new_flags, level, 0); |
be1a5564 MF |
1061 | if (ret) |
1062 | return ret; | |
5d4f98a2 YZ |
1063 | } |
1064 | } else { | |
1065 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
1066 | if (root->root_key.objectid == | |
1067 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1068 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1069 | else |
e339a6b0 | 1070 | ret = btrfs_inc_ref(trans, root, cow, 0); |
79787eaa | 1071 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 1072 | ret = btrfs_dec_ref(trans, root, buf, 1); |
79787eaa | 1073 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 | 1074 | } |
01d58472 | 1075 | clean_tree_block(trans, root->fs_info, buf); |
f0486c68 | 1076 | *last_ref = 1; |
5d4f98a2 YZ |
1077 | } |
1078 | return 0; | |
1079 | } | |
1080 | ||
d352ac68 | 1081 | /* |
d397712b CM |
1082 | * does the dirty work in cow of a single block. The parent block (if |
1083 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1084 | * dirty and returned locked. If you modify the block it needs to be marked | |
1085 | * dirty again. | |
d352ac68 CM |
1086 | * |
1087 | * search_start -- an allocation hint for the new block | |
1088 | * | |
d397712b CM |
1089 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1090 | * bytes the allocator should try to find free next to the block it returns. | |
1091 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1092 | */ |
d397712b | 1093 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1094 | struct btrfs_root *root, |
1095 | struct extent_buffer *buf, | |
1096 | struct extent_buffer *parent, int parent_slot, | |
1097 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1098 | u64 search_start, u64 empty_size) |
02217ed2 | 1099 | { |
5d4f98a2 | 1100 | struct btrfs_disk_key disk_key; |
5f39d397 | 1101 | struct extent_buffer *cow; |
be1a5564 | 1102 | int level, ret; |
f0486c68 | 1103 | int last_ref = 0; |
925baedd | 1104 | int unlock_orig = 0; |
5d4f98a2 | 1105 | u64 parent_start; |
7bb86316 | 1106 | |
925baedd CM |
1107 | if (*cow_ret == buf) |
1108 | unlock_orig = 1; | |
1109 | ||
b9447ef8 | 1110 | btrfs_assert_tree_locked(buf); |
925baedd | 1111 | |
27cdeb70 MX |
1112 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1113 | trans->transid != root->fs_info->running_transaction->transid); | |
1114 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && | |
1115 | trans->transid != root->last_trans); | |
5f39d397 | 1116 | |
7bb86316 | 1117 | level = btrfs_header_level(buf); |
31840ae1 | 1118 | |
5d4f98a2 YZ |
1119 | if (level == 0) |
1120 | btrfs_item_key(buf, &disk_key, 0); | |
1121 | else | |
1122 | btrfs_node_key(buf, &disk_key, 0); | |
1123 | ||
1124 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
1125 | if (parent) | |
1126 | parent_start = parent->start; | |
1127 | else | |
1128 | parent_start = 0; | |
1129 | } else | |
1130 | parent_start = 0; | |
1131 | ||
4d75f8a9 DS |
1132 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1133 | root->root_key.objectid, &disk_key, level, | |
1134 | search_start, empty_size); | |
54aa1f4d CM |
1135 | if (IS_ERR(cow)) |
1136 | return PTR_ERR(cow); | |
6702ed49 | 1137 | |
b4ce94de CM |
1138 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1139 | ||
5f39d397 | 1140 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 1141 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1142 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1143 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1144 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1145 | BTRFS_HEADER_FLAG_RELOC); | |
1146 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1147 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1148 | else | |
1149 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1150 | |
0a4e5586 | 1151 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(), |
2b82032c YZ |
1152 | BTRFS_FSID_SIZE); |
1153 | ||
be1a5564 | 1154 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1155 | if (ret) { |
79787eaa | 1156 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1157 | return ret; |
1158 | } | |
1a40e23b | 1159 | |
27cdeb70 | 1160 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1161 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b Z |
1162 | if (ret) { |
1163 | btrfs_abort_transaction(trans, root, ret); | |
83d4cfd4 | 1164 | return ret; |
93314e3b | 1165 | } |
83d4cfd4 | 1166 | } |
3fd0a558 | 1167 | |
02217ed2 | 1168 | if (buf == root->node) { |
925baedd | 1169 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1170 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1171 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1172 | parent_start = buf->start; | |
1173 | else | |
1174 | parent_start = 0; | |
925baedd | 1175 | |
5f39d397 | 1176 | extent_buffer_get(cow); |
90f8d62e | 1177 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1178 | rcu_assign_pointer(root->node, cow); |
925baedd | 1179 | |
f0486c68 | 1180 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1181 | last_ref); |
5f39d397 | 1182 | free_extent_buffer(buf); |
0b86a832 | 1183 | add_root_to_dirty_list(root); |
02217ed2 | 1184 | } else { |
5d4f98a2 YZ |
1185 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1186 | parent_start = parent->start; | |
1187 | else | |
1188 | parent_start = 0; | |
1189 | ||
1190 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e | 1191 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
c8cc6341 | 1192 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1193 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1194 | cow->start); |
74493f7a CM |
1195 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1196 | trans->transid); | |
d6025579 | 1197 | btrfs_mark_buffer_dirty(parent); |
5de865ee FDBM |
1198 | if (last_ref) { |
1199 | ret = tree_mod_log_free_eb(root->fs_info, buf); | |
1200 | if (ret) { | |
1201 | btrfs_abort_transaction(trans, root, ret); | |
1202 | return ret; | |
1203 | } | |
1204 | } | |
f0486c68 | 1205 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1206 | last_ref); |
02217ed2 | 1207 | } |
925baedd CM |
1208 | if (unlock_orig) |
1209 | btrfs_tree_unlock(buf); | |
3083ee2e | 1210 | free_extent_buffer_stale(buf); |
ccd467d6 | 1211 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1212 | *cow_ret = cow; |
02217ed2 CM |
1213 | return 0; |
1214 | } | |
1215 | ||
5d9e75c4 JS |
1216 | /* |
1217 | * returns the logical address of the oldest predecessor of the given root. | |
1218 | * entries older than time_seq are ignored. | |
1219 | */ | |
1220 | static struct tree_mod_elem * | |
1221 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1222 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1223 | { |
1224 | struct tree_mod_elem *tm; | |
1225 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1226 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1227 | int looped = 0; |
1228 | ||
1229 | if (!time_seq) | |
35a3621b | 1230 | return NULL; |
5d9e75c4 JS |
1231 | |
1232 | /* | |
298cfd36 CR |
1233 | * the very last operation that's logged for a root is the |
1234 | * replacement operation (if it is replaced at all). this has | |
1235 | * the logical address of the *new* root, making it the very | |
1236 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1237 | */ |
1238 | while (1) { | |
1239 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1240 | time_seq); | |
1241 | if (!looped && !tm) | |
35a3621b | 1242 | return NULL; |
5d9e75c4 | 1243 | /* |
28da9fb4 JS |
1244 | * if there are no tree operation for the oldest root, we simply |
1245 | * return it. this should only happen if that (old) root is at | |
1246 | * level 0. | |
5d9e75c4 | 1247 | */ |
28da9fb4 JS |
1248 | if (!tm) |
1249 | break; | |
5d9e75c4 | 1250 | |
28da9fb4 JS |
1251 | /* |
1252 | * if there's an operation that's not a root replacement, we | |
1253 | * found the oldest version of our root. normally, we'll find a | |
1254 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1255 | */ | |
5d9e75c4 JS |
1256 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1257 | break; | |
1258 | ||
1259 | found = tm; | |
1260 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1261 | looped = 1; |
1262 | } | |
1263 | ||
a95236d9 JS |
1264 | /* if there's no old root to return, return what we found instead */ |
1265 | if (!found) | |
1266 | found = tm; | |
1267 | ||
5d9e75c4 JS |
1268 | return found; |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1273 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1274 | * time_seq). |
1275 | */ | |
1276 | static void | |
f1ca7e98 JB |
1277 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1278 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1279 | { |
1280 | u32 n; | |
1281 | struct rb_node *next; | |
1282 | struct tree_mod_elem *tm = first_tm; | |
1283 | unsigned long o_dst; | |
1284 | unsigned long o_src; | |
1285 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1286 | ||
1287 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1288 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1289 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1290 | /* |
1291 | * all the operations are recorded with the operator used for | |
1292 | * the modification. as we're going backwards, we do the | |
1293 | * opposite of each operation here. | |
1294 | */ | |
1295 | switch (tm->op) { | |
1296 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1297 | BUG_ON(tm->slot < n); | |
1c697d4a | 1298 | /* Fallthrough */ |
95c80bb1 | 1299 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1300 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1301 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1302 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1303 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1304 | tm->generation); | |
4c3e6969 | 1305 | n++; |
5d9e75c4 JS |
1306 | break; |
1307 | case MOD_LOG_KEY_REPLACE: | |
1308 | BUG_ON(tm->slot >= n); | |
1309 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1310 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1311 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1312 | tm->generation); | |
1313 | break; | |
1314 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1315 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1316 | n--; |
1317 | break; | |
1318 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1319 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1320 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1321 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1322 | tm->move.nr_items * p_size); |
1323 | break; | |
1324 | case MOD_LOG_ROOT_REPLACE: | |
1325 | /* | |
1326 | * this operation is special. for roots, this must be | |
1327 | * handled explicitly before rewinding. | |
1328 | * for non-roots, this operation may exist if the node | |
1329 | * was a root: root A -> child B; then A gets empty and | |
1330 | * B is promoted to the new root. in the mod log, we'll | |
1331 | * have a root-replace operation for B, a tree block | |
1332 | * that is no root. we simply ignore that operation. | |
1333 | */ | |
1334 | break; | |
1335 | } | |
1336 | next = rb_next(&tm->node); | |
1337 | if (!next) | |
1338 | break; | |
1339 | tm = container_of(next, struct tree_mod_elem, node); | |
298cfd36 | 1340 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1341 | break; |
1342 | } | |
f1ca7e98 | 1343 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1344 | btrfs_set_header_nritems(eb, n); |
1345 | } | |
1346 | ||
47fb091f | 1347 | /* |
01327610 | 1348 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1349 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1350 | * returned buffer is always read-locked. If the returned buffer is not the | |
1351 | * input buffer, the lock on the input buffer is released and the input buffer | |
1352 | * is freed (its refcount is decremented). | |
1353 | */ | |
5d9e75c4 | 1354 | static struct extent_buffer * |
9ec72677 JB |
1355 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1356 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1357 | { |
1358 | struct extent_buffer *eb_rewin; | |
1359 | struct tree_mod_elem *tm; | |
1360 | ||
1361 | if (!time_seq) | |
1362 | return eb; | |
1363 | ||
1364 | if (btrfs_header_level(eb) == 0) | |
1365 | return eb; | |
1366 | ||
1367 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1368 | if (!tm) | |
1369 | return eb; | |
1370 | ||
9ec72677 JB |
1371 | btrfs_set_path_blocking(path); |
1372 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1373 | ||
5d9e75c4 JS |
1374 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1375 | BUG_ON(tm->slot != 0); | |
b9ef22de FX |
1376 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start, |
1377 | eb->len); | |
db7f3436 | 1378 | if (!eb_rewin) { |
9ec72677 | 1379 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1380 | free_extent_buffer(eb); |
1381 | return NULL; | |
1382 | } | |
5d9e75c4 JS |
1383 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1384 | btrfs_set_header_backref_rev(eb_rewin, | |
1385 | btrfs_header_backref_rev(eb)); | |
1386 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1387 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1388 | } else { |
1389 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1390 | if (!eb_rewin) { |
9ec72677 | 1391 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1392 | free_extent_buffer(eb); |
1393 | return NULL; | |
1394 | } | |
5d9e75c4 JS |
1395 | } |
1396 | ||
9ec72677 JB |
1397 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1398 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1399 | free_extent_buffer(eb); |
1400 | ||
47fb091f JS |
1401 | extent_buffer_get(eb_rewin); |
1402 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1403 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1404 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1405 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1406 | |
1407 | return eb_rewin; | |
1408 | } | |
1409 | ||
8ba97a15 JS |
1410 | /* |
1411 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1412 | * value. If there are no changes, the current root->root_node is returned. If | |
1413 | * anything changed in between, there's a fresh buffer allocated on which the | |
1414 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1415 | * Returns NULL on error (with no locks held). | |
1416 | */ | |
5d9e75c4 JS |
1417 | static inline struct extent_buffer * |
1418 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1419 | { | |
1420 | struct tree_mod_elem *tm; | |
30b0463a JS |
1421 | struct extent_buffer *eb = NULL; |
1422 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1423 | struct extent_buffer *old; |
a95236d9 | 1424 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1425 | u64 old_generation = 0; |
a95236d9 | 1426 | u64 logical; |
5d9e75c4 | 1427 | |
30b0463a JS |
1428 | eb_root = btrfs_read_lock_root_node(root); |
1429 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); | |
5d9e75c4 | 1430 | if (!tm) |
30b0463a | 1431 | return eb_root; |
5d9e75c4 | 1432 | |
a95236d9 JS |
1433 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1434 | old_root = &tm->old_root; | |
1435 | old_generation = tm->generation; | |
1436 | logical = old_root->logical; | |
1437 | } else { | |
30b0463a | 1438 | logical = eb_root->start; |
a95236d9 | 1439 | } |
5d9e75c4 | 1440 | |
a95236d9 | 1441 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 | 1442 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1443 | btrfs_tree_read_unlock(eb_root); |
1444 | free_extent_buffer(eb_root); | |
ce86cd59 | 1445 | old = read_tree_block(root, logical, 0); |
64c043de LB |
1446 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1447 | if (!IS_ERR(old)) | |
1448 | free_extent_buffer(old); | |
efe120a0 FH |
1449 | btrfs_warn(root->fs_info, |
1450 | "failed to read tree block %llu from get_old_root", logical); | |
834328a8 | 1451 | } else { |
7bfdcf7f LB |
1452 | eb = btrfs_clone_extent_buffer(old); |
1453 | free_extent_buffer(old); | |
834328a8 JS |
1454 | } |
1455 | } else if (old_root) { | |
30b0463a JS |
1456 | btrfs_tree_read_unlock(eb_root); |
1457 | free_extent_buffer(eb_root); | |
b9ef22de FX |
1458 | eb = alloc_dummy_extent_buffer(root->fs_info, logical, |
1459 | root->nodesize); | |
834328a8 | 1460 | } else { |
9ec72677 | 1461 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1462 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1463 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1464 | free_extent_buffer(eb_root); |
834328a8 JS |
1465 | } |
1466 | ||
8ba97a15 JS |
1467 | if (!eb) |
1468 | return NULL; | |
d6381084 | 1469 | extent_buffer_get(eb); |
8ba97a15 | 1470 | btrfs_tree_read_lock(eb); |
a95236d9 | 1471 | if (old_root) { |
5d9e75c4 JS |
1472 | btrfs_set_header_bytenr(eb, eb->start); |
1473 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1474 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1475 | btrfs_set_header_level(eb, old_root->level); |
1476 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1477 | } |
28da9fb4 | 1478 | if (tm) |
f1ca7e98 | 1479 | __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1480 | else |
1481 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1482 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1483 | |
1484 | return eb; | |
1485 | } | |
1486 | ||
5b6602e7 JS |
1487 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1488 | { | |
1489 | struct tree_mod_elem *tm; | |
1490 | int level; | |
30b0463a | 1491 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1492 | |
30b0463a | 1493 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1494 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1495 | level = tm->old_root.level; | |
1496 | } else { | |
30b0463a | 1497 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1498 | } |
30b0463a | 1499 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1500 | |
1501 | return level; | |
1502 | } | |
1503 | ||
5d4f98a2 YZ |
1504 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1505 | struct btrfs_root *root, | |
1506 | struct extent_buffer *buf) | |
1507 | { | |
fccb84c9 | 1508 | if (btrfs_test_is_dummy_root(root)) |
faa2dbf0 | 1509 | return 0; |
fccb84c9 | 1510 | |
f1ebcc74 LB |
1511 | /* ensure we can see the force_cow */ |
1512 | smp_rmb(); | |
1513 | ||
1514 | /* | |
1515 | * We do not need to cow a block if | |
1516 | * 1) this block is not created or changed in this transaction; | |
1517 | * 2) this block does not belong to TREE_RELOC tree; | |
1518 | * 3) the root is not forced COW. | |
1519 | * | |
1520 | * What is forced COW: | |
01327610 | 1521 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1522 | * after we've finished coping src root, we must COW the shared |
1523 | * block to ensure the metadata consistency. | |
1524 | */ | |
5d4f98a2 YZ |
1525 | if (btrfs_header_generation(buf) == trans->transid && |
1526 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1527 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1528 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1529 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1530 | return 0; |
1531 | return 1; | |
1532 | } | |
1533 | ||
d352ac68 CM |
1534 | /* |
1535 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1536 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1537 | * once per transaction, as long as it hasn't been written yet |
1538 | */ | |
d397712b | 1539 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1540 | struct btrfs_root *root, struct extent_buffer *buf, |
1541 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1542 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1543 | { |
1544 | u64 search_start; | |
f510cfec | 1545 | int ret; |
dc17ff8f | 1546 | |
31b1a2bd JL |
1547 | if (trans->transaction != root->fs_info->running_transaction) |
1548 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1549 | trans->transid, |
6702ed49 | 1550 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1551 | |
1552 | if (trans->transid != root->fs_info->generation) | |
1553 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1554 | trans->transid, root->fs_info->generation); |
dc17ff8f | 1555 | |
5d4f98a2 | 1556 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1557 | trans->dirty = true; |
6702ed49 CM |
1558 | *cow_ret = buf; |
1559 | return 0; | |
1560 | } | |
c487685d | 1561 | |
ee22184b | 1562 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1563 | |
1564 | if (parent) | |
1565 | btrfs_set_lock_blocking(parent); | |
1566 | btrfs_set_lock_blocking(buf); | |
1567 | ||
f510cfec | 1568 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1569 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1570 | |
1571 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1572 | ||
f510cfec | 1573 | return ret; |
6702ed49 CM |
1574 | } |
1575 | ||
d352ac68 CM |
1576 | /* |
1577 | * helper function for defrag to decide if two blocks pointed to by a | |
1578 | * node are actually close by | |
1579 | */ | |
6b80053d | 1580 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1581 | { |
6b80053d | 1582 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1583 | return 1; |
6b80053d | 1584 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1585 | return 1; |
1586 | return 0; | |
1587 | } | |
1588 | ||
081e9573 CM |
1589 | /* |
1590 | * compare two keys in a memcmp fashion | |
1591 | */ | |
1592 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1593 | { | |
1594 | struct btrfs_key k1; | |
1595 | ||
1596 | btrfs_disk_key_to_cpu(&k1, disk); | |
1597 | ||
20736aba | 1598 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1599 | } |
1600 | ||
f3465ca4 JB |
1601 | /* |
1602 | * same as comp_keys only with two btrfs_key's | |
1603 | */ | |
5d4f98a2 | 1604 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1605 | { |
1606 | if (k1->objectid > k2->objectid) | |
1607 | return 1; | |
1608 | if (k1->objectid < k2->objectid) | |
1609 | return -1; | |
1610 | if (k1->type > k2->type) | |
1611 | return 1; | |
1612 | if (k1->type < k2->type) | |
1613 | return -1; | |
1614 | if (k1->offset > k2->offset) | |
1615 | return 1; | |
1616 | if (k1->offset < k2->offset) | |
1617 | return -1; | |
1618 | return 0; | |
1619 | } | |
081e9573 | 1620 | |
d352ac68 CM |
1621 | /* |
1622 | * this is used by the defrag code to go through all the | |
1623 | * leaves pointed to by a node and reallocate them so that | |
1624 | * disk order is close to key order | |
1625 | */ | |
6702ed49 | 1626 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1627 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1628 | int start_slot, u64 *last_ret, |
a6b6e75e | 1629 | struct btrfs_key *progress) |
6702ed49 | 1630 | { |
6b80053d | 1631 | struct extent_buffer *cur; |
6702ed49 | 1632 | u64 blocknr; |
ca7a79ad | 1633 | u64 gen; |
e9d0b13b CM |
1634 | u64 search_start = *last_ret; |
1635 | u64 last_block = 0; | |
6702ed49 CM |
1636 | u64 other; |
1637 | u32 parent_nritems; | |
6702ed49 CM |
1638 | int end_slot; |
1639 | int i; | |
1640 | int err = 0; | |
f2183bde | 1641 | int parent_level; |
6b80053d CM |
1642 | int uptodate; |
1643 | u32 blocksize; | |
081e9573 CM |
