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> |
adf02123 | 22 | #include <linux/mm.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); | |
310712b2 OS |
31 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
32 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
33 | int data_size, int extend); | |
5f39d397 | 34 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
35 | struct btrfs_fs_info *fs_info, |
36 | struct extent_buffer *dst, | |
971a1f66 | 37 | struct extent_buffer *src, int empty); |
5f39d397 | 38 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 39 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
40 | struct extent_buffer *dst_buf, |
41 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
42 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
43 | int level, int slot); | |
5de865ee | 44 | static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
f230475e | 45 | struct extent_buffer *eb); |
d97e63b6 | 46 | |
df24a2b9 | 47 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 48 | { |
e2c89907 | 49 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
50 | } |
51 | ||
b4ce94de CM |
52 | /* |
53 | * set all locked nodes in the path to blocking locks. This should | |
54 | * be done before scheduling | |
55 | */ | |
56 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
57 | { | |
58 | int i; | |
59 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
60 | if (!p->nodes[i] || !p->locks[i]) |
61 | continue; | |
62 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
63 | if (p->locks[i] == BTRFS_READ_LOCK) | |
64 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
65 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
66 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
67 | } |
68 | } | |
69 | ||
70 | /* | |
71 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
72 | * |
73 | * held is used to keep lockdep happy, when lockdep is enabled | |
74 | * we set held to a blocking lock before we go around and | |
75 | * retake all the spinlocks in the path. You can safely use NULL | |
76 | * for held | |
b4ce94de | 77 | */ |
4008c04a | 78 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 79 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
80 | { |
81 | int i; | |
4008c04a | 82 | |
bd681513 CM |
83 | if (held) { |
84 | btrfs_set_lock_blocking_rw(held, held_rw); | |
85 | if (held_rw == BTRFS_WRITE_LOCK) | |
86 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
87 | else if (held_rw == BTRFS_READ_LOCK) | |
88 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
89 | } | |
4008c04a | 90 | btrfs_set_path_blocking(p); |
4008c04a CM |
91 | |
92 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
93 | if (p->nodes[i] && p->locks[i]) { |
94 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
95 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
96 | p->locks[i] = BTRFS_WRITE_LOCK; | |
97 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
98 | p->locks[i] = BTRFS_READ_LOCK; | |
99 | } | |
b4ce94de | 100 | } |
4008c04a | 101 | |
4008c04a | 102 | if (held) |
bd681513 | 103 | btrfs_clear_lock_blocking_rw(held, held_rw); |
b4ce94de CM |
104 | } |
105 | ||
d352ac68 | 106 | /* this also releases the path */ |
df24a2b9 | 107 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 108 | { |
ff175d57 JJ |
109 | if (!p) |
110 | return; | |
b3b4aa74 | 111 | btrfs_release_path(p); |
df24a2b9 | 112 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
113 | } |
114 | ||
d352ac68 CM |
115 | /* |
116 | * path release drops references on the extent buffers in the path | |
117 | * and it drops any locks held by this path | |
118 | * | |
119 | * It is safe to call this on paths that no locks or extent buffers held. | |
120 | */ | |
b3b4aa74 | 121 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
122 | { |
123 | int i; | |
a2135011 | 124 | |
234b63a0 | 125 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 126 | p->slots[i] = 0; |
eb60ceac | 127 | if (!p->nodes[i]) |
925baedd CM |
128 | continue; |
129 | if (p->locks[i]) { | |
bd681513 | 130 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
131 | p->locks[i] = 0; |
132 | } | |
5f39d397 | 133 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 134 | p->nodes[i] = NULL; |
eb60ceac CM |
135 | } |
136 | } | |
137 | ||
d352ac68 CM |
138 | /* |
139 | * safely gets a reference on the root node of a tree. A lock | |
140 | * is not taken, so a concurrent writer may put a different node | |
141 | * at the root of the tree. See btrfs_lock_root_node for the | |
142 | * looping required. | |
143 | * | |
144 | * The extent buffer returned by this has a reference taken, so | |
145 | * it won't disappear. It may stop being the root of the tree | |
146 | * at any time because there are no locks held. | |
147 | */ | |
925baedd CM |
148 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
149 | { | |
150 | struct extent_buffer *eb; | |
240f62c8 | 151 | |
3083ee2e JB |
152 | while (1) { |
153 | rcu_read_lock(); | |
154 | eb = rcu_dereference(root->node); | |
155 | ||
156 | /* | |
157 | * RCU really hurts here, we could free up the root node because | |
01327610 | 158 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
159 | * the inc_not_zero dance and if it doesn't work then |
160 | * synchronize_rcu and try again. | |
161 | */ | |
162 | if (atomic_inc_not_zero(&eb->refs)) { | |
163 | rcu_read_unlock(); | |
164 | break; | |
165 | } | |
166 | rcu_read_unlock(); | |
167 | synchronize_rcu(); | |
168 | } | |
925baedd CM |
169 | return eb; |
170 | } | |
171 | ||
d352ac68 CM |
172 | /* loop around taking references on and locking the root node of the |
173 | * tree until you end up with a lock on the root. A locked buffer | |
174 | * is returned, with a reference held. | |
175 | */ | |
925baedd CM |
176 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
177 | { | |
178 | struct extent_buffer *eb; | |
179 | ||
d397712b | 180 | while (1) { |
925baedd CM |
181 | eb = btrfs_root_node(root); |
182 | btrfs_tree_lock(eb); | |
240f62c8 | 183 | if (eb == root->node) |
925baedd | 184 | break; |
925baedd CM |
185 | btrfs_tree_unlock(eb); |
186 | free_extent_buffer(eb); | |
187 | } | |
188 | return eb; | |
189 | } | |
190 | ||
bd681513 CM |
191 | /* loop around taking references on and locking the root node of the |
192 | * tree until you end up with a lock on the root. A locked buffer | |
193 | * is returned, with a reference held. | |
194 | */ | |
84f7d8e6 | 195 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
196 | { |
197 | struct extent_buffer *eb; | |
198 | ||
199 | while (1) { | |
200 | eb = btrfs_root_node(root); | |
201 | btrfs_tree_read_lock(eb); | |
202 | if (eb == root->node) | |
203 | break; | |
204 | btrfs_tree_read_unlock(eb); | |
205 | free_extent_buffer(eb); | |
206 | } | |
207 | return eb; | |
208 | } | |
209 | ||
d352ac68 CM |
210 | /* cowonly root (everything not a reference counted cow subvolume), just get |
211 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
212 | * get properly updated on disk. | |
213 | */ | |
0b86a832 CM |
214 | static void add_root_to_dirty_list(struct btrfs_root *root) |
215 | { | |
0b246afa JM |
216 | struct btrfs_fs_info *fs_info = root->fs_info; |
217 | ||
e7070be1 JB |
218 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
219 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
220 | return; | |
221 | ||
0b246afa | 222 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
223 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
224 | /* Want the extent tree to be the last on the list */ | |
225 | if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
226 | list_move_tail(&root->dirty_list, | |
0b246afa | 227 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
228 | else |
229 | list_move(&root->dirty_list, | |
0b246afa | 230 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 231 | } |
0b246afa | 232 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
233 | } |
234 | ||
d352ac68 CM |
235 | /* |
236 | * used by snapshot creation to make a copy of a root for a tree with | |
237 | * a given objectid. The buffer with the new root node is returned in | |
238 | * cow_ret, and this func returns zero on success or a negative error code. | |
239 | */ | |
be20aa9d CM |
240 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
241 | struct btrfs_root *root, | |
242 | struct extent_buffer *buf, | |
243 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
244 | { | |
0b246afa | 245 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 246 | struct extent_buffer *cow; |
be20aa9d CM |
247 | int ret = 0; |
248 | int level; | |
5d4f98a2 | 249 | struct btrfs_disk_key disk_key; |
be20aa9d | 250 | |
27cdeb70 | 251 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 252 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
253 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
254 | trans->transid != root->last_trans); | |
be20aa9d CM |
255 | |
256 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
257 | if (level == 0) |
258 | btrfs_item_key(buf, &disk_key, 0); | |
259 | else | |
260 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 261 | |
4d75f8a9 DS |
262 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
263 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 264 | if (IS_ERR(cow)) |
be20aa9d CM |
265 | return PTR_ERR(cow); |
266 | ||
58e8012c | 267 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
268 | btrfs_set_header_bytenr(cow, cow->start); |
269 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
270 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
271 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
272 | BTRFS_HEADER_FLAG_RELOC); | |
273 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
274 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
275 | else | |
276 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 277 | |
0b246afa | 278 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 279 | |
be20aa9d | 280 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 281 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 282 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 283 | else |
e339a6b0 | 284 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 285 | |
be20aa9d CM |
286 | if (ret) |
287 | return ret; | |
288 | ||
289 | btrfs_mark_buffer_dirty(cow); | |
290 | *cow_ret = cow; | |
291 | return 0; | |
292 | } | |
293 | ||
bd989ba3 JS |
294 | enum mod_log_op { |
295 | MOD_LOG_KEY_REPLACE, | |
296 | MOD_LOG_KEY_ADD, | |
297 | MOD_LOG_KEY_REMOVE, | |
298 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
299 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
300 | MOD_LOG_MOVE_KEYS, | |
301 | MOD_LOG_ROOT_REPLACE, | |
302 | }; | |
303 | ||
304 | struct tree_mod_move { | |
305 | int dst_slot; | |
306 | int nr_items; | |
307 | }; | |
308 | ||
309 | struct tree_mod_root { | |
310 | u64 logical; | |
311 | u8 level; | |
312 | }; | |
313 | ||
314 | struct tree_mod_elem { | |
315 | struct rb_node node; | |
298cfd36 | 316 | u64 logical; |
097b8a7c | 317 | u64 seq; |
bd989ba3 JS |
318 | enum mod_log_op op; |
319 | ||
320 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
321 | int slot; | |
322 | ||
323 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
324 | u64 generation; | |
325 | ||
326 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
327 | struct btrfs_disk_key key; | |
328 | u64 blockptr; | |
329 | ||
330 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
331 | struct tree_mod_move move; | |
332 | ||
333 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
334 | struct tree_mod_root old_root; | |
335 | }; | |
336 | ||
097b8a7c | 337 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 338 | { |
097b8a7c | 339 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
340 | } |
341 | ||
097b8a7c JS |
342 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
343 | { | |
344 | read_unlock(&fs_info->tree_mod_log_lock); | |
345 | } | |
346 | ||
347 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
348 | { | |
349 | write_lock(&fs_info->tree_mod_log_lock); | |
350 | } | |
351 | ||
352 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
353 | { | |
354 | write_unlock(&fs_info->tree_mod_log_lock); | |
355 | } | |
356 | ||
fc36ed7e | 357 | /* |
fcebe456 | 358 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 359 | */ |
fcebe456 | 360 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
361 | { |
362 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
363 | } | |
364 | ||
097b8a7c JS |
365 | /* |
366 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
367 | * passed does not already have a sequence number set. So when a caller expects | |
368 | * to record tree modifications, it should ensure to set elem->seq to zero | |
369 | * before calling btrfs_get_tree_mod_seq. | |
370 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
371 | * blocker was added. | |
372 | */ | |
373 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
374 | struct seq_list *elem) | |
bd989ba3 | 375 | { |
097b8a7c | 376 | tree_mod_log_write_lock(fs_info); |
bd989ba3 | 377 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 378 | if (!elem->seq) { |
fcebe456 | 379 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
380 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
381 | } | |
bd989ba3 | 382 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
383 | tree_mod_log_write_unlock(fs_info); |
384 | ||
fcebe456 | 385 | return elem->seq; |
bd989ba3 JS |
386 | } |
387 | ||
388 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
389 | struct seq_list *elem) | |
390 | { | |
391 | struct rb_root *tm_root; | |
392 | struct rb_node *node; | |
393 | struct rb_node *next; | |
394 | struct seq_list *cur_elem; | |
395 | struct tree_mod_elem *tm; | |
396 | u64 min_seq = (u64)-1; | |
397 | u64 seq_putting = elem->seq; | |
398 | ||
399 | if (!seq_putting) | |
400 | return; | |
401 | ||
bd989ba3 JS |
402 | spin_lock(&fs_info->tree_mod_seq_lock); |
403 | list_del(&elem->list); | |
097b8a7c | 404 | elem->seq = 0; |
bd989ba3 JS |
405 | |
406 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 407 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
408 | if (seq_putting > cur_elem->seq) { |
409 | /* | |
410 | * blocker with lower sequence number exists, we | |
411 | * cannot remove anything from the log | |
412 | */ | |
097b8a7c JS |
413 | spin_unlock(&fs_info->tree_mod_seq_lock); |
414 | return; | |
bd989ba3 JS |
415 | } |
416 | min_seq = cur_elem->seq; | |
417 | } | |
418 | } | |
097b8a7c JS |
419 | spin_unlock(&fs_info->tree_mod_seq_lock); |
420 | ||
bd989ba3 JS |
421 | /* |
422 | * anything that's lower than the lowest existing (read: blocked) | |
423 | * sequence number can be removed from the tree. | |
424 | */ | |
097b8a7c | 425 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
426 | tm_root = &fs_info->tree_mod_log; |
427 | for (node = rb_first(tm_root); node; node = next) { | |
428 | next = rb_next(node); | |
6b4df8b6 | 429 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 430 | if (tm->seq > min_seq) |
bd989ba3 JS |
431 | continue; |
432 | rb_erase(node, tm_root); | |
bd989ba3 JS |
433 | kfree(tm); |
434 | } | |
097b8a7c | 435 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
436 | } |
437 | ||
438 | /* | |
439 | * key order of the log: | |
298cfd36 | 440 | * node/leaf start address -> sequence |
bd989ba3 | 441 | * |
298cfd36 CR |
442 | * The 'start address' is the logical address of the *new* root node |
443 | * for root replace operations, or the logical address of the affected | |
444 | * block for all other operations. | |
5de865ee FDBM |
445 | * |
446 | * Note: must be called with write lock (tree_mod_log_write_lock). | |
bd989ba3 JS |
447 | */ |
448 | static noinline int | |
449 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
450 | { | |
451 | struct rb_root *tm_root; | |
452 | struct rb_node **new; | |
453 | struct rb_node *parent = NULL; | |
454 | struct tree_mod_elem *cur; | |
c8cc6341 | 455 | |
fcebe456 | 456 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 457 | |
bd989ba3 JS |
458 | tm_root = &fs_info->tree_mod_log; |
459 | new = &tm_root->rb_node; | |
460 | while (*new) { | |
6b4df8b6 | 461 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 462 | parent = *new; |
298cfd36 | 463 | if (cur->logical < tm->logical) |
bd989ba3 | 464 | new = &((*new)->rb_left); |
298cfd36 | 465 | else if (cur->logical > tm->logical) |
bd989ba3 | 466 | new = &((*new)->rb_right); |
097b8a7c | 467 | else if (cur->seq < tm->seq) |
bd989ba3 | 468 | new = &((*new)->rb_left); |
097b8a7c | 469 | else if (cur->seq > tm->seq) |
bd989ba3 | 470 | new = &((*new)->rb_right); |
5de865ee FDBM |
471 | else |
472 | return -EEXIST; | |
bd989ba3 JS |
473 | } |
474 | ||
475 | rb_link_node(&tm->node, parent, new); | |
476 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 477 | return 0; |
bd989ba3 JS |
478 | } |
479 | ||
097b8a7c JS |
480 | /* |
481 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
482 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
483 | * this until all tree mod log insertions are recorded in the rb tree and then | |
484 | * call tree_mod_log_write_unlock() to release. | |
485 | */ | |
e9b7fd4d JS |
486 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
487 | struct extent_buffer *eb) { | |