1644 | int progress_passed = 0; |
1645 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1646 | |
5708b959 | 1647 | parent_level = btrfs_header_level(parent); |
5708b959 | 1648 | |
6c1500f2 JL |
1649 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1650 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1651 | |
6b80053d | 1652 | parent_nritems = btrfs_header_nritems(parent); |
707e8a07 | 1653 | blocksize = root->nodesize; |
5dfe2be7 | 1654 | end_slot = parent_nritems - 1; |
6702ed49 | 1655 | |
5dfe2be7 | 1656 | if (parent_nritems <= 1) |
6702ed49 CM |
1657 | return 0; |
1658 | ||
b4ce94de CM |
1659 | btrfs_set_lock_blocking(parent); |
1660 | ||
5dfe2be7 | 1661 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 1662 | int close = 1; |
a6b6e75e | 1663 | |
081e9573 CM |
1664 | btrfs_node_key(parent, &disk_key, i); |
1665 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1666 | continue; | |
1667 | ||
1668 | progress_passed = 1; | |
6b80053d | 1669 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1670 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1671 | if (last_block == 0) |
1672 | last_block = blocknr; | |
5708b959 | 1673 | |
6702ed49 | 1674 | if (i > 0) { |
6b80053d CM |
1675 | other = btrfs_node_blockptr(parent, i - 1); |
1676 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1677 | } |
5dfe2be7 | 1678 | if (!close && i < end_slot) { |
6b80053d CM |
1679 | other = btrfs_node_blockptr(parent, i + 1); |
1680 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1681 | } |
e9d0b13b CM |
1682 | if (close) { |
1683 | last_block = blocknr; | |
6702ed49 | 1684 | continue; |
e9d0b13b | 1685 | } |
6702ed49 | 1686 | |
01d58472 | 1687 | cur = btrfs_find_tree_block(root->fs_info, blocknr); |
6b80053d | 1688 | if (cur) |
b9fab919 | 1689 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1690 | else |
1691 | uptodate = 0; | |
5708b959 | 1692 | if (!cur || !uptodate) { |
6b80053d | 1693 | if (!cur) { |
ce86cd59 | 1694 | cur = read_tree_block(root, blocknr, gen); |
64c043de LB |
1695 | if (IS_ERR(cur)) { |
1696 | return PTR_ERR(cur); | |
1697 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1698 | free_extent_buffer(cur); |
97d9a8a4 | 1699 | return -EIO; |
416bc658 | 1700 | } |
6b80053d | 1701 | } else if (!uptodate) { |
018642a1 TI |
1702 | err = btrfs_read_buffer(cur, gen); |
1703 | if (err) { | |
1704 | free_extent_buffer(cur); | |
1705 | return err; | |
1706 | } | |
f2183bde | 1707 | } |
6702ed49 | 1708 | } |
e9d0b13b | 1709 | if (search_start == 0) |
6b80053d | 1710 | search_start = last_block; |
e9d0b13b | 1711 | |
e7a84565 | 1712 | btrfs_tree_lock(cur); |
b4ce94de | 1713 | btrfs_set_lock_blocking(cur); |
6b80053d | 1714 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1715 | &cur, search_start, |
6b80053d | 1716 | min(16 * blocksize, |
9fa8cfe7 | 1717 | (end_slot - i) * blocksize)); |
252c38f0 | 1718 | if (err) { |
e7a84565 | 1719 | btrfs_tree_unlock(cur); |
6b80053d | 1720 | free_extent_buffer(cur); |
6702ed49 | 1721 | break; |
252c38f0 | 1722 | } |
e7a84565 CM |
1723 | search_start = cur->start; |
1724 | last_block = cur->start; | |
f2183bde | 1725 | *last_ret = search_start; |
e7a84565 CM |
1726 | btrfs_tree_unlock(cur); |
1727 | free_extent_buffer(cur); | |
6702ed49 CM |
1728 | } |
1729 | return err; | |
1730 | } | |
1731 | ||
74123bd7 CM |
1732 | /* |
1733 | * The leaf data grows from end-to-front in the node. | |
1734 | * this returns the address of the start of the last item, | |
1735 | * which is the stop of the leaf data stack | |
1736 | */ | |
123abc88 | 1737 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1738 | struct extent_buffer *leaf) |
be0e5c09 | 1739 | { |
5f39d397 | 1740 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1741 | if (nr == 0) |
123abc88 | 1742 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1743 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1744 | } |
1745 | ||
aa5d6bed | 1746 | |
74123bd7 | 1747 | /* |
5f39d397 CM |
1748 | * search for key in the extent_buffer. The items start at offset p, |
1749 | * and they are item_size apart. There are 'max' items in p. | |
1750 | * | |
74123bd7 CM |
1751 | * the slot in the array is returned via slot, and it points to |
1752 | * the place where you would insert key if it is not found in | |
1753 | * the array. | |
1754 | * | |
1755 | * slot may point to max if the key is bigger than all of the keys | |
1756 | */ | |
e02119d5 CM |
1757 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1758 | unsigned long p, | |
1759 | int item_size, struct btrfs_key *key, | |
1760 | int max, int *slot) | |
be0e5c09 CM |
1761 | { |
1762 | int low = 0; | |
1763 | int high = max; | |
1764 | int mid; | |
1765 | int ret; | |
479965d6 | 1766 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1767 | struct btrfs_disk_key unaligned; |
1768 | unsigned long offset; | |
5f39d397 CM |
1769 | char *kaddr = NULL; |
1770 | unsigned long map_start = 0; | |
1771 | unsigned long map_len = 0; | |
479965d6 | 1772 | int err; |
be0e5c09 | 1773 | |
d397712b | 1774 | while (low < high) { |
be0e5c09 | 1775 | mid = (low + high) / 2; |
5f39d397 CM |
1776 | offset = p + mid * item_size; |
1777 | ||
a6591715 | 1778 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1779 | (offset + sizeof(struct btrfs_disk_key)) > |
1780 | map_start + map_len) { | |
934d375b CM |
1781 | |
1782 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1783 | sizeof(struct btrfs_disk_key), |
a6591715 | 1784 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1785 | |
1786 | if (!err) { | |
1787 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1788 | map_start); | |
415b35a5 | 1789 | } else if (err == 1) { |
479965d6 CM |
1790 | read_extent_buffer(eb, &unaligned, |
1791 | offset, sizeof(unaligned)); | |
1792 | tmp = &unaligned; | |
415b35a5 LB |
1793 | } else { |
1794 | return err; | |
479965d6 | 1795 | } |
5f39d397 | 1796 | |
5f39d397 CM |
1797 | } else { |
1798 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1799 | map_start); | |
1800 | } | |
be0e5c09 CM |
1801 | ret = comp_keys(tmp, key); |
1802 | ||
1803 | if (ret < 0) | |
1804 | low = mid + 1; | |
1805 | else if (ret > 0) | |
1806 | high = mid; | |
1807 | else { | |
1808 | *slot = mid; | |
1809 | return 0; | |
1810 | } | |
1811 | } | |
1812 | *slot = low; | |
1813 | return 1; | |
1814 | } | |
1815 | ||
97571fd0 CM |
1816 | /* |
1817 | * simple bin_search frontend that does the right thing for | |
1818 | * leaves vs nodes | |
1819 | */ | |
5f39d397 CM |
1820 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1821 | int level, int *slot) | |
be0e5c09 | 1822 | { |
f775738f | 1823 | if (level == 0) |
5f39d397 CM |
1824 | return generic_bin_search(eb, |
1825 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1826 | sizeof(struct btrfs_item), |
5f39d397 | 1827 | key, btrfs_header_nritems(eb), |
7518a238 | 1828 | slot); |
f775738f | 1829 | else |
5f39d397 CM |
1830 | return generic_bin_search(eb, |
1831 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1832 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1833 | key, btrfs_header_nritems(eb), |
7518a238 | 1834 | slot); |
be0e5c09 CM |
1835 | } |
1836 | ||
5d4f98a2 YZ |
1837 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1838 | int level, int *slot) | |
1839 | { | |
1840 | return bin_search(eb, key, level, slot); | |
1841 | } | |
1842 | ||
f0486c68 YZ |
1843 | static void root_add_used(struct btrfs_root *root, u32 size) |
1844 | { | |
1845 | spin_lock(&root->accounting_lock); | |
1846 | btrfs_set_root_used(&root->root_item, | |
1847 | btrfs_root_used(&root->root_item) + size); | |
1848 | spin_unlock(&root->accounting_lock); | |
1849 | } | |
1850 | ||
1851 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1852 | { | |
1853 | spin_lock(&root->accounting_lock); | |
1854 | btrfs_set_root_used(&root->root_item, | |
1855 | btrfs_root_used(&root->root_item) - size); | |
1856 | spin_unlock(&root->accounting_lock); | |
1857 | } | |
1858 | ||
d352ac68 CM |
1859 | /* given a node and slot number, this reads the blocks it points to. The |
1860 | * extent buffer is returned with a reference taken (but unlocked). | |
1861 | * NULL is returned on error. | |
1862 | */ | |
e02119d5 | 1863 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1864 | struct extent_buffer *parent, int slot) |
bb803951 | 1865 | { |
ca7a79ad | 1866 | int level = btrfs_header_level(parent); |
416bc658 JB |
1867 | struct extent_buffer *eb; |
1868 | ||
bb803951 CM |
1869 | if (slot < 0) |
1870 | return NULL; | |
5f39d397 | 1871 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1872 | return NULL; |
ca7a79ad CM |
1873 | |
1874 | BUG_ON(level == 0); | |
1875 | ||
416bc658 | 1876 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
416bc658 | 1877 | btrfs_node_ptr_generation(parent, slot)); |
64c043de LB |
1878 | if (IS_ERR(eb) || !extent_buffer_uptodate(eb)) { |
1879 | if (!IS_ERR(eb)) | |
1880 | free_extent_buffer(eb); | |
416bc658 JB |
1881 | eb = NULL; |
1882 | } | |
1883 | ||
1884 | return eb; | |
bb803951 CM |
1885 | } |
1886 | ||
d352ac68 CM |
1887 | /* |
1888 | * node level balancing, used to make sure nodes are in proper order for | |
1889 | * item deletion. We balance from the top down, so we have to make sure | |
1890 | * that a deletion won't leave an node completely empty later on. | |
1891 | */ | |
e02119d5 | 1892 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1893 | struct btrfs_root *root, |
1894 | struct btrfs_path *path, int level) | |
bb803951 | 1895 | { |
5f39d397 CM |
1896 | struct extent_buffer *right = NULL; |
1897 | struct extent_buffer *mid; | |
1898 | struct extent_buffer *left = NULL; | |
1899 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1900 | int ret = 0; |
1901 | int wret; | |
1902 | int pslot; | |
bb803951 | 1903 | int orig_slot = path->slots[level]; |
79f95c82 | 1904 | u64 orig_ptr; |
bb803951 CM |
1905 | |
1906 | if (level == 0) | |
1907 | return 0; | |
1908 | ||
5f39d397 | 1909 | mid = path->nodes[level]; |
b4ce94de | 1910 | |
bd681513 CM |
1911 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1912 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1913 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1914 | ||
1d4f8a0c | 1915 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1916 | |
a05a9bb1 | 1917 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1918 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1919 | pslot = path->slots[level + 1]; |
1920 | } | |
bb803951 | 1921 | |
40689478 CM |
1922 | /* |
1923 | * deal with the case where there is only one pointer in the root | |
1924 | * by promoting the node below to a root | |
1925 | */ | |
5f39d397 CM |
1926 | if (!parent) { |
1927 | struct extent_buffer *child; | |
bb803951 | 1928 | |
5f39d397 | 1929 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1930 | return 0; |
1931 | ||
1932 | /* promote the child to a root */ | |
5f39d397 | 1933 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1934 | if (!child) { |
1935 | ret = -EROFS; | |
34d97007 | 1936 | btrfs_handle_fs_error(root->fs_info, ret, NULL); |
305a26af MF |
1937 | goto enospc; |
1938 | } | |
1939 | ||
925baedd | 1940 | btrfs_tree_lock(child); |
b4ce94de | 1941 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1942 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1943 | if (ret) { |
1944 | btrfs_tree_unlock(child); | |
1945 | free_extent_buffer(child); | |
1946 | goto enospc; | |
1947 | } | |
2f375ab9 | 1948 | |
90f8d62e | 1949 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1950 | rcu_assign_pointer(root->node, child); |
925baedd | 1951 | |
0b86a832 | 1952 | add_root_to_dirty_list(root); |
925baedd | 1953 | btrfs_tree_unlock(child); |
b4ce94de | 1954 | |
925baedd | 1955 | path->locks[level] = 0; |
bb803951 | 1956 | path->nodes[level] = NULL; |
01d58472 | 1957 | clean_tree_block(trans, root->fs_info, mid); |
925baedd | 1958 | btrfs_tree_unlock(mid); |
bb803951 | 1959 | /* once for the path */ |
5f39d397 | 1960 | free_extent_buffer(mid); |
f0486c68 YZ |
1961 | |
1962 | root_sub_used(root, mid->len); | |
5581a51a | 1963 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1964 | /* once for the root ptr */ |
3083ee2e | 1965 | free_extent_buffer_stale(mid); |
f0486c68 | 1966 | return 0; |
bb803951 | 1967 | } |
5f39d397 | 1968 | if (btrfs_header_nritems(mid) > |
123abc88 | 1969 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1970 | return 0; |
1971 | ||
5f39d397 CM |
1972 | left = read_node_slot(root, parent, pslot - 1); |
1973 | if (left) { | |
925baedd | 1974 | btrfs_tree_lock(left); |
b4ce94de | 1975 | btrfs_set_lock_blocking(left); |
5f39d397 | 1976 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1977 | parent, pslot - 1, &left); |
54aa1f4d CM |
1978 | if (wret) { |
1979 | ret = wret; | |
1980 | goto enospc; | |
1981 | } | |
2cc58cf2 | 1982 | } |
5f39d397 CM |
1983 | right = read_node_slot(root, parent, pslot + 1); |
1984 | if (right) { | |
925baedd | 1985 | btrfs_tree_lock(right); |
b4ce94de | 1986 | btrfs_set_lock_blocking(right); |
5f39d397 | 1987 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1988 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1989 | if (wret) { |
1990 | ret = wret; | |
1991 | goto enospc; | |
1992 | } | |
1993 | } | |
1994 | ||
1995 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1996 | if (left) { |
1997 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1998 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1999 | if (wret < 0) |
2000 | ret = wret; | |
bb803951 | 2001 | } |
79f95c82 CM |
2002 | |
2003 | /* | |
2004 | * then try to empty the right most buffer into the middle | |
2005 | */ | |
5f39d397 | 2006 | if (right) { |
971a1f66 | 2007 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 2008 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 2009 | ret = wret; |
5f39d397 | 2010 | if (btrfs_header_nritems(right) == 0) { |
01d58472 | 2011 | clean_tree_block(trans, root->fs_info, right); |
925baedd | 2012 | btrfs_tree_unlock(right); |
afe5fea7 | 2013 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 2014 | root_sub_used(root, right->len); |
5581a51a | 2015 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 2016 | free_extent_buffer_stale(right); |
f0486c68 | 2017 | right = NULL; |
bb803951 | 2018 | } else { |
5f39d397 CM |
2019 | struct btrfs_disk_key right_key; |
2020 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 2021 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2022 | pslot + 1, 0); |
5f39d397 CM |
2023 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
2024 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
2025 | } |
2026 | } | |
5f39d397 | 2027 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
2028 | /* |
2029 | * we're not allowed to leave a node with one item in the | |
2030 | * tree during a delete. A deletion from lower in the tree | |
2031 | * could try to delete the only pointer in this node. | |
2032 | * So, pull some keys from the left. | |
2033 | * There has to be a left pointer at this point because | |
2034 | * otherwise we would have pulled some pointers from the | |
2035 | * right | |
2036 | */ | |
305a26af MF |
2037 | if (!left) { |
2038 | ret = -EROFS; | |
34d97007 | 2039 | btrfs_handle_fs_error(root->fs_info, ret, NULL); |
305a26af MF |
2040 | goto enospc; |
2041 | } | |
5f39d397 | 2042 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 2043 | if (wret < 0) { |
79f95c82 | 2044 | ret = wret; |
54aa1f4d CM |
2045 | goto enospc; |
2046 | } | |
bce4eae9 CM |
2047 | if (wret == 1) { |
2048 | wret = push_node_left(trans, root, left, mid, 1); | |
2049 | if (wret < 0) | |
2050 | ret = wret; | |
2051 | } | |
79f95c82 CM |
2052 | BUG_ON(wret == 1); |
2053 | } | |
5f39d397 | 2054 | if (btrfs_header_nritems(mid) == 0) { |
01d58472 | 2055 | clean_tree_block(trans, root->fs_info, mid); |
925baedd | 2056 | btrfs_tree_unlock(mid); |
afe5fea7 | 2057 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2058 | root_sub_used(root, mid->len); |
5581a51a | 2059 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2060 | free_extent_buffer_stale(mid); |
f0486c68 | 2061 | mid = NULL; |
79f95c82 CM |
2062 | } else { |
2063 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2064 | struct btrfs_disk_key mid_key; |
2065 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 2066 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 2067 | pslot, 0); |
5f39d397 CM |
2068 | btrfs_set_node_key(parent, &mid_key, pslot); |
2069 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2070 | } |
bb803951 | 2071 | |
79f95c82 | 2072 | /* update the path */ |
5f39d397 CM |
2073 | if (left) { |
2074 | if (btrfs_header_nritems(left) > orig_slot) { | |
2075 | extent_buffer_get(left); | |
925baedd | 2076 | /* left was locked after cow */ |
5f39d397 | 2077 | path->nodes[level] = left; |
bb803951 CM |
2078 | path->slots[level + 1] -= 1; |
2079 | path->slots[level] = orig_slot; | |
925baedd CM |
2080 | if (mid) { |
2081 | btrfs_tree_unlock(mid); | |
5f39d397 | 2082 | free_extent_buffer(mid); |
925baedd | 2083 | } |
bb803951 | 2084 | } else { |
5f39d397 | 2085 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2086 | path->slots[level] = orig_slot; |
2087 | } | |
2088 | } | |
79f95c82 | 2089 | /* double check we haven't messed things up */ |
e20d96d6 | 2090 | if (orig_ptr != |
5f39d397 | 2091 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2092 | BUG(); |
54aa1f4d | 2093 | enospc: |
925baedd CM |
2094 | if (right) { |
2095 | btrfs_tree_unlock(right); | |
5f39d397 | 2096 | free_extent_buffer(right); |
925baedd CM |
2097 | } |
2098 | if (left) { | |
2099 | if (path->nodes[level] != left) | |
2100 | btrfs_tree_unlock(left); | |
5f39d397 | 2101 | free_extent_buffer(left); |
925baedd | 2102 | } |
bb803951 CM |
2103 | return ret; |
2104 | } | |
2105 | ||
d352ac68 CM |
2106 | /* Node balancing for insertion. Here we only split or push nodes around |
2107 | * when they are completely full. This is also done top down, so we | |
2108 | * have to be pessimistic. | |
2109 | */ | |
d397712b | 2110 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2111 | struct btrfs_root *root, |
2112 | struct btrfs_path *path, int level) | |
e66f709b | 2113 | { |
5f39d397 CM |
2114 | struct extent_buffer *right = NULL; |
2115 | struct extent_buffer *mid; | |
2116 | struct extent_buffer *left = NULL; | |
2117 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2118 | int ret = 0; |
2119 | int wret; | |
2120 | int pslot; | |
2121 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2122 | |
2123 | if (level == 0) | |
2124 | return 1; | |
2125 | ||
5f39d397 | 2126 | mid = path->nodes[level]; |
7bb86316 | 2127 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2128 | |
a05a9bb1 | 2129 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2130 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2131 | pslot = path->slots[level + 1]; |
2132 | } | |
e66f709b | 2133 | |
5f39d397 | 2134 | if (!parent) |
e66f709b | 2135 | return 1; |
e66f709b | 2136 | |
5f39d397 | 2137 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
2138 | |
2139 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2140 | if (left) { |
e66f709b | 2141 | u32 left_nr; |
925baedd CM |
2142 | |
2143 | btrfs_tree_lock(left); | |
b4ce94de CM |
2144 | btrfs_set_lock_blocking(left); |
2145 | ||
5f39d397 | 2146 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
2147 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2148 | wret = 1; | |
2149 | } else { | |
5f39d397 | 2150 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2151 | pslot - 1, &left); |
54aa1f4d CM |
2152 | if (ret) |
2153 | wret = 1; | |
2154 | else { | |
54aa1f4d | 2155 | wret = push_node_left(trans, root, |
971a1f66 | 2156 | left, mid, 0); |
54aa1f4d | 2157 | } |
33ade1f8 | 2158 | } |
e66f709b CM |
2159 | if (wret < 0) |
2160 | ret = wret; | |
2161 | if (wret == 0) { | |
5f39d397 | 2162 | struct btrfs_disk_key disk_key; |
e66f709b | 2163 | orig_slot += left_nr; |
5f39d397 | 2164 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 2165 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2166 | pslot, 0); |
5f39d397 CM |
2167 | btrfs_set_node_key(parent, &disk_key, pslot); |
2168 | btrfs_mark_buffer_dirty(parent); | |
2169 | if (btrfs_header_nritems(left) > orig_slot) { | |
2170 | path->nodes[level] = left; | |
e66f709b CM |
2171 | path->slots[level + 1] -= 1; |
2172 | path->slots[level] = orig_slot; | |
925baedd | 2173 | btrfs_tree_unlock(mid); |
5f39d397 | 2174 | free_extent_buffer(mid); |
e66f709b CM |
2175 | } else { |
2176 | orig_slot -= | |
5f39d397 | 2177 | btrfs_header_nritems(left); |
e66f709b | 2178 | path->slots[level] = orig_slot; |
925baedd | 2179 | btrfs_tree_unlock(left); |
5f39d397 | 2180 | free_extent_buffer(left); |
e66f709b | 2181 | } |
e66f709b CM |
2182 | return 0; |
2183 | } | |
925baedd | 2184 | btrfs_tree_unlock(left); |
5f39d397 | 2185 | free_extent_buffer(left); |
e66f709b | 2186 | } |
925baedd | 2187 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
2188 | |
2189 | /* | |
2190 | * then try to empty the right most buffer into the middle | |
2191 | */ | |
5f39d397 | 2192 | if (right) { |
33ade1f8 | 2193 | u32 right_nr; |
b4ce94de | 2194 | |
925baedd | 2195 | btrfs_tree_lock(right); |
b4ce94de CM |
2196 | btrfs_set_lock_blocking(right); |
2197 | ||
5f39d397 | 2198 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2199 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2200 | wret = 1; | |
2201 | } else { | |
5f39d397 CM |
2202 | ret = btrfs_cow_block(trans, root, right, |
2203 | parent, pslot + 1, | |
9fa8cfe7 | 2204 | &right); |
54aa1f4d CM |
2205 | if (ret) |
2206 | wret = 1; | |
2207 | else { | |
54aa1f4d | 2208 | wret = balance_node_right(trans, root, |
5f39d397 | 2209 | right, mid); |
54aa1f4d | 2210 | } |
33ade1f8 | 2211 | } |
e66f709b CM |
2212 | if (wret < 0) |
2213 | ret = wret; | |
2214 | if (wret == 0) { | |
5f39d397 CM |
2215 | struct btrfs_disk_key disk_key; |
2216 | ||
2217 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2218 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2219 | pslot + 1, 0); |
5f39d397 CM |
2220 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2221 | btrfs_mark_buffer_dirty(parent); | |
2222 | ||
2223 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2224 | path->nodes[level] = right; | |
e66f709b CM |
2225 | path->slots[level + 1] += 1; |
2226 | path->slots[level] = orig_slot - | |
5f39d397 | 2227 | btrfs_header_nritems(mid); |
925baedd | 2228 | btrfs_tree_unlock(mid); |
5f39d397 | 2229 | free_extent_buffer(mid); |
e66f709b | 2230 | } else { |
925baedd | 2231 | btrfs_tree_unlock(right); |
5f39d397 | 2232 | free_extent_buffer(right); |
e66f709b | 2233 | } |
e66f709b CM |
2234 | return 0; |
2235 | } | |
925baedd | 2236 | btrfs_tree_unlock(right); |
5f39d397 | 2237 | free_extent_buffer(right); |
e66f709b | 2238 | } |
e66f709b CM |
2239 | return 1; |
2240 | } | |
2241 | ||
3c69faec | 2242 | /* |
d352ac68 CM |
2243 | * readahead one full node of leaves, finding things that are close |
2244 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2245 | */ |
c8c42864 CM |
2246 | static void reada_for_search(struct btrfs_root *root, |