488 | smp_mb(); | |
489 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
490 | return 1; | |
097b8a7c JS |
491 | if (eb && btrfs_header_level(eb) == 0) |
492 | return 1; | |
5de865ee FDBM |
493 | |
494 | tree_mod_log_write_lock(fs_info); | |
495 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { | |
496 | tree_mod_log_write_unlock(fs_info); | |
497 | return 1; | |
498 | } | |
499 | ||
e9b7fd4d JS |
500 | return 0; |
501 | } | |
502 | ||
5de865ee FDBM |
503 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
504 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
505 | struct extent_buffer *eb) | |
506 | { | |
507 | smp_mb(); | |
508 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
509 | return 0; | |
510 | if (eb && btrfs_header_level(eb) == 0) | |
511 | return 0; | |
512 | ||
513 | return 1; | |
514 | } | |
515 | ||
516 | static struct tree_mod_elem * | |
517 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
518 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 519 | { |
097b8a7c | 520 | struct tree_mod_elem *tm; |
bd989ba3 | 521 | |
c8cc6341 JB |
522 | tm = kzalloc(sizeof(*tm), flags); |
523 | if (!tm) | |
5de865ee | 524 | return NULL; |
bd989ba3 | 525 | |
298cfd36 | 526 | tm->logical = eb->start; |
bd989ba3 JS |
527 | if (op != MOD_LOG_KEY_ADD) { |
528 | btrfs_node_key(eb, &tm->key, slot); | |
529 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
530 | } | |
531 | tm->op = op; | |
532 | tm->slot = slot; | |
533 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 534 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 535 | |
5de865ee | 536 | return tm; |
097b8a7c JS |
537 | } |
538 | ||
539 | static noinline int | |
c8cc6341 JB |
540 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
541 | struct extent_buffer *eb, int slot, | |
542 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 543 | { |
5de865ee FDBM |
544 | struct tree_mod_elem *tm; |
545 | int ret; | |
546 | ||
547 | if (!tree_mod_need_log(fs_info, eb)) | |
548 | return 0; | |
549 | ||
550 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
551 | if (!tm) | |
552 | return -ENOMEM; | |
553 | ||
554 | if (tree_mod_dont_log(fs_info, eb)) { | |
555 | kfree(tm); | |
097b8a7c | 556 | return 0; |
5de865ee FDBM |
557 | } |
558 | ||
559 | ret = __tree_mod_log_insert(fs_info, tm); | |
560 | tree_mod_log_write_unlock(fs_info); | |
561 | if (ret) | |
562 | kfree(tm); | |
097b8a7c | 563 | |
5de865ee | 564 | return ret; |
097b8a7c JS |
565 | } |
566 | ||
6074d45f DS |
567 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
568 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 569 | { |
5de865ee FDBM |
570 | struct tree_mod_elem *tm = NULL; |
571 | struct tree_mod_elem **tm_list = NULL; | |
572 | int ret = 0; | |
bd989ba3 | 573 | int i; |
5de865ee | 574 | int locked = 0; |
bd989ba3 | 575 | |
6074d45f | 576 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 577 | return 0; |
bd989ba3 | 578 | |
176ef8f5 | 579 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
580 | if (!tm_list) |
581 | return -ENOMEM; | |
582 | ||
176ef8f5 | 583 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
584 | if (!tm) { |
585 | ret = -ENOMEM; | |
586 | goto free_tms; | |
587 | } | |
588 | ||
298cfd36 | 589 | tm->logical = eb->start; |
5de865ee FDBM |
590 | tm->slot = src_slot; |
591 | tm->move.dst_slot = dst_slot; | |
592 | tm->move.nr_items = nr_items; | |
593 | tm->op = MOD_LOG_MOVE_KEYS; | |
594 | ||
595 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
596 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 597 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
598 | if (!tm_list[i]) { |
599 | ret = -ENOMEM; | |
600 | goto free_tms; | |
601 | } | |
602 | } | |
603 | ||
6074d45f | 604 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
605 | goto free_tms; |
606 | locked = 1; | |
607 | ||
01763a2e JS |
608 | /* |
609 | * When we override something during the move, we log these removals. | |
610 | * This can only happen when we move towards the beginning of the | |
611 | * buffer, i.e. dst_slot < src_slot. | |
612 | */ | |
bd989ba3 | 613 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 614 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
615 | if (ret) |
616 | goto free_tms; | |
bd989ba3 JS |
617 | } |
618 | ||
6074d45f | 619 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
620 | if (ret) |
621 | goto free_tms; | |
6074d45f | 622 | tree_mod_log_write_unlock(eb->fs_info); |
5de865ee | 623 | kfree(tm_list); |
f395694c | 624 | |
5de865ee FDBM |
625 | return 0; |
626 | free_tms: | |
627 | for (i = 0; i < nr_items; i++) { | |
628 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 629 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
630 | kfree(tm_list[i]); |
631 | } | |
632 | if (locked) | |
6074d45f | 633 | tree_mod_log_write_unlock(eb->fs_info); |
5de865ee FDBM |
634 | kfree(tm_list); |
635 | kfree(tm); | |
bd989ba3 | 636 | |
5de865ee | 637 | return ret; |
bd989ba3 JS |
638 | } |
639 | ||
5de865ee FDBM |
640 | static inline int |
641 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
642 | struct tree_mod_elem **tm_list, | |
643 | int nritems) | |
097b8a7c | 644 | { |
5de865ee | 645 | int i, j; |
097b8a7c JS |
646 | int ret; |
647 | ||
097b8a7c | 648 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
649 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
650 | if (ret) { | |
651 | for (j = nritems - 1; j > i; j--) | |
652 | rb_erase(&tm_list[j]->node, | |
653 | &fs_info->tree_mod_log); | |
654 | return ret; | |
655 | } | |
097b8a7c | 656 | } |
5de865ee FDBM |
657 | |
658 | return 0; | |
097b8a7c JS |
659 | } |
660 | ||
bd989ba3 JS |
661 | static noinline int |
662 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
663 | struct extent_buffer *old_root, | |
bcc8e07f | 664 | struct extent_buffer *new_root, |
90f8d62e | 665 | int log_removal) |
bd989ba3 | 666 | { |
5de865ee FDBM |
667 | struct tree_mod_elem *tm = NULL; |
668 | struct tree_mod_elem **tm_list = NULL; | |
669 | int nritems = 0; | |
670 | int ret = 0; | |
671 | int i; | |
bd989ba3 | 672 | |
5de865ee | 673 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
674 | return 0; |
675 | ||
5de865ee FDBM |
676 | if (log_removal && btrfs_header_level(old_root) > 0) { |
677 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 678 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 679 | GFP_NOFS); |
5de865ee FDBM |
680 | if (!tm_list) { |
681 | ret = -ENOMEM; | |
682 | goto free_tms; | |
683 | } | |
684 | for (i = 0; i < nritems; i++) { | |
685 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 686 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
687 | if (!tm_list[i]) { |
688 | ret = -ENOMEM; | |
689 | goto free_tms; | |
690 | } | |
691 | } | |
692 | } | |
d9abbf1c | 693 | |
bcc8e07f | 694 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
695 | if (!tm) { |
696 | ret = -ENOMEM; | |
697 | goto free_tms; | |
698 | } | |
bd989ba3 | 699 | |
298cfd36 | 700 | tm->logical = new_root->start; |
bd989ba3 JS |
701 | tm->old_root.logical = old_root->start; |
702 | tm->old_root.level = btrfs_header_level(old_root); | |
703 | tm->generation = btrfs_header_generation(old_root); | |
704 | tm->op = MOD_LOG_ROOT_REPLACE; | |
705 | ||
5de865ee FDBM |
706 | if (tree_mod_dont_log(fs_info, NULL)) |
707 | goto free_tms; | |
708 | ||
709 | if (tm_list) | |
710 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
711 | if (!ret) | |
712 | ret = __tree_mod_log_insert(fs_info, tm); | |
713 | ||
714 | tree_mod_log_write_unlock(fs_info); | |
715 | if (ret) | |
716 | goto free_tms; | |
717 | kfree(tm_list); | |
718 | ||
719 | return ret; | |
720 | ||
721 | free_tms: | |
722 | if (tm_list) { | |
723 | for (i = 0; i < nritems; i++) | |
724 | kfree(tm_list[i]); | |
725 | kfree(tm_list); | |
726 | } | |
727 | kfree(tm); | |
728 | ||
729 | return ret; | |
bd989ba3 JS |
730 | } |
731 | ||
732 | static struct tree_mod_elem * | |
733 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
734 | int smallest) | |
735 | { | |
736 | struct rb_root *tm_root; | |
737 | struct rb_node *node; | |
738 | struct tree_mod_elem *cur = NULL; | |
739 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 740 | |
097b8a7c | 741 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
742 | tm_root = &fs_info->tree_mod_log; |
743 | node = tm_root->rb_node; | |
744 | while (node) { | |
6b4df8b6 | 745 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 746 | if (cur->logical < start) { |
bd989ba3 | 747 | node = node->rb_left; |
298cfd36 | 748 | } else if (cur->logical > start) { |
bd989ba3 | 749 | node = node->rb_right; |
097b8a7c | 750 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
751 | node = node->rb_left; |
752 | } else if (!smallest) { | |
753 | /* we want the node with the highest seq */ | |
754 | if (found) | |
097b8a7c | 755 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
756 | found = cur; |
757 | node = node->rb_left; | |
097b8a7c | 758 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
759 | /* we want the node with the smallest seq */ |
760 | if (found) | |
097b8a7c | 761 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
762 | found = cur; |
763 | node = node->rb_right; | |
764 | } else { | |
765 | found = cur; | |
766 | break; | |
767 | } | |
768 | } | |
097b8a7c | 769 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
770 | |
771 | return found; | |
772 | } | |
773 | ||
774 | /* | |
775 | * this returns the element from the log with the smallest time sequence | |
776 | * value that's in the log (the oldest log item). any element with a time | |
777 | * sequence lower than min_seq will be ignored. | |
778 | */ | |
779 | static struct tree_mod_elem * | |
780 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
781 | u64 min_seq) | |
782 | { | |
783 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
784 | } | |
785 | ||
786 | /* | |
787 | * this returns the element from the log with the largest time sequence | |
788 | * value that's in the log (the most recent log item). any element with | |
789 | * a time sequence lower than min_seq will be ignored. | |
790 | */ | |
791 | static struct tree_mod_elem * | |
792 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
793 | { | |
794 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
795 | } | |
796 | ||
5de865ee | 797 | static noinline int |
bd989ba3 JS |
798 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
799 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 800 | unsigned long src_offset, int nr_items) |
bd989ba3 | 801 | { |
5de865ee FDBM |
802 | int ret = 0; |
803 | struct tree_mod_elem **tm_list = NULL; | |
804 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 805 | int i; |
5de865ee | 806 | int locked = 0; |
bd989ba3 | 807 | |
5de865ee FDBM |
808 | if (!tree_mod_need_log(fs_info, NULL)) |
809 | return 0; | |
bd989ba3 | 810 | |
c8cc6341 | 811 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
812 | return 0; |
813 | ||
31e818fe | 814 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
815 | GFP_NOFS); |
816 | if (!tm_list) | |
817 | return -ENOMEM; | |
bd989ba3 | 818 | |
5de865ee FDBM |
819 | tm_list_add = tm_list; |
820 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 821 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
822 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
823 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
824 | if (!tm_list_rem[i]) { | |
825 | ret = -ENOMEM; | |
826 | goto free_tms; | |
827 | } | |
828 | ||
829 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
830 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
831 | if (!tm_list_add[i]) { | |
832 | ret = -ENOMEM; | |
833 | goto free_tms; | |
834 | } | |
835 | } | |
836 | ||
837 | if (tree_mod_dont_log(fs_info, NULL)) | |
838 | goto free_tms; | |
839 | locked = 1; | |
840 | ||
841 | for (i = 0; i < nr_items; i++) { | |
842 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
843 | if (ret) | |
844 | goto free_tms; | |
845 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
846 | if (ret) | |
847 | goto free_tms; | |
bd989ba3 | 848 | } |
5de865ee FDBM |
849 | |
850 | tree_mod_log_write_unlock(fs_info); | |
851 | kfree(tm_list); | |
852 | ||
853 | return 0; | |
854 | ||
855 | free_tms: | |
856 | for (i = 0; i < nr_items * 2; i++) { | |
857 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
858 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
859 | kfree(tm_list[i]); | |
860 | } | |
861 | if (locked) | |
862 | tree_mod_log_write_unlock(fs_info); | |
863 | kfree(tm_list); | |
864 | ||
865 | return ret; | |
bd989ba3 JS |
866 | } |
867 | ||
868 | static inline void | |
869 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
870 | int dst_offset, int src_offset, int nr_items) | |
871 | { | |
872 | int ret; | |
6074d45f | 873 | ret = tree_mod_log_insert_move(dst, dst_offset, src_offset, nr_items); |
bd989ba3 JS |
874 | BUG_ON(ret < 0); |
875 | } | |
876 | ||
3ac6de1a DS |
877 | static noinline void tree_mod_log_set_node_key(struct extent_buffer *eb, |
878 | int slot, int atomic) | |
bd989ba3 JS |
879 | { |
880 | int ret; | |
881 | ||
3ac6de1a | 882 | ret = tree_mod_log_insert_key(eb->fs_info, eb, slot, |
c8cc6341 JB |
883 | MOD_LOG_KEY_REPLACE, |
884 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
885 | BUG_ON(ret < 0); |
886 | } | |
887 | ||
5de865ee | 888 | static noinline int |
097b8a7c | 889 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) |
bd989ba3 | 890 | { |
5de865ee FDBM |
891 | struct tree_mod_elem **tm_list = NULL; |
892 | int nritems = 0; | |
893 | int i; | |
894 | int ret = 0; | |
895 | ||
896 | if (btrfs_header_level(eb) == 0) | |
897 | return 0; | |
898 | ||
899 | if (!tree_mod_need_log(fs_info, NULL)) | |
900 | return 0; | |
901 | ||
902 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 903 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
904 | if (!tm_list) |
905 | return -ENOMEM; | |
906 | ||
907 | for (i = 0; i < nritems; i++) { | |
908 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
909 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
910 | if (!tm_list[i]) { | |
911 | ret = -ENOMEM; | |
912 | goto free_tms; | |
913 | } | |
914 | } | |
915 | ||
e9b7fd4d | 916 | if (tree_mod_dont_log(fs_info, eb)) |
5de865ee FDBM |
917 | goto free_tms; |
918 | ||
919 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
920 | tree_mod_log_write_unlock(fs_info); | |
921 | if (ret) | |
922 | goto free_tms; | |
923 | kfree(tm_list); | |
924 | ||
925 | return 0; | |
926 | ||
927 | free_tms: | |
928 | for (i = 0; i < nritems; i++) | |
929 | kfree(tm_list[i]); | |
930 | kfree(tm_list); | |
931 | ||
932 | return ret; | |
bd989ba3 JS |
933 | } |
934 | ||
097b8a7c | 935 | static noinline void |
bd989ba3 | 936 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
937 | struct extent_buffer *new_root_node, |
938 | int log_removal) | |
bd989ba3 JS |
939 | { |
940 | int ret; | |
bd989ba3 | 941 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
bcc8e07f | 942 | new_root_node, log_removal); |
bd989ba3 JS |
943 | BUG_ON(ret < 0); |
944 | } | |
945 | ||
5d4f98a2 YZ |
946 | /* |
947 | * check if the tree block can be shared by multiple trees | |
948 | */ | |
949 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
950 | struct extent_buffer *buf) | |
951 | { | |
952 | /* | |
01327610 | 953 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
954 | * are never shared. If a block was allocated after the last |
955 | * snapshot and the block was not allocated by tree relocation, | |
956 | * we know the block is not shared. | |
957 | */ | |
27cdeb70 | 958 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
959 | buf != root->node && buf != root->commit_root && |
960 | (btrfs_header_generation(buf) <= | |
961 | btrfs_root_last_snapshot(&root->root_item) || | |
962 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
963 | return 1; | |
964 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 965 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
966 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
967 | return 1; | |
968 | #endif | |
969 | return 0; | |
970 | } | |
971 | ||
972 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
973 | struct btrfs_root *root, | |
974 | struct extent_buffer *buf, | |
f0486c68 YZ |
975 | struct extent_buffer *cow, |
976 | int *last_ref) | |
5d4f98a2 | 977 | { |
0b246afa | 978 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
979 | u64 refs; |
980 | u64 owner; | |
981 | u64 flags; | |
982 | u64 new_flags = 0; | |
983 | int ret; | |
984 | ||
985 | /* | |
986 | * Backrefs update rules: | |
987 | * | |
988 | * Always use full backrefs for extent pointers in tree block | |
989 | * allocated by tree relocation. | |
990 | * | |
991 | * If a shared tree block is no longer referenced by its owner | |
992 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
993 | * use full backrefs for extent pointers in tree block. | |
994 | * | |
995 | * If a tree block is been relocating | |
996 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
997 | * use full backrefs for extent pointers in tree block. | |
998 | * The reason for this is some operations (such as drop tree) | |
999 | * are only allowed for blocks use full backrefs. | |
1000 | */ | |
1001 | ||
1002 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 1003 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
1004 | btrfs_header_level(buf), 1, |
1005 | &refs, &flags); | |
be1a5564 MF |
1006 | if (ret) |
1007 | return ret; | |
e5df9573 MF |
1008 | if (refs == 0) { |
1009 | ret = -EROFS; | |
0b246afa | 1010 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
1011 | return ret; |
1012 | } | |
5d4f98a2 YZ |
1013 | } else { |
1014 | refs = 1; | |
1015 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
1016 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1017 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1018 | else | |
1019 | flags = 0; | |
1020 | } | |
1021 | ||
1022 | owner = btrfs_header_owner(buf); | |
1023 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
1024 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
1025 | ||
1026 | if (refs > 1) { | |
1027 | if ((owner == root->root_key.objectid || | |
1028 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
1029 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 1030 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
1031 | if (ret) |
1032 | return ret; | |
5d4f98a2 YZ |
1033 | |
1034 | if (root->root_key.objectid == | |
1035 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 1036 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
1037 | if (ret) |
1038 | return ret; | |
e339a6b0 | 1039 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
1040 | if (ret) |
1041 | return ret; | |
5d4f98a2 YZ |
1042 | } |
1043 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1044 | } else { | |
1045 | ||
1046 | if (root->root_key.objectid == | |
1047 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1048 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1049 | else |
e339a6b0 | 1050 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
1051 | if (ret) |
1052 | return ret; | |
5d4f98a2 YZ |
1053 | } |
1054 | if (new_flags != 0) { | |
b1c79e09 JB |
1055 | int level = btrfs_header_level(buf); |
1056 | ||
2ff7e61e | 1057 | ret = btrfs_set_disk_extent_flags(trans, fs_info, |
5d4f98a2 YZ |
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); |
692826b2 JM |
1071 | if (ret) |
1072 | return ret; | |
e339a6b0 | 1073 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
1074 | if (ret) |
1075 | return ret; | |
5d4f98a2 | 1076 | } |
7c302b49 | 1077 | clean_tree_block(fs_info, buf); |
f0486c68 | 1078 | *last_ref = 1; |
5d4f98a2 YZ |
1079 | } |
1080 | return 0; | |
1081 | } | |
1082 | ||
d352ac68 | 1083 | /* |
d397712b CM |
1084 | * does the dirty work in cow of a single block. The parent block (if |
1085 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1086 | * dirty and returned locked. If you modify the block it needs to be marked | |
1087 | * dirty again. | |
d352ac68 CM |
1088 | * |
1089 | * search_start -- an allocation hint for the new block | |
1090 | * | |
d397712b CM |
1091 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1092 | * bytes the allocator should try to find free next to the block it returns. | |
1093 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1094 | */ |
d397712b | 1095 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1096 | struct btrfs_root *root, |
1097 | struct extent_buffer *buf, | |
1098 | struct extent_buffer *parent, int parent_slot, | |
1099 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1100 | u64 search_start, u64 empty_size) |
02217ed2 | 1101 | { |
0b246afa | 1102 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1103 | struct btrfs_disk_key disk_key; |
5f39d397 | 1104 | struct extent_buffer *cow; |
be1a5564 | 1105 | int level, ret; |
f0486c68 | 1106 | int last_ref = 0; |
925baedd | 1107 | int unlock_orig = 0; |
0f5053eb | 1108 | u64 parent_start = 0; |
7bb86316 | 1109 | |
925baedd CM |
1110 | if (*cow_ret == buf) |
1111 | unlock_orig = 1; | |
1112 | ||
b9447ef8 | 1113 | btrfs_assert_tree_locked(buf); |
925baedd | 1114 | |
27cdeb70 | 1115 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1116 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1117 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1118 | trans->transid != root->last_trans); | |
5f39d397 | 1119 | |
7bb86316 | 1120 | level = btrfs_header_level(buf); |
31840ae1 | 1121 | |
5d4f98a2 YZ |
1122 | if (level == 0) |
1123 | btrfs_item_key(buf, &disk_key, 0); | |
1124 | else | |
1125 | btrfs_node_key(buf, &disk_key, 0); | |
1126 | ||
0f5053eb GR |
1127 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1128 | parent_start = parent->start; | |
5d4f98a2 | 1129 | |
4d75f8a9 DS |
1130 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1131 | root->root_key.objectid, &disk_key, level, | |
1132 | search_start, empty_size); | |
54aa1f4d CM |
1133 | if (IS_ERR(cow)) |
1134 | return PTR_ERR(cow); | |
6702ed49 | 1135 | |
b4ce94de CM |
1136 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1137 | ||
58e8012c | 1138 | copy_extent_buffer_full(cow, buf); |
db94535d | 1139 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1140 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1141 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1142 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1143 | BTRFS_HEADER_FLAG_RELOC); | |
1144 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1145 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1146 | else | |
1147 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1148 | |
0b246afa | 1149 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 1150 | |
be1a5564 | 1151 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1152 | if (ret) { |
66642832 | 1153 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1154 | return ret; |
1155 | } | |
1a40e23b | 1156 | |
27cdeb70 | 1157 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1158 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1159 | if (ret) { |
66642832 | 1160 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1161 | return ret; |
93314e3b | 1162 | } |
83d4cfd4 | 1163 | } |
3fd0a558 | 1164 | |
02217ed2 | 1165 | if (buf == root->node) { |
925baedd | 1166 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1167 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1168 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1169 | parent_start = buf->start; | |
925baedd | 1170 | |
5f39d397 | 1171 | extent_buffer_get(cow); |
90f8d62e | 1172 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1173 | rcu_assign_pointer(root->node, cow); |
925baedd | 1174 | |
f0486c68 | 1175 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1176 | last_ref); |
5f39d397 | 1177 | free_extent_buffer(buf); |
0b86a832 | 1178 | add_root_to_dirty_list(root); |
02217ed2 | 1179 | } else { |
5d4f98a2 | 1180 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
0b246afa | 1181 | tree_mod_log_insert_key(fs_info, parent, parent_slot, |
c8cc6341 | 1182 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1183 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1184 | cow->start); |
74493f7a CM |
1185 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1186 | trans->transid); | |
d6025579 | 1187 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1188 | if (last_ref) { |
0b246afa | 1189 | ret = tree_mod_log_free_eb(fs_info, buf); |
5de865ee | 1190 | if (ret) { |
66642832 | 1191 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1192 | return ret; |
1193 | } | |
1194 | } | |
f0486c68 | 1195 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1196 | last_ref); |
02217ed2 | 1197 | } |
925baedd CM |
1198 | if (unlock_orig) |
1199 | btrfs_tree_unlock(buf); | |
3083ee2e | 1200 | free_extent_buffer_stale(buf); |
ccd467d6 | 1201 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1202 | *cow_ret = cow; |
02217ed2 CM |
1203 | return 0; |
1204 | } | |
1205 | ||
5d9e75c4 JS |
1206 | /* |
1207 | * returns the logical address of the oldest predecessor of the given root. | |
1208 | * entries older than time_seq are ignored. | |
1209 | */ | |
1210 | static struct tree_mod_elem * | |
1211 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1212 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1213 | { |
1214 | struct tree_mod_elem *tm; | |
1215 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1216 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1217 | int looped = 0; |
1218 | ||
1219 | if (!time_seq) | |
35a3621b | 1220 | return NULL; |
5d9e75c4 JS |
1221 | |
1222 | /* | |
298cfd36 CR |
1223 | * the very last operation that's logged for a root is the |
1224 | * replacement operation (if it is replaced at all). this has | |
1225 | * the logical address of the *new* root, making it the very | |
1226 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1227 | */ |
1228 | while (1) { | |
1229 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1230 | time_seq); | |
1231 | if (!looped && !tm) | |
35a3621b | 1232 | return NULL; |
5d9e75c4 | 1233 | /* |
28da9fb4 JS |
1234 | * if there are no tree operation for the oldest root, we simply |
1235 | * return it. this should only happen if that (old) root is at | |
1236 | * level 0. | |
5d9e75c4 | 1237 | */ |
28da9fb4 JS |
1238 | if (!tm) |
1239 | break; | |
5d9e75c4 | 1240 | |
28da9fb4 JS |
1241 | /* |
1242 | * if there's an operation that's not a root replacement, we | |
1243 | * found the oldest version of our root. normally, we'll find a | |
1244 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1245 | */ | |
5d9e75c4 JS |
1246 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1247 | break; | |
1248 | ||
1249 | found = tm; | |
1250 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1251 | looped = 1; |
1252 | } | |
1253 | ||
a95236d9 JS |
1254 | /* if there's no old root to return, return what we found instead */ |
1255 | if (!found) | |
1256 | found = tm; | |
1257 | ||
5d9e75c4 JS |
1258 | return found; |
1259 | } | |
1260 | ||
1261 | /* | |
1262 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1263 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1264 | * time_seq). |
1265 | */ | |
1266 | static void | |
f1ca7e98 JB |
1267 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1268 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1269 | { |
1270 | u32 n; | |
1271 | struct rb_node *next; | |
1272 | struct tree_mod_elem *tm = first_tm; | |
1273 | unsigned long o_dst; | |
1274 | unsigned long o_src; | |
1275 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1276 | ||
1277 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1278 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1279 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1280 | /* |
1281 | * all the operations are recorded with the operator used for | |
1282 | * the modification. as we're going backwards, we do the | |
1283 | * opposite of each operation here. | |
1284 | */ | |
1285 | switch (tm->op) { | |
1286 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1287 | BUG_ON(tm->slot < n); | |
1c697d4a | 1288 | /* Fallthrough */ |
95c80bb1 | 1289 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1290 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1291 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1292 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1293 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1294 | tm->generation); | |
4c3e6969 | 1295 | n++; |
5d9e75c4 JS |
1296 | break; |
1297 | case MOD_LOG_KEY_REPLACE: | |
1298 | BUG_ON(tm->slot >= n); | |
1299 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1300 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1301 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1302 | tm->generation); | |
1303 | break; | |
1304 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1305 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1306 | n--; |
1307 | break; | |
1308 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1309 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1310 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1311 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1312 | tm->move.nr_items * p_size); |
1313 | break; | |
1314 | case MOD_LOG_ROOT_REPLACE: | |
1315 | /* | |
1316 | * this operation is special. for roots, this must be | |
1317 | * handled explicitly before rewinding. | |
1318 | * for non-roots, this operation may exist if the node | |
1319 | * was a root: root A -> child B; then A gets empty and | |
1320 | * B is promoted to the new root. in the mod log, we'll | |
1321 | * have a root-replace operation for B, a tree block | |
1322 | * that is no root. we simply ignore that operation. | |
1323 | */ | |
1324 | break; | |
1325 | } | |
1326 | next = rb_next(&tm->node); | |
1327 | if (!next) | |
1328 | break; | |
6b4df8b6 | 1329 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1330 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1331 | break; |
1332 | } | |
f1ca7e98 | 1333 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1334 | btrfs_set_header_nritems(eb, n); |
1335 | } | |
1336 | ||
47fb091f | 1337 | /* |
01327610 | 1338 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1339 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1340 | * returned buffer is always read-locked. If the returned buffer is not the | |
1341 | * input buffer, the lock on the input buffer is released and the input buffer | |
1342 | * is freed (its refcount is decremented). | |
1343 | */ | |
5d9e75c4 | 1344 | static struct extent_buffer * |
9ec72677 JB |
1345 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1346 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1347 | { |
1348 | struct extent_buffer *eb_rewin; | |
1349 | struct tree_mod_elem *tm; | |
1350 | ||
1351 | if (!time_seq) | |
1352 | return eb; | |
1353 | ||
1354 | if (btrfs_header_level(eb) == 0) | |
1355 | return eb; | |
1356 | ||
1357 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1358 | if (!tm) | |
1359 | return eb; | |
1360 | ||
9ec72677 JB |
1361 | btrfs_set_path_blocking(path); |
1362 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1363 | ||
5d9e75c4 JS |
1364 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1365 | BUG_ON(tm->slot != 0); | |
da17066c | 1366 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1367 | if (!eb_rewin) { |
9ec72677 | 1368 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1369 | free_extent_buffer(eb); |
1370 | return NULL; | |
1371 | } | |
5d9e75c4 JS |
1372 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1373 | btrfs_set_header_backref_rev(eb_rewin, | |
1374 | btrfs_header_backref_rev(eb)); | |
1375 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1376 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1377 | } else { |
1378 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1379 | if (!eb_rewin) { |
9ec72677 | 1380 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1381 | free_extent_buffer(eb); |
1382 | return NULL; | |
1383 | } | |
5d9e75c4 JS |
1384 | } |
1385 | ||
9ec72677 JB |
1386 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1387 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1388 | free_extent_buffer(eb); |
1389 | ||
47fb091f JS |
1390 | extent_buffer_get(eb_rewin); |
1391 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1392 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1393 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1394 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1395 | |
1396 | return eb_rewin; | |
1397 | } | |
1398 | ||
8ba97a15 JS |
1399 | /* |
1400 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1401 | * value. If there are no changes, the current root->root_node is returned. If | |
1402 | * anything changed in between, there's a fresh buffer allocated on which the | |
1403 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1404 | * Returns NULL on error (with no locks held). | |
1405 | */ | |
5d9e75c4 JS |
1406 | static inline struct extent_buffer * |
1407 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1408 | { | |
0b246afa | 1409 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1410 | struct tree_mod_elem *tm; |
30b0463a JS |
1411 | struct extent_buffer *eb = NULL; |
1412 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1413 | struct extent_buffer *old; |
a95236d9 | 1414 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1415 | u64 old_generation = 0; |
a95236d9 | 1416 | u64 logical; |
5d9e75c4 | 1417 | |
30b0463a | 1418 | eb_root = btrfs_read_lock_root_node(root); |
0b246afa | 1419 | tm = __tree_mod_log_oldest_root(fs_info, eb_root, time_seq); |
5d9e75c4 | 1420 | if (!tm) |
30b0463a | 1421 | return eb_root; |
5d9e75c4 | 1422 | |
a95236d9 JS |
1423 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1424 | old_root = &tm->old_root; | |
1425 | old_generation = tm->generation; | |
1426 | logical = old_root->logical; | |
1427 | } else { | |
30b0463a | 1428 | logical = eb_root->start; |
a95236d9 | 1429 | } |
5d9e75c4 | 1430 | |
0b246afa | 1431 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1432 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1433 | btrfs_tree_read_unlock(eb_root); |
1434 | free_extent_buffer(eb_root); | |
2ff7e61e | 1435 | old = read_tree_block(fs_info, logical, 0); |
64c043de LB |
1436 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1437 | if (!IS_ERR(old)) | |
1438 | free_extent_buffer(old); | |
0b246afa JM |
1439 | btrfs_warn(fs_info, |
1440 | "failed to read tree block %llu from get_old_root", | |