2247 | struct btrfs_path *path, | |
2248 | int level, int slot, u64 objectid) | |
3c69faec | 2249 | { |
5f39d397 | 2250 | struct extent_buffer *node; |
01f46658 | 2251 | struct btrfs_disk_key disk_key; |
3c69faec | 2252 | u32 nritems; |
3c69faec | 2253 | u64 search; |
a7175319 | 2254 | u64 target; |
6b80053d | 2255 | u64 nread = 0; |
cb25c2ea | 2256 | u64 gen; |
5f39d397 | 2257 | struct extent_buffer *eb; |
6b80053d CM |
2258 | u32 nr; |
2259 | u32 blocksize; | |
2260 | u32 nscan = 0; | |
db94535d | 2261 | |
a6b6e75e | 2262 | if (level != 1) |
6702ed49 CM |
2263 | return; |
2264 | ||
2265 | if (!path->nodes[level]) | |
3c69faec CM |
2266 | return; |
2267 | ||
5f39d397 | 2268 | node = path->nodes[level]; |
925baedd | 2269 | |
3c69faec | 2270 | search = btrfs_node_blockptr(node, slot); |
707e8a07 | 2271 | blocksize = root->nodesize; |
01d58472 | 2272 | eb = btrfs_find_tree_block(root->fs_info, search); |
5f39d397 CM |
2273 | if (eb) { |
2274 | free_extent_buffer(eb); | |
3c69faec CM |
2275 | return; |
2276 | } | |
2277 | ||
a7175319 | 2278 | target = search; |
6b80053d | 2279 | |
5f39d397 | 2280 | nritems = btrfs_header_nritems(node); |
6b80053d | 2281 | nr = slot; |
25b8b936 | 2282 | |
d397712b | 2283 | while (1) { |
e4058b54 | 2284 | if (path->reada == READA_BACK) { |
6b80053d CM |
2285 | if (nr == 0) |
2286 | break; | |
2287 | nr--; | |
e4058b54 | 2288 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2289 | nr++; |
2290 | if (nr >= nritems) | |
2291 | break; | |
3c69faec | 2292 | } |
e4058b54 | 2293 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2294 | btrfs_node_key(node, &disk_key, nr); |
2295 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2296 | break; | |
2297 | } | |
6b80053d | 2298 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2299 | if ((search <= target && target - search <= 65536) || |
2300 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2301 | gen = btrfs_node_ptr_generation(node, nr); |
d3e46fea | 2302 | readahead_tree_block(root, search); |
6b80053d CM |
2303 | nread += blocksize; |
2304 | } | |
2305 | nscan++; | |
a7175319 | 2306 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2307 | break; |
3c69faec CM |
2308 | } |
2309 | } | |
925baedd | 2310 | |
0b08851f JB |
2311 | static noinline void reada_for_balance(struct btrfs_root *root, |
2312 | struct btrfs_path *path, int level) | |
b4ce94de CM |
2313 | { |
2314 | int slot; | |
2315 | int nritems; | |
2316 | struct extent_buffer *parent; | |
2317 | struct extent_buffer *eb; | |
2318 | u64 gen; | |
2319 | u64 block1 = 0; | |
2320 | u64 block2 = 0; | |
b4ce94de | 2321 | |
8c594ea8 | 2322 | parent = path->nodes[level + 1]; |
b4ce94de | 2323 | if (!parent) |
0b08851f | 2324 | return; |
b4ce94de CM |
2325 | |
2326 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2327 | slot = path->slots[level + 1]; |
b4ce94de CM |
2328 | |
2329 | if (slot > 0) { | |
2330 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2331 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
01d58472 | 2332 | eb = btrfs_find_tree_block(root->fs_info, block1); |
b9fab919 CM |
2333 | /* |
2334 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2335 | * don't want to return eagain here. That will loop | |
2336 | * forever | |
2337 | */ | |
2338 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2339 | block1 = 0; |
2340 | free_extent_buffer(eb); | |
2341 | } | |
8c594ea8 | 2342 | if (slot + 1 < nritems) { |
b4ce94de CM |
2343 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2344 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
01d58472 | 2345 | eb = btrfs_find_tree_block(root->fs_info, block2); |
b9fab919 | 2346 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2347 | block2 = 0; |
2348 | free_extent_buffer(eb); | |
2349 | } | |
8c594ea8 | 2350 | |
0b08851f | 2351 | if (block1) |
d3e46fea | 2352 | readahead_tree_block(root, block1); |
0b08851f | 2353 | if (block2) |
d3e46fea | 2354 | readahead_tree_block(root, block2); |
b4ce94de CM |
2355 | } |
2356 | ||
2357 | ||
d352ac68 | 2358 | /* |
d397712b CM |
2359 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2360 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2361 | * operations on the tree might require changing key pointers higher up in the | |
2362 | * tree. | |
d352ac68 | 2363 | * |
d397712b CM |
2364 | * callers might also have set path->keep_locks, which tells this code to keep |
2365 | * the lock if the path points to the last slot in the block. This is part of | |
2366 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2367 | * |
d397712b CM |
2368 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2369 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2370 | */ |
e02119d5 | 2371 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2372 | int lowest_unlock, int min_write_lock_level, |
2373 | int *write_lock_level) | |
925baedd CM |
2374 | { |
2375 | int i; | |
2376 | int skip_level = level; | |
051e1b9f | 2377 | int no_skips = 0; |
925baedd CM |
2378 | struct extent_buffer *t; |
2379 | ||
2380 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2381 | if (!path->nodes[i]) | |
2382 | break; | |
2383 | if (!path->locks[i]) | |
2384 | break; | |
051e1b9f | 2385 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2386 | skip_level = i + 1; |
2387 | continue; | |
2388 | } | |
051e1b9f | 2389 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2390 | u32 nritems; |
2391 | t = path->nodes[i]; | |
2392 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2393 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2394 | skip_level = i + 1; |
2395 | continue; | |
2396 | } | |
2397 | } | |
051e1b9f CM |
2398 | if (skip_level < i && i >= lowest_unlock) |
2399 | no_skips = 1; | |
2400 | ||
925baedd CM |
2401 | t = path->nodes[i]; |
2402 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2403 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2404 | path->locks[i] = 0; |
f7c79f30 CM |
2405 | if (write_lock_level && |
2406 | i > min_write_lock_level && | |
2407 | i <= *write_lock_level) { | |
2408 | *write_lock_level = i - 1; | |
2409 | } | |
925baedd CM |
2410 | } |
2411 | } | |
2412 | } | |
2413 | ||
b4ce94de CM |
2414 | /* |
2415 | * This releases any locks held in the path starting at level and | |
2416 | * going all the way up to the root. | |
2417 | * | |
2418 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2419 | * corner cases, such as COW of the block at slot zero in the node. This | |
2420 | * ignores those rules, and it should only be called when there are no | |
2421 | * more updates to be done higher up in the tree. | |
2422 | */ | |
2423 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2424 | { | |
2425 | int i; | |
2426 | ||
09a2a8f9 | 2427 | if (path->keep_locks) |
b4ce94de CM |
2428 | return; |
2429 | ||
2430 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2431 | if (!path->nodes[i]) | |
12f4dacc | 2432 | continue; |
b4ce94de | 2433 | if (!path->locks[i]) |
12f4dacc | 2434 | continue; |
bd681513 | 2435 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2436 | path->locks[i] = 0; |
2437 | } | |
2438 | } | |
2439 | ||
c8c42864 CM |
2440 | /* |
2441 | * helper function for btrfs_search_slot. The goal is to find a block | |
2442 | * in cache without setting the path to blocking. If we find the block | |
2443 | * we return zero and the path is unchanged. | |
2444 | * | |
2445 | * If we can't find the block, we set the path blocking and do some | |
2446 | * reada. -EAGAIN is returned and the search must be repeated. | |
2447 | */ | |
2448 | static int | |
2449 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2450 | struct btrfs_root *root, struct btrfs_path *p, | |
2451 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2452 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2453 | { |
2454 | u64 blocknr; | |
2455 | u64 gen; | |
c8c42864 CM |
2456 | struct extent_buffer *b = *eb_ret; |
2457 | struct extent_buffer *tmp; | |
76a05b35 | 2458 | int ret; |
c8c42864 CM |
2459 | |
2460 | blocknr = btrfs_node_blockptr(b, slot); | |
2461 | gen = btrfs_node_ptr_generation(b, slot); | |
c8c42864 | 2462 | |
01d58472 | 2463 | tmp = btrfs_find_tree_block(root->fs_info, blocknr); |
cb44921a | 2464 | if (tmp) { |
b9fab919 | 2465 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2466 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2467 | *eb_ret = tmp; | |
2468 | return 0; | |
2469 | } | |
2470 | ||
2471 | /* the pages were up to date, but we failed | |
2472 | * the generation number check. Do a full | |
2473 | * read for the generation number that is correct. | |
2474 | * We must do this without dropping locks so | |
2475 | * we can trust our generation number | |
2476 | */ | |
2477 | btrfs_set_path_blocking(p); | |
2478 | ||
2479 | /* now we're allowed to do a blocking uptodate check */ | |
2480 | ret = btrfs_read_buffer(tmp, gen); | |
2481 | if (!ret) { | |
2482 | *eb_ret = tmp; | |
2483 | return 0; | |
cb44921a | 2484 | } |
bdf7c00e JB |
2485 | free_extent_buffer(tmp); |
2486 | btrfs_release_path(p); | |
2487 | return -EIO; | |
c8c42864 CM |
2488 | } |
2489 | ||
2490 | /* | |
2491 | * reduce lock contention at high levels | |
2492 | * of the btree by dropping locks before | |
76a05b35 CM |
2493 | * we read. Don't release the lock on the current |
2494 | * level because we need to walk this node to figure | |
2495 | * out which blocks to read. | |
c8c42864 | 2496 | */ |
8c594ea8 CM |
2497 | btrfs_unlock_up_safe(p, level + 1); |
2498 | btrfs_set_path_blocking(p); | |
2499 | ||
cb44921a | 2500 | free_extent_buffer(tmp); |
e4058b54 | 2501 | if (p->reada != READA_NONE) |
c8c42864 CM |
2502 | reada_for_search(root, p, level, slot, key->objectid); |
2503 | ||
b3b4aa74 | 2504 | btrfs_release_path(p); |
76a05b35 CM |
2505 | |
2506 | ret = -EAGAIN; | |
ce86cd59 | 2507 | tmp = read_tree_block(root, blocknr, 0); |
64c043de | 2508 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2509 | /* |
2510 | * If the read above didn't mark this buffer up to date, | |
2511 | * it will never end up being up to date. Set ret to EIO now | |
2512 | * and give up so that our caller doesn't loop forever | |
2513 | * on our EAGAINs. | |
2514 | */ | |
b9fab919 | 2515 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2516 | ret = -EIO; |
c8c42864 | 2517 | free_extent_buffer(tmp); |
c871b0f2 LB |
2518 | } else { |
2519 | ret = PTR_ERR(tmp); | |
76a05b35 CM |
2520 | } |
2521 | return ret; | |
c8c42864 CM |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * helper function for btrfs_search_slot. This does all of the checks | |
2526 | * for node-level blocks and does any balancing required based on | |
2527 | * the ins_len. | |
2528 | * | |
2529 | * If no extra work was required, zero is returned. If we had to | |
2530 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2531 | * start over | |
2532 | */ | |
2533 | static int | |
2534 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2535 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2536 | struct extent_buffer *b, int level, int ins_len, |
2537 | int *write_lock_level) | |
c8c42864 CM |
2538 | { |
2539 | int ret; | |
2540 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2541 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2542 | int sret; | |
2543 | ||
bd681513 CM |
2544 | if (*write_lock_level < level + 1) { |
2545 | *write_lock_level = level + 1; | |
2546 | btrfs_release_path(p); | |
2547 | goto again; | |
2548 | } | |
2549 | ||
c8c42864 | 2550 | btrfs_set_path_blocking(p); |
0b08851f | 2551 | reada_for_balance(root, p, level); |
c8c42864 | 2552 | sret = split_node(trans, root, p, level); |
bd681513 | 2553 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2554 | |
2555 | BUG_ON(sret > 0); | |
2556 | if (sret) { | |
2557 | ret = sret; | |
2558 | goto done; | |
2559 | } | |
2560 | b = p->nodes[level]; | |
2561 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2562 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2563 | int sret; |
2564 | ||
bd681513 CM |
2565 | if (*write_lock_level < level + 1) { |
2566 | *write_lock_level = level + 1; | |
2567 | btrfs_release_path(p); | |
2568 | goto again; | |
2569 | } | |
2570 | ||
c8c42864 | 2571 | btrfs_set_path_blocking(p); |
0b08851f | 2572 | reada_for_balance(root, p, level); |
c8c42864 | 2573 | sret = balance_level(trans, root, p, level); |
bd681513 | 2574 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2575 | |
2576 | if (sret) { | |
2577 | ret = sret; | |
2578 | goto done; | |
2579 | } | |
2580 | b = p->nodes[level]; | |
2581 | if (!b) { | |
b3b4aa74 | 2582 | btrfs_release_path(p); |
c8c42864 CM |
2583 | goto again; |
2584 | } | |
2585 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2586 | } | |
2587 | return 0; | |
2588 | ||
2589 | again: | |
2590 | ret = -EAGAIN; | |
2591 | done: | |
2592 | return ret; | |
2593 | } | |
2594 | ||
d7396f07 FDBM |
2595 | static void key_search_validate(struct extent_buffer *b, |
2596 | struct btrfs_key *key, | |
2597 | int level) | |
2598 | { | |
2599 | #ifdef CONFIG_BTRFS_ASSERT | |
2600 | struct btrfs_disk_key disk_key; | |
2601 | ||
2602 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2603 | ||
2604 | if (level == 0) | |
2605 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2606 | offsetof(struct btrfs_leaf, items[0].key), | |
2607 | sizeof(disk_key))); | |
2608 | else | |
2609 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2610 | offsetof(struct btrfs_node, ptrs[0].key), | |
2611 | sizeof(disk_key))); | |
2612 | #endif | |
2613 | } | |
2614 | ||
2615 | static int key_search(struct extent_buffer *b, struct btrfs_key *key, | |
2616 | int level, int *prev_cmp, int *slot) | |
2617 | { | |
2618 | if (*prev_cmp != 0) { | |
2619 | *prev_cmp = bin_search(b, key, level, slot); | |
2620 | return *prev_cmp; | |
2621 | } | |
2622 | ||
2623 | key_search_validate(b, key, level); | |
2624 | *slot = 0; | |
2625 | ||
2626 | return 0; | |
2627 | } | |
2628 | ||
381cf658 | 2629 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2630 | u64 iobjectid, u64 ioff, u8 key_type, |
2631 | struct btrfs_key *found_key) | |
2632 | { | |
2633 | int ret; | |
2634 | struct btrfs_key key; | |
2635 | struct extent_buffer *eb; | |
381cf658 DS |
2636 | |
2637 | ASSERT(path); | |
1d4c08e0 | 2638 | ASSERT(found_key); |
e33d5c3d KN |
2639 | |
2640 | key.type = key_type; | |
2641 | key.objectid = iobjectid; | |
2642 | key.offset = ioff; | |
2643 | ||
2644 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2645 | if (ret < 0) |
e33d5c3d KN |
2646 | return ret; |
2647 | ||
2648 | eb = path->nodes[0]; | |
2649 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2650 | ret = btrfs_next_leaf(fs_root, path); | |
2651 | if (ret) | |
2652 | return ret; | |
2653 | eb = path->nodes[0]; | |
2654 | } | |
2655 | ||
2656 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2657 | if (found_key->type != key.type || | |
2658 | found_key->objectid != key.objectid) | |
2659 | return 1; | |
2660 | ||
2661 | return 0; | |
2662 | } | |
2663 | ||
74123bd7 CM |
2664 | /* |
2665 | * look for key in the tree. path is filled in with nodes along the way | |
2666 | * if key is found, we return zero and you can find the item in the leaf | |
2667 | * level of the path (level 0) | |
2668 | * | |
2669 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2670 | * be inserted, and 1 is returned. If there are other errors during the |
2671 | * search a negative error number is returned. | |
97571fd0 CM |
2672 | * |
2673 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2674 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2675 | * possible) | |
74123bd7 | 2676 | */ |
e089f05c CM |
2677 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2678 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2679 | ins_len, int cow) | |
be0e5c09 | 2680 | { |
5f39d397 | 2681 | struct extent_buffer *b; |
be0e5c09 CM |
2682 | int slot; |
2683 | int ret; | |
33c66f43 | 2684 | int err; |
be0e5c09 | 2685 | int level; |
925baedd | 2686 | int lowest_unlock = 1; |
bd681513 CM |
2687 | int root_lock; |
2688 | /* everything at write_lock_level or lower must be write locked */ | |
2689 | int write_lock_level = 0; | |
9f3a7427 | 2690 | u8 lowest_level = 0; |
f7c79f30 | 2691 | int min_write_lock_level; |
d7396f07 | 2692 | int prev_cmp; |
9f3a7427 | 2693 | |
6702ed49 | 2694 | lowest_level = p->lowest_level; |
323ac95b | 2695 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2696 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2697 | BUG_ON(!cow && ins_len); |
25179201 | 2698 | |
bd681513 | 2699 | if (ins_len < 0) { |
925baedd | 2700 | lowest_unlock = 2; |
65b51a00 | 2701 | |
bd681513 CM |
2702 | /* when we are removing items, we might have to go up to level |
2703 | * two as we update tree pointers Make sure we keep write | |
2704 | * for those levels as well | |
2705 | */ | |
2706 | write_lock_level = 2; | |
2707 | } else if (ins_len > 0) { | |
2708 | /* | |
2709 | * for inserting items, make sure we have a write lock on | |
2710 | * level 1 so we can update keys | |
2711 | */ | |
2712 | write_lock_level = 1; | |
2713 | } | |
2714 | ||
2715 | if (!cow) | |
2716 | write_lock_level = -1; | |
2717 | ||
09a2a8f9 | 2718 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2719 | write_lock_level = BTRFS_MAX_LEVEL; |
2720 | ||
f7c79f30 CM |
2721 | min_write_lock_level = write_lock_level; |
2722 | ||
bb803951 | 2723 | again: |
d7396f07 | 2724 | prev_cmp = -1; |
bd681513 CM |
2725 | /* |
2726 | * we try very hard to do read locks on the root | |
2727 | */ | |
2728 | root_lock = BTRFS_READ_LOCK; | |
2729 | level = 0; | |
5d4f98a2 | 2730 | if (p->search_commit_root) { |
bd681513 CM |
2731 | /* |
2732 | * the commit roots are read only | |
2733 | * so we always do read locks | |
2734 | */ | |
3f8a18cc JB |
2735 | if (p->need_commit_sem) |
2736 | down_read(&root->fs_info->commit_root_sem); | |
5d4f98a2 YZ |
2737 | b = root->commit_root; |
2738 | extent_buffer_get(b); | |
bd681513 | 2739 | level = btrfs_header_level(b); |
3f8a18cc JB |
2740 | if (p->need_commit_sem) |
2741 | up_read(&root->fs_info->commit_root_sem); | |
5d4f98a2 | 2742 | if (!p->skip_locking) |
bd681513 | 2743 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2744 | } else { |
bd681513 | 2745 | if (p->skip_locking) { |
5d4f98a2 | 2746 | b = btrfs_root_node(root); |
bd681513 CM |
2747 | level = btrfs_header_level(b); |
2748 | } else { | |
2749 | /* we don't know the level of the root node | |
2750 | * until we actually have it read locked | |
2751 | */ | |
2752 | b = btrfs_read_lock_root_node(root); | |
2753 | level = btrfs_header_level(b); | |
2754 | if (level <= write_lock_level) { | |
2755 | /* whoops, must trade for write lock */ | |
2756 | btrfs_tree_read_unlock(b); | |
2757 | free_extent_buffer(b); | |
2758 | b = btrfs_lock_root_node(root); | |
2759 | root_lock = BTRFS_WRITE_LOCK; | |
2760 | ||
2761 | /* the level might have changed, check again */ | |
2762 | level = btrfs_header_level(b); | |
2763 | } | |
2764 | } | |
5d4f98a2 | 2765 | } |
bd681513 CM |
2766 | p->nodes[level] = b; |
2767 | if (!p->skip_locking) | |
2768 | p->locks[level] = root_lock; | |
925baedd | 2769 | |
eb60ceac | 2770 | while (b) { |
5f39d397 | 2771 | level = btrfs_header_level(b); |
65b51a00 CM |
2772 | |
2773 | /* | |
2774 | * setup the path here so we can release it under lock | |
2775 | * contention with the cow code | |
2776 | */ | |
02217ed2 | 2777 | if (cow) { |
c8c42864 CM |
2778 | /* |
2779 | * if we don't really need to cow this block | |
2780 | * then we don't want to set the path blocking, | |
2781 | * so we test it here | |
2782 | */ | |
64c12921 JM |
2783 | if (!should_cow_block(trans, root, b)) { |
2784 | trans->dirty = true; | |
65b51a00 | 2785 | goto cow_done; |
64c12921 | 2786 | } |
5d4f98a2 | 2787 | |
bd681513 CM |
2788 | /* |
2789 | * must have write locks on this node and the | |
2790 | * parent | |
2791 | */ | |
5124e00e JB |
2792 | if (level > write_lock_level || |
2793 | (level + 1 > write_lock_level && | |
2794 | level + 1 < BTRFS_MAX_LEVEL && | |
2795 | p->nodes[level + 1])) { | |
bd681513 CM |
2796 | write_lock_level = level + 1; |
2797 | btrfs_release_path(p); | |
2798 | goto again; | |
2799 | } | |
2800 | ||
160f4089 | 2801 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2802 | err = btrfs_cow_block(trans, root, b, |
2803 | p->nodes[level + 1], | |
2804 | p->slots[level + 1], &b); | |
2805 | if (err) { | |
33c66f43 | 2806 | ret = err; |
65b51a00 | 2807 | goto done; |
54aa1f4d | 2808 | } |
02217ed2 | 2809 | } |
65b51a00 | 2810 | cow_done: |
eb60ceac | 2811 | p->nodes[level] = b; |
bd681513 | 2812 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2813 | |
2814 | /* | |
2815 | * we have a lock on b and as long as we aren't changing | |
2816 | * the tree, there is no way to for the items in b to change. | |
2817 | * It is safe to drop the lock on our parent before we | |
2818 | * go through the expensive btree search on b. | |
2819 | * | |
eb653de1 FDBM |
2820 | * If we're inserting or deleting (ins_len != 0), then we might |
2821 | * be changing slot zero, which may require changing the parent. | |
2822 | * So, we can't drop the lock until after we know which slot | |
2823 | * we're operating on. | |
b4ce94de | 2824 | */ |
eb653de1 FDBM |
2825 | if (!ins_len && !p->keep_locks) { |
2826 | int u = level + 1; | |
2827 | ||
2828 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2829 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2830 | p->locks[u] = 0; | |
2831 | } | |
2832 | } | |
b4ce94de | 2833 | |
d7396f07 | 2834 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2835 | if (ret < 0) |
2836 | goto done; | |
b4ce94de | 2837 | |
5f39d397 | 2838 | if (level != 0) { |
33c66f43 YZ |
2839 | int dec = 0; |
2840 | if (ret && slot > 0) { | |
2841 | dec = 1; | |
be0e5c09 | 2842 | slot -= 1; |
33c66f43 | 2843 | } |
be0e5c09 | 2844 | p->slots[level] = slot; |
33c66f43 | 2845 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2846 | ins_len, &write_lock_level); |
33c66f43 | 2847 | if (err == -EAGAIN) |
c8c42864 | 2848 | goto again; |
33c66f43 YZ |
2849 | if (err) { |
2850 | ret = err; | |
c8c42864 | 2851 | goto done; |
33c66f43 | 2852 | } |
c8c42864 CM |
2853 | b = p->nodes[level]; |
2854 | slot = p->slots[level]; | |
b4ce94de | 2855 | |
bd681513 CM |
2856 | /* |
2857 | * slot 0 is special, if we change the key | |
2858 | * we have to update the parent pointer | |
2859 | * which means we must have a write lock | |
2860 | * on the parent | |
2861 | */ | |
eb653de1 | 2862 | if (slot == 0 && ins_len && |
bd681513 CM |
2863 | write_lock_level < level + 1) { |
2864 | write_lock_level = level + 1; | |
2865 | btrfs_release_path(p); | |
2866 | goto again; | |
2867 | } | |
2868 | ||
f7c79f30 CM |
2869 | unlock_up(p, level, lowest_unlock, |
2870 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2871 | |
925baedd | 2872 | if (level == lowest_level) { |
33c66f43 YZ |
2873 | if (dec) |
2874 | p->slots[level]++; | |
5b21f2ed | 2875 | goto done; |
925baedd | 2876 | } |
ca7a79ad | 2877 | |
33c66f43 | 2878 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2879 | &b, level, slot, key, 0); |
33c66f43 | 2880 | if (err == -EAGAIN) |
c8c42864 | 2881 | goto again; |
33c66f43 YZ |
2882 | if (err) { |
2883 | ret = err; | |
76a05b35 | 2884 | goto done; |
33c66f43 | 2885 | } |
76a05b35 | 2886 | |
b4ce94de | 2887 | if (!p->skip_locking) { |
bd681513 CM |
2888 | level = btrfs_header_level(b); |
2889 | if (level <= write_lock_level) { | |
2890 | err = btrfs_try_tree_write_lock(b); | |
2891 | if (!err) { | |
2892 | btrfs_set_path_blocking(p); | |
2893 | btrfs_tree_lock(b); | |
2894 | btrfs_clear_path_blocking(p, b, | |
2895 | BTRFS_WRITE_LOCK); | |
2896 | } | |
2897 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2898 | } else { | |
f82c458a | 2899 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2900 | if (!err) { |
2901 | btrfs_set_path_blocking(p); | |
2902 | btrfs_tree_read_lock(b); | |
2903 | btrfs_clear_path_blocking(p, b, | |
2904 | BTRFS_READ_LOCK); | |
2905 | } | |
2906 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2907 | } |
bd681513 | 2908 | p->nodes[level] = b; |
b4ce94de | 2909 | } |
be0e5c09 CM |
2910 | } else { |
2911 | p->slots[level] = slot; | |
87b29b20 YZ |
2912 | if (ins_len > 0 && |
2913 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2914 | if (write_lock_level < 1) { |
2915 | write_lock_level = 1; | |
2916 | btrfs_release_path(p); | |
2917 | goto again; | |
2918 | } | |
2919 | ||
b4ce94de | 2920 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2921 | err = split_leaf(trans, root, key, |
2922 | p, ins_len, ret == 0); | |
bd681513 | 2923 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2924 | |
33c66f43 YZ |
2925 | BUG_ON(err > 0); |
2926 | if (err) { | |
2927 | ret = err; | |
65b51a00 CM |
2928 | goto done; |
2929 | } | |
5c680ed6 | 2930 | } |
459931ec | 2931 | if (!p->search_for_split) |
f7c79f30 CM |
2932 | unlock_up(p, level, lowest_unlock, |
2933 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2934 | goto done; |
be0e5c09 CM |
2935 | } |
2936 | } | |
65b51a00 CM |
2937 | ret = 1; |
2938 | done: | |
b4ce94de CM |
2939 | /* |
2940 | * we don't really know what they plan on doing with the path | |
2941 | * from here on, so for now just mark it as blocking | |
2942 | */ | |
b9473439 CM |
2943 | if (!p->leave_spinning) |
2944 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2945 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2946 | btrfs_release_path(p); |
65b51a00 | 2947 | return ret; |
be0e5c09 CM |
2948 | } |
2949 | ||
5d9e75c4 JS |
2950 | /* |
2951 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2952 | * current state of the tree together with the operations recorded in the tree | |
2953 | * modification log to search for the key in a previous version of this tree, as | |
2954 | * denoted by the time_seq parameter. | |
2955 | * | |
2956 | * Naturally, there is no support for insert, delete or cow operations. | |
2957 | * | |
2958 | * The resulting path and return value will be set up as if we called | |
2959 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2960 | */ | |
2961 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2962 | struct btrfs_path *p, u64 time_seq) | |
2963 | { | |
2964 | struct extent_buffer *b; | |
2965 | int slot; | |
2966 | int ret; | |
2967 | int err; | |
2968 | int level; | |
2969 | int lowest_unlock = 1; | |
2970 | u8 lowest_level = 0; | |
d4b4087c | 2971 | int prev_cmp = -1; |
5d9e75c4 JS |
2972 | |
2973 | lowest_level = p->lowest_level; | |
2974 | WARN_ON(p->nodes[0] != NULL); | |
2975 | ||
2976 | if (p->search_commit_root) { | |
2977 | BUG_ON(time_seq); | |
2978 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2979 | } | |
2980 | ||
2981 | again: | |
5d9e75c4 | 2982 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2983 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2984 | p->locks[level] = BTRFS_READ_LOCK; |
2985 | ||
2986 | while (b) { | |
2987 | level = btrfs_header_level(b); | |
2988 | p->nodes[level] = b; | |
2989 | btrfs_clear_path_blocking(p, NULL, 0); | |
2990 | ||
2991 | /* | |
2992 | * we have a lock on b and as long as we aren't changing | |
2993 | * the tree, there is no way to for the items in b to change. | |
2994 | * It is safe to drop the lock on our parent before we | |
2995 | * go through the expensive btree search on b. | |
2996 | */ | |
2997 | btrfs_unlock_up_safe(p, level + 1); | |
2998 | ||
d4b4087c | 2999 | /* |
01327610 | 3000 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
3001 | * time. |
3002 | */ | |
3003 | prev_cmp = -1; | |
d7396f07 | 3004 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
3005 | |
3006 | if (level != 0) { | |
3007 | int dec = 0; | |
3008 | if (ret && slot > 0) { | |
3009 | dec = 1; | |
3010 | slot -= 1; | |
3011 | } | |
3012 | p->slots[level] = slot; | |
3013 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3014 | ||
3015 | if (level == lowest_level) { | |
3016 | if (dec) | |
3017 | p->slots[level]++; | |
3018 | goto done; | |
3019 | } | |
3020 | ||
3021 | err = read_block_for_search(NULL, root, p, &b, level, | |
3022 | slot, key, time_seq); | |
3023 | if (err == -EAGAIN) | |
3024 | goto again; | |
3025 | if (err) { | |
3026 | ret = err; | |
3027 | goto done; | |
3028 | } | |
3029 | ||
3030 | level = btrfs_header_level(b); | |
f82c458a | 3031 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3032 | if (!err) { |
3033 | btrfs_set_path_blocking(p); | |
3034 | btrfs_tree_read_lock(b); | |
3035 | btrfs_clear_path_blocking(p, b, | |
3036 | BTRFS_READ_LOCK); | |
3037 | } | |
9ec72677 | 3038 | b = tree_mod_log_rewind(root->fs_info, p, b, time_seq); |
db7f3436 JB |
3039 | if (!b) { |
3040 | ret = -ENOMEM; | |
3041 | goto done; | |
3042 | } | |
5d9e75c4 JS |
3043 | p->locks[level] = BTRFS_READ_LOCK; |
3044 | p->nodes[level] = b; | |
5d9e75c4 JS |
3045 | } else { |
3046 | p->slots[level] = slot; | |
3047 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3048 | goto done; | |
3049 | } | |
3050 | } | |
3051 | ret = 1; | |
3052 | done: | |
3053 | if (!p->leave_spinning) | |
3054 | btrfs_set_path_blocking(p); | |
3055 | if (ret < 0) | |
3056 | btrfs_release_path(p); | |
3057 | ||
3058 | return ret; | |
3059 | } | |
3060 | ||
2f38b3e1 AJ |
3061 | /* |
3062 | * helper to use instead of search slot if no exact match is needed but | |
3063 | * instead the next or previous item should be returned. | |
3064 | * When find_higher is true, the next higher item is returned, the next lower | |
3065 | * otherwise. | |
3066 | * When return_any and find_higher are both true, and no higher item is found, | |
3067 | * return the next lower instead. | |
3068 | * When return_any is true and find_higher is false, and no lower item is found, | |
3069 | * return the next higher instead. | |
3070 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3071 | * < 0 on error | |
3072 | */ | |
3073 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
3074 | struct btrfs_key *key, struct btrfs_path *p, | |
3075 | int find_higher, int return_any) | |
3076 | { | |
3077 | int ret; | |
3078 | struct extent_buffer *leaf; | |
3079 | ||
3080 | again: | |
3081 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3082 | if (ret <= 0) | |
3083 | return ret; | |
3084 | /* | |
3085 | * a return value of 1 means the path is at the position where the | |
3086 | * item should be inserted. Normally this is the next bigger item, | |
3087 | * but in case the previous item is the last in a leaf, path points | |
3088 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3089 | * item. | |
3090 | */ | |
3091 | leaf = p->nodes[0]; | |
3092 | ||
3093 | if (find_higher) { | |
3094 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3095 | ret = btrfs_next_leaf(root, p); | |
3096 | if (ret <= 0) | |
3097 | return ret; | |
3098 | if (!return_any) | |
3099 | return 1; | |
3100 | /* | |
3101 | * no higher item found, return the next | |
3102 | * lower instead | |
3103 | */ | |
3104 | return_any = 0; | |
3105 | find_higher = 0; | |
3106 | btrfs_release_path(p); | |
3107 | goto again; | |
3108 | } | |
3109 | } else { | |
e6793769 AJ |
3110 | if (p->slots[0] == 0) { |
3111 | ret = btrfs_prev_leaf(root, p); | |
3112 | if (ret < 0) | |
3113 | return ret; | |
3114 | if (!ret) { | |
23c6bf6a FDBM |
3115 | leaf = p->nodes[0]; |
3116 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3117 | p->slots[0]--; | |
e6793769 | 3118 | return 0; |
2f38b3e1 | 3119 | } |
e6793769 AJ |
3120 | if (!return_any) |
3121 | return 1; | |
3122 | /* | |
3123 | * no lower item found, return the next | |
3124 | * higher instead | |
3125 | */ | |
3126 | return_any = 0; | |
3127 | find_higher = 1; | |
3128 | btrfs_release_path(p); | |
3129 | goto again; | |
3130 | } else { | |
2f38b3e1 AJ |
3131 | --p->slots[0]; |
3132 | } | |
3133 | } | |
3134 | return 0; | |
3135 | } | |
3136 | ||
74123bd7 CM |
3137 | /* |
3138 | * adjust the pointers going up the tree, starting at level | |
3139 | * making sure the right key of each node is points to 'key'. | |
3140 | * This is used after shifting pointers to the left, so it stops | |
3141 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3142 | * higher levels | |
aa5d6bed | 3143 | * |
74123bd7 | 3144 | */ |
b7a0365e DD |
3145 | static void fixup_low_keys(struct btrfs_fs_info *fs_info, |
3146 | struct btrfs_path *path, | |
143bede5 | 3147 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3148 | { |
3149 | int i; | |
5f39d397 CM |
3150 | struct extent_buffer *t; |
3151 | ||
234b63a0 | 3152 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3153 | int tslot = path->slots[i]; |
eb60ceac | 3154 | if (!path->nodes[i]) |
be0e5c09 | 3155 | break; |
5f39d397 | 3156 | t = path->nodes[i]; |
b7a0365e | 3157 | tree_mod_log_set_node_key(fs_info, t, tslot, 1); |
5f39d397 | 3158 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3159 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3160 | if (tslot != 0) |
3161 | break; | |
3162 | } | |
3163 | } | |
3164 | ||
31840ae1 ZY |
3165 | /* |
3166 | * update item key. | |
3167 | * | |
3168 | * This function isn't completely safe. It's the caller's responsibility | |
3169 | * that the new key won't break the order | |
3170 | */ | |
b7a0365e DD |
3171 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3172 | struct btrfs_path *path, | |
143bede5 | 3173 | struct btrfs_key *new_key) |
31840ae1 ZY |
3174 | { |
3175 | struct btrfs_disk_key disk_key; | |
3176 | struct extent_buffer *eb; | |
3177 | int slot; | |
3178 | ||
3179 | eb = path->nodes[0]; | |
3180 | slot = path->slots[0]; | |
3181 | if (slot > 0) { | |
3182 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3183 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3184 | } |
3185 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3186 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3187 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3188 | } |
3189 | ||
3190 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3191 | btrfs_set_item_key(eb, &disk_key, slot); | |
3192 | btrfs_mark_buffer_dirty(eb); | |
3193 | if (slot == 0) | |
b7a0365e | 3194 | fixup_low_keys(fs_info, path, &disk_key, 1); |
31840ae1 ZY |
3195 | } |
3196 | ||
74123bd7 CM |
3197 | /* |
3198 | * try to push data from one node into the next node left in the | |
79f95c82 | 3199 | * tree. |
aa5d6bed CM |
3200 | * |
3201 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3202 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3203 | */ |
98ed5174 CM |
3204 | static int push_node_left(struct btrfs_trans_handle *trans, |
3205 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 3206 | struct extent_buffer *src, int empty) |
be0e5c09 | 3207 | { |
be0e5c09 | 3208 | int push_items = 0; |
bb803951 CM |
3209 | int src_nritems; |
3210 | int dst_nritems; | |
aa5d6bed | 3211 | int ret = 0; |
be0e5c09 | 3212 | |
5f39d397 CM |
3213 | src_nritems = btrfs_header_nritems(src); |
3214 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3215 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
3216 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3217 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3218 | |
bce4eae9 | 3219 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3220 | return 1; |
3221 | ||
d397712b | 3222 | if (push_items <= 0) |
be0e5c09 CM |
3223 | return 1; |
3224 | ||
bce4eae9 | 3225 | if (empty) { |
971a1f66 | 3226 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3227 | if (push_items < src_nritems) { |
3228 | /* leave at least 8 pointers in the node if | |
3229 | * we aren't going to empty it | |
3230 | */ | |
3231 | if (src_nritems - push_items < 8) { | |
3232 | if (push_items <= 8) | |
3233 | return 1; | |
3234 | push_items -= 8; | |
3235 | } | |
3236 | } | |
3237 | } else | |
3238 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3239 | |
5de865ee FDBM |
3240 | ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
3241 | push_items); | |
3242 | if (ret) { | |
3243 | btrfs_abort_transaction(trans, root, ret); | |
3244 | return ret; | |
3245 | } | |
5f39d397 CM |
3246 | copy_extent_buffer(dst, src, |
3247 | btrfs_node_key_ptr_offset(dst_nritems), | |
3248 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3249 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3250 | |
bb803951 | 3251 | if (push_items < src_nritems) { |
57911b8b JS |
3252 | /* |
3253 | * don't call tree_mod_log_eb_move here, key removal was already | |
3254 | * fully logged by tree_mod_log_eb_copy above. | |
3255 | */ | |
5f39d397 CM |
3256 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3257 | btrfs_node_key_ptr_offset(push_items), | |
3258 | (src_nritems - push_items) * | |
3259 | sizeof(struct btrfs_key_ptr)); | |
3260 | } | |
3261 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3262 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3263 | btrfs_mark_buffer_dirty(src); | |
3264 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3265 | |
79f95c82 CM |
3266 | return ret; |
3267 | } | |
3268 | ||
3269 | /* | |
3270 | * try to push data from one node into the next node right in the | |
3271 | * tree. | |
3272 | * | |
3273 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3274 | * error, and > 0 if there was no room in the right hand block. | |
3275 | * | |
3276 | * this will only push up to 1/2 the contents of the left node over | |
3277 | */ | |
5f39d397 CM |
3278 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3279 | struct btrfs_root *root, | |
3280 | struct extent_buffer *dst, | |
3281 | struct extent_buffer *src) | |
79f95c82 | 3282 | { |
79f95c82 CM |
3283 | int push_items = 0; |
3284 | int max_push; | |
3285 | int src_nritems; | |
3286 | int dst_nritems; | |
3287 | int ret = 0; | |
79f95c82 | 3288 | |
7bb86316 CM |
3289 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3290 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3291 | ||
5f39d397 CM |
3292 | src_nritems = btrfs_header_nritems(src); |
3293 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3294 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3295 | if (push_items <= 0) |
79f95c82 | 3296 | return 1; |
bce4eae9 | 3297 | |
d397712b | 3298 | if (src_nritems < 4) |
bce4eae9 | 3299 | return 1; |
79f95c82 CM |
3300 | |
3301 | max_push = src_nritems / 2 + 1; | |
3302 | /* don't try to empty the node */ | |
d397712b | 3303 | if (max_push >= src_nritems) |
79f95c82 | 3304 | return 1; |
252c38f0 | 3305 | |
79f95c82 CM |
3306 | if (max_push < push_items) |
3307 | push_items = max_push; | |
3308 | ||
f230475e | 3309 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3310 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3311 | btrfs_node_key_ptr_offset(0), | |
3312 | (dst_nritems) * | |
3313 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3314 | |
5de865ee FDBM |
3315 | ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
3316 | src_nritems - push_items, push_items); | |
3317 | if (ret) { | |
3318 | btrfs_abort_transaction(trans, root, ret); | |
3319 | return ret; | |
3320 | } | |
5f39d397 CM |
3321 | copy_extent_buffer(dst, src, |
3322 | btrfs_node_key_ptr_offset(0), | |
3323 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3324 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3325 | |
5f39d397 CM |
3326 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3327 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3328 | |
5f39d397 CM |
3329 | btrfs_mark_buffer_dirty(src); |
3330 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3331 | |
aa5d6bed | 3332 | return ret; |
be0e5c09 CM |
3333 | } |
3334 | ||
97571fd0 CM |
3335 | /* |
3336 | * helper function to insert a new root level in the tree. | |
3337 | * A new node is allocated, and a single item is inserted to | |
3338 | * point to the existing root | |
aa5d6bed CM |
3339 | * |
3340 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3341 | */ |
d397712b | 3342 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3343 | struct btrfs_root *root, |
fdd99c72 | 3344 | struct btrfs_path *path, int level) |
5c680ed6 | 3345 | { |
7bb86316 | 3346 | u64 lower_gen; |
5f39d397 CM |
3347 | struct extent_buffer *lower; |
3348 | struct extent_buffer *c; | |
925baedd | 3349 | struct extent_buffer *old; |
5f39d397 | 3350 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3351 | |
3352 | BUG_ON(path->nodes[level]); | |
3353 | BUG_ON(path->nodes[level-1] != root->node); | |
3354 | ||
7bb86316 CM |
3355 | lower = path->nodes[level-1]; |
3356 | if (level == 1) | |
3357 | btrfs_item_key(lower, &lower_key, 0); | |
3358 | else | |
3359 | btrfs_node_key(lower, &lower_key, 0); | |
3360 | ||
4d75f8a9 DS |
3361 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3362 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3363 | if (IS_ERR(c)) |
3364 | return PTR_ERR(c); | |
925baedd | 3365 | |
f0486c68 YZ |
3366 | root_add_used(root, root->nodesize); |
3367 | ||
5d4f98a2 | 3368 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3369 | btrfs_set_header_nritems(c, 1); |
3370 | btrfs_set_header_level(c, level); | |
db94535d | 3371 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3372 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3373 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3374 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3375 | |
0a4e5586 | 3376 | write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(), |
5f39d397 | 3377 | BTRFS_FSID_SIZE); |
e17cade2 CM |
3378 | |
3379 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 3380 | btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE); |
e17cade2 | 3381 | |
5f39d397 | 3382 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3383 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3384 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3385 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3386 | |
3387 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3388 | |
5f39d397 | 3389 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3390 | |
925baedd | 3391 | old = root->node; |
fdd99c72 | 3392 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3393 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3394 | |
3395 | /* the super has an extra ref to root->node */ | |
3396 | free_extent_buffer(old); | |
3397 | ||
0b86a832 | 3398 | add_root_to_dirty_list(root); |
5f39d397 CM |
3399 | extent_buffer_get(c); |
3400 | path->nodes[level] = c; | |
95449a16 | 3401 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3402 | path->slots[level] = 0; |
3403 | return 0; | |
3404 | } | |
3405 | ||
74123bd7 CM |
3406 | /* |
3407 | * worker function to insert a single pointer in a node. | |
3408 | * the node should have enough room for the pointer already | |
97571fd0 | 3409 | * |
74123bd7 CM |
3410 | * slot and level indicate where you want the key to go, and |
3411 | * blocknr is the block the key points to. | |
3412 | */ | |
143bede5 JM |
3413 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3414 | struct btrfs_root *root, struct btrfs_path *path, | |
3415 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3416 | int slot, int level) |
74123bd7 | 3417 | { |
5f39d397 | 3418 | struct extent_buffer *lower; |
74123bd7 | 3419 | int nritems; |
f3ea38da | 3420 | int ret; |
5c680ed6 CM |
3421 | |
3422 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3423 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3424 | lower = path->nodes[level]; |
3425 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3426 | BUG_ON(slot > nritems); |
143bede5 | 3427 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3428 | if (slot != nritems) { |
c3e06965 | 3429 | if (level) |
f3ea38da JS |
3430 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3431 | slot, nritems - slot); | |
5f39d397 CM |
3432 | memmove_extent_buffer(lower, |
3433 | btrfs_node_key_ptr_offset(slot + 1), | |
3434 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3435 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3436 | } |
c3e06965 | 3437 | if (level) { |
f3ea38da | 3438 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
c8cc6341 | 3439 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3440 | BUG_ON(ret < 0); |
3441 | } | |
5f39d397 | 3442 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3443 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3444 | WARN_ON(trans->transid == 0); |
3445 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3446 | btrfs_set_header_nritems(lower, nritems + 1); |
3447 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3448 | } |
3449 | ||
97571fd0 CM |
3450 | /* |
3451 | * split the node at the specified level in path in two. | |
3452 | * The path is corrected to point to the appropriate node after the split | |
3453 | * | |
3454 | * Before splitting this tries to make some room in the node by pushing | |
3455 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3456 | * |
3457 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3458 | */ |
e02119d5 CM |
3459 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3460 | struct btrfs_root *root, | |
3461 | struct btrfs_path *path, int level) | |
be0e5c09 | 3462 | { |
5f39d397 CM |
3463 | struct extent_buffer *c; |
3464 | struct extent_buffer *split; | |
3465 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3466 | int mid; |
5c680ed6 | 3467 | int ret; |
7518a238 | 3468 | u32 c_nritems; |
eb60ceac | 3469 | |
5f39d397 | 3470 | c = path->nodes[level]; |
7bb86316 | 3471 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3472 | if (c == root->node) { |
d9abbf1c | 3473 | /* |
90f8d62e JS |
3474 | * trying to split the root, lets make a new one |
3475 | * | |
fdd99c72 | 3476 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3477 | * insert_new_root, because that root buffer will be kept as a |
3478 | * normal node. We are going to log removal of half of the | |
3479 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3480 | * tree lock on the buffer, which is why we cannot race with | |
3481 | * other tree_mod_log users. | |
d9abbf1c | 3482 | */ |
fdd99c72 | 3483 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3484 | if (ret) |
3485 | return ret; | |
b3612421 | 3486 | } else { |
e66f709b | 3487 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3488 | c = path->nodes[level]; |