1441 | logical); | |
834328a8 | 1442 | } else { |
7bfdcf7f LB |
1443 | eb = btrfs_clone_extent_buffer(old); |
1444 | free_extent_buffer(old); | |
834328a8 JS |
1445 | } |
1446 | } else if (old_root) { | |
30b0463a JS |
1447 | btrfs_tree_read_unlock(eb_root); |
1448 | free_extent_buffer(eb_root); | |
0b246afa | 1449 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1450 | } else { |
9ec72677 | 1451 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1452 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1453 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1454 | free_extent_buffer(eb_root); |
834328a8 JS |
1455 | } |
1456 | ||
8ba97a15 JS |
1457 | if (!eb) |
1458 | return NULL; | |
d6381084 | 1459 | extent_buffer_get(eb); |
8ba97a15 | 1460 | btrfs_tree_read_lock(eb); |
a95236d9 | 1461 | if (old_root) { |
5d9e75c4 JS |
1462 | btrfs_set_header_bytenr(eb, eb->start); |
1463 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1464 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1465 | btrfs_set_header_level(eb, old_root->level); |
1466 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1467 | } |
28da9fb4 | 1468 | if (tm) |
0b246afa | 1469 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1470 | else |
1471 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1472 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1473 | |
1474 | return eb; | |
1475 | } | |
1476 | ||
5b6602e7 JS |
1477 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1478 | { | |
1479 | struct tree_mod_elem *tm; | |
1480 | int level; | |
30b0463a | 1481 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1482 | |
30b0463a | 1483 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1484 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1485 | level = tm->old_root.level; | |
1486 | } else { | |
30b0463a | 1487 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1488 | } |
30b0463a | 1489 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1490 | |
1491 | return level; | |
1492 | } | |
1493 | ||
5d4f98a2 YZ |
1494 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1495 | struct btrfs_root *root, | |
1496 | struct extent_buffer *buf) | |
1497 | { | |
f5ee5c9a | 1498 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1499 | return 0; |
fccb84c9 | 1500 | |
f1ebcc74 LB |
1501 | /* ensure we can see the force_cow */ |
1502 | smp_rmb(); | |
1503 | ||
1504 | /* | |
1505 | * We do not need to cow a block if | |
1506 | * 1) this block is not created or changed in this transaction; | |
1507 | * 2) this block does not belong to TREE_RELOC tree; | |
1508 | * 3) the root is not forced COW. | |
1509 | * | |
1510 | * What is forced COW: | |
01327610 | 1511 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1512 | * after we've finished coping src root, we must COW the shared |
1513 | * block to ensure the metadata consistency. | |
1514 | */ | |
5d4f98a2 YZ |
1515 | if (btrfs_header_generation(buf) == trans->transid && |
1516 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1517 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1518 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1519 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1520 | return 0; |
1521 | return 1; | |
1522 | } | |
1523 | ||
d352ac68 CM |
1524 | /* |
1525 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1526 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1527 | * once per transaction, as long as it hasn't been written yet |
1528 | */ | |
d397712b | 1529 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1530 | struct btrfs_root *root, struct extent_buffer *buf, |
1531 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1532 | struct extent_buffer **cow_ret) |
6702ed49 | 1533 | { |
0b246afa | 1534 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1535 | u64 search_start; |
f510cfec | 1536 | int ret; |
dc17ff8f | 1537 | |
0b246afa | 1538 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1539 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1540 | trans->transid, |
0b246afa | 1541 | fs_info->running_transaction->transid); |
31b1a2bd | 1542 | |
0b246afa | 1543 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1544 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1545 | trans->transid, fs_info->generation); |
dc17ff8f | 1546 | |
5d4f98a2 | 1547 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1548 | trans->dirty = true; |
6702ed49 CM |
1549 | *cow_ret = buf; |
1550 | return 0; | |
1551 | } | |
c487685d | 1552 | |
ee22184b | 1553 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1554 | |
1555 | if (parent) | |
1556 | btrfs_set_lock_blocking(parent); | |
1557 | btrfs_set_lock_blocking(buf); | |
1558 | ||
f510cfec | 1559 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1560 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1561 | |
1562 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1563 | ||
f510cfec | 1564 | return ret; |
6702ed49 CM |
1565 | } |
1566 | ||
d352ac68 CM |
1567 | /* |
1568 | * helper function for defrag to decide if two blocks pointed to by a | |
1569 | * node are actually close by | |
1570 | */ | |
6b80053d | 1571 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1572 | { |
6b80053d | 1573 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1574 | return 1; |
6b80053d | 1575 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1576 | return 1; |
1577 | return 0; | |
1578 | } | |
1579 | ||
081e9573 CM |
1580 | /* |
1581 | * compare two keys in a memcmp fashion | |
1582 | */ | |
310712b2 OS |
1583 | static int comp_keys(const struct btrfs_disk_key *disk, |
1584 | const struct btrfs_key *k2) | |
081e9573 CM |
1585 | { |
1586 | struct btrfs_key k1; | |
1587 | ||
1588 | btrfs_disk_key_to_cpu(&k1, disk); | |
1589 | ||
20736aba | 1590 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1591 | } |
1592 | ||
f3465ca4 JB |
1593 | /* |
1594 | * same as comp_keys only with two btrfs_key's | |
1595 | */ | |
310712b2 | 1596 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1597 | { |
1598 | if (k1->objectid > k2->objectid) | |
1599 | return 1; | |
1600 | if (k1->objectid < k2->objectid) | |
1601 | return -1; | |
1602 | if (k1->type > k2->type) | |
1603 | return 1; | |
1604 | if (k1->type < k2->type) | |
1605 | return -1; | |
1606 | if (k1->offset > k2->offset) | |
1607 | return 1; | |
1608 | if (k1->offset < k2->offset) | |
1609 | return -1; | |
1610 | return 0; | |
1611 | } | |
081e9573 | 1612 | |
d352ac68 CM |
1613 | /* |
1614 | * this is used by the defrag code to go through all the | |
1615 | * leaves pointed to by a node and reallocate them so that | |
1616 | * disk order is close to key order | |
1617 | */ | |
6702ed49 | 1618 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1619 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1620 | int start_slot, u64 *last_ret, |
a6b6e75e | 1621 | struct btrfs_key *progress) |
6702ed49 | 1622 | { |
0b246afa | 1623 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1624 | struct extent_buffer *cur; |
6702ed49 | 1625 | u64 blocknr; |
ca7a79ad | 1626 | u64 gen; |
e9d0b13b CM |
1627 | u64 search_start = *last_ret; |
1628 | u64 last_block = 0; | |
6702ed49 CM |
1629 | u64 other; |
1630 | u32 parent_nritems; | |
6702ed49 CM |
1631 | int end_slot; |
1632 | int i; | |
1633 | int err = 0; | |
f2183bde | 1634 | int parent_level; |
6b80053d CM |
1635 | int uptodate; |
1636 | u32 blocksize; | |
081e9573 CM |
1637 | int progress_passed = 0; |
1638 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1639 | |
5708b959 | 1640 | parent_level = btrfs_header_level(parent); |
5708b959 | 1641 | |
0b246afa JM |
1642 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1643 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1644 | |
6b80053d | 1645 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1646 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1647 | end_slot = parent_nritems - 1; |
6702ed49 | 1648 | |
5dfe2be7 | 1649 | if (parent_nritems <= 1) |
6702ed49 CM |
1650 | return 0; |
1651 | ||
b4ce94de CM |
1652 | btrfs_set_lock_blocking(parent); |
1653 | ||
5dfe2be7 | 1654 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 1655 | int close = 1; |
a6b6e75e | 1656 | |
081e9573 CM |
1657 | btrfs_node_key(parent, &disk_key, i); |
1658 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1659 | continue; | |
1660 | ||
1661 | progress_passed = 1; | |
6b80053d | 1662 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1663 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1664 | if (last_block == 0) |
1665 | last_block = blocknr; | |
5708b959 | 1666 | |
6702ed49 | 1667 | if (i > 0) { |
6b80053d CM |
1668 | other = btrfs_node_blockptr(parent, i - 1); |
1669 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1670 | } |
5dfe2be7 | 1671 | if (!close && i < end_slot) { |
6b80053d CM |
1672 | other = btrfs_node_blockptr(parent, i + 1); |
1673 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1674 | } |
e9d0b13b CM |
1675 | if (close) { |
1676 | last_block = blocknr; | |
6702ed49 | 1677 | continue; |
e9d0b13b | 1678 | } |
6702ed49 | 1679 | |
0b246afa | 1680 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1681 | if (cur) |
b9fab919 | 1682 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1683 | else |
1684 | uptodate = 0; | |
5708b959 | 1685 | if (!cur || !uptodate) { |
6b80053d | 1686 | if (!cur) { |
2ff7e61e | 1687 | cur = read_tree_block(fs_info, blocknr, gen); |
64c043de LB |
1688 | if (IS_ERR(cur)) { |
1689 | return PTR_ERR(cur); | |
1690 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1691 | free_extent_buffer(cur); |
97d9a8a4 | 1692 | return -EIO; |
416bc658 | 1693 | } |
6b80053d | 1694 | } else if (!uptodate) { |
018642a1 TI |
1695 | err = btrfs_read_buffer(cur, gen); |
1696 | if (err) { | |
1697 | free_extent_buffer(cur); | |
1698 | return err; | |
1699 | } | |
f2183bde | 1700 | } |
6702ed49 | 1701 | } |
e9d0b13b | 1702 | if (search_start == 0) |
6b80053d | 1703 | search_start = last_block; |
e9d0b13b | 1704 | |
e7a84565 | 1705 | btrfs_tree_lock(cur); |
b4ce94de | 1706 | btrfs_set_lock_blocking(cur); |
6b80053d | 1707 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1708 | &cur, search_start, |
6b80053d | 1709 | min(16 * blocksize, |
9fa8cfe7 | 1710 | (end_slot - i) * blocksize)); |
252c38f0 | 1711 | if (err) { |
e7a84565 | 1712 | btrfs_tree_unlock(cur); |
6b80053d | 1713 | free_extent_buffer(cur); |
6702ed49 | 1714 | break; |
252c38f0 | 1715 | } |
e7a84565 CM |
1716 | search_start = cur->start; |
1717 | last_block = cur->start; | |
f2183bde | 1718 | *last_ret = search_start; |
e7a84565 CM |
1719 | btrfs_tree_unlock(cur); |
1720 | free_extent_buffer(cur); | |
6702ed49 CM |
1721 | } |
1722 | return err; | |
1723 | } | |
1724 | ||
74123bd7 | 1725 | /* |
5f39d397 CM |
1726 | * search for key in the extent_buffer. The items start at offset p, |
1727 | * and they are item_size apart. There are 'max' items in p. | |
1728 | * | |
74123bd7 CM |
1729 | * the slot in the array is returned via slot, and it points to |
1730 | * the place where you would insert key if it is not found in | |
1731 | * the array. | |
1732 | * | |
1733 | * slot may point to max if the key is bigger than all of the keys | |
1734 | */ | |
e02119d5 | 1735 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1736 | unsigned long p, int item_size, |
1737 | const struct btrfs_key *key, | |
e02119d5 | 1738 | int max, int *slot) |
be0e5c09 CM |
1739 | { |
1740 | int low = 0; | |
1741 | int high = max; | |
1742 | int mid; | |
1743 | int ret; | |
479965d6 | 1744 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1745 | struct btrfs_disk_key unaligned; |
1746 | unsigned long offset; | |
5f39d397 CM |
1747 | char *kaddr = NULL; |
1748 | unsigned long map_start = 0; | |
1749 | unsigned long map_len = 0; | |
479965d6 | 1750 | int err; |
be0e5c09 | 1751 | |
5e24e9af LB |
1752 | if (low > high) { |
1753 | btrfs_err(eb->fs_info, | |
1754 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1755 | __func__, low, high, eb->start, | |
1756 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1757 | return -EINVAL; | |
1758 | } | |
1759 | ||
d397712b | 1760 | while (low < high) { |
be0e5c09 | 1761 | mid = (low + high) / 2; |
5f39d397 CM |
1762 | offset = p + mid * item_size; |
1763 | ||
a6591715 | 1764 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1765 | (offset + sizeof(struct btrfs_disk_key)) > |
1766 | map_start + map_len) { | |
934d375b CM |
1767 | |
1768 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1769 | sizeof(struct btrfs_disk_key), |
a6591715 | 1770 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1771 | |
1772 | if (!err) { | |
1773 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1774 | map_start); | |
415b35a5 | 1775 | } else if (err == 1) { |
479965d6 CM |
1776 | read_extent_buffer(eb, &unaligned, |
1777 | offset, sizeof(unaligned)); | |
1778 | tmp = &unaligned; | |
415b35a5 LB |
1779 | } else { |
1780 | return err; | |
479965d6 | 1781 | } |
5f39d397 | 1782 | |
5f39d397 CM |
1783 | } else { |
1784 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1785 | map_start); | |
1786 | } | |
be0e5c09 CM |
1787 | ret = comp_keys(tmp, key); |
1788 | ||
1789 | if (ret < 0) | |
1790 | low = mid + 1; | |
1791 | else if (ret > 0) | |
1792 | high = mid; | |
1793 | else { | |
1794 | *slot = mid; | |
1795 | return 0; | |
1796 | } | |
1797 | } | |
1798 | *slot = low; | |
1799 | return 1; | |
1800 | } | |
1801 | ||
97571fd0 CM |
1802 | /* |
1803 | * simple bin_search frontend that does the right thing for | |
1804 | * leaves vs nodes | |
1805 | */ | |
a74b35ec NB |
1806 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1807 | int level, int *slot) | |
be0e5c09 | 1808 | { |
f775738f | 1809 | if (level == 0) |
5f39d397 CM |
1810 | return generic_bin_search(eb, |
1811 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1812 | sizeof(struct btrfs_item), |
5f39d397 | 1813 | key, btrfs_header_nritems(eb), |
7518a238 | 1814 | slot); |
f775738f | 1815 | else |
5f39d397 CM |
1816 | return generic_bin_search(eb, |
1817 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1818 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1819 | key, btrfs_header_nritems(eb), |
7518a238 | 1820 | slot); |
be0e5c09 CM |
1821 | } |
1822 | ||
f0486c68 YZ |
1823 | static void root_add_used(struct btrfs_root *root, u32 size) |
1824 | { | |
1825 | spin_lock(&root->accounting_lock); | |
1826 | btrfs_set_root_used(&root->root_item, | |
1827 | btrfs_root_used(&root->root_item) + size); | |
1828 | spin_unlock(&root->accounting_lock); | |
1829 | } | |
1830 | ||
1831 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1832 | { | |
1833 | spin_lock(&root->accounting_lock); | |
1834 | btrfs_set_root_used(&root->root_item, | |
1835 | btrfs_root_used(&root->root_item) - size); | |
1836 | spin_unlock(&root->accounting_lock); | |
1837 | } | |
1838 | ||
d352ac68 CM |
1839 | /* given a node and slot number, this reads the blocks it points to. The |
1840 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1841 | */ |
2ff7e61e JM |
1842 | static noinline struct extent_buffer * |
1843 | read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent, | |
1844 | int slot) | |
bb803951 | 1845 | { |
ca7a79ad | 1846 | int level = btrfs_header_level(parent); |
416bc658 JB |
1847 | struct extent_buffer *eb; |
1848 | ||
fb770ae4 LB |
1849 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1850 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1851 | |
1852 | BUG_ON(level == 0); | |
1853 | ||
2ff7e61e | 1854 | eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot), |
416bc658 | 1855 | btrfs_node_ptr_generation(parent, slot)); |
fb770ae4 LB |
1856 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1857 | free_extent_buffer(eb); | |
1858 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1859 | } |
1860 | ||
1861 | return eb; | |
bb803951 CM |
1862 | } |
1863 | ||
d352ac68 CM |
1864 | /* |
1865 | * node level balancing, used to make sure nodes are in proper order for | |
1866 | * item deletion. We balance from the top down, so we have to make sure | |
1867 | * that a deletion won't leave an node completely empty later on. | |
1868 | */ | |
e02119d5 | 1869 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1870 | struct btrfs_root *root, |
1871 | struct btrfs_path *path, int level) | |
bb803951 | 1872 | { |
0b246afa | 1873 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1874 | struct extent_buffer *right = NULL; |
1875 | struct extent_buffer *mid; | |
1876 | struct extent_buffer *left = NULL; | |
1877 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1878 | int ret = 0; |
1879 | int wret; | |
1880 | int pslot; | |
bb803951 | 1881 | int orig_slot = path->slots[level]; |
79f95c82 | 1882 | u64 orig_ptr; |
bb803951 CM |
1883 | |
1884 | if (level == 0) | |
1885 | return 0; | |
1886 | ||
5f39d397 | 1887 | mid = path->nodes[level]; |
b4ce94de | 1888 | |
bd681513 CM |
1889 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1890 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1891 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1892 | ||
1d4f8a0c | 1893 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1894 | |
a05a9bb1 | 1895 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1896 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1897 | pslot = path->slots[level + 1]; |