3489 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3490 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3491 | return 0; |
54aa1f4d CM |
3492 | if (ret < 0) |
3493 | return ret; | |
be0e5c09 | 3494 | } |
e66f709b | 3495 | |
5f39d397 | 3496 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3497 | mid = (c_nritems + 1) / 2; |
3498 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3499 | |
4d75f8a9 DS |
3500 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3501 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3502 | if (IS_ERR(split)) |
3503 | return PTR_ERR(split); | |
3504 | ||
f0486c68 YZ |
3505 | root_add_used(root, root->nodesize); |
3506 | ||
5d4f98a2 | 3507 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3508 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3509 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3510 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3511 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3512 | btrfs_set_header_owner(split, root->root_key.objectid); |
3513 | write_extent_buffer(split, root->fs_info->fsid, | |
0a4e5586 | 3514 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 | 3515 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
b308bc2f | 3516 | btrfs_header_chunk_tree_uuid(split), |
e17cade2 | 3517 | BTRFS_UUID_SIZE); |
54aa1f4d | 3518 | |
5de865ee FDBM |
3519 | ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0, |
3520 | mid, c_nritems - mid); | |
3521 | if (ret) { | |
3522 | btrfs_abort_transaction(trans, root, ret); | |
3523 | return ret; | |
3524 | } | |
5f39d397 CM |
3525 | copy_extent_buffer(split, c, |
3526 | btrfs_node_key_ptr_offset(0), | |
3527 | btrfs_node_key_ptr_offset(mid), | |
3528 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3529 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3530 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3531 | ret = 0; |
3532 | ||
5f39d397 CM |
3533 | btrfs_mark_buffer_dirty(c); |
3534 | btrfs_mark_buffer_dirty(split); | |
3535 | ||
143bede5 | 3536 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3537 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3538 | |
5de08d7d | 3539 | if (path->slots[level] >= mid) { |
5c680ed6 | 3540 | path->slots[level] -= mid; |
925baedd | 3541 | btrfs_tree_unlock(c); |
5f39d397 CM |
3542 | free_extent_buffer(c); |
3543 | path->nodes[level] = split; | |
5c680ed6 CM |
3544 | path->slots[level + 1] += 1; |
3545 | } else { | |
925baedd | 3546 | btrfs_tree_unlock(split); |
5f39d397 | 3547 | free_extent_buffer(split); |
be0e5c09 | 3548 | } |
aa5d6bed | 3549 | return ret; |
be0e5c09 CM |
3550 | } |
3551 | ||
74123bd7 CM |
3552 | /* |
3553 | * how many bytes are required to store the items in a leaf. start | |
3554 | * and nr indicate which items in the leaf to check. This totals up the | |
3555 | * space used both by the item structs and the item data | |
3556 | */ | |
5f39d397 | 3557 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3558 | { |
41be1f3b JB |
3559 | struct btrfs_item *start_item; |
3560 | struct btrfs_item *end_item; | |
3561 | struct btrfs_map_token token; | |
be0e5c09 | 3562 | int data_len; |
5f39d397 | 3563 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3564 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3565 | |
3566 | if (!nr) | |
3567 | return 0; | |
41be1f3b | 3568 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3569 | start_item = btrfs_item_nr(start); |
3570 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3571 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3572 | btrfs_token_item_size(l, start_item, &token); | |
3573 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3574 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3575 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3576 | return data_len; |
3577 | } | |
3578 | ||
d4dbff95 CM |
3579 | /* |
3580 | * The space between the end of the leaf items and | |
3581 | * the start of the leaf data. IOW, how much room | |
3582 | * the leaf has left for both items and data | |
3583 | */ | |
d397712b | 3584 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3585 | struct extent_buffer *leaf) |
d4dbff95 | 3586 | { |
5f39d397 CM |
3587 | int nritems = btrfs_header_nritems(leaf); |
3588 | int ret; | |
3589 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3590 | if (ret < 0) { | |
efe120a0 FH |
3591 | btrfs_crit(root->fs_info, |
3592 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
ae2f5411 | 3593 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3594 | leaf_space_used(leaf, 0, nritems), nritems); |
3595 | } | |
3596 | return ret; | |
d4dbff95 CM |
3597 | } |
3598 | ||
99d8f83c CM |
3599 | /* |
3600 | * min slot controls the lowest index we're willing to push to the | |
3601 | * right. We'll push up to and including min_slot, but no lower | |
3602 | */ | |
44871b1b CM |
3603 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3604 | struct btrfs_root *root, | |
3605 | struct btrfs_path *path, | |
3606 | int data_size, int empty, | |
3607 | struct extent_buffer *right, | |
99d8f83c CM |
3608 | int free_space, u32 left_nritems, |
3609 | u32 min_slot) | |
00ec4c51 | 3610 | { |
5f39d397 | 3611 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3612 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3613 | struct btrfs_map_token token; |
5f39d397 | 3614 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3615 | int slot; |
34a38218 | 3616 | u32 i; |
00ec4c51 CM |
3617 | int push_space = 0; |
3618 | int push_items = 0; | |
0783fcfc | 3619 | struct btrfs_item *item; |
34a38218 | 3620 | u32 nr; |
7518a238 | 3621 | u32 right_nritems; |
5f39d397 | 3622 | u32 data_end; |
db94535d | 3623 | u32 this_item_size; |
00ec4c51 | 3624 | |
cfed81a0 CM |
3625 | btrfs_init_map_token(&token); |
3626 | ||
34a38218 CM |
3627 | if (empty) |
3628 | nr = 0; | |
3629 | else | |
99d8f83c | 3630 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3631 | |
31840ae1 | 3632 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3633 | push_space += data_size; |
31840ae1 | 3634 | |
44871b1b | 3635 | slot = path->slots[1]; |
34a38218 CM |
3636 | i = left_nritems - 1; |
3637 | while (i >= nr) { | |
dd3cc16b | 3638 | item = btrfs_item_nr(i); |
db94535d | 3639 | |
31840ae1 ZY |
3640 | if (!empty && push_items > 0) { |
3641 | if (path->slots[0] > i) | |
3642 | break; | |
3643 | if (path->slots[0] == i) { | |
3644 | int space = btrfs_leaf_free_space(root, left); | |
3645 | if (space + push_space * 2 > free_space) | |
3646 | break; | |
3647 | } | |
3648 | } | |
3649 | ||
00ec4c51 | 3650 | if (path->slots[0] == i) |
87b29b20 | 3651 | push_space += data_size; |
db94535d | 3652 | |
db94535d CM |
3653 | this_item_size = btrfs_item_size(left, item); |
3654 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3655 | break; |
31840ae1 | 3656 | |
00ec4c51 | 3657 | push_items++; |
db94535d | 3658 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3659 | if (i == 0) |
3660 | break; | |
3661 | i--; | |
db94535d | 3662 | } |
5f39d397 | 3663 | |
925baedd CM |
3664 | if (push_items == 0) |
3665 | goto out_unlock; | |
5f39d397 | 3666 | |
6c1500f2 | 3667 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3668 | |
00ec4c51 | 3669 | /* push left to right */ |
5f39d397 | 3670 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3671 | |
5f39d397 | 3672 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3673 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3674 | |
00ec4c51 | 3675 | /* make room in the right data area */ |
5f39d397 CM |
3676 | data_end = leaf_data_end(root, right); |
3677 | memmove_extent_buffer(right, | |
3678 | btrfs_leaf_data(right) + data_end - push_space, | |
3679 | btrfs_leaf_data(right) + data_end, | |
3680 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3681 | ||
00ec4c51 | 3682 | /* copy from the left data area */ |
5f39d397 | 3683 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3684 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3685 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3686 | push_space); | |
5f39d397 CM |
3687 | |
3688 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3689 | btrfs_item_nr_offset(0), | |
3690 | right_nritems * sizeof(struct btrfs_item)); | |
3691 | ||
00ec4c51 | 3692 | /* copy the items from left to right */ |
5f39d397 CM |
3693 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3694 | btrfs_item_nr_offset(left_nritems - push_items), | |
3695 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3696 | |
3697 | /* update the item pointers */ | |
7518a238 | 3698 | right_nritems += push_items; |
5f39d397 | 3699 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3700 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3701 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3702 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3703 | push_space -= btrfs_token_item_size(right, item, &token); |
3704 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3705 | } |
3706 | ||
7518a238 | 3707 | left_nritems -= push_items; |
5f39d397 | 3708 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3709 | |
34a38218 CM |
3710 | if (left_nritems) |
3711 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3712 | else |
01d58472 | 3713 | clean_tree_block(trans, root->fs_info, left); |
f0486c68 | 3714 | |
5f39d397 | 3715 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3716 | |
5f39d397 CM |
3717 | btrfs_item_key(right, &disk_key, 0); |
3718 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3719 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3720 | |
00ec4c51 | 3721 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3722 | if (path->slots[0] >= left_nritems) { |
3723 | path->slots[0] -= left_nritems; | |
925baedd | 3724 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
01d58472 | 3725 | clean_tree_block(trans, root->fs_info, path->nodes[0]); |
925baedd | 3726 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3727 | free_extent_buffer(path->nodes[0]); |
3728 | path->nodes[0] = right; | |
00ec4c51 CM |
3729 | path->slots[1] += 1; |
3730 | } else { | |
925baedd | 3731 | btrfs_tree_unlock(right); |
5f39d397 | 3732 | free_extent_buffer(right); |
00ec4c51 CM |
3733 | } |
3734 | return 0; | |
925baedd CM |
3735 | |
3736 | out_unlock: | |
3737 | btrfs_tree_unlock(right); | |
3738 | free_extent_buffer(right); | |
3739 | return 1; | |
00ec4c51 | 3740 | } |
925baedd | 3741 | |
44871b1b CM |
3742 | /* |
3743 | * push some data in the path leaf to the right, trying to free up at | |
3744 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3745 | * | |
3746 | * returns 1 if the push failed because the other node didn't have enough | |
3747 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3748 | * |
3749 | * this will push starting from min_slot to the end of the leaf. It won't | |
3750 | * push any slot lower than min_slot | |
44871b1b CM |
3751 | */ |
3752 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3753 | *root, struct btrfs_path *path, |
3754 | int min_data_size, int data_size, | |
3755 | int empty, u32 min_slot) | |
44871b1b CM |
3756 | { |
3757 | struct extent_buffer *left = path->nodes[0]; | |
3758 | struct extent_buffer *right; | |
3759 | struct extent_buffer *upper; | |
3760 | int slot; | |
3761 | int free_space; | |
3762 | u32 left_nritems; | |
3763 | int ret; | |
3764 | ||
3765 | if (!path->nodes[1]) | |
3766 | return 1; | |
3767 | ||
3768 | slot = path->slots[1]; | |
3769 | upper = path->nodes[1]; | |
3770 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3771 | return 1; | |
3772 | ||
3773 | btrfs_assert_tree_locked(path->nodes[1]); | |
3774 | ||
3775 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3776 | if (right == NULL) |
3777 | return 1; | |
3778 | ||
44871b1b CM |
3779 | btrfs_tree_lock(right); |
3780 | btrfs_set_lock_blocking(right); | |
3781 | ||
3782 | free_space = btrfs_leaf_free_space(root, right); | |
3783 | if (free_space < data_size) | |
3784 | goto out_unlock; | |
3785 | ||
3786 | /* cow and double check */ | |
3787 | ret = btrfs_cow_block(trans, root, right, upper, | |
3788 | slot + 1, &right); | |
3789 | if (ret) | |
3790 | goto out_unlock; | |
3791 | ||
3792 | free_space = btrfs_leaf_free_space(root, right); | |
3793 | if (free_space < data_size) | |
3794 | goto out_unlock; | |
3795 | ||
3796 | left_nritems = btrfs_header_nritems(left); | |
3797 | if (left_nritems == 0) | |
3798 | goto out_unlock; | |
3799 | ||
2ef1fed2 FDBM |
3800 | if (path->slots[0] == left_nritems && !empty) { |
3801 | /* Key greater than all keys in the leaf, right neighbor has | |
3802 | * enough room for it and we're not emptying our leaf to delete | |
3803 | * it, therefore use right neighbor to insert the new item and | |
3804 | * no need to touch/dirty our left leaft. */ | |
3805 | btrfs_tree_unlock(left); | |
3806 | free_extent_buffer(left); | |
3807 | path->nodes[0] = right; | |
3808 | path->slots[0] = 0; | |
3809 | path->slots[1]++; | |
3810 | return 0; | |
3811 | } | |
3812 | ||
99d8f83c CM |
3813 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3814 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3815 | out_unlock: |
3816 | btrfs_tree_unlock(right); | |
3817 | free_extent_buffer(right); | |
3818 | return 1; | |
3819 | } | |
3820 | ||
74123bd7 CM |
3821 | /* |
3822 | * push some data in the path leaf to the left, trying to free up at | |
3823 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3824 | * |
3825 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3826 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3827 | * items | |
74123bd7 | 3828 | */ |
44871b1b CM |
3829 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3830 | struct btrfs_root *root, | |
3831 | struct btrfs_path *path, int data_size, | |
3832 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3833 | int free_space, u32 right_nritems, |
3834 | u32 max_slot) | |
be0e5c09 | 3835 | { |
5f39d397 CM |
3836 | struct btrfs_disk_key disk_key; |
3837 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3838 | int i; |
be0e5c09 CM |
3839 | int push_space = 0; |
3840 | int push_items = 0; | |
0783fcfc | 3841 | struct btrfs_item *item; |
7518a238 | 3842 | u32 old_left_nritems; |
34a38218 | 3843 | u32 nr; |
aa5d6bed | 3844 | int ret = 0; |
db94535d CM |
3845 | u32 this_item_size; |
3846 | u32 old_left_item_size; | |
cfed81a0 CM |
3847 | struct btrfs_map_token token; |
3848 | ||
3849 | btrfs_init_map_token(&token); | |
be0e5c09 | 3850 | |
34a38218 | 3851 | if (empty) |
99d8f83c | 3852 | nr = min(right_nritems, max_slot); |
34a38218 | 3853 | else |
99d8f83c | 3854 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3855 | |
3856 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3857 | item = btrfs_item_nr(i); |
db94535d | 3858 | |
31840ae1 ZY |
3859 | if (!empty && push_items > 0) { |
3860 | if (path->slots[0] < i) | |
3861 | break; | |
3862 | if (path->slots[0] == i) { | |
3863 | int space = btrfs_leaf_free_space(root, right); | |
3864 | if (space + push_space * 2 > free_space) | |
3865 | break; | |
3866 | } | |
3867 | } | |
3868 | ||
be0e5c09 | 3869 | if (path->slots[0] == i) |
87b29b20 | 3870 | push_space += data_size; |
db94535d CM |
3871 | |
3872 | this_item_size = btrfs_item_size(right, item); | |
3873 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3874 | break; |
db94535d | 3875 | |
be0e5c09 | 3876 | push_items++; |
db94535d CM |
3877 | push_space += this_item_size + sizeof(*item); |
3878 | } | |
3879 | ||
be0e5c09 | 3880 | if (push_items == 0) { |
925baedd CM |
3881 | ret = 1; |
3882 | goto out; | |
be0e5c09 | 3883 | } |
fae7f21c | 3884 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3885 | |
be0e5c09 | 3886 | /* push data from right to left */ |
5f39d397 CM |
3887 | copy_extent_buffer(left, right, |
3888 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3889 | btrfs_item_nr_offset(0), | |
3890 | push_items * sizeof(struct btrfs_item)); | |
3891 | ||
123abc88 | 3892 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3893 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3894 | |
3895 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3896 | leaf_data_end(root, left) - push_space, |
3897 | btrfs_leaf_data(right) + | |
5f39d397 | 3898 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3899 | push_space); |
5f39d397 | 3900 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3901 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3902 | |
db94535d | 3903 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3904 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3905 | u32 ioff; |
db94535d | 3906 | |
dd3cc16b | 3907 | item = btrfs_item_nr(i); |
db94535d | 3908 | |
cfed81a0 CM |
3909 | ioff = btrfs_token_item_offset(left, item, &token); |
3910 | btrfs_set_token_item_offset(left, item, | |
3911 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3912 | &token); | |
be0e5c09 | 3913 | } |
5f39d397 | 3914 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3915 | |
3916 | /* fixup right node */ | |
31b1a2bd JL |
3917 | if (push_items > right_nritems) |
3918 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3919 | right_nritems); |
34a38218 CM |
3920 | |
3921 | if (push_items < right_nritems) { | |
3922 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3923 | leaf_data_end(root, right); | |
3924 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3925 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3926 | btrfs_leaf_data(right) + | |
3927 | leaf_data_end(root, right), push_space); | |
3928 | ||
3929 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3930 | btrfs_item_nr_offset(push_items), |
3931 | (btrfs_header_nritems(right) - push_items) * | |
3932 | sizeof(struct btrfs_item)); | |
34a38218 | 3933 | } |
eef1c494 Y |
3934 | right_nritems -= push_items; |
3935 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3936 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 3937 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3938 | item = btrfs_item_nr(i); |
db94535d | 3939 | |
cfed81a0 CM |
3940 | push_space = push_space - btrfs_token_item_size(right, |
3941 | item, &token); | |
3942 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3943 | } |
eb60ceac | 3944 | |
5f39d397 | 3945 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3946 | if (right_nritems) |
3947 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3948 | else |
01d58472 | 3949 | clean_tree_block(trans, root->fs_info, right); |
098f59c2 | 3950 | |
5f39d397 | 3951 | btrfs_item_key(right, &disk_key, 0); |
b7a0365e | 3952 | fixup_low_keys(root->fs_info, path, &disk_key, 1); |
be0e5c09 CM |
3953 | |
3954 | /* then fixup the leaf pointer in the path */ | |
3955 | if (path->slots[0] < push_items) { | |
3956 | path->slots[0] += old_left_nritems; | |
925baedd | 3957 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3958 | free_extent_buffer(path->nodes[0]); |
3959 | path->nodes[0] = left; | |
be0e5c09 CM |
3960 | path->slots[1] -= 1; |
3961 | } else { | |
925baedd | 3962 | btrfs_tree_unlock(left); |
5f39d397 | 3963 | free_extent_buffer(left); |
be0e5c09 CM |
3964 | path->slots[0] -= push_items; |
3965 | } | |
eb60ceac | 3966 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3967 | return ret; |
925baedd CM |
3968 | out: |
3969 | btrfs_tree_unlock(left); | |
3970 | free_extent_buffer(left); | |
3971 | return ret; | |
be0e5c09 CM |
3972 | } |
3973 | ||
44871b1b CM |
3974 | /* |
3975 | * push some data in the path leaf to the left, trying to free up at | |
3976 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3977 | * |
3978 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3979 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3980 | * items | |
44871b1b CM |
3981 | */ |
3982 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3983 | *root, struct btrfs_path *path, int min_data_size, |
3984 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3985 | { |
3986 | struct extent_buffer *right = path->nodes[0]; | |
3987 | struct extent_buffer *left; | |
3988 | int slot; | |
3989 | int free_space; | |
3990 | u32 right_nritems; | |
3991 | int ret = 0; | |
3992 | ||
3993 | slot = path->slots[1]; | |
3994 | if (slot == 0) | |
3995 | return 1; | |
3996 | if (!path->nodes[1]) | |
3997 | return 1; | |
3998 | ||
3999 | right_nritems = btrfs_header_nritems(right); | |
4000 | if (right_nritems == 0) | |
4001 | return 1; | |
4002 | ||
4003 | btrfs_assert_tree_locked(path->nodes[1]); | |
4004 | ||
4005 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
4006 | if (left == NULL) |
4007 | return 1; | |
4008 | ||
44871b1b CM |
4009 | btrfs_tree_lock(left); |
4010 | btrfs_set_lock_blocking(left); | |
4011 | ||
4012 | free_space = btrfs_leaf_free_space(root, left); | |
4013 | if (free_space < data_size) { | |
4014 | ret = 1; | |
4015 | goto out; | |
4016 | } | |
4017 | ||
4018 | /* cow and double check */ | |
4019 | ret = btrfs_cow_block(trans, root, left, | |
4020 | path->nodes[1], slot - 1, &left); | |
4021 | if (ret) { | |
4022 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4023 | if (ret == -ENOSPC) |
4024 | ret = 1; | |
44871b1b CM |
4025 | goto out; |
4026 | } | |
4027 | ||
4028 | free_space = btrfs_leaf_free_space(root, left); | |
4029 | if (free_space < data_size) { | |
4030 | ret = 1; | |
4031 | goto out; | |
4032 | } | |
4033 | ||
99d8f83c CM |
4034 | return __push_leaf_left(trans, root, path, min_data_size, |
4035 | empty, left, free_space, right_nritems, | |
4036 | max_slot); | |
44871b1b CM |
4037 | out: |
4038 | btrfs_tree_unlock(left); | |
4039 | free_extent_buffer(left); | |
4040 | return ret; | |
4041 | } | |
4042 | ||
4043 | /* | |
4044 | * split the path's leaf in two, making sure there is at least data_size | |
4045 | * available for the resulting leaf level of the path. | |
44871b1b | 4046 | */ |
143bede5 JM |
4047 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
4048 | struct btrfs_root *root, | |
4049 | struct btrfs_path *path, | |
4050 | struct extent_buffer *l, | |
4051 | struct extent_buffer *right, | |
4052 | int slot, int mid, int nritems) | |
44871b1b CM |
4053 | { |
4054 | int data_copy_size; | |
4055 | int rt_data_off; | |
4056 | int i; | |
44871b1b | 4057 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4058 | struct btrfs_map_token token; |
4059 | ||
4060 | btrfs_init_map_token(&token); | |
44871b1b CM |
4061 | |
4062 | nritems = nritems - mid; | |
4063 | btrfs_set_header_nritems(right, nritems); | |
4064 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
4065 | ||
4066 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4067 | btrfs_item_nr_offset(mid), | |
4068 | nritems * sizeof(struct btrfs_item)); | |
4069 | ||
4070 | copy_extent_buffer(right, l, | |
4071 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
4072 | data_copy_size, btrfs_leaf_data(l) + | |
4073 | leaf_data_end(root, l), data_copy_size); | |
4074 | ||
4075 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
4076 | btrfs_item_end_nr(l, mid); | |
4077 | ||
4078 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4079 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4080 | u32 ioff; |
4081 | ||
cfed81a0 CM |
4082 | ioff = btrfs_token_item_offset(right, item, &token); |