1898 | } | |
bb803951 | 1899 | |
40689478 CM |
1900 | /* |
1901 | * deal with the case where there is only one pointer in the root | |
1902 | * by promoting the node below to a root | |
1903 | */ | |
5f39d397 CM |
1904 | if (!parent) { |
1905 | struct extent_buffer *child; | |
bb803951 | 1906 | |
5f39d397 | 1907 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1908 | return 0; |
1909 | ||
1910 | /* promote the child to a root */ | |
2ff7e61e | 1911 | child = read_node_slot(fs_info, mid, 0); |
fb770ae4 LB |
1912 | if (IS_ERR(child)) { |
1913 | ret = PTR_ERR(child); | |
0b246afa | 1914 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1915 | goto enospc; |
1916 | } | |
1917 | ||
925baedd | 1918 | btrfs_tree_lock(child); |
b4ce94de | 1919 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1920 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1921 | if (ret) { |
1922 | btrfs_tree_unlock(child); | |
1923 | free_extent_buffer(child); | |
1924 | goto enospc; | |
1925 | } | |
2f375ab9 | 1926 | |
90f8d62e | 1927 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1928 | rcu_assign_pointer(root->node, child); |
925baedd | 1929 | |
0b86a832 | 1930 | add_root_to_dirty_list(root); |
925baedd | 1931 | btrfs_tree_unlock(child); |
b4ce94de | 1932 | |
925baedd | 1933 | path->locks[level] = 0; |
bb803951 | 1934 | path->nodes[level] = NULL; |
7c302b49 | 1935 | clean_tree_block(fs_info, mid); |
925baedd | 1936 | btrfs_tree_unlock(mid); |
bb803951 | 1937 | /* once for the path */ |
5f39d397 | 1938 | free_extent_buffer(mid); |
f0486c68 YZ |
1939 | |
1940 | root_sub_used(root, mid->len); | |
5581a51a | 1941 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1942 | /* once for the root ptr */ |
3083ee2e | 1943 | free_extent_buffer_stale(mid); |
f0486c68 | 1944 | return 0; |
bb803951 | 1945 | } |
5f39d397 | 1946 | if (btrfs_header_nritems(mid) > |
0b246afa | 1947 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1948 | return 0; |
1949 | ||
2ff7e61e | 1950 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
1951 | if (IS_ERR(left)) |
1952 | left = NULL; | |
1953 | ||
5f39d397 | 1954 | if (left) { |
925baedd | 1955 | btrfs_tree_lock(left); |
b4ce94de | 1956 | btrfs_set_lock_blocking(left); |
5f39d397 | 1957 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1958 | parent, pslot - 1, &left); |
54aa1f4d CM |
1959 | if (wret) { |
1960 | ret = wret; | |
1961 | goto enospc; | |
1962 | } | |
2cc58cf2 | 1963 | } |
fb770ae4 | 1964 | |
2ff7e61e | 1965 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
1966 | if (IS_ERR(right)) |
1967 | right = NULL; | |
1968 | ||
5f39d397 | 1969 | if (right) { |
925baedd | 1970 | btrfs_tree_lock(right); |
b4ce94de | 1971 | btrfs_set_lock_blocking(right); |
5f39d397 | 1972 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1973 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1974 | if (wret) { |
1975 | ret = wret; | |
1976 | goto enospc; | |
1977 | } | |
1978 | } | |
1979 | ||
1980 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1981 | if (left) { |
1982 | orig_slot += btrfs_header_nritems(left); | |
2ff7e61e | 1983 | wret = push_node_left(trans, fs_info, left, mid, 1); |
79f95c82 CM |
1984 | if (wret < 0) |
1985 | ret = wret; | |
bb803951 | 1986 | } |
79f95c82 CM |
1987 | |
1988 | /* | |
1989 | * then try to empty the right most buffer into the middle | |
1990 | */ | |
5f39d397 | 1991 | if (right) { |
2ff7e61e | 1992 | wret = push_node_left(trans, fs_info, mid, right, 1); |
54aa1f4d | 1993 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1994 | ret = wret; |
5f39d397 | 1995 | if (btrfs_header_nritems(right) == 0) { |
7c302b49 | 1996 | clean_tree_block(fs_info, right); |
925baedd | 1997 | btrfs_tree_unlock(right); |
afe5fea7 | 1998 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1999 | root_sub_used(root, right->len); |
5581a51a | 2000 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 2001 | free_extent_buffer_stale(right); |
f0486c68 | 2002 | right = NULL; |
bb803951 | 2003 | } else { |
5f39d397 CM |
2004 | struct btrfs_disk_key right_key; |
2005 | btrfs_node_key(right, &right_key, 0); | |
3ac6de1a | 2006 | tree_mod_log_set_node_key(parent, pslot + 1, 0); |
5f39d397 CM |
2007 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
2008 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
2009 | } |
2010 | } | |
5f39d397 | 2011 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
2012 | /* |
2013 | * we're not allowed to leave a node with one item in the | |
2014 | * tree during a delete. A deletion from lower in the tree | |
2015 | * could try to delete the only pointer in this node. | |
2016 | * So, pull some keys from the left. | |
2017 | * There has to be a left pointer at this point because | |
2018 | * otherwise we would have pulled some pointers from the | |
2019 | * right | |
2020 | */ | |
305a26af MF |
2021 | if (!left) { |
2022 | ret = -EROFS; | |
0b246afa | 2023 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
2024 | goto enospc; |
2025 | } | |
2ff7e61e | 2026 | wret = balance_node_right(trans, fs_info, mid, left); |
54aa1f4d | 2027 | if (wret < 0) { |
79f95c82 | 2028 | ret = wret; |
54aa1f4d CM |
2029 | goto enospc; |
2030 | } | |
bce4eae9 | 2031 | if (wret == 1) { |
2ff7e61e | 2032 | wret = push_node_left(trans, fs_info, left, mid, 1); |
bce4eae9 CM |
2033 | if (wret < 0) |
2034 | ret = wret; | |
2035 | } | |
79f95c82 CM |
2036 | BUG_ON(wret == 1); |
2037 | } | |
5f39d397 | 2038 | if (btrfs_header_nritems(mid) == 0) { |
7c302b49 | 2039 | clean_tree_block(fs_info, mid); |
925baedd | 2040 | btrfs_tree_unlock(mid); |
afe5fea7 | 2041 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2042 | root_sub_used(root, mid->len); |
5581a51a | 2043 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2044 | free_extent_buffer_stale(mid); |
f0486c68 | 2045 | mid = NULL; |
79f95c82 CM |
2046 | } else { |
2047 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2048 | struct btrfs_disk_key mid_key; |
2049 | btrfs_node_key(mid, &mid_key, 0); | |
3ac6de1a | 2050 | tree_mod_log_set_node_key(parent, pslot, 0); |
5f39d397 CM |
2051 | btrfs_set_node_key(parent, &mid_key, pslot); |
2052 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2053 | } |
bb803951 | 2054 | |
79f95c82 | 2055 | /* update the path */ |
5f39d397 CM |
2056 | if (left) { |
2057 | if (btrfs_header_nritems(left) > orig_slot) { | |
2058 | extent_buffer_get(left); | |
925baedd | 2059 | /* left was locked after cow */ |
5f39d397 | 2060 | path->nodes[level] = left; |
bb803951 CM |
2061 | path->slots[level + 1] -= 1; |
2062 | path->slots[level] = orig_slot; | |
925baedd CM |
2063 | if (mid) { |
2064 | btrfs_tree_unlock(mid); | |
5f39d397 | 2065 | free_extent_buffer(mid); |
925baedd | 2066 | } |
bb803951 | 2067 | } else { |
5f39d397 | 2068 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2069 | path->slots[level] = orig_slot; |
2070 | } | |
2071 | } | |
79f95c82 | 2072 | /* double check we haven't messed things up */ |
e20d96d6 | 2073 | if (orig_ptr != |
5f39d397 | 2074 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2075 | BUG(); |
54aa1f4d | 2076 | enospc: |
925baedd CM |
2077 | if (right) { |
2078 | btrfs_tree_unlock(right); | |
5f39d397 | 2079 | free_extent_buffer(right); |
925baedd CM |
2080 | } |
2081 | if (left) { | |
2082 | if (path->nodes[level] != left) | |
2083 | btrfs_tree_unlock(left); | |
5f39d397 | 2084 | free_extent_buffer(left); |
925baedd | 2085 | } |
bb803951 CM |
2086 | return ret; |
2087 | } | |
2088 | ||
d352ac68 CM |
2089 | /* Node balancing for insertion. Here we only split or push nodes around |
2090 | * when they are completely full. This is also done top down, so we | |
2091 | * have to be pessimistic. | |
2092 | */ | |
d397712b | 2093 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2094 | struct btrfs_root *root, |
2095 | struct btrfs_path *path, int level) | |
e66f709b | 2096 | { |
0b246afa | 2097 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2098 | struct extent_buffer *right = NULL; |
2099 | struct extent_buffer *mid; | |
2100 | struct extent_buffer *left = NULL; | |
2101 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2102 | int ret = 0; |
2103 | int wret; | |
2104 | int pslot; | |
2105 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2106 | |
2107 | if (level == 0) | |
2108 | return 1; | |
2109 | ||
5f39d397 | 2110 | mid = path->nodes[level]; |
7bb86316 | 2111 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2112 | |
a05a9bb1 | 2113 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2114 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2115 | pslot = path->slots[level + 1]; |
2116 | } | |
e66f709b | 2117 | |
5f39d397 | 2118 | if (!parent) |
e66f709b | 2119 | return 1; |
e66f709b | 2120 | |
2ff7e61e | 2121 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
2122 | if (IS_ERR(left)) |
2123 | left = NULL; | |
e66f709b CM |
2124 | |
2125 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2126 | if (left) { |
e66f709b | 2127 | u32 left_nr; |
925baedd CM |
2128 | |
2129 | btrfs_tree_lock(left); | |
b4ce94de CM |
2130 | btrfs_set_lock_blocking(left); |
2131 | ||
5f39d397 | 2132 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2133 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2134 | wret = 1; |
2135 | } else { | |
5f39d397 | 2136 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2137 | pslot - 1, &left); |
54aa1f4d CM |
2138 | if (ret) |
2139 | wret = 1; | |
2140 | else { | |
2ff7e61e | 2141 | wret = push_node_left(trans, fs_info, |
971a1f66 | 2142 | left, mid, 0); |
54aa1f4d | 2143 | } |
33ade1f8 | 2144 | } |
e66f709b CM |
2145 | if (wret < 0) |
2146 | ret = wret; | |
2147 | if (wret == 0) { | |
5f39d397 | 2148 | struct btrfs_disk_key disk_key; |
e66f709b | 2149 | orig_slot += left_nr; |
5f39d397 | 2150 | btrfs_node_key(mid, &disk_key, 0); |
3ac6de1a | 2151 | tree_mod_log_set_node_key(parent, pslot, 0); |
5f39d397 CM |
2152 | btrfs_set_node_key(parent, &disk_key, pslot); |
2153 | btrfs_mark_buffer_dirty(parent); | |
2154 | if (btrfs_header_nritems(left) > orig_slot) { | |
2155 | path->nodes[level] = left; | |
e66f709b CM |
2156 | path->slots[level + 1] -= 1; |
2157 | path->slots[level] = orig_slot; | |
925baedd | 2158 | btrfs_tree_unlock(mid); |
5f39d397 | 2159 | free_extent_buffer(mid); |
e66f709b CM |
2160 | } else { |
2161 | orig_slot -= | |
5f39d397 | 2162 | btrfs_header_nritems(left); |
e66f709b | 2163 | path->slots[level] = orig_slot; |
925baedd | 2164 | btrfs_tree_unlock(left); |
5f39d397 | 2165 | free_extent_buffer(left); |
e66f709b | 2166 | } |
e66f709b CM |
2167 | return 0; |
2168 | } | |
925baedd | 2169 | btrfs_tree_unlock(left); |
5f39d397 | 2170 | free_extent_buffer(left); |
e66f709b | 2171 | } |
2ff7e61e | 2172 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
2173 | if (IS_ERR(right)) |
2174 | right = NULL; | |
e66f709b CM |
2175 | |
2176 | /* | |
2177 | * then try to empty the right most buffer into the middle | |
2178 | */ | |
5f39d397 | 2179 | if (right) { |
33ade1f8 | 2180 | u32 right_nr; |
b4ce94de | 2181 | |
925baedd | 2182 | btrfs_tree_lock(right); |
b4ce94de CM |
2183 | btrfs_set_lock_blocking(right); |
2184 | ||
5f39d397 | 2185 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2186 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2187 | wret = 1; |
2188 | } else { | |
5f39d397 CM |
2189 | ret = btrfs_cow_block(trans, root, right, |
2190 | parent, pslot + 1, | |
9fa8cfe7 | 2191 | &right); |
54aa1f4d CM |
2192 | if (ret) |
2193 | wret = 1; | |
2194 | else { | |
2ff7e61e | 2195 | wret = balance_node_right(trans, fs_info, |
5f39d397 | 2196 | right, mid); |
54aa1f4d | 2197 | } |
33ade1f8 | 2198 | } |
e66f709b CM |
2199 | if (wret < 0) |
2200 | ret = wret; | |
2201 | if (wret == 0) { | |
5f39d397 CM |
2202 | struct btrfs_disk_key disk_key; |
2203 | ||
2204 | btrfs_node_key(right, &disk_key, 0); | |
3ac6de1a | 2205 | tree_mod_log_set_node_key(parent, pslot + 1, 0); |
5f39d397 CM |
2206 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2207 | btrfs_mark_buffer_dirty(parent); | |
2208 | ||
2209 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2210 | path->nodes[level] = right; | |
e66f709b CM |
2211 | path->slots[level + 1] += 1; |
2212 | path->slots[level] = orig_slot - | |
5f39d397 | 2213 | btrfs_header_nritems(mid); |
925baedd | 2214 | btrfs_tree_unlock(mid); |
5f39d397 | 2215 | free_extent_buffer(mid); |
e66f709b | 2216 | } else { |
925baedd | 2217 | btrfs_tree_unlock(right); |
5f39d397 | 2218 | free_extent_buffer(right); |
e66f709b | 2219 | } |
e66f709b CM |
2220 | return 0; |
2221 | } | |
925baedd | 2222 | btrfs_tree_unlock(right); |
5f39d397 | 2223 | free_extent_buffer(right); |
e66f709b | 2224 | } |
e66f709b CM |
2225 | return 1; |
2226 | } | |
2227 | ||
3c69faec | 2228 | /* |
d352ac68 CM |
2229 | * readahead one full node of leaves, finding things that are close |
2230 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2231 | */ |
2ff7e61e | 2232 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2233 | struct btrfs_path *path, |
2234 | int level, int slot, u64 objectid) | |
3c69faec | 2235 | { |
5f39d397 | 2236 | struct extent_buffer *node; |
01f46658 | 2237 | struct btrfs_disk_key disk_key; |
3c69faec | 2238 | u32 nritems; |
3c69faec | 2239 | u64 search; |
a7175319 | 2240 | u64 target; |
6b80053d | 2241 | u64 nread = 0; |
5f39d397 | 2242 | struct extent_buffer *eb; |
6b80053d CM |
2243 | u32 nr; |
2244 | u32 blocksize; | |
2245 | u32 nscan = 0; | |
db94535d | 2246 | |
a6b6e75e | 2247 | if (level != 1) |
6702ed49 CM |
2248 | return; |
2249 | ||
2250 | if (!path->nodes[level]) | |
3c69faec CM |
2251 | return; |
2252 | ||
5f39d397 | 2253 | node = path->nodes[level]; |
925baedd | 2254 | |
3c69faec | 2255 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2256 | blocksize = fs_info->nodesize; |
2257 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2258 | if (eb) { |
2259 | free_extent_buffer(eb); | |
3c69faec CM |
2260 | return; |
2261 | } | |
2262 | ||
a7175319 | 2263 | target = search; |
6b80053d | 2264 | |
5f39d397 | 2265 | nritems = btrfs_header_nritems(node); |
6b80053d | 2266 | nr = slot; |
25b8b936 | 2267 | |
d397712b | 2268 | while (1) { |
e4058b54 | 2269 | if (path->reada == READA_BACK) { |
6b80053d CM |
2270 | if (nr == 0) |
2271 | break; | |
2272 | nr--; | |
e4058b54 | 2273 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2274 | nr++; |
2275 | if (nr >= nritems) | |
2276 | break; | |
3c69faec | 2277 | } |
e4058b54 | 2278 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2279 | btrfs_node_key(node, &disk_key, nr); |
2280 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2281 | break; | |
2282 | } | |
6b80053d | 2283 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2284 | if ((search <= target && target - search <= 65536) || |
2285 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2286 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2287 | nread += blocksize; |
2288 | } | |
2289 | nscan++; | |
a7175319 | 2290 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2291 | break; |
3c69faec CM |
2292 | } |
2293 | } | |
925baedd | 2294 | |
2ff7e61e | 2295 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2296 | struct btrfs_path *path, int level) |
b4ce94de CM |
2297 | { |
2298 | int slot; | |
2299 | int nritems; | |
2300 | struct extent_buffer *parent; | |
2301 | struct extent_buffer *eb; | |
2302 | u64 gen; | |
2303 | u64 block1 = 0; | |
2304 | u64 block2 = 0; | |
b4ce94de | 2305 | |
8c594ea8 | 2306 | parent = path->nodes[level + 1]; |
b4ce94de | 2307 | if (!parent) |
0b08851f | 2308 | return; |
b4ce94de CM |
2309 | |
2310 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2311 | slot = path->slots[level + 1]; |
b4ce94de CM |
2312 | |
2313 | if (slot > 0) { | |
2314 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2315 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2316 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2317 | /* |
2318 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2319 | * don't want to return eagain here. That will loop | |
2320 | * forever | |
2321 | */ | |
2322 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2323 | block1 = 0; |
2324 | free_extent_buffer(eb); | |
2325 | } | |
8c594ea8 | 2326 | if (slot + 1 < nritems) { |
b4ce94de CM |
2327 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2328 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2329 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2330 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2331 | block2 = 0; |
2332 | free_extent_buffer(eb); | |
2333 | } | |
8c594ea8 | 2334 | |
0b08851f | 2335 | if (block1) |
2ff7e61e | 2336 | readahead_tree_block(fs_info, block1); |
0b08851f | 2337 | if (block2) |
2ff7e61e | 2338 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2339 | } |
2340 | ||
2341 | ||
d352ac68 | 2342 | /* |
d397712b CM |
2343 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2344 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2345 | * operations on the tree might require changing key pointers higher up in the | |
2346 | * tree. | |
d352ac68 | 2347 | * |
d397712b CM |
2348 | * callers might also have set path->keep_locks, which tells this code to keep |
2349 | * the lock if the path points to the last slot in the block. This is part of | |
2350 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2351 | * |