4083 | btrfs_set_token_item_offset(right, item, | |
4084 | ioff + rt_data_off, &token); | |
44871b1b CM |
4085 | } |
4086 | ||
44871b1b | 4087 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4088 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 4089 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4090 | path->slots[1] + 1, 1); |
44871b1b CM |
4091 | |
4092 | btrfs_mark_buffer_dirty(right); | |
4093 | btrfs_mark_buffer_dirty(l); | |
4094 | BUG_ON(path->slots[0] != slot); | |
4095 | ||
44871b1b CM |
4096 | if (mid <= slot) { |
4097 | btrfs_tree_unlock(path->nodes[0]); | |
4098 | free_extent_buffer(path->nodes[0]); | |
4099 | path->nodes[0] = right; | |
4100 | path->slots[0] -= mid; | |
4101 | path->slots[1] += 1; | |
4102 | } else { | |
4103 | btrfs_tree_unlock(right); | |
4104 | free_extent_buffer(right); | |
4105 | } | |
4106 | ||
4107 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4108 | } |
4109 | ||
99d8f83c CM |
4110 | /* |
4111 | * double splits happen when we need to insert a big item in the middle | |
4112 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4113 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4114 | * A B C | |
4115 | * | |
4116 | * We avoid this by trying to push the items on either side of our target | |
4117 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4118 | * completely. | |
4119 | */ | |
4120 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4121 | struct btrfs_root *root, | |
4122 | struct btrfs_path *path, | |
4123 | int data_size) | |
4124 | { | |
4125 | int ret; | |
4126 | int progress = 0; | |
4127 | int slot; | |
4128 | u32 nritems; | |
5a4267ca | 4129 | int space_needed = data_size; |
99d8f83c CM |
4130 | |
4131 | slot = path->slots[0]; | |
5a4267ca FDBM |
4132 | if (slot < btrfs_header_nritems(path->nodes[0])) |
4133 | space_needed -= btrfs_leaf_free_space(root, path->nodes[0]); | |
99d8f83c CM |
4134 | |
4135 | /* | |
4136 | * try to push all the items after our slot into the | |
4137 | * right leaf | |
4138 | */ | |
5a4267ca | 4139 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4140 | if (ret < 0) |
4141 | return ret; | |
4142 | ||
4143 | if (ret == 0) | |
4144 | progress++; | |
4145 | ||
4146 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4147 | /* | |
4148 | * our goal is to get our slot at the start or end of a leaf. If | |
4149 | * we've done so we're done | |
4150 | */ | |
4151 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4152 | return 0; | |
4153 | ||
4154 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4155 | return 0; | |
4156 | ||
4157 | /* try to push all the items before our slot into the next leaf */ | |
4158 | slot = path->slots[0]; | |
5a4267ca | 4159 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4160 | if (ret < 0) |
4161 | return ret; | |
4162 | ||
4163 | if (ret == 0) | |
4164 | progress++; | |
4165 | ||
4166 | if (progress) | |
4167 | return 0; | |
4168 | return 1; | |
4169 | } | |
4170 | ||
74123bd7 CM |
4171 | /* |
4172 | * split the path's leaf in two, making sure there is at least data_size | |
4173 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4174 | * |
4175 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4176 | */ |
e02119d5 CM |
4177 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4178 | struct btrfs_root *root, | |
4179 | struct btrfs_key *ins_key, | |
4180 | struct btrfs_path *path, int data_size, | |
4181 | int extend) | |
be0e5c09 | 4182 | { |
5d4f98a2 | 4183 | struct btrfs_disk_key disk_key; |
5f39d397 | 4184 | struct extent_buffer *l; |
7518a238 | 4185 | u32 nritems; |
eb60ceac CM |
4186 | int mid; |
4187 | int slot; | |
5f39d397 | 4188 | struct extent_buffer *right; |
b7a0365e | 4189 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4190 | int ret = 0; |
aa5d6bed | 4191 | int wret; |
5d4f98a2 | 4192 | int split; |
cc0c5538 | 4193 | int num_doubles = 0; |
99d8f83c | 4194 | int tried_avoid_double = 0; |
aa5d6bed | 4195 | |
a5719521 YZ |
4196 | l = path->nodes[0]; |
4197 | slot = path->slots[0]; | |
4198 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
4199 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
4200 | return -EOVERFLOW; | |
4201 | ||
40689478 | 4202 | /* first try to make some room by pushing left and right */ |
33157e05 | 4203 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4204 | int space_needed = data_size; |
4205 | ||
4206 | if (slot < btrfs_header_nritems(l)) | |
4207 | space_needed -= btrfs_leaf_free_space(root, l); | |
4208 | ||
4209 | wret = push_leaf_right(trans, root, path, space_needed, | |
4210 | space_needed, 0, 0); | |
d397712b | 4211 | if (wret < 0) |
eaee50e8 | 4212 | return wret; |
3685f791 | 4213 | if (wret) { |
5a4267ca FDBM |
4214 | wret = push_leaf_left(trans, root, path, space_needed, |
4215 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4216 | if (wret < 0) |
4217 | return wret; | |
4218 | } | |
4219 | l = path->nodes[0]; | |
aa5d6bed | 4220 | |
3685f791 | 4221 | /* did the pushes work? */ |
87b29b20 | 4222 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 4223 | return 0; |
3326d1b0 | 4224 | } |
aa5d6bed | 4225 | |
5c680ed6 | 4226 | if (!path->nodes[1]) { |
fdd99c72 | 4227 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4228 | if (ret) |
4229 | return ret; | |
4230 | } | |
cc0c5538 | 4231 | again: |
5d4f98a2 | 4232 | split = 1; |
cc0c5538 | 4233 | l = path->nodes[0]; |
eb60ceac | 4234 | slot = path->slots[0]; |
5f39d397 | 4235 | nritems = btrfs_header_nritems(l); |
d397712b | 4236 | mid = (nritems + 1) / 2; |
54aa1f4d | 4237 | |
5d4f98a2 YZ |
4238 | if (mid <= slot) { |
4239 | if (nritems == 1 || | |
4240 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
4241 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4242 | if (slot >= nritems) { | |
4243 | split = 0; | |
4244 | } else { | |
4245 | mid = slot; | |
4246 | if (mid != nritems && | |
4247 | leaf_space_used(l, mid, nritems - mid) + | |
4248 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4249 | if (data_size && !tried_avoid_double) |
4250 | goto push_for_double; | |
5d4f98a2 YZ |
4251 | split = 2; |
4252 | } | |
4253 | } | |
4254 | } | |
4255 | } else { | |
4256 | if (leaf_space_used(l, 0, mid) + data_size > | |
4257 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4258 | if (!extend && data_size && slot == 0) { | |
4259 | split = 0; | |
4260 | } else if ((extend || !data_size) && slot == 0) { | |
4261 | mid = 1; | |
4262 | } else { | |
4263 | mid = slot; | |
4264 | if (mid != nritems && | |
4265 | leaf_space_used(l, mid, nritems - mid) + | |
4266 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4267 | if (data_size && !tried_avoid_double) |
4268 | goto push_for_double; | |
67871254 | 4269 | split = 2; |
5d4f98a2 YZ |
4270 | } |
4271 | } | |
4272 | } | |
4273 | } | |
4274 | ||
4275 | if (split == 0) | |
4276 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4277 | else | |
4278 | btrfs_item_key(l, &disk_key, mid); | |
4279 | ||
4d75f8a9 DS |
4280 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4281 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4282 | if (IS_ERR(right)) |
5f39d397 | 4283 | return PTR_ERR(right); |
f0486c68 | 4284 | |
707e8a07 | 4285 | root_add_used(root, root->nodesize); |
5f39d397 CM |
4286 | |
4287 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 4288 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4289 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4290 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4291 | btrfs_set_header_owner(right, root->root_key.objectid); |
4292 | btrfs_set_header_level(right, 0); | |
b7a0365e | 4293 | write_extent_buffer(right, fs_info->fsid, |
0a4e5586 | 4294 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 | 4295 | |
b7a0365e | 4296 | write_extent_buffer(right, fs_info->chunk_tree_uuid, |
b308bc2f | 4297 | btrfs_header_chunk_tree_uuid(right), |
e17cade2 | 4298 | BTRFS_UUID_SIZE); |
44871b1b | 4299 | |
5d4f98a2 YZ |
4300 | if (split == 0) { |
4301 | if (mid <= slot) { | |
4302 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4303 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4304 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4305 | btrfs_tree_unlock(path->nodes[0]); |
4306 | free_extent_buffer(path->nodes[0]); | |
4307 | path->nodes[0] = right; | |
4308 | path->slots[0] = 0; | |
4309 | path->slots[1] += 1; | |
4310 | } else { | |
4311 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4312 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4313 | path->slots[1], 1); |
5d4f98a2 YZ |
4314 | btrfs_tree_unlock(path->nodes[0]); |
4315 | free_extent_buffer(path->nodes[0]); | |
4316 | path->nodes[0] = right; | |
4317 | path->slots[0] = 0; | |
143bede5 | 4318 | if (path->slots[1] == 0) |
b7a0365e | 4319 | fixup_low_keys(fs_info, path, &disk_key, 1); |
d4dbff95 | 4320 | } |
5d4f98a2 YZ |
4321 | btrfs_mark_buffer_dirty(right); |
4322 | return ret; | |
d4dbff95 | 4323 | } |
74123bd7 | 4324 | |
143bede5 | 4325 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4326 | |
5d4f98a2 | 4327 | if (split == 2) { |
cc0c5538 CM |
4328 | BUG_ON(num_doubles != 0); |
4329 | num_doubles++; | |
4330 | goto again; | |
a429e513 | 4331 | } |
44871b1b | 4332 | |
143bede5 | 4333 | return 0; |
99d8f83c CM |
4334 | |
4335 | push_for_double: | |
4336 | push_for_double_split(trans, root, path, data_size); | |
4337 | tried_avoid_double = 1; | |
4338 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4339 | return 0; | |
4340 | goto again; | |
be0e5c09 CM |
4341 | } |
4342 | ||
ad48fd75 YZ |
4343 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4344 | struct btrfs_root *root, | |
4345 | struct btrfs_path *path, int ins_len) | |
459931ec | 4346 | { |
ad48fd75 | 4347 | struct btrfs_key key; |
459931ec | 4348 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4349 | struct btrfs_file_extent_item *fi; |
4350 | u64 extent_len = 0; | |
4351 | u32 item_size; | |
4352 | int ret; | |
459931ec CM |
4353 | |
4354 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4355 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4356 | ||
4357 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4358 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4359 | ||
4360 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4361 | return 0; | |
459931ec CM |
4362 | |
4363 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4364 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4365 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4366 | struct btrfs_file_extent_item); | |
4367 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4368 | } | |
b3b4aa74 | 4369 | btrfs_release_path(path); |
459931ec | 4370 | |
459931ec | 4371 | path->keep_locks = 1; |
ad48fd75 YZ |
4372 | path->search_for_split = 1; |
4373 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4374 | path->search_for_split = 0; |
a8df6fe6 FM |
4375 | if (ret > 0) |
4376 | ret = -EAGAIN; | |
ad48fd75 YZ |
4377 | if (ret < 0) |
4378 | goto err; | |
459931ec | 4379 | |
ad48fd75 YZ |
4380 | ret = -EAGAIN; |
4381 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4382 | /* if our item isn't there, return now */ |
4383 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4384 | goto err; |
4385 | ||
109f6aef CM |
4386 | /* the leaf has changed, it now has room. return now */ |
4387 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4388 | goto err; | |
4389 | ||
ad48fd75 YZ |
4390 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4391 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4392 | struct btrfs_file_extent_item); | |
4393 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4394 | goto err; | |
459931ec CM |
4395 | } |
4396 | ||
b9473439 | 4397 | btrfs_set_path_blocking(path); |
ad48fd75 | 4398 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4399 | if (ret) |
4400 | goto err; | |
459931ec | 4401 | |
ad48fd75 | 4402 | path->keep_locks = 0; |
b9473439 | 4403 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4404 | return 0; |
4405 | err: | |
4406 | path->keep_locks = 0; | |
4407 | return ret; | |
4408 | } | |
4409 | ||
4410 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4411 | struct btrfs_root *root, | |
4412 | struct btrfs_path *path, | |
4413 | struct btrfs_key *new_key, | |
4414 | unsigned long split_offset) | |
4415 | { | |
4416 | struct extent_buffer *leaf; | |
4417 | struct btrfs_item *item; | |
4418 | struct btrfs_item *new_item; | |
4419 | int slot; | |
4420 | char *buf; | |
4421 | u32 nritems; | |
4422 | u32 item_size; | |
4423 | u32 orig_offset; | |
4424 | struct btrfs_disk_key disk_key; | |
4425 | ||
b9473439 CM |
4426 | leaf = path->nodes[0]; |
4427 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4428 | ||
b4ce94de CM |
4429 | btrfs_set_path_blocking(path); |
4430 | ||
dd3cc16b | 4431 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4432 | orig_offset = btrfs_item_offset(leaf, item); |
4433 | item_size = btrfs_item_size(leaf, item); | |
4434 | ||
459931ec | 4435 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4436 | if (!buf) |
4437 | return -ENOMEM; | |
4438 | ||
459931ec CM |
4439 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4440 | path->slots[0]), item_size); | |
459931ec | 4441 | |
ad48fd75 | 4442 | slot = path->slots[0] + 1; |
459931ec | 4443 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4444 | if (slot != nritems) { |
4445 | /* shift the items */ | |
4446 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4447 | btrfs_item_nr_offset(slot), |
4448 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4449 | } |
4450 | ||
4451 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4452 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4453 | ||
dd3cc16b | 4454 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4455 | |
4456 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4457 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4458 | ||
4459 | btrfs_set_item_offset(leaf, item, | |
4460 | orig_offset + item_size - split_offset); | |
4461 | btrfs_set_item_size(leaf, item, split_offset); | |
4462 | ||
4463 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4464 | ||
4465 | /* write the data for the start of the original item */ | |
4466 | write_extent_buffer(leaf, buf, | |
4467 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4468 | split_offset); | |
4469 | ||
4470 | /* write the data for the new item */ | |
4471 | write_extent_buffer(leaf, buf + split_offset, | |
4472 | btrfs_item_ptr_offset(leaf, slot), | |
4473 | item_size - split_offset); | |
4474 | btrfs_mark_buffer_dirty(leaf); | |
4475 | ||
ad48fd75 | 4476 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4477 | kfree(buf); |
ad48fd75 YZ |
4478 | return 0; |
4479 | } | |
4480 | ||
4481 | /* | |
4482 | * This function splits a single item into two items, | |
4483 | * giving 'new_key' to the new item and splitting the | |
4484 | * old one at split_offset (from the start of the item). | |
4485 | * | |
4486 | * The path may be released by this operation. After | |
4487 | * the split, the path is pointing to the old item. The | |
4488 | * new item is going to be in the same node as the old one. | |
4489 | * | |
4490 | * Note, the item being split must be smaller enough to live alone on | |
4491 | * a tree block with room for one extra struct btrfs_item | |
4492 | * | |
4493 | * This allows us to split the item in place, keeping a lock on the | |
4494 | * leaf the entire time. | |
4495 | */ | |
4496 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4497 | struct btrfs_root *root, | |
4498 | struct btrfs_path *path, | |
4499 | struct btrfs_key *new_key, | |
4500 | unsigned long split_offset) | |
4501 | { | |
4502 | int ret; | |
4503 | ret = setup_leaf_for_split(trans, root, path, | |
4504 | sizeof(struct btrfs_item)); | |
4505 | if (ret) | |
4506 | return ret; | |
4507 | ||
4508 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4509 | return ret; |
4510 | } | |
4511 | ||
ad48fd75 YZ |
4512 | /* |
4513 | * This function duplicate a item, giving 'new_key' to the new item. | |
4514 | * It guarantees both items live in the same tree leaf and the new item | |
4515 | * is contiguous with the original item. | |
4516 | * | |
4517 | * This allows us to split file extent in place, keeping a lock on the | |
4518 | * leaf the entire time. | |
4519 | */ | |
4520 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4521 | struct btrfs_root *root, | |
4522 | struct btrfs_path *path, | |
4523 | struct btrfs_key *new_key) | |
4524 | { | |
4525 | struct extent_buffer *leaf; | |
4526 | int ret; | |
4527 | u32 item_size; | |
4528 | ||
4529 | leaf = path->nodes[0]; | |
4530 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4531 | ret = setup_leaf_for_split(trans, root, path, | |
4532 | item_size + sizeof(struct btrfs_item)); | |
4533 | if (ret) | |
4534 | return ret; | |
4535 | ||
4536 | path->slots[0]++; | |
afe5fea7 | 4537 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4538 | item_size, item_size + |
4539 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4540 | leaf = path->nodes[0]; |
4541 | memcpy_extent_buffer(leaf, | |
4542 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4543 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4544 | item_size); | |
4545 | return 0; | |
4546 | } | |
4547 | ||
d352ac68 CM |
4548 | /* |
4549 | * make the item pointed to by the path smaller. new_size indicates | |
4550 | * how small to make it, and from_end tells us if we just chop bytes | |
4551 | * off the end of the item or if we shift the item to chop bytes off | |
4552 | * the front. | |
4553 | */ | |
afe5fea7 | 4554 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4555 | u32 new_size, int from_end) |
b18c6685 | 4556 | { |
b18c6685 | 4557 | int slot; |
5f39d397 CM |
4558 | struct extent_buffer *leaf; |
4559 | struct btrfs_item *item; | |
b18c6685 CM |
4560 | u32 nritems; |
4561 | unsigned int data_end; | |
4562 | unsigned int old_data_start; | |
4563 | unsigned int old_size; | |
4564 | unsigned int size_diff; | |
4565 | int i; | |
cfed81a0 CM |
4566 | struct btrfs_map_token token; |
4567 | ||
4568 | btrfs_init_map_token(&token); | |
b18c6685 | 4569 | |
5f39d397 | 4570 | leaf = path->nodes[0]; |
179e29e4 CM |
4571 | slot = path->slots[0]; |
4572 | ||
4573 | old_size = btrfs_item_size_nr(leaf, slot); | |
4574 | if (old_size == new_size) | |
143bede5 | 4575 | return; |
b18c6685 | 4576 | |
5f39d397 | 4577 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4578 | data_end = leaf_data_end(root, leaf); |
4579 | ||
5f39d397 | 4580 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4581 | |
b18c6685 CM |
4582 | size_diff = old_size - new_size; |
4583 | ||
4584 | BUG_ON(slot < 0); | |
4585 | BUG_ON(slot >= nritems); | |
4586 | ||
4587 | /* | |
4588 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4589 | */ | |
4590 | /* first correct the data pointers */ | |
4591 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4592 | u32 ioff; |
dd3cc16b | 4593 | item = btrfs_item_nr(i); |
db94535d | 4594 | |
cfed81a0 CM |
4595 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4596 | btrfs_set_token_item_offset(leaf, item, | |
4597 | ioff + size_diff, &token); | |
b18c6685 | 4598 | } |
db94535d | 4599 | |
b18c6685 | 4600 | /* shift the data */ |
179e29e4 CM |
4601 | if (from_end) { |
4602 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4603 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4604 | data_end, old_data_start + new_size - data_end); | |
4605 | } else { | |
4606 | struct btrfs_disk_key disk_key; | |
4607 | u64 offset; | |
4608 | ||
4609 | btrfs_item_key(leaf, &disk_key, slot); | |
4610 | ||
4611 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4612 | unsigned long ptr; | |
4613 | struct btrfs_file_extent_item *fi; | |
4614 | ||
4615 | fi = btrfs_item_ptr(leaf, slot, | |
4616 | struct btrfs_file_extent_item); | |
4617 | fi = (struct btrfs_file_extent_item *)( | |
4618 | (unsigned long)fi - size_diff); | |
4619 | ||
4620 | if (btrfs_file_extent_type(leaf, fi) == | |
4621 | BTRFS_FILE_EXTENT_INLINE) { | |
4622 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4623 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4624 | (unsigned long)fi, |
7ec20afb | 4625 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4626 | } |
4627 | } | |
4628 | ||
4629 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4630 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4631 | data_end, old_data_start - data_end); | |
4632 | ||
4633 | offset = btrfs_disk_key_offset(&disk_key); | |
4634 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4635 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4636 | if (slot == 0) | |
b7a0365e | 4637 | fixup_low_keys(root->fs_info, path, &disk_key, 1); |
179e29e4 | 4638 | } |
5f39d397 | 4639 | |
dd3cc16b | 4640 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4641 | btrfs_set_item_size(leaf, item, new_size); |
4642 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4643 | |
5f39d397 CM |
4644 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4645 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4646 | BUG(); |
5f39d397 | 4647 | } |
b18c6685 CM |
4648 | } |
4649 | ||
d352ac68 | 4650 | /* |
8f69dbd2 | 4651 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4652 | */ |
4b90c680 | 4653 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4654 | u32 data_size) |
6567e837 | 4655 | { |
6567e837 | 4656 | int slot; |
5f39d397 CM |
4657 | struct extent_buffer *leaf; |
4658 | struct btrfs_item *item; | |
6567e837 CM |
4659 | u32 nritems; |
4660 | unsigned int data_end; | |
4661 | unsigned int old_data; | |
4662 | unsigned int old_size; | |
4663 | int i; | |
cfed81a0 CM |
4664 | struct btrfs_map_token token; |
4665 | ||
4666 | btrfs_init_map_token(&token); | |
6567e837 | 4667 | |
5f39d397 | 4668 | leaf = path->nodes[0]; |
6567e837 | 4669 | |
5f39d397 | 4670 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4671 | data_end = leaf_data_end(root, leaf); |
4672 | ||
5f39d397 CM |
4673 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4674 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4675 | BUG(); |
5f39d397 | 4676 | } |
6567e837 | 4677 | slot = path->slots[0]; |
5f39d397 | 4678 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4679 | |
4680 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4681 | if (slot >= nritems) { |
4682 | btrfs_print_leaf(root, leaf); | |
efe120a0 | 4683 | btrfs_crit(root->fs_info, "slot %d too large, nritems %d", |
d397712b | 4684 | slot, nritems); |
3326d1b0 CM |
4685 | BUG_ON(1); |
4686 | } | |
6567e837 CM |
4687 | |
4688 | /* | |
4689 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4690 | */ | |
4691 | /* first correct the data pointers */ | |
4692 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4693 | u32 ioff; |
dd3cc16b | 4694 | item = btrfs_item_nr(i); |
db94535d | 4695 | |
cfed81a0 CM |
4696 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4697 | btrfs_set_token_item_offset(leaf, item, | |
4698 | ioff - data_size, &token); | |
6567e837 | 4699 | } |
5f39d397 | 4700 | |
6567e837 | 4701 | /* shift the data */ |
5f39d397 | 4702 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4703 | data_end - data_size, btrfs_leaf_data(leaf) + |
4704 | data_end, old_data - data_end); | |
5f39d397 | 4705 | |
6567e837 | 4706 | data_end = old_data; |
5f39d397 | 4707 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4708 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4709 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4710 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4711 | |
5f39d397 CM |
4712 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4713 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4714 | BUG(); |
5f39d397 | 4715 | } |
6567e837 CM |
4716 | } |
4717 | ||
74123bd7 | 4718 | /* |
44871b1b CM |
4719 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4720 | * to save stack depth by doing the bulk of the work in a function | |
4721 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4722 | */ |
afe5fea7 | 4723 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4724 | struct btrfs_key *cpu_key, u32 *data_size, |
4725 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4726 | { |
5f39d397 | 4727 | struct btrfs_item *item; |
9c58309d | 4728 | int i; |
7518a238 | 4729 | u32 nritems; |
be0e5c09 | 4730 | unsigned int data_end; |
e2fa7227 | 4731 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4732 | struct extent_buffer *leaf; |
4733 | int slot; | |
cfed81a0 CM |
4734 | struct btrfs_map_token token; |
4735 | ||
24cdc847 FM |
4736 | if (path->slots[0] == 0) { |
4737 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b7a0365e | 4738 | fixup_low_keys(root->fs_info, path, &disk_key, 1); |
24cdc847 FM |
4739 | } |
4740 | btrfs_unlock_up_safe(path, 1); | |
4741 | ||
cfed81a0 | 4742 | btrfs_init_map_token(&token); |
e2fa7227 | 4743 | |
5f39d397 | 4744 | leaf = path->nodes[0]; |
44871b1b | 4745 | slot = path->slots[0]; |
74123bd7 | 4746 | |
5f39d397 | 4747 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4748 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4749 | |
f25956cc | 4750 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4751 | btrfs_print_leaf(root, leaf); |
efe120a0 | 4752 | btrfs_crit(root->fs_info, "not enough freespace need %u have %d", |
9c58309d | 4753 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4754 | BUG(); |
d4dbff95 | 4755 | } |
5f39d397 | 4756 | |
be0e5c09 | 4757 | if (slot != nritems) { |
5f39d397 | 4758 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4759 | |
5f39d397 CM |
4760 | if (old_data < data_end) { |
4761 | btrfs_print_leaf(root, leaf); | |
efe120a0 | 4762 | btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d", |
5f39d397 CM |
4763 | slot, old_data, data_end); |
4764 | BUG_ON(1); | |
4765 | } | |
be0e5c09 CM |
4766 | /* |
4767 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4768 | */ | |
4769 | /* first correct the data pointers */ | |
0783fcfc | 4770 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4771 | u32 ioff; |
db94535d | 4772 | |
dd3cc16b | 4773 | item = btrfs_item_nr( i); |
cfed81a0 CM |
4774 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4775 | btrfs_set_token_item_offset(leaf, item, | |
4776 | ioff - total_data, &token); | |
0783fcfc | 4777 | } |
be0e5c09 | 4778 | /* shift the items */ |
9c58309d | 4779 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4780 | btrfs_item_nr_offset(slot), |
d6025579 | 4781 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4782 | |
4783 | /* shift the data */ | |
5f39d397 | 4784 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4785 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4786 | data_end, old_data - data_end); |
be0e5c09 CM |
4787 | data_end = old_data; |
4788 | } | |
5f39d397 | 4789 | |
62e2749e | 4790 | /* setup the item for the new data */ |
9c58309d CM |
4791 | for (i = 0; i < nr; i++) { |
4792 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4793 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4794 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4795 | btrfs_set_token_item_offset(leaf, item, |
4796 | data_end - data_size[i], &token); | |
9c58309d | 4797 | data_end -= data_size[i]; |
cfed81a0 | 4798 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4799 | } |
44871b1b | 4800 | |
9c58309d | 4801 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4802 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4803 | |
5f39d397 CM |
4804 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4805 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4806 | BUG(); |
5f39d397 | 4807 | } |
44871b1b CM |
4808 | } |
4809 | ||
4810 | /* | |
4811 | * Given a key and some data, insert items into the tree. | |
4812 | * This does all the path init required, making room in the tree if needed. | |
4813 | */ | |
4814 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4815 | struct btrfs_root *root, | |
4816 | struct btrfs_path *path, | |
4817 | struct btrfs_key *cpu_key, u32 *data_size, | |
4818 | int nr) | |
4819 | { | |
44871b1b CM |
4820 | int ret = 0; |
4821 | int slot; | |
4822 | int i; | |
4823 | u32 total_size = 0; | |
4824 | u32 total_data = 0; | |
4825 | ||
4826 | for (i = 0; i < nr; i++) | |
4827 | total_data += data_size[i]; | |
4828 | ||
4829 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4830 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4831 | if (ret == 0) | |
4832 | return -EEXIST; | |
4833 | if (ret < 0) | |
143bede5 | 4834 | return ret; |
44871b1b | 4835 | |
44871b1b CM |
4836 | slot = path->slots[0]; |
4837 | BUG_ON(slot < 0); | |
4838 | ||
afe5fea7 | 4839 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4840 | total_data, total_size, nr); |
143bede5 | 4841 | return 0; |
62e2749e CM |
4842 | } |
4843 | ||
4844 | /* | |
4845 | * Given a key and some data, insert an item into the tree. | |
4846 | * This does all the path init required, making room in the tree if needed. | |
4847 | */ | |
e089f05c CM |
4848 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4849 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4850 | data_size) | |
62e2749e CM |
4851 | { |
4852 | int ret = 0; | |
2c90e5d6 | 4853 | struct btrfs_path *path; |
5f39d397 CM |
4854 | struct extent_buffer *leaf; |
4855 | unsigned long ptr; | |
62e2749e | 4856 | |
2c90e5d6 | 4857 | path = btrfs_alloc_path(); |
db5b493a TI |
4858 | if (!path) |
4859 | return -ENOMEM; | |
2c90e5d6 | 4860 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4861 | if (!ret) { |
5f39d397 CM |
4862 | leaf = path->nodes[0]; |
4863 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4864 | write_extent_buffer(leaf, data, ptr, data_size); | |
4865 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4866 | } |
2c90e5d6 | 4867 | btrfs_free_path(path); |
aa5d6bed | 4868 | return ret; |
be0e5c09 CM |
4869 | } |
4870 | ||
74123bd7 | 4871 | /* |
5de08d7d | 4872 | * delete the pointer from a given node. |
74123bd7 | 4873 | * |
d352ac68 CM |
4874 | * the tree should have been previously balanced so the deletion does not |
4875 | * empty a node. | |
74123bd7 | 4876 | */ |
afe5fea7 TI |
4877 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4878 | int level, int slot) | |
be0e5c09 | 4879 | { |
5f39d397 | 4880 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4881 | u32 nritems; |
f3ea38da | 4882 | int ret; |
be0e5c09 | 4883 | |
5f39d397 | 4884 | nritems = btrfs_header_nritems(parent); |
d397712b | 4885 | if (slot != nritems - 1) { |
0e411ece | 4886 | if (level) |
f3ea38da JS |
4887 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4888 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4889 | memmove_extent_buffer(parent, |
4890 | btrfs_node_key_ptr_offset(slot), | |
4891 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4892 | sizeof(struct btrfs_key_ptr) * |
4893 | (nritems - slot - 1)); | |
57ba86c0 CM |
4894 | } else if (level) { |
4895 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
c8cc6341 | 4896 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4897 | BUG_ON(ret < 0); |
bb803951 | 4898 | } |
f3ea38da | 4899 | |
7518a238 | 4900 | nritems--; |
5f39d397 | 4901 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4902 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4903 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4904 | /* just turn the root into a leaf and break */ |
5f39d397 | 4905 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4906 | } else if (slot == 0) { |
5f39d397 CM |
4907 | struct btrfs_disk_key disk_key; |
4908 | ||
4909 | btrfs_node_key(parent, &disk_key, 0); | |
b7a0365e | 4910 | fixup_low_keys(root->fs_info, path, &disk_key, level + 1); |
be0e5c09 | 4911 | } |
d6025579 | 4912 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4913 | } |
4914 | ||
323ac95b CM |
4915 | /* |
4916 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4917 | * path->nodes[1]. |
323ac95b CM |
4918 | * |
4919 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4920 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4921 | * | |
4922 | * The path must have already been setup for deleting the leaf, including | |
4923 | * all the proper balancing. path->nodes[1] must be locked. | |
4924 | */ | |
143bede5 JM |
4925 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4926 | struct btrfs_root *root, | |
4927 | struct btrfs_path *path, | |
4928 | struct extent_buffer *leaf) | |
323ac95b | 4929 | { |
5d4f98a2 | 4930 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4931 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4932 | |
4d081c41 CM |
4933 | /* |
4934 | * btrfs_free_extent is expensive, we want to make sure we | |
4935 | * aren't holding any locks when we call it | |
4936 | */ | |
4937 | btrfs_unlock_up_safe(path, 0); | |
4938 | ||
f0486c68 YZ |
4939 | root_sub_used(root, leaf->len); |
4940 | ||
3083ee2e | 4941 | extent_buffer_get(leaf); |
5581a51a | 4942 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4943 | free_extent_buffer_stale(leaf); |
323ac95b | 4944 | } |
74123bd7 CM |
4945 | /* |
4946 | * delete the item at the leaf level in path. If that empties | |
4947 | * the leaf, remove it from the tree | |
4948 | */ | |
85e21bac CM |
4949 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4950 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4951 | { |
5f39d397 CM |
4952 | struct extent_buffer *leaf; |
4953 | struct btrfs_item *item; | |
ce0eac2a AM |
4954 | u32 last_off; |
4955 | u32 dsize = 0; | |
aa5d6bed CM |
4956 | int ret = 0; |
4957 | int wret; | |
85e21bac | 4958 | int i; |
7518a238 | 4959 | u32 nritems; |
cfed81a0 CM |
4960 | struct btrfs_map_token token; |
4961 | ||
4962 | btrfs_init_map_token(&token); | |
be0e5c09 | 4963 | |
5f39d397 | 4964 | leaf = path->nodes[0]; |
85e21bac CM |
4965 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4966 | ||
4967 | for (i = 0; i < nr; i++) | |
4968 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4969 | ||
5f39d397 | 4970 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4971 | |
85e21bac | 4972 | if (slot + nr != nritems) { |
123abc88 | 4973 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4974 | |
4975 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4976 | data_end + dsize, |
4977 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4978 | last_off - data_end); |
5f39d397 | 4979 | |
85e21bac | 4980 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4981 | u32 ioff; |
db94535d | 4982 | |
dd3cc16b | 4983 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4984 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4985 | btrfs_set_token_item_offset(leaf, item, | |
4986 | ioff + dsize, &token); | |
0783fcfc | 4987 | } |
db94535d | 4988 | |
5f39d397 | 4989 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4990 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4991 | sizeof(struct btrfs_item) * |
85e21bac | 4992 | (nritems - slot - nr)); |
be0e5c09 | 4993 | } |
85e21bac CM |
4994 | btrfs_set_header_nritems(leaf, nritems - nr); |
4995 | nritems -= nr; | |
5f39d397 | 4996 | |
74123bd7 | 4997 | /* delete the leaf if we've emptied it */ |
7518a238 | 4998 | if (nritems == 0) { |
5f39d397 CM |
4999 | if (leaf == root->node) { |
5000 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5001 | } else { |
f0486c68 | 5002 | btrfs_set_path_blocking(path); |
01d58472 | 5003 | clean_tree_block(trans, root->fs_info, leaf); |
143bede5 | 5004 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5005 | } |
be0e5c09 | 5006 | } else { |
7518a238 | 5007 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5008 | if (slot == 0) { |
5f39d397 CM |
5009 | struct btrfs_disk_key disk_key; |
5010 | ||
5011 | btrfs_item_key(leaf, &disk_key, 0); | |
b7a0365e | 5012 | fixup_low_keys(root->fs_info, path, &disk_key, 1); |
aa5d6bed | 5013 | } |
aa5d6bed | 5014 | |
74123bd7 | 5015 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 5016 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
5017 | /* push_leaf_left fixes the path. |
5018 | * make sure the path still points to our leaf | |
5019 | * for possible call to del_ptr below | |
5020 | */ | |
4920c9ac | 5021 | slot = path->slots[1]; |
5f39d397 CM |
5022 | extent_buffer_get(leaf); |
5023 | ||
b9473439 | 5024 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5025 | wret = push_leaf_left(trans, root, path, 1, 1, |
5026 | 1, (u32)-1); | |
54aa1f4d | 5027 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5028 | ret = wret; |
5f39d397 CM |
5029 | |
5030 | if (path->nodes[0] == leaf && | |
5031 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5032 | wret = push_leaf_right(trans, root, path, 1, |
5033 | 1, 1, 0); | |
54aa1f4d | 5034 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5035 | ret = wret; |
5036 | } | |
5f39d397 CM |
5037 | |
5038 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5039 | path->slots[1] = slot; |
143bede5 | 5040 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5041 | free_extent_buffer(leaf); |
143bede5 | 5042 | ret = 0; |
5de08d7d | 5043 | } else { |
925baedd CM |
5044 | /* if we're still in the path, make sure |
5045 | * we're dirty. Otherwise, one of the | |
5046 | * push_leaf functions must have already | |
5047 | * dirtied this buffer | |
5048 | */ | |
5049 | if (path->nodes[0] == leaf) | |
5050 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5051 | free_extent_buffer(leaf); |
be0e5c09 | 5052 | } |
d5719762 | 5053 | } else { |
5f39d397 | 5054 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5055 | } |
5056 | } | |
aa5d6bed | 5057 | return ret; |
be0e5c09 CM |
5058 | } |
5059 | ||
7bb86316 | 5060 | /* |
925baedd | 5061 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5062 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5063 | * returns < 0 on io errors. | |
d352ac68 CM |
5064 | * |
5065 | * This may release the path, and so you may lose any locks held at the | |
5066 | * time you call it. | |
7bb86316 | 5067 | */ |
16e7549f | 5068 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5069 | { |
925baedd CM |
5070 | struct btrfs_key key; |
5071 | struct btrfs_disk_key found_key; | |
5072 | int ret; | |
7bb86316 | 5073 | |
925baedd | 5074 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5075 | |
e8b0d724 | 5076 | if (key.offset > 0) { |
925baedd | 5077 | key.offset--; |
e8b0d724 | 5078 | } else if (key.type > 0) { |
925baedd | 5079 | key.type--; |
e8b0d724 FDBM |
5080 | key.offset = (u64)-1; |
5081 | } else if (key.objectid > 0) { | |
925baedd | 5082 | key.objectid--; |
e8b0d724 FDBM |
5083 | key.type = (u8)-1; |
5084 | key.offset = (u64)-1; | |
5085 | } else { | |
925baedd | 5086 | return 1; |
e8b0d724 | 5087 | } |
7bb86316 | 5088 | |
b3b4aa74 | 5089 | btrfs_release_path(path); |
925baedd CM |
5090 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5091 | if (ret < 0) | |
5092 | return ret; | |
5093 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5094 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5095 | /* |
5096 | * We might have had an item with the previous key in the tree right | |
5097 | * before we released our path. And after we released our path, that | |
5098 | * item might have been pushed to the first slot (0) of the leaf we | |
5099 | * were holding due to a tree balance. Alternatively, an item with the | |
5100 | * previous key can exist as the only element of a leaf (big fat item). | |
5101 | * Therefore account for these 2 cases, so that our callers (like | |
5102 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5103 | * the previous key we computed above. | |
5104 | */ | |
5105 | if (ret <= 0) | |
925baedd CM |
5106 | return 0; |
5107 | return 1; | |
7bb86316 CM |
5108 | } |
5109 | ||
3f157a2f CM |
5110 | /* |
5111 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5112 | * for nodes or leaves that are have a minimum transaction id. |
5113 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5114 | * |
5115 | * This does not cow, but it does stuff the starting key it finds back | |
5116 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5117 | * key and get a writable path. | |
5118 | * | |
5119 | * This does lock as it descends, and path->keep_locks should be set | |
5120 | * to 1 by the caller. | |
5121 | * | |
5122 | * This honors path->lowest_level to prevent descent past a given level | |
5123 | * of the tree. | |
5124 | * | |
d352ac68 CM |
5125 | * min_trans indicates the oldest transaction that you are interested |
5126 | * in walking through. Any nodes or leaves older than min_trans are | |
5127 | * skipped over (without reading them). | |
5128 | * | |
3f157a2f CM |
5129 | * returns zero if something useful was found, < 0 on error and 1 if there |
5130 | * was nothing in the tree that matched the search criteria. | |
5131 | */ | |
5132 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5133 | struct btrfs_path *path, |
3f157a2f CM |
5134 | u64 min_trans) |
5135 | { | |
5136 | struct extent_buffer *cur; | |
5137 | struct btrfs_key found_key; | |
5138 | int slot; | |
9652480b | 5139 | int sret; |
3f157a2f CM |
5140 | u32 nritems; |
5141 | int level; | |
5142 | int ret = 1; | |
f98de9b9 | 5143 | int keep_locks = path->keep_locks; |
3f157a2f | 5144 | |
f98de9b9 | 5145 | path->keep_locks = 1; |
3f157a2f | 5146 | again: |
bd681513 | 5147 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5148 | level = btrfs_header_level(cur); |
e02119d5 | 5149 | WARN_ON(path->nodes[level]); |
3f157a2f | 5150 | path->nodes[level] = cur; |
bd681513 | 5151 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5152 | |
5153 | if (btrfs_header_generation(cur) < min_trans) { | |
5154 | ret = 1; | |
5155 | goto out; | |
5156 | } | |
d397712b | 5157 | while (1) { |
3f157a2f CM |
5158 | nritems = btrfs_header_nritems(cur); |
5159 | level = btrfs_header_level(cur); | |
9652480b | 5160 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 5161 | |
323ac95b CM |
5162 | /* at the lowest level, we're done, setup the path and exit */ |
5163 | if (level == path->lowest_level) { | |
e02119d5 CM |
5164 | if (slot >= nritems) |
5165 | goto find_next_key; | |
3f157a2f CM |
5166 | ret = 0; |
5167 | path->slots[level] = slot; | |
5168 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5169 | goto out; | |
5170 | } | |
9652480b Y |
5171 | if (sret && slot > 0) |
5172 | slot--; | |
3f157a2f | 5173 | /* |
de78b51a ES |
5174 | * check this node pointer against the min_trans parameters. |
5175 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5176 | */ |
d397712b | 5177 | while (slot < nritems) { |
3f157a2f | 5178 | u64 gen; |
e02119d5 | 5179 | |
3f157a2f CM |
5180 | gen = btrfs_node_ptr_generation(cur, slot); |
5181 | if (gen < min_trans) { | |
5182 | slot++; | |
5183 | continue; | |
5184 | } | |
de78b51a | 5185 | break; |
3f157a2f | 5186 | } |
e02119d5 | 5187 | find_next_key: |
3f157a2f CM |
5188 | /* |
5189 | * we didn't find a candidate key in this node, walk forward | |
5190 | * and find another one | |
5191 | */ | |
5192 | if (slot >= nritems) { | |
e02119d5 | 5193 | path->slots[level] = slot; |
b4ce94de | 5194 | btrfs_set_path_blocking(path); |
e02119d5 | 5195 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5196 | min_trans); |
e02119d5 | 5197 | if (sret == 0) { |
b3b4aa74 | 5198 | btrfs_release_path(path); |
3f157a2f CM |
5199 | goto again; |
5200 | } else { | |
5201 | goto out; | |
5202 | } | |
5203 | } | |
5204 | /* save our key for returning back */ | |
5205 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5206 | path->slots[level] = slot; | |
5207 | if (level == path->lowest_level) { | |
5208 | ret = 0; | |
3f157a2f CM |
5209 | goto out; |
5210 | } | |
b4ce94de | 5211 | btrfs_set_path_blocking(path); |
3f157a2f | 5212 | cur = read_node_slot(root, cur, slot); |
79787eaa | 5213 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 5214 | |
bd681513 | 5215 | btrfs_tree_read_lock(cur); |
b4ce94de | 5216 | |
bd681513 | 5217 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5218 | path->nodes[level - 1] = cur; |
f7c79f30 | 5219 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5220 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5221 | } |
5222 | out: | |
f98de9b9 FM |
5223 | path->keep_locks = keep_locks; |
5224 | if (ret == 0) { | |
5225 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5226 | btrfs_set_path_blocking(path); | |
3f157a2f | 5227 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5228 | } |
3f157a2f CM |
5229 | return ret; |
5230 | } | |
5231 | ||
7069830a AB |
5232 | static void tree_move_down(struct btrfs_root *root, |
5233 | struct btrfs_path *path, | |
5234 | int *level, int root_level) | |
5235 | { | |
74dd17fb | 5236 | BUG_ON(*level == 0); |
7069830a AB |
5237 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
5238 | path->slots[*level]); | |
5239 | path->slots[*level - 1] = 0; | |
5240 | (*level)--; | |
5241 | } | |
5242 | ||
5243 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
5244 | struct btrfs_path *path, | |
5245 | int *level, int root_level) | |
5246 | { | |
5247 | int ret = 0; | |
5248 | int nritems; | |
5249 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5250 | ||
5251 | path->slots[*level]++; | |
5252 | ||
74dd17fb | 5253 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5254 | if (*level == root_level) |
5255 | return -1; | |
5256 | ||
5257 | /* move upnext */ | |
5258 | path->slots[*level] = 0; | |
5259 | free_extent_buffer(path->nodes[*level]); | |
5260 | path->nodes[*level] = NULL; | |
5261 | (*level)++; | |
5262 | path->slots[*level]++; | |
5263 | ||
5264 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5265 | ret = 1; | |
5266 | } | |
5267 | return ret; | |
5268 | } | |
5269 | ||
5270 | /* | |
5271 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5272 | * or down. | |
5273 | */ | |
5274 | static int tree_advance(struct btrfs_root *root, | |
5275 | struct btrfs_path *path, | |
5276 | int *level, int root_level, | |
5277 | int allow_down, | |
5278 | struct btrfs_key *key) | |
5279 | { | |
5280 | int ret; | |
5281 | ||
5282 | if (*level == 0 || !allow_down) { | |
5283 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
5284 | } else { | |
5285 | tree_move_down(root, path, level, root_level); | |
5286 | ret = 0; | |
5287 | } | |
5288 | if (ret >= 0) { | |
5289 | if (*level == 0) | |