d397712b CM |
2352 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2353 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2354 | */ |
e02119d5 | 2355 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2356 | int lowest_unlock, int min_write_lock_level, |
2357 | int *write_lock_level) | |
925baedd CM |
2358 | { |
2359 | int i; | |
2360 | int skip_level = level; | |
051e1b9f | 2361 | int no_skips = 0; |
925baedd CM |
2362 | struct extent_buffer *t; |
2363 | ||
2364 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2365 | if (!path->nodes[i]) | |
2366 | break; | |
2367 | if (!path->locks[i]) | |
2368 | break; | |
051e1b9f | 2369 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2370 | skip_level = i + 1; |
2371 | continue; | |
2372 | } | |
051e1b9f | 2373 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2374 | u32 nritems; |
2375 | t = path->nodes[i]; | |
2376 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2377 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2378 | skip_level = i + 1; |
2379 | continue; | |
2380 | } | |
2381 | } | |
051e1b9f CM |
2382 | if (skip_level < i && i >= lowest_unlock) |
2383 | no_skips = 1; | |
2384 | ||
925baedd CM |
2385 | t = path->nodes[i]; |
2386 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2387 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2388 | path->locks[i] = 0; |
f7c79f30 CM |
2389 | if (write_lock_level && |
2390 | i > min_write_lock_level && | |
2391 | i <= *write_lock_level) { | |
2392 | *write_lock_level = i - 1; | |
2393 | } | |
925baedd CM |
2394 | } |
2395 | } | |
2396 | } | |
2397 | ||
b4ce94de CM |
2398 | /* |
2399 | * This releases any locks held in the path starting at level and | |
2400 | * going all the way up to the root. | |
2401 | * | |
2402 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2403 | * corner cases, such as COW of the block at slot zero in the node. This | |
2404 | * ignores those rules, and it should only be called when there are no | |
2405 | * more updates to be done higher up in the tree. | |
2406 | */ | |
2407 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2408 | { | |
2409 | int i; | |
2410 | ||
09a2a8f9 | 2411 | if (path->keep_locks) |
b4ce94de CM |
2412 | return; |
2413 | ||
2414 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2415 | if (!path->nodes[i]) | |
12f4dacc | 2416 | continue; |
b4ce94de | 2417 | if (!path->locks[i]) |
12f4dacc | 2418 | continue; |
bd681513 | 2419 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2420 | path->locks[i] = 0; |
2421 | } | |
2422 | } | |
2423 | ||
c8c42864 CM |
2424 | /* |
2425 | * helper function for btrfs_search_slot. The goal is to find a block | |
2426 | * in cache without setting the path to blocking. If we find the block | |
2427 | * we return zero and the path is unchanged. | |
2428 | * | |
2429 | * If we can't find the block, we set the path blocking and do some | |
2430 | * reada. -EAGAIN is returned and the search must be repeated. | |
2431 | */ | |
2432 | static int | |
d07b8528 LB |
2433 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2434 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2435 | const struct btrfs_key *key) |
c8c42864 | 2436 | { |
0b246afa | 2437 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2438 | u64 blocknr; |
2439 | u64 gen; | |
c8c42864 CM |
2440 | struct extent_buffer *b = *eb_ret; |
2441 | struct extent_buffer *tmp; | |
76a05b35 | 2442 | int ret; |
c8c42864 CM |
2443 | |
2444 | blocknr = btrfs_node_blockptr(b, slot); | |
2445 | gen = btrfs_node_ptr_generation(b, slot); | |
c8c42864 | 2446 | |
0b246afa | 2447 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2448 | if (tmp) { |
b9fab919 | 2449 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2450 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2451 | *eb_ret = tmp; | |
2452 | return 0; | |
2453 | } | |
2454 | ||
2455 | /* the pages were up to date, but we failed | |
2456 | * the generation number check. Do a full | |
2457 | * read for the generation number that is correct. | |
2458 | * We must do this without dropping locks so | |
2459 | * we can trust our generation number | |
2460 | */ | |
2461 | btrfs_set_path_blocking(p); | |
2462 | ||
2463 | /* now we're allowed to do a blocking uptodate check */ | |
2464 | ret = btrfs_read_buffer(tmp, gen); | |
2465 | if (!ret) { | |
2466 | *eb_ret = tmp; | |
2467 | return 0; | |
cb44921a | 2468 | } |
bdf7c00e JB |
2469 | free_extent_buffer(tmp); |
2470 | btrfs_release_path(p); | |
2471 | return -EIO; | |
c8c42864 CM |
2472 | } |
2473 | ||
2474 | /* | |
2475 | * reduce lock contention at high levels | |
2476 | * of the btree by dropping locks before | |
76a05b35 CM |
2477 | * we read. Don't release the lock on the current |
2478 | * level because we need to walk this node to figure | |
2479 | * out which blocks to read. | |
c8c42864 | 2480 | */ |
8c594ea8 CM |
2481 | btrfs_unlock_up_safe(p, level + 1); |
2482 | btrfs_set_path_blocking(p); | |
2483 | ||
cb44921a | 2484 | free_extent_buffer(tmp); |
e4058b54 | 2485 | if (p->reada != READA_NONE) |
2ff7e61e | 2486 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2487 | |
b3b4aa74 | 2488 | btrfs_release_path(p); |
76a05b35 CM |
2489 | |
2490 | ret = -EAGAIN; | |
2ff7e61e | 2491 | tmp = read_tree_block(fs_info, blocknr, 0); |
64c043de | 2492 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2493 | /* |
2494 | * If the read above didn't mark this buffer up to date, | |
2495 | * it will never end up being up to date. Set ret to EIO now | |
2496 | * and give up so that our caller doesn't loop forever | |
2497 | * on our EAGAINs. | |
2498 | */ | |
b9fab919 | 2499 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2500 | ret = -EIO; |
c8c42864 | 2501 | free_extent_buffer(tmp); |
c871b0f2 LB |
2502 | } else { |
2503 | ret = PTR_ERR(tmp); | |
76a05b35 CM |
2504 | } |
2505 | return ret; | |
c8c42864 CM |
2506 | } |
2507 | ||
2508 | /* | |
2509 | * helper function for btrfs_search_slot. This does all of the checks | |
2510 | * for node-level blocks and does any balancing required based on | |
2511 | * the ins_len. | |
2512 | * | |
2513 | * If no extra work was required, zero is returned. If we had to | |
2514 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2515 | * start over | |
2516 | */ | |
2517 | static int | |
2518 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2519 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2520 | struct extent_buffer *b, int level, int ins_len, |
2521 | int *write_lock_level) | |
c8c42864 | 2522 | { |
0b246afa | 2523 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2524 | int ret; |
0b246afa | 2525 | |
c8c42864 | 2526 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2527 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2528 | int sret; |
2529 | ||
bd681513 CM |
2530 | if (*write_lock_level < level + 1) { |
2531 | *write_lock_level = level + 1; | |
2532 | btrfs_release_path(p); | |
2533 | goto again; | |
2534 | } | |
2535 | ||
c8c42864 | 2536 | btrfs_set_path_blocking(p); |
2ff7e61e | 2537 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2538 | sret = split_node(trans, root, p, level); |
bd681513 | 2539 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2540 | |
2541 | BUG_ON(sret > 0); | |
2542 | if (sret) { | |
2543 | ret = sret; | |
2544 | goto done; | |
2545 | } | |
2546 | b = p->nodes[level]; | |
2547 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2548 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2549 | int sret; |
2550 | ||
bd681513 CM |
2551 | if (*write_lock_level < level + 1) { |
2552 | *write_lock_level = level + 1; | |
2553 | btrfs_release_path(p); | |
2554 | goto again; | |
2555 | } | |
2556 | ||
c8c42864 | 2557 | btrfs_set_path_blocking(p); |
2ff7e61e | 2558 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2559 | sret = balance_level(trans, root, p, level); |
bd681513 | 2560 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2561 | |
2562 | if (sret) { | |
2563 | ret = sret; | |
2564 | goto done; | |
2565 | } | |
2566 | b = p->nodes[level]; | |
2567 | if (!b) { | |
b3b4aa74 | 2568 | btrfs_release_path(p); |
c8c42864 CM |
2569 | goto again; |
2570 | } | |
2571 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2572 | } | |
2573 | return 0; | |
2574 | ||
2575 | again: | |
2576 | ret = -EAGAIN; | |
2577 | done: | |
2578 | return ret; | |
2579 | } | |
2580 | ||
d7396f07 | 2581 | static void key_search_validate(struct extent_buffer *b, |
310712b2 | 2582 | const struct btrfs_key *key, |
d7396f07 FDBM |
2583 | int level) |
2584 | { | |
2585 | #ifdef CONFIG_BTRFS_ASSERT | |
2586 | struct btrfs_disk_key disk_key; | |
2587 | ||
2588 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2589 | ||
2590 | if (level == 0) | |
2591 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2592 | offsetof(struct btrfs_leaf, items[0].key), | |
2593 | sizeof(disk_key))); | |
2594 | else | |
2595 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2596 | offsetof(struct btrfs_node, ptrs[0].key), | |
2597 | sizeof(disk_key))); | |
2598 | #endif | |
2599 | } | |
2600 | ||
310712b2 | 2601 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2602 | int level, int *prev_cmp, int *slot) |
2603 | { | |
2604 | if (*prev_cmp != 0) { | |
a74b35ec | 2605 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2606 | return *prev_cmp; |
2607 | } | |
2608 | ||
2609 | key_search_validate(b, key, level); | |
2610 | *slot = 0; | |
2611 | ||
2612 | return 0; | |
2613 | } | |
2614 | ||
381cf658 | 2615 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2616 | u64 iobjectid, u64 ioff, u8 key_type, |
2617 | struct btrfs_key *found_key) | |
2618 | { | |
2619 | int ret; | |
2620 | struct btrfs_key key; | |
2621 | struct extent_buffer *eb; | |
381cf658 DS |
2622 | |
2623 | ASSERT(path); | |
1d4c08e0 | 2624 | ASSERT(found_key); |
e33d5c3d KN |
2625 | |
2626 | key.type = key_type; | |
2627 | key.objectid = iobjectid; | |
2628 | key.offset = ioff; | |
2629 | ||
2630 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2631 | if (ret < 0) |
e33d5c3d KN |
2632 | return ret; |
2633 | ||
2634 | eb = path->nodes[0]; | |
2635 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2636 | ret = btrfs_next_leaf(fs_root, path); | |
2637 | if (ret) | |
2638 | return ret; | |
2639 | eb = path->nodes[0]; | |
2640 | } | |
2641 | ||
2642 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2643 | if (found_key->type != key.type || | |
2644 | found_key->objectid != key.objectid) | |
2645 | return 1; | |
2646 | ||
2647 | return 0; | |
2648 | } | |
2649 | ||
74123bd7 | 2650 | /* |
4271ecea NB |
2651 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2652 | * modifications to preserve tree invariants. | |
74123bd7 | 2653 | * |
4271ecea NB |
2654 | * @trans: Handle of transaction, used when modifying the tree |
2655 | * @p: Holds all btree nodes along the search path | |
2656 | * @root: The root node of the tree | |
2657 | * @key: The key we are looking for | |
2658 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2659 | * deletions it's -1. 0 for plain searches | |
2660 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2661 | * when modifying the tree. | |
97571fd0 | 2662 | * |
4271ecea NB |
2663 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2664 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2665 | * | |
2666 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2667 | * of the path (level 0) | |
2668 | * | |
2669 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2670 | * points to the slot where it should be inserted | |
2671 | * | |
2672 | * If an error is encountered while searching the tree a negative error number | |
2673 | * is returned | |
74123bd7 | 2674 | */ |
310712b2 OS |
2675 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2676 | const struct btrfs_key *key, struct btrfs_path *p, | |
2677 | int ins_len, int cow) | |
be0e5c09 | 2678 | { |
0b246afa | 2679 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2680 | struct extent_buffer *b; |
be0e5c09 CM |
2681 | int slot; |
2682 | int ret; | |
33c66f43 | 2683 | int err; |
be0e5c09 | 2684 | int level; |
925baedd | 2685 | int lowest_unlock = 1; |
bd681513 CM |
2686 | int root_lock; |
2687 | /* everything at write_lock_level or lower must be write locked */ | |
2688 | int write_lock_level = 0; | |
9f3a7427 | 2689 | u8 lowest_level = 0; |
f7c79f30 | 2690 | int min_write_lock_level; |
d7396f07 | 2691 | int prev_cmp; |
9f3a7427 | 2692 | |
6702ed49 | 2693 | lowest_level = p->lowest_level; |
323ac95b | 2694 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2695 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2696 | BUG_ON(!cow && ins_len); |
25179201 | 2697 | |
bd681513 | 2698 | if (ins_len < 0) { |
925baedd | 2699 | lowest_unlock = 2; |
65b51a00 | 2700 | |
bd681513 CM |
2701 | /* when we are removing items, we might have to go up to level |
2702 | * two as we update tree pointers Make sure we keep write | |
2703 | * for those levels as well | |
2704 | */ | |
2705 | write_lock_level = 2; | |
2706 | } else if (ins_len > 0) { | |
2707 | /* | |
2708 | * for inserting items, make sure we have a write lock on | |
2709 | * level 1 so we can update keys | |
2710 | */ | |
2711 | write_lock_level = 1; | |
2712 | } | |
2713 | ||
2714 | if (!cow) | |
2715 | write_lock_level = -1; | |
2716 | ||
09a2a8f9 | 2717 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2718 | write_lock_level = BTRFS_MAX_LEVEL; |
2719 | ||
f7c79f30 CM |
2720 | min_write_lock_level = write_lock_level; |
2721 | ||
bb803951 | 2722 | again: |
d7396f07 | 2723 | prev_cmp = -1; |
bd681513 CM |
2724 | /* |
2725 | * we try very hard to do read locks on the root | |
2726 | */ | |
2727 | root_lock = BTRFS_READ_LOCK; | |
2728 | level = 0; | |
5d4f98a2 | 2729 | if (p->search_commit_root) { |
bd681513 CM |
2730 | /* |
2731 | * the commit roots are read only | |
2732 | * so we always do read locks | |
2733 | */ | |
3f8a18cc | 2734 | if (p->need_commit_sem) |
0b246afa | 2735 | down_read(&fs_info->commit_root_sem); |
5d4f98a2 YZ |
2736 | b = root->commit_root; |
2737 | extent_buffer_get(b); | |
bd681513 | 2738 | level = btrfs_header_level(b); |
3f8a18cc | 2739 | if (p->need_commit_sem) |
0b246afa | 2740 | up_read(&fs_info->commit_root_sem); |
5d4f98a2 | 2741 | if (!p->skip_locking) |
bd681513 | 2742 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2743 | } else { |
bd681513 | 2744 | if (p->skip_locking) { |
5d4f98a2 | 2745 | b = btrfs_root_node(root); |
bd681513 CM |
2746 | level = btrfs_header_level(b); |
2747 | } else { | |
2748 | /* we don't know the level of the root node | |
2749 | * until we actually have it read locked | |
2750 | */ | |
2751 | b = btrfs_read_lock_root_node(root); | |
2752 | level = btrfs_header_level(b); | |
2753 | if (level <= write_lock_level) { | |
2754 | /* whoops, must trade for write lock */ | |
2755 | btrfs_tree_read_unlock(b); | |
2756 | free_extent_buffer(b); | |
2757 | b = btrfs_lock_root_node(root); | |
2758 | root_lock = BTRFS_WRITE_LOCK; | |
2759 | ||
2760 | /* the level might have changed, check again */ | |
2761 | level = btrfs_header_level(b); | |
2762 | } | |
2763 | } | |
5d4f98a2 | 2764 | } |
bd681513 CM |
2765 | p->nodes[level] = b; |
2766 | if (!p->skip_locking) | |
2767 | p->locks[level] = root_lock; | |
925baedd | 2768 | |
eb60ceac | 2769 | while (b) { |
5f39d397 | 2770 | level = btrfs_header_level(b); |
65b51a00 CM |
2771 | |
2772 | /* | |
2773 | * setup the path here so we can release it under lock | |
2774 | * contention with the cow code | |
2775 | */ | |
02217ed2 | 2776 | if (cow) { |
9ea2c7c9 NB |
2777 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2778 | ||
c8c42864 CM |
2779 | /* |
2780 | * if we don't really need to cow this block | |
2781 | * then we don't want to set the path blocking, | |
2782 | * so we test it here | |
2783 | */ | |
64c12921 JM |
2784 | if (!should_cow_block(trans, root, b)) { |
2785 | trans->dirty = true; | |
65b51a00 | 2786 | goto cow_done; |
64c12921 | 2787 | } |
5d4f98a2 | 2788 | |
bd681513 CM |
2789 | /* |
2790 | * must have write locks on this node and the | |
2791 | * parent | |
2792 | */ | |
5124e00e JB |
2793 | if (level > write_lock_level || |
2794 | (level + 1 > write_lock_level && | |
2795 | level + 1 < BTRFS_MAX_LEVEL && | |
2796 | p->nodes[level + 1])) { | |
bd681513 CM |
2797 | write_lock_level = level + 1; |
2798 | btrfs_release_path(p); | |
2799 | goto again; | |
2800 | } | |
2801 | ||
160f4089 | 2802 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2803 | if (last_level) |
2804 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2805 | &b); | |
2806 | else | |
2807 | err = btrfs_cow_block(trans, root, b, | |
2808 | p->nodes[level + 1], | |
2809 | p->slots[level + 1], &b); | |
33c66f43 | 2810 | if (err) { |
33c66f43 | 2811 | ret = err; |
65b51a00 | 2812 | goto done; |
54aa1f4d | 2813 | } |
02217ed2 | 2814 | } |
65b51a00 | 2815 | cow_done: |
eb60ceac | 2816 | p->nodes[level] = b; |
bd681513 | 2817 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2818 | |
2819 | /* | |
2820 | * we have a lock on b and as long as we aren't changing | |
2821 | * the tree, there is no way to for the items in b to change. | |
2822 | * It is safe to drop the lock on our parent before we | |
2823 | * go through the expensive btree search on b. | |
2824 | * | |
eb653de1 FDBM |