5290 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5291 | path->slots[*level]); | |
5292 | else | |
5293 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5294 | path->slots[*level]); | |
5295 | } | |
5296 | return ret; | |
5297 | } | |
5298 | ||
5299 | static int tree_compare_item(struct btrfs_root *left_root, | |
5300 | struct btrfs_path *left_path, | |
5301 | struct btrfs_path *right_path, | |
5302 | char *tmp_buf) | |
5303 | { | |
5304 | int cmp; | |
5305 | int len1, len2; | |
5306 | unsigned long off1, off2; | |
5307 | ||
5308 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5309 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5310 | if (len1 != len2) | |
5311 | return 1; | |
5312 | ||
5313 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5314 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5315 | right_path->slots[0]); | |
5316 | ||
5317 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5318 | ||
5319 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5320 | if (cmp) | |
5321 | return 1; | |
5322 | return 0; | |
5323 | } | |
5324 | ||
5325 | #define ADVANCE 1 | |
5326 | #define ADVANCE_ONLY_NEXT -1 | |
5327 | ||
5328 | /* | |
5329 | * This function compares two trees and calls the provided callback for | |
5330 | * every changed/new/deleted item it finds. | |
5331 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5332 | * the compare pretty fast on snapshotted subvolumes. | |
5333 | * | |
5334 | * This currently works on commit roots only. As commit roots are read only, | |
5335 | * we don't do any locking. The commit roots are protected with transactions. | |
5336 | * Transactions are ended and rejoined when a commit is tried in between. | |
5337 | * | |
5338 | * This function checks for modifications done to the trees while comparing. | |
5339 | * If it detects a change, it aborts immediately. | |
5340 | */ | |
5341 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5342 | struct btrfs_root *right_root, | |
5343 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5344 | { | |
5345 | int ret; | |
5346 | int cmp; | |
7069830a AB |
5347 | struct btrfs_path *left_path = NULL; |
5348 | struct btrfs_path *right_path = NULL; | |
5349 | struct btrfs_key left_key; | |
5350 | struct btrfs_key right_key; | |
5351 | char *tmp_buf = NULL; | |
5352 | int left_root_level; | |
5353 | int right_root_level; | |
5354 | int left_level; | |
5355 | int right_level; | |
5356 | int left_end_reached; | |
5357 | int right_end_reached; | |
5358 | int advance_left; | |
5359 | int advance_right; | |
5360 | u64 left_blockptr; | |
5361 | u64 right_blockptr; | |
6baa4293 FM |
5362 | u64 left_gen; |
5363 | u64 right_gen; | |
7069830a AB |
5364 | |
5365 | left_path = btrfs_alloc_path(); | |
5366 | if (!left_path) { | |
5367 | ret = -ENOMEM; | |
5368 | goto out; | |
5369 | } | |
5370 | right_path = btrfs_alloc_path(); | |
5371 | if (!right_path) { | |
5372 | ret = -ENOMEM; | |
5373 | goto out; | |
5374 | } | |
5375 | ||
8f282f71 | 5376 | tmp_buf = kmalloc(left_root->nodesize, GFP_KERNEL | __GFP_NOWARN); |
7069830a | 5377 | if (!tmp_buf) { |
8f282f71 DS |
5378 | tmp_buf = vmalloc(left_root->nodesize); |
5379 | if (!tmp_buf) { | |
5380 | ret = -ENOMEM; | |
5381 | goto out; | |
5382 | } | |
7069830a AB |
5383 | } |
5384 | ||
5385 | left_path->search_commit_root = 1; | |
5386 | left_path->skip_locking = 1; | |
5387 | right_path->search_commit_root = 1; | |
5388 | right_path->skip_locking = 1; | |
5389 | ||
7069830a AB |
5390 | /* |
5391 | * Strategy: Go to the first items of both trees. Then do | |
5392 | * | |
5393 | * If both trees are at level 0 | |
5394 | * Compare keys of current items | |
5395 | * If left < right treat left item as new, advance left tree | |
5396 | * and repeat | |
5397 | * If left > right treat right item as deleted, advance right tree | |
5398 | * and repeat | |
5399 | * If left == right do deep compare of items, treat as changed if | |
5400 | * needed, advance both trees and repeat | |
5401 | * If both trees are at the same level but not at level 0 | |
5402 | * Compare keys of current nodes/leafs | |
5403 | * If left < right advance left tree and repeat | |
5404 | * If left > right advance right tree and repeat | |
5405 | * If left == right compare blockptrs of the next nodes/leafs | |
5406 | * If they match advance both trees but stay at the same level | |
5407 | * and repeat | |
5408 | * If they don't match advance both trees while allowing to go | |
5409 | * deeper and repeat | |
5410 | * If tree levels are different | |
5411 | * Advance the tree that needs it and repeat | |
5412 | * | |
5413 | * Advancing a tree means: | |
5414 | * If we are at level 0, try to go to the next slot. If that's not | |
5415 | * possible, go one level up and repeat. Stop when we found a level | |
5416 | * where we could go to the next slot. We may at this point be on a | |
5417 | * node or a leaf. | |
5418 | * | |
5419 | * If we are not at level 0 and not on shared tree blocks, go one | |
5420 | * level deeper. | |
5421 | * | |
5422 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5423 | * the right if possible or go up and right. | |
5424 | */ | |
5425 | ||
3f8a18cc | 5426 | down_read(&left_root->fs_info->commit_root_sem); |
7069830a AB |
5427 | left_level = btrfs_header_level(left_root->commit_root); |
5428 | left_root_level = left_level; | |
5429 | left_path->nodes[left_level] = left_root->commit_root; | |
5430 | extent_buffer_get(left_path->nodes[left_level]); | |
5431 | ||
5432 | right_level = btrfs_header_level(right_root->commit_root); | |
5433 | right_root_level = right_level; | |
5434 | right_path->nodes[right_level] = right_root->commit_root; | |
5435 | extent_buffer_get(right_path->nodes[right_level]); | |
3f8a18cc | 5436 | up_read(&left_root->fs_info->commit_root_sem); |
7069830a AB |
5437 | |
5438 | if (left_level == 0) | |
5439 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5440 | &left_key, left_path->slots[left_level]); | |
5441 | else | |
5442 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5443 | &left_key, left_path->slots[left_level]); | |
5444 | if (right_level == 0) | |
5445 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5446 | &right_key, right_path->slots[right_level]); | |
5447 | else | |
5448 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5449 | &right_key, right_path->slots[right_level]); | |
5450 | ||
5451 | left_end_reached = right_end_reached = 0; | |
5452 | advance_left = advance_right = 0; | |
5453 | ||
5454 | while (1) { | |
7069830a AB |
5455 | if (advance_left && !left_end_reached) { |
5456 | ret = tree_advance(left_root, left_path, &left_level, | |
5457 | left_root_level, | |
5458 | advance_left != ADVANCE_ONLY_NEXT, | |
5459 | &left_key); | |
5460 | if (ret < 0) | |
5461 | left_end_reached = ADVANCE; | |
5462 | advance_left = 0; | |
5463 | } | |
5464 | if (advance_right && !right_end_reached) { | |
5465 | ret = tree_advance(right_root, right_path, &right_level, | |
5466 | right_root_level, | |
5467 | advance_right != ADVANCE_ONLY_NEXT, | |
5468 | &right_key); | |
5469 | if (ret < 0) | |
5470 | right_end_reached = ADVANCE; | |
5471 | advance_right = 0; | |
5472 | } | |
5473 | ||
5474 | if (left_end_reached && right_end_reached) { | |
5475 | ret = 0; | |
5476 | goto out; | |
5477 | } else if (left_end_reached) { | |
5478 | if (right_level == 0) { | |
5479 | ret = changed_cb(left_root, right_root, | |
5480 | left_path, right_path, | |
5481 | &right_key, | |
5482 | BTRFS_COMPARE_TREE_DELETED, | |
5483 | ctx); | |
5484 | if (ret < 0) | |
5485 | goto out; | |
5486 | } | |
5487 | advance_right = ADVANCE; | |
5488 | continue; | |
5489 | } else if (right_end_reached) { | |
5490 | if (left_level == 0) { | |
5491 | ret = changed_cb(left_root, right_root, | |
5492 | left_path, right_path, | |
5493 | &left_key, | |
5494 | BTRFS_COMPARE_TREE_NEW, | |
5495 | ctx); | |
5496 | if (ret < 0) | |
5497 | goto out; | |
5498 | } | |
5499 | advance_left = ADVANCE; | |
5500 | continue; | |
5501 | } | |
5502 | ||
5503 | if (left_level == 0 && right_level == 0) { | |
5504 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5505 | if (cmp < 0) { | |
5506 | ret = changed_cb(left_root, right_root, | |
5507 | left_path, right_path, | |
5508 | &left_key, | |
5509 | BTRFS_COMPARE_TREE_NEW, | |
5510 | ctx); | |
5511 | if (ret < 0) | |
5512 | goto out; | |
5513 | advance_left = ADVANCE; | |
5514 | } else if (cmp > 0) { | |
5515 | ret = changed_cb(left_root, right_root, | |
5516 | left_path, right_path, | |
5517 | &right_key, | |
5518 | BTRFS_COMPARE_TREE_DELETED, | |
5519 | ctx); | |
5520 | if (ret < 0) | |
5521 | goto out; | |
5522 | advance_right = ADVANCE; | |
5523 | } else { | |
b99d9a6a | 5524 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5525 | |
74dd17fb | 5526 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5527 | ret = tree_compare_item(left_root, left_path, |
5528 | right_path, tmp_buf); | |
ba5e8f2e | 5529 | if (ret) |
b99d9a6a | 5530 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5531 | else |
b99d9a6a | 5532 | result = BTRFS_COMPARE_TREE_SAME; |
ba5e8f2e JB |
5533 | ret = changed_cb(left_root, right_root, |
5534 | left_path, right_path, | |
b99d9a6a | 5535 | &left_key, result, ctx); |
ba5e8f2e JB |
5536 | if (ret < 0) |
5537 | goto out; | |
7069830a AB |
5538 | advance_left = ADVANCE; |
5539 | advance_right = ADVANCE; | |
5540 | } | |
5541 | } else if (left_level == right_level) { | |
5542 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5543 | if (cmp < 0) { | |
5544 | advance_left = ADVANCE; | |
5545 | } else if (cmp > 0) { | |
5546 | advance_right = ADVANCE; | |
5547 | } else { | |
5548 | left_blockptr = btrfs_node_blockptr( | |
5549 | left_path->nodes[left_level], | |
5550 | left_path->slots[left_level]); | |
5551 | right_blockptr = btrfs_node_blockptr( | |
5552 | right_path->nodes[right_level], | |
5553 | right_path->slots[right_level]); | |
6baa4293 FM |
5554 | left_gen = btrfs_node_ptr_generation( |
5555 | left_path->nodes[left_level], | |
5556 | left_path->slots[left_level]); | |
5557 | right_gen = btrfs_node_ptr_generation( | |
5558 | right_path->nodes[right_level], | |
5559 | right_path->slots[right_level]); | |
5560 | if (left_blockptr == right_blockptr && | |
5561 | left_gen == right_gen) { | |
7069830a AB |
5562 | /* |
5563 | * As we're on a shared block, don't | |
5564 | * allow to go deeper. | |
5565 | */ | |
5566 | advance_left = ADVANCE_ONLY_NEXT; | |
5567 | advance_right = ADVANCE_ONLY_NEXT; | |
5568 | } else { | |
5569 | advance_left = ADVANCE; | |
5570 | advance_right = ADVANCE; | |
5571 | } | |
5572 | } | |
5573 | } else if (left_level < right_level) { | |
5574 | advance_right = ADVANCE; | |
5575 | } else { | |
5576 | advance_left = ADVANCE; | |
5577 | } | |
5578 | } | |
5579 | ||
5580 | out: | |
5581 | btrfs_free_path(left_path); | |
5582 | btrfs_free_path(right_path); | |
8f282f71 | 5583 | kvfree(tmp_buf); |
7069830a AB |
5584 | return ret; |
5585 | } | |
5586 | ||
3f157a2f CM |
5587 | /* |
5588 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5589 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5590 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5591 | * |
5592 | * 0 is returned if another key is found, < 0 if there are any errors | |
5593 | * and 1 is returned if there are no higher keys in the tree | |
5594 | * | |
5595 | * path->keep_locks should be set to 1 on the search made before | |
5596 | * calling this function. | |
5597 | */ | |
e7a84565 | 5598 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5599 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5600 | { |
e7a84565 CM |
5601 | int slot; |
5602 | struct extent_buffer *c; | |
5603 | ||
934d375b | 5604 | WARN_ON(!path->keep_locks); |
d397712b | 5605 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5606 | if (!path->nodes[level]) |
5607 | return 1; | |
5608 | ||
5609 | slot = path->slots[level] + 1; | |
5610 | c = path->nodes[level]; | |
3f157a2f | 5611 | next: |
e7a84565 | 5612 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5613 | int ret; |
5614 | int orig_lowest; | |
5615 | struct btrfs_key cur_key; | |
5616 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5617 | !path->nodes[level + 1]) | |
e7a84565 | 5618 | return 1; |
33c66f43 YZ |
5619 | |
5620 | if (path->locks[level + 1]) { | |
5621 | level++; | |
5622 | continue; | |
5623 | } | |
5624 | ||
5625 | slot = btrfs_header_nritems(c) - 1; | |
5626 | if (level == 0) | |
5627 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5628 | else | |
5629 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5630 | ||
5631 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5632 | btrfs_release_path(path); |
33c66f43 YZ |
5633 | path->lowest_level = level; |
5634 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5635 | 0, 0); | |
5636 | path->lowest_level = orig_lowest; | |
5637 | if (ret < 0) | |
5638 | return ret; | |
5639 | ||
5640 | c = path->nodes[level]; | |
5641 | slot = path->slots[level]; | |
5642 | if (ret == 0) | |
5643 | slot++; | |
5644 | goto next; | |
e7a84565 | 5645 | } |
33c66f43 | 5646 | |
e7a84565 CM |
5647 | if (level == 0) |
5648 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5649 | else { |
3f157a2f CM |
5650 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5651 | ||
3f157a2f CM |
5652 | if (gen < min_trans) { |
5653 | slot++; | |
5654 | goto next; | |
5655 | } | |
e7a84565 | 5656 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5657 | } |
e7a84565 CM |
5658 | return 0; |
5659 | } | |
5660 | return 1; | |
5661 | } | |
5662 | ||
97571fd0 | 5663 | /* |
925baedd | 5664 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5665 | * returns 0 if it found something or 1 if there are no greater leaves. |
5666 | * returns < 0 on io errors. | |
97571fd0 | 5667 | */ |
234b63a0 | 5668 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5669 | { |
5670 | return btrfs_next_old_leaf(root, path, 0); | |
5671 | } | |
5672 | ||
5673 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5674 | u64 time_seq) | |
d97e63b6 CM |
5675 | { |
5676 | int slot; | |
8e73f275 | 5677 | int level; |
5f39d397 | 5678 | struct extent_buffer *c; |
8e73f275 | 5679 | struct extent_buffer *next; |
925baedd CM |
5680 | struct btrfs_key key; |
5681 | u32 nritems; | |
5682 | int ret; | |
8e73f275 | 5683 | int old_spinning = path->leave_spinning; |
bd681513 | 5684 | int next_rw_lock = 0; |
925baedd CM |
5685 | |
5686 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5687 | if (nritems == 0) |
925baedd | 5688 | return 1; |
925baedd | 5689 | |
8e73f275 CM |
5690 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5691 | again: | |
5692 | level = 1; | |
5693 | next = NULL; | |
bd681513 | 5694 | next_rw_lock = 0; |
b3b4aa74 | 5695 | btrfs_release_path(path); |
8e73f275 | 5696 | |
a2135011 | 5697 | path->keep_locks = 1; |
31533fb2 | 5698 | path->leave_spinning = 1; |
8e73f275 | 5699 | |
3d7806ec JS |
5700 | if (time_seq) |
5701 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5702 | else | |
5703 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5704 | path->keep_locks = 0; |
5705 | ||
5706 | if (ret < 0) | |
5707 | return ret; | |
5708 | ||
a2135011 | 5709 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5710 | /* |
5711 | * by releasing the path above we dropped all our locks. A balance | |
5712 | * could have added more items next to the key that used to be | |
5713 | * at the very end of the block. So, check again here and | |
5714 | * advance the path if there are now more items available. | |
5715 | */ | |
a2135011 | 5716 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5717 | if (ret == 0) |
5718 | path->slots[0]++; | |
8e73f275 | 5719 | ret = 0; |
925baedd CM |
5720 | goto done; |
5721 | } | |
0b43e04f LB |
5722 | /* |
5723 | * So the above check misses one case: | |
5724 | * - after releasing the path above, someone has removed the item that | |
5725 | * used to be at the very end of the block, and balance between leafs | |
5726 | * gets another one with bigger key.offset to replace it. | |
5727 | * | |
5728 | * This one should be returned as well, or we can get leaf corruption | |
5729 | * later(esp. in __btrfs_drop_extents()). | |
5730 | * | |
5731 | * And a bit more explanation about this check, | |
5732 | * with ret > 0, the key isn't found, the path points to the slot | |
5733 | * where it should be inserted, so the path->slots[0] item must be the | |
5734 | * bigger one. | |
5735 | */ | |
5736 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5737 | ret = 0; | |
5738 | goto done; | |
5739 | } | |
d97e63b6 | 5740 | |
d397712b | 5741 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5742 | if (!path->nodes[level]) { |
5743 | ret = 1; | |
5744 | goto done; | |
5745 | } | |
5f39d397 | 5746 | |
d97e63b6 CM |
5747 | slot = path->slots[level] + 1; |
5748 | c = path->nodes[level]; | |
5f39d397 | 5749 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5750 | level++; |
8e73f275 CM |
5751 | if (level == BTRFS_MAX_LEVEL) { |
5752 | ret = 1; | |
5753 | goto done; | |
5754 | } | |
d97e63b6 CM |
5755 | continue; |
5756 | } | |
5f39d397 | 5757 | |
925baedd | 5758 | if (next) { |
bd681513 | 5759 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5760 | free_extent_buffer(next); |
925baedd | 5761 | } |
5f39d397 | 5762 | |
8e73f275 | 5763 | next = c; |
bd681513 | 5764 | next_rw_lock = path->locks[level]; |
8e73f275 | 5765 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5766 | slot, &key, 0); |
8e73f275 CM |
5767 | if (ret == -EAGAIN) |
5768 | goto again; | |
5f39d397 | 5769 | |
76a05b35 | 5770 | if (ret < 0) { |
b3b4aa74 | 5771 | btrfs_release_path(path); |
76a05b35 CM |
5772 | goto done; |
5773 | } | |
5774 | ||
5cd57b2c | 5775 | if (!path->skip_locking) { |
bd681513 | 5776 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5777 | if (!ret && time_seq) { |
5778 | /* | |
5779 | * If we don't get the lock, we may be racing | |
5780 | * with push_leaf_left, holding that lock while | |
5781 | * itself waiting for the leaf we've currently | |
5782 | * locked. To solve this situation, we give up | |
5783 | * on our lock and cycle. | |
5784 | */ | |
cf538830 | 5785 | free_extent_buffer(next); |
d42244a0 JS |
5786 | btrfs_release_path(path); |
5787 | cond_resched(); | |
5788 | goto again; | |
5789 | } | |
8e73f275 CM |
5790 | if (!ret) { |
5791 | btrfs_set_path_blocking(path); | |
bd681513 | 5792 | btrfs_tree_read_lock(next); |
31533fb2 | 5793 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5794 | BTRFS_READ_LOCK); |
8e73f275 | 5795 | } |
31533fb2 | 5796 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5797 | } |
d97e63b6 CM |
5798 | break; |
5799 | } | |
5800 | path->slots[level] = slot; | |
d397712b | 5801 | while (1) { |
d97e63b6 CM |
5802 | level--; |
5803 | c = path->nodes[level]; | |
925baedd | 5804 | if (path->locks[level]) |
bd681513 | 5805 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5806 | |
5f39d397 | 5807 | free_extent_buffer(c); |
d97e63b6 CM |
5808 | path->nodes[level] = next; |
5809 | path->slots[level] = 0; | |
a74a4b97 | 5810 | if (!path->skip_locking) |
bd681513 | 5811 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5812 | if (!level) |
5813 | break; | |
b4ce94de | 5814 | |
8e73f275 | 5815 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5816 | 0, &key, 0); |
8e73f275 CM |
5817 | if (ret == -EAGAIN) |
5818 | goto again; | |
5819 | ||
76a05b35 | 5820 | if (ret < 0) { |
b3b4aa74 | 5821 | btrfs_release_path(path); |
76a05b35 CM |
5822 | goto done; |
5823 | } | |
5824 | ||
5cd57b2c | 5825 | if (!path->skip_locking) { |
bd681513 | 5826 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5827 | if (!ret) { |
5828 | btrfs_set_path_blocking(path); | |
bd681513 | 5829 | btrfs_tree_read_lock(next); |
31533fb2 | 5830 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5831 | BTRFS_READ_LOCK); |
5832 | } | |
31533fb2 | 5833 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5834 | } |
d97e63b6 | 5835 | } |
8e73f275 | 5836 | ret = 0; |
925baedd | 5837 | done: |
f7c79f30 | 5838 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5839 | path->leave_spinning = old_spinning; |
5840 | if (!old_spinning) | |
5841 | btrfs_set_path_blocking(path); | |
5842 | ||
5843 | return ret; | |
d97e63b6 | 5844 | } |
0b86a832 | 5845 | |
3f157a2f CM |
5846 | /* |
5847 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5848 | * searching until it gets past min_objectid or finds an item of 'type' | |
5849 | * | |
5850 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5851 | */ | |
0b86a832 CM |
5852 | int btrfs_previous_item(struct btrfs_root *root, |
5853 | struct btrfs_path *path, u64 min_objectid, | |
5854 | int type) | |
5855 | { | |
5856 | struct btrfs_key found_key; | |
5857 | struct extent_buffer *leaf; | |
e02119d5 | 5858 | u32 nritems; |
0b86a832 CM |
5859 | int ret; |
5860 | ||
d397712b | 5861 | while (1) { |
0b86a832 | 5862 | if (path->slots[0] == 0) { |
b4ce94de | 5863 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5864 | ret = btrfs_prev_leaf(root, path); |
5865 | if (ret != 0) | |
5866 | return ret; | |
5867 | } else { | |
5868 | path->slots[0]--; | |
5869 | } | |
5870 | leaf = path->nodes[0]; | |
e02119d5 CM |
5871 | nritems = btrfs_header_nritems(leaf); |
5872 | if (nritems == 0) | |
5873 | return 1; | |
5874 | if (path->slots[0] == nritems) | |
5875 | path->slots[0]--; | |
5876 | ||
0b86a832 | 5877 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5878 | if (found_key.objectid < min_objectid) |
5879 | break; | |
0a4eefbb YZ |
5880 | if (found_key.type == type) |
5881 | return 0; | |
e02119d5 CM |
5882 | if (found_key.objectid == min_objectid && |
5883 | found_key.type < type) | |
5884 | break; | |
0b86a832 CM |
5885 | } |
5886 | return 1; | |
5887 | } | |
ade2e0b3 WS |
5888 | |
5889 | /* | |
5890 | * search in extent tree to find a previous Metadata/Data extent item with | |
5891 | * min objecitd. | |
5892 | * | |
5893 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5894 | */ | |
5895 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5896 | struct btrfs_path *path, u64 min_objectid) | |
5897 | { | |
5898 | struct btrfs_key found_key; | |
5899 | struct extent_buffer *leaf; | |
5900 | u32 nritems; | |
5901 | int ret; | |
5902 | ||
5903 | while (1) { | |
5904 | if (path->slots[0] == 0) { | |
5905 | btrfs_set_path_blocking(path); | |
5906 | ret = btrfs_prev_leaf(root, path); | |
5907 | if (ret != 0) | |
5908 | return ret; | |
5909 | } else { | |
5910 | path->slots[0]--; | |
5911 | } | |
5912 | leaf = path->nodes[0]; | |
5913 | nritems = btrfs_header_nritems(leaf); | |
5914 | if (nritems == 0) | |
5915 | return 1; | |
5916 | if (path->slots[0] == nritems) | |
5917 | path->slots[0]--; | |
5918 | ||
5919 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5920 | if (found_key.objectid < min_objectid) | |
5921 | break; | |
5922 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5923 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5924 | return 0; | |
5925 | if (found_key.objectid == min_objectid && | |
5926 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5927 | break; | |
5928 | } | |
5929 | return 1; | |
5930 | } |