2825 | * If we're inserting or deleting (ins_len != 0), then we might |
2826 | * be changing slot zero, which may require changing the parent. | |
2827 | * So, we can't drop the lock until after we know which slot | |
2828 | * we're operating on. | |
b4ce94de | 2829 | */ |
eb653de1 FDBM |
2830 | if (!ins_len && !p->keep_locks) { |
2831 | int u = level + 1; | |
2832 | ||
2833 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2834 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2835 | p->locks[u] = 0; | |
2836 | } | |
2837 | } | |
b4ce94de | 2838 | |
d7396f07 | 2839 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2840 | if (ret < 0) |
2841 | goto done; | |
b4ce94de | 2842 | |
5f39d397 | 2843 | if (level != 0) { |
33c66f43 YZ |
2844 | int dec = 0; |
2845 | if (ret && slot > 0) { | |
2846 | dec = 1; | |
be0e5c09 | 2847 | slot -= 1; |
33c66f43 | 2848 | } |
be0e5c09 | 2849 | p->slots[level] = slot; |
33c66f43 | 2850 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2851 | ins_len, &write_lock_level); |
33c66f43 | 2852 | if (err == -EAGAIN) |
c8c42864 | 2853 | goto again; |
33c66f43 YZ |
2854 | if (err) { |
2855 | ret = err; | |
c8c42864 | 2856 | goto done; |
33c66f43 | 2857 | } |
c8c42864 CM |
2858 | b = p->nodes[level]; |
2859 | slot = p->slots[level]; | |
b4ce94de | 2860 | |
bd681513 CM |
2861 | /* |
2862 | * slot 0 is special, if we change the key | |
2863 | * we have to update the parent pointer | |
2864 | * which means we must have a write lock | |
2865 | * on the parent | |
2866 | */ | |
eb653de1 | 2867 | if (slot == 0 && ins_len && |
bd681513 CM |
2868 | write_lock_level < level + 1) { |
2869 | write_lock_level = level + 1; | |
2870 | btrfs_release_path(p); | |
2871 | goto again; | |
2872 | } | |
2873 | ||
f7c79f30 CM |
2874 | unlock_up(p, level, lowest_unlock, |
2875 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2876 | |
925baedd | 2877 | if (level == lowest_level) { |
33c66f43 YZ |
2878 | if (dec) |
2879 | p->slots[level]++; | |
5b21f2ed | 2880 | goto done; |
925baedd | 2881 | } |
ca7a79ad | 2882 | |
d07b8528 | 2883 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2884 | slot, key); |
33c66f43 | 2885 | if (err == -EAGAIN) |
c8c42864 | 2886 | goto again; |
33c66f43 YZ |
2887 | if (err) { |
2888 | ret = err; | |
76a05b35 | 2889 | goto done; |
33c66f43 | 2890 | } |
76a05b35 | 2891 | |
b4ce94de | 2892 | if (!p->skip_locking) { |
bd681513 CM |
2893 | level = btrfs_header_level(b); |
2894 | if (level <= write_lock_level) { | |
2895 | err = btrfs_try_tree_write_lock(b); | |
2896 | if (!err) { | |
2897 | btrfs_set_path_blocking(p); | |
2898 | btrfs_tree_lock(b); | |
2899 | btrfs_clear_path_blocking(p, b, | |
2900 | BTRFS_WRITE_LOCK); | |
2901 | } | |
2902 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2903 | } else { | |
f82c458a | 2904 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2905 | if (!err) { |
2906 | btrfs_set_path_blocking(p); | |
2907 | btrfs_tree_read_lock(b); | |
2908 | btrfs_clear_path_blocking(p, b, | |
2909 | BTRFS_READ_LOCK); | |
2910 | } | |
2911 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2912 | } |
bd681513 | 2913 | p->nodes[level] = b; |
b4ce94de | 2914 | } |
be0e5c09 CM |
2915 | } else { |
2916 | p->slots[level] = slot; | |
87b29b20 | 2917 | if (ins_len > 0 && |
2ff7e61e | 2918 | btrfs_leaf_free_space(fs_info, b) < ins_len) { |
bd681513 CM |
2919 | if (write_lock_level < 1) { |
2920 | write_lock_level = 1; | |
2921 | btrfs_release_path(p); | |
2922 | goto again; | |
2923 | } | |
2924 | ||
b4ce94de | 2925 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2926 | err = split_leaf(trans, root, key, |
2927 | p, ins_len, ret == 0); | |
bd681513 | 2928 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2929 | |
33c66f43 YZ |
2930 | BUG_ON(err > 0); |
2931 | if (err) { | |
2932 | ret = err; | |
65b51a00 CM |
2933 | goto done; |
2934 | } | |
5c680ed6 | 2935 | } |
459931ec | 2936 | if (!p->search_for_split) |
f7c79f30 CM |
2937 | unlock_up(p, level, lowest_unlock, |
2938 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2939 | goto done; |
be0e5c09 CM |
2940 | } |
2941 | } | |
65b51a00 CM |
2942 | ret = 1; |
2943 | done: | |
b4ce94de CM |
2944 | /* |
2945 | * we don't really know what they plan on doing with the path | |
2946 | * from here on, so for now just mark it as blocking | |
2947 | */ | |
b9473439 CM |
2948 | if (!p->leave_spinning) |
2949 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2950 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2951 | btrfs_release_path(p); |
65b51a00 | 2952 | return ret; |
be0e5c09 CM |
2953 | } |
2954 | ||
5d9e75c4 JS |
2955 | /* |
2956 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2957 | * current state of the tree together with the operations recorded in the tree | |
2958 | * modification log to search for the key in a previous version of this tree, as | |
2959 | * denoted by the time_seq parameter. | |
2960 | * | |
2961 | * Naturally, there is no support for insert, delete or cow operations. | |
2962 | * | |
2963 | * The resulting path and return value will be set up as if we called | |
2964 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2965 | */ | |
310712b2 | 2966 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2967 | struct btrfs_path *p, u64 time_seq) |
2968 | { | |
0b246afa | 2969 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2970 | struct extent_buffer *b; |
2971 | int slot; | |
2972 | int ret; | |
2973 | int err; | |
2974 | int level; | |
2975 | int lowest_unlock = 1; | |
2976 | u8 lowest_level = 0; | |
d4b4087c | 2977 | int prev_cmp = -1; |
5d9e75c4 JS |
2978 | |
2979 | lowest_level = p->lowest_level; | |
2980 | WARN_ON(p->nodes[0] != NULL); | |
2981 | ||
2982 | if (p->search_commit_root) { | |
2983 | BUG_ON(time_seq); | |
2984 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2985 | } | |
2986 | ||
2987 | again: | |
5d9e75c4 | 2988 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2989 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2990 | p->locks[level] = BTRFS_READ_LOCK; |
2991 | ||
2992 | while (b) { | |
2993 | level = btrfs_header_level(b); | |
2994 | p->nodes[level] = b; | |
2995 | btrfs_clear_path_blocking(p, NULL, 0); | |
2996 | ||
2997 | /* | |
2998 | * we have a lock on b and as long as we aren't changing | |
2999 | * the tree, there is no way to for the items in b to change. | |
3000 | * It is safe to drop the lock on our parent before we | |
3001 | * go through the expensive btree search on b. | |
3002 | */ | |
3003 | btrfs_unlock_up_safe(p, level + 1); | |
3004 | ||
d4b4087c | 3005 | /* |
01327610 | 3006 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
3007 | * time. |
3008 | */ | |
3009 | prev_cmp = -1; | |
d7396f07 | 3010 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
3011 | |
3012 | if (level != 0) { | |
3013 | int dec = 0; | |
3014 | if (ret && slot > 0) { | |
3015 | dec = 1; | |
3016 | slot -= 1; | |
3017 | } | |
3018 | p->slots[level] = slot; | |
3019 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3020 | ||
3021 | if (level == lowest_level) { | |
3022 | if (dec) | |
3023 | p->slots[level]++; | |
3024 | goto done; | |
3025 | } | |
3026 | ||
d07b8528 | 3027 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 3028 | slot, key); |
5d9e75c4 JS |
3029 | if (err == -EAGAIN) |
3030 | goto again; | |
3031 | if (err) { | |
3032 | ret = err; | |
3033 | goto done; | |
3034 | } | |
3035 | ||
3036 | level = btrfs_header_level(b); | |
f82c458a | 3037 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3038 | if (!err) { |
3039 | btrfs_set_path_blocking(p); | |
3040 | btrfs_tree_read_lock(b); | |
3041 | btrfs_clear_path_blocking(p, b, | |
3042 | BTRFS_READ_LOCK); | |
3043 | } | |
0b246afa | 3044 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3045 | if (!b) { |
3046 | ret = -ENOMEM; | |
3047 | goto done; | |
3048 | } | |
5d9e75c4 JS |
3049 | p->locks[level] = BTRFS_READ_LOCK; |
3050 | p->nodes[level] = b; | |
5d9e75c4 JS |
3051 | } else { |
3052 | p->slots[level] = slot; | |
3053 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3054 | goto done; | |
3055 | } | |
3056 | } | |
3057 | ret = 1; | |
3058 | done: | |
3059 | if (!p->leave_spinning) | |
3060 | btrfs_set_path_blocking(p); | |
3061 | if (ret < 0) | |
3062 | btrfs_release_path(p); | |
3063 | ||
3064 | return ret; | |
3065 | } | |
3066 | ||
2f38b3e1 AJ |
3067 | /* |
3068 | * helper to use instead of search slot if no exact match is needed but | |
3069 | * instead the next or previous item should be returned. | |
3070 | * When find_higher is true, the next higher item is returned, the next lower | |
3071 | * otherwise. | |
3072 | * When return_any and find_higher are both true, and no higher item is found, | |
3073 | * return the next lower instead. | |
3074 | * When return_any is true and find_higher is false, and no lower item is found, | |
3075 | * return the next higher instead. | |
3076 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3077 | * < 0 on error | |
3078 | */ | |
3079 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3080 | const struct btrfs_key *key, |
3081 | struct btrfs_path *p, int find_higher, | |
3082 | int return_any) | |
2f38b3e1 AJ |
3083 | { |
3084 | int ret; | |
3085 | struct extent_buffer *leaf; | |
3086 | ||
3087 | again: | |
3088 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3089 | if (ret <= 0) | |
3090 | return ret; | |
3091 | /* | |
3092 | * a return value of 1 means the path is at the position where the | |
3093 | * item should be inserted. Normally this is the next bigger item, | |
3094 | * but in case the previous item is the last in a leaf, path points | |
3095 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3096 | * item. | |
3097 | */ | |
3098 | leaf = p->nodes[0]; | |
3099 | ||
3100 | if (find_higher) { | |
3101 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3102 | ret = btrfs_next_leaf(root, p); | |
3103 | if (ret <= 0) | |
3104 | return ret; | |
3105 | if (!return_any) | |
3106 | return 1; | |
3107 | /* | |
3108 | * no higher item found, return the next | |
3109 | * lower instead | |
3110 | */ | |
3111 | return_any = 0; | |
3112 | find_higher = 0; | |
3113 | btrfs_release_path(p); | |
3114 | goto again; | |
3115 | } | |
3116 | } else { | |
e6793769 AJ |
3117 | if (p->slots[0] == 0) { |
3118 | ret = btrfs_prev_leaf(root, p); | |
3119 | if (ret < 0) | |
3120 | return ret; | |
3121 | if (!ret) { | |
23c6bf6a FDBM |
3122 | leaf = p->nodes[0]; |
3123 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3124 | p->slots[0]--; | |
e6793769 | 3125 | return 0; |
2f38b3e1 | 3126 | } |
e6793769 AJ |
3127 | if (!return_any) |
3128 | return 1; | |
3129 | /* | |
3130 | * no lower item found, return the next | |
3131 | * higher instead | |
3132 | */ | |
3133 | return_any = 0; | |
3134 | find_higher = 1; | |
3135 | btrfs_release_path(p); | |
3136 | goto again; | |
3137 | } else { | |
2f38b3e1 AJ |
3138 | --p->slots[0]; |
3139 | } | |
3140 | } | |
3141 | return 0; | |
3142 | } | |
3143 | ||
74123bd7 CM |
3144 | /* |
3145 | * adjust the pointers going up the tree, starting at level | |
3146 | * making sure the right key of each node is points to 'key'. | |
3147 | * This is used after shifting pointers to the left, so it stops | |
3148 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3149 | * higher levels | |
aa5d6bed | 3150 | * |
74123bd7 | 3151 | */ |
b7a0365e DD |
3152 | static void fixup_low_keys(struct btrfs_fs_info *fs_info, |
3153 | struct btrfs_path *path, | |
143bede5 | 3154 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3155 | { |
3156 | int i; | |
5f39d397 CM |
3157 | struct extent_buffer *t; |
3158 | ||
234b63a0 | 3159 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3160 | int tslot = path->slots[i]; |
eb60ceac | 3161 | if (!path->nodes[i]) |
be0e5c09 | 3162 | break; |
5f39d397 | 3163 | t = path->nodes[i]; |
3ac6de1a | 3164 | tree_mod_log_set_node_key(t, tslot, 1); |
5f39d397 | 3165 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3166 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3167 | if (tslot != 0) |
3168 | break; | |
3169 | } | |
3170 | } | |
3171 | ||
31840ae1 ZY |
3172 | /* |
3173 | * update item key. | |
3174 | * | |
3175 | * This function isn't completely safe. It's the caller's responsibility | |
3176 | * that the new key won't break the order | |
3177 | */ | |
b7a0365e DD |
3178 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3179 | struct btrfs_path *path, | |
310712b2 | 3180 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3181 | { |
3182 | struct btrfs_disk_key disk_key; | |
3183 | struct extent_buffer *eb; | |
3184 | int slot; | |
3185 | ||
3186 | eb = path->nodes[0]; | |
3187 | slot = path->slots[0]; | |
3188 | if (slot > 0) { | |
3189 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3190 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3191 | } |
3192 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3193 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3194 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3195 | } |
3196 | ||
3197 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3198 | btrfs_set_item_key(eb, &disk_key, slot); | |
3199 | btrfs_mark_buffer_dirty(eb); | |
3200 | if (slot == 0) | |
b7a0365e | 3201 | fixup_low_keys(fs_info, path, &disk_key, 1); |
31840ae1 ZY |
3202 | } |
3203 | ||
74123bd7 CM |
3204 | /* |
3205 | * try to push data from one node into the next node left in the | |
79f95c82 | 3206 | * tree. |
aa5d6bed CM |
3207 | * |
3208 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3209 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3210 | */ |
98ed5174 | 3211 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
3212 | struct btrfs_fs_info *fs_info, |
3213 | struct extent_buffer *dst, | |
971a1f66 | 3214 | struct extent_buffer *src, int empty) |
be0e5c09 | 3215 | { |
be0e5c09 | 3216 | int push_items = 0; |
bb803951 CM |
3217 | int src_nritems; |
3218 | int dst_nritems; | |
aa5d6bed | 3219 | int ret = 0; |
be0e5c09 | 3220 | |
5f39d397 CM |
3221 | src_nritems = btrfs_header_nritems(src); |
3222 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3223 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3224 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3225 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3226 | |
bce4eae9 | 3227 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3228 | return 1; |
3229 | ||
d397712b | 3230 | if (push_items <= 0) |
be0e5c09 CM |
3231 | return 1; |
3232 | ||
bce4eae9 | 3233 | if (empty) { |
971a1f66 | 3234 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3235 | if (push_items < src_nritems) { |
3236 | /* leave at least 8 pointers in the node if | |
3237 | * we aren't going to empty it | |
3238 | */ | |
3239 | if (src_nritems - push_items < 8) { | |
3240 | if (push_items <= 8) | |
3241 | return 1; | |
3242 | push_items -= 8; | |
3243 | } | |
3244 | } | |
3245 | } else | |
3246 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3247 | |
0b246afa | 3248 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3249 | push_items); |
3250 | if (ret) { | |
66642832 | 3251 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3252 | return ret; |
3253 | } | |
5f39d397 CM |
3254 | copy_extent_buffer(dst, src, |
3255 | btrfs_node_key_ptr_offset(dst_nritems), | |
3256 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3257 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3258 | |
bb803951 | 3259 | if (push_items < src_nritems) { |
57911b8b JS |
3260 | /* |
3261 | * don't call tree_mod_log_eb_move here, key removal was already | |
3262 | * fully logged by tree_mod_log_eb_copy above. | |
3263 | */ | |
5f39d397 CM |
3264 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3265 | btrfs_node_key_ptr_offset(push_items), | |
3266 | (src_nritems - push_items) * | |
3267 | sizeof(struct btrfs_key_ptr)); | |
3268 | } | |
3269 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3270 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3271 | btrfs_mark_buffer_dirty(src); | |
3272 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3273 | |
79f95c82 CM |
3274 | return ret; |
3275 | } | |
3276 | ||
3277 | /* | |
3278 | * try to push data from one node into the next node right in the | |
3279 | * tree. | |
3280 | * | |
3281 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3282 | * error, and > 0 if there was no room in the right hand block. | |
3283 | * | |
3284 | * this will only push up to 1/2 the contents of the left node over | |
3285 | */ | |
5f39d397 | 3286 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 3287 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
3288 | struct extent_buffer *dst, |
3289 | struct extent_buffer *src) | |
79f95c82 | 3290 | { |
79f95c82 CM |
3291 | int push_items = 0; |
3292 | int max_push; | |
3293 | int src_nritems; | |
3294 | int dst_nritems; | |
3295 | int ret = 0; | |
79f95c82 | 3296 | |
7bb86316 CM |
3297 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3298 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3299 | ||
5f39d397 CM |
3300 | src_nritems = btrfs_header_nritems(src); |
3301 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3302 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3303 | if (push_items <= 0) |
79f95c82 | 3304 | return 1; |
bce4eae9 | 3305 | |
d397712b | 3306 | if (src_nritems < 4) |
bce4eae9 | 3307 | return 1; |
79f95c82 CM |
3308 | |
3309 | max_push = src_nritems / 2 + 1; | |
3310 | /* don't try to empty the node */ | |
d397712b | 3311 | if (max_push >= src_nritems) |
79f95c82 | 3312 | return 1; |
252c38f0 | 3313 | |
79f95c82 CM |
3314 | if (max_push < push_items) |
3315 | push_items = max_push; | |
3316 | ||
0b246afa | 3317 | tree_mod_log_eb_move(fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3318 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3319 | btrfs_node_key_ptr_offset(0), | |
3320 | (dst_nritems) * | |
3321 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3322 | |
0b246afa | 3323 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3324 | src_nritems - push_items, push_items); |
3325 | if (ret) { | |
66642832 | 3326 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3327 | return ret; |
3328 | } | |
5f39d397 CM |
3329 | copy_extent_buffer(dst, src, |
3330 | btrfs_node_key_ptr_offset(0), | |
3331 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3332 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3333 | |
5f39d397 CM |
3334 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3335 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3336 | |
5f39d397 CM |
3337 | btrfs_mark_buffer_dirty(src); |
3338 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3339 | |
aa5d6bed | 3340 | return ret; |
be0e5c09 CM |
3341 | } |
3342 | ||
97571fd0 CM |
3343 | /* |
3344 | * helper function to insert a new root level in the tree. | |
3345 | * A new node is allocated, and a single item is inserted to | |
3346 | * point to the existing root | |
aa5d6bed CM |
3347 | * |
3348 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3349 | */ |
d397712b | 3350 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3351 | struct btrfs_root *root, |
fdd99c72 | 3352 | struct btrfs_path *path, int level) |
5c680ed6 | 3353 | { |
0b246afa | 3354 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3355 | u64 lower_gen; |
5f39d397 CM |
3356 | struct extent_buffer *lower; |
3357 | struct extent_buffer *c; | |
925baedd | 3358 | struct extent_buffer *old; |
5f39d397 | 3359 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3360 | |
3361 | BUG_ON(path->nodes[level]); | |
3362 | BUG_ON(path->nodes[level-1] != root->node); | |
3363 | ||
7bb86316 CM |
3364 | lower = path->nodes[level-1]; |
3365 | if (level == 1) | |
3366 | btrfs_item_key(lower, &lower_key, 0); | |
3367 | else | |
3368 | btrfs_node_key(lower, &lower_key, 0); | |
3369 | ||
4d75f8a9 DS |
3370 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3371 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3372 | if (IS_ERR(c)) |
3373 | return PTR_ERR(c); | |
925baedd | 3374 | |
0b246afa | 3375 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3376 | |
b159fa28 | 3377 | memzero_extent_buffer(c, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3378 | btrfs_set_header_nritems(c, 1); |
3379 | btrfs_set_header_level(c, level); | |
db94535d | 3380 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3381 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3382 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3383 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3384 | |
0b246afa JM |
3385 | write_extent_buffer_fsid(c, fs_info->fsid); |
3386 | write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid); | |
e17cade2 | 3387 | |
5f39d397 | 3388 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3389 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3390 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3391 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3392 | |
3393 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3394 | |
5f39d397 | 3395 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3396 | |
925baedd | 3397 | old = root->node; |
fdd99c72 | 3398 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3399 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3400 | |
3401 | /* the super has an extra ref to root->node */ | |
3402 | free_extent_buffer(old); | |
3403 | ||
0b86a832 | 3404 | add_root_to_dirty_list(root); |
5f39d397 CM |
3405 | extent_buffer_get(c); |
3406 | path->nodes[level] = c; | |
95449a16 | 3407 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3408 | path->slots[level] = 0; |
3409 | return 0; | |
3410 | } | |
3411 | ||
74123bd7 CM |
3412 | /* |
3413 | * worker function to insert a single pointer in a node. | |
3414 | * the node should have enough room for the pointer already | |
97571fd0 | 3415 | * |
74123bd7 CM |
3416 | * slot and level indicate where you want the key to go, and |
3417 | * blocknr is the block the key points to. | |
3418 | */ | |
143bede5 | 3419 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2ff7e61e | 3420 | struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 3421 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3422 | int slot, int level) |
74123bd7 | 3423 | { |
5f39d397 | 3424 | struct extent_buffer *lower; |
74123bd7 | 3425 | int nritems; |
f3ea38da | 3426 | int ret; |
5c680ed6 CM |
3427 | |
3428 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3429 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3430 | lower = path->nodes[level]; |
3431 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3432 | BUG_ON(slot > nritems); |
0b246afa | 3433 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3434 | if (slot != nritems) { |
c3e06965 | 3435 | if (level) |
0b246afa | 3436 | tree_mod_log_eb_move(fs_info, lower, slot + 1, |
f3ea38da | 3437 | slot, nritems - slot); |
5f39d397 CM |
3438 | memmove_extent_buffer(lower, |
3439 | btrfs_node_key_ptr_offset(slot + 1), | |
3440 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3441 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3442 | } |
c3e06965 | 3443 | if (level) { |
0b246afa | 3444 | ret = tree_mod_log_insert_key(fs_info, lower, slot, |
c8cc6341 | 3445 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3446 | BUG_ON(ret < 0); |
3447 | } | |
5f39d397 | 3448 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3449 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3450 | WARN_ON(trans->transid == 0); |
3451 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3452 | btrfs_set_header_nritems(lower, nritems + 1); |
3453 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3454 | } |
3455 | ||
97571fd0 CM |
3456 | /* |
3457 | * split the node at the specified level in path in two. | |
3458 | * The path is corrected to point to the appropriate node after the split | |
3459 | * | |
3460 | * Before splitting this tries to make some room in the node by pushing | |
3461 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3462 | * |
3463 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3464 | */ |
e02119d5 CM |
3465 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3466 | struct btrfs_root *root, | |
3467 | struct btrfs_path *path, int level) | |
be0e5c09 | 3468 | { |
0b246afa | 3469 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3470 | struct extent_buffer *c; |
3471 | struct extent_buffer *split; | |
3472 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3473 | int mid; |
5c680ed6 | 3474 | int ret; |
7518a238 | 3475 | u32 c_nritems; |
eb60ceac | 3476 | |
5f39d397 | 3477 | c = path->nodes[level]; |
7bb86316 | 3478 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3479 | if (c == root->node) { |
d9abbf1c | 3480 | /* |
90f8d62e JS |
3481 | * trying to split the root, lets make a new one |
3482 | * | |
fdd99c72 | 3483 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3484 | * insert_new_root, because that root buffer will be kept as a |
3485 | * normal node. We are going to log removal of half of the | |
3486 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3487 | * tree lock on the buffer, which is why we cannot race with | |
3488 | * other tree_mod_log users. | |
d9abbf1c | 3489 | */ |
fdd99c72 | 3490 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3491 | if (ret) |
3492 | return ret; | |
b3612421 | 3493 | } else { |
e66f709b | 3494 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3495 | c = path->nodes[level]; |
3496 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3497 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3498 | return 0; |
54aa1f4d CM |
3499 | if (ret < 0) |
3500 | return ret; | |
be0e5c09 | 3501 | } |
e66f709b | 3502 | |
5f39d397 | 3503 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3504 | mid = (c_nritems + 1) / 2; |
3505 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3506 | |
4d75f8a9 DS |
3507 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3508 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3509 | if (IS_ERR(split)) |
3510 | return PTR_ERR(split); | |
3511 | ||
0b246afa | 3512 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3513 | |
b159fa28 | 3514 | memzero_extent_buffer(split, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3515 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3516 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3517 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3518 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3519 | btrfs_set_header_owner(split, root->root_key.objectid); |
0b246afa JM |
3520 | write_extent_buffer_fsid(split, fs_info->fsid); |
3521 | write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid); | |
54aa1f4d | 3522 | |
0b246afa | 3523 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3524 | if (ret) { |
66642832 | 3525 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3526 | return ret; |
3527 | } | |
5f39d397 CM |
3528 | copy_extent_buffer(split, c, |
3529 | btrfs_node_key_ptr_offset(0), | |
3530 | btrfs_node_key_ptr_offset(mid), | |
3531 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3532 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3533 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3534 | ret = 0; |
3535 | ||
5f39d397 CM |
3536 | btrfs_mark_buffer_dirty(c); |
3537 | btrfs_mark_buffer_dirty(split); | |
3538 | ||
2ff7e61e | 3539 | insert_ptr(trans, fs_info, path, &disk_key, split->start, |
c3e06965 | 3540 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3541 | |
5de08d7d | 3542 | if (path->slots[level] >= mid) { |
5c680ed6 | 3543 | path->slots[level] -= mid; |
925baedd | 3544 | btrfs_tree_unlock(c); |
5f39d397 CM |
3545 | free_extent_buffer(c); |
3546 | path->nodes[level] = split; | |
5c680ed6 CM |
3547 | path->slots[level + 1] += 1; |
3548 | } else { | |
925baedd | 3549 | btrfs_tree_unlock(split); |
5f39d397 | 3550 | free_extent_buffer(split); |
be0e5c09 | 3551 | } |
aa5d6bed | 3552 | return ret; |
be0e5c09 CM |
3553 | } |
3554 | ||
74123bd7 CM |
3555 | /* |
3556 | * how many bytes are required to store the items in a leaf. start | |
3557 | * and nr indicate which items in the leaf to check. This totals up the | |
3558 | * space used both by the item structs and the item data | |
3559 | */ | |
5f39d397 | 3560 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3561 | { |
41be1f3b JB |
3562 | struct btrfs_item *start_item; |
3563 | struct btrfs_item *end_item; | |
3564 | struct btrfs_map_token token; | |
be0e5c09 | 3565 | int data_len; |
5f39d397 | 3566 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3567 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3568 | |
3569 | if (!nr) | |
3570 | return 0; | |
41be1f3b | 3571 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3572 | start_item = btrfs_item_nr(start); |
3573 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3574 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3575 | btrfs_token_item_size(l, start_item, &token); | |
3576 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3577 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3578 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3579 | return data_len; |
3580 | } | |
3581 | ||
d4dbff95 CM |
3582 | /* |
3583 | * The space between the end of the leaf items and | |
3584 | * the start of the leaf data. IOW, how much room | |
3585 | * the leaf has left for both items and data | |
3586 | */ | |
2ff7e61e | 3587 | noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info, |
e02119d5 | 3588 | struct extent_buffer *leaf) |
d4dbff95 | 3589 | { |
5f39d397 CM |
3590 | int nritems = btrfs_header_nritems(leaf); |
3591 | int ret; | |
0b246afa JM |
3592 | |
3593 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3594 | if (ret < 0) { |
0b246afa JM |
3595 | btrfs_crit(fs_info, |
3596 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3597 | ret, | |
3598 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3599 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3600 | } |
3601 | return ret; | |
d4dbff95 CM |
3602 | } |
3603 | ||
99d8f83c CM |
3604 | /* |
3605 | * min slot controls the lowest index we're willing to push to the | |
3606 | * right. We'll push up to and including min_slot, but no lower | |
3607 | */ | |
1e47eef2 | 3608 | static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3609 | struct btrfs_path *path, |
3610 | int data_size, int empty, | |
3611 | struct extent_buffer *right, | |
99d8f83c CM |
3612 | int free_space, u32 left_nritems, |
3613 | u32 min_slot) | |
00ec4c51 | 3614 | { |
5f39d397 | 3615 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3616 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3617 | struct btrfs_map_token token; |
5f39d397 | 3618 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3619 | int slot; |
34a38218 | 3620 | u32 i; |
00ec4c51 CM |
3621 | int push_space = 0; |
3622 | int push_items = 0; | |
0783fcfc | 3623 | struct btrfs_item *item; |
34a38218 | 3624 | u32 nr; |
7518a238 | 3625 | u32 right_nritems; |
5f39d397 | 3626 | u32 data_end; |
db94535d | 3627 | u32 this_item_size; |
00ec4c51 | 3628 | |
cfed81a0 CM |
3629 | btrfs_init_map_token(&token); |
3630 | ||
34a38218 CM |
3631 | if (empty) |
3632 | nr = 0; | |
3633 | else | |
99d8f83c | 3634 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3635 | |
31840ae1 | 3636 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3637 | push_space += data_size; |
31840ae1 | 3638 | |
44871b1b | 3639 | slot = path->slots[1]; |
34a38218 CM |
3640 | i = left_nritems - 1; |
3641 | while (i >= nr) { | |
dd3cc16b | 3642 | item = btrfs_item_nr(i); |
db94535d | 3643 | |
31840ae1 ZY |
3644 | if (!empty && push_items > 0) { |
3645 | if (path->slots[0] > i) | |
3646 | break; | |
3647 | if (path->slots[0] == i) { | |
2ff7e61e | 3648 | int space = btrfs_leaf_free_space(fs_info, left); |
31840ae1 ZY |
3649 | if (space + push_space * 2 > free_space) |
3650 | break; | |
3651 | } | |
3652 | } | |
3653 | ||
00ec4c51 | 3654 | if (path->slots[0] == i) |
87b29b20 | 3655 | push_space += data_size; |
db94535d | 3656 | |
db94535d CM |
3657 | this_item_size = btrfs_item_size(left, item); |
3658 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3659 | break; |
31840ae1 | 3660 | |
00ec4c51 | 3661 | push_items++; |
db94535d | 3662 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3663 | if (i == 0) |
3664 | break; | |
3665 | i--; | |
db94535d | 3666 | } |
5f39d397 | 3667 | |
925baedd CM |
3668 | if (push_items == 0) |
3669 | goto out_unlock; | |
5f39d397 | 3670 | |
6c1500f2 | 3671 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3672 | |
00ec4c51 | 3673 | /* push left to right */ |
5f39d397 | 3674 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3675 | |
5f39d397 | 3676 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
2ff7e61e | 3677 | push_space -= leaf_data_end(fs_info, left); |
5f39d397 | 3678 | |
00ec4c51 | 3679 | /* make room in the right data area */ |
2ff7e61e | 3680 | data_end = leaf_data_end(fs_info, right); |
5f39d397 | 3681 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3682 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3683 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3684 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3685 | |
00ec4c51 | 3686 | /* copy from the left data area */ |
3d9ec8c4 | 3687 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3688 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3689 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left), |
d6025579 | 3690 | push_space); |
5f39d397 CM |
3691 | |
3692 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3693 | btrfs_item_nr_offset(0), | |
3694 | right_nritems * sizeof(struct btrfs_item)); | |
3695 | ||
00ec4c51 | 3696 | /* copy the items from left to right */ |
5f39d397 CM |
3697 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3698 | btrfs_item_nr_offset(left_nritems - push_items), | |
3699 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3700 | |
3701 | /* update the item pointers */ | |
7518a238 | 3702 | right_nritems += push_items; |
5f39d397 | 3703 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3704 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3705 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3706 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3707 | push_space -= btrfs_token_item_size(right, item, &token); |
3708 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3709 | } |
3710 | ||
7518a238 | 3711 | left_nritems -= push_items; |
5f39d397 | 3712 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3713 | |
34a38218 CM |
3714 | if (left_nritems) |
3715 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3716 | else |
7c302b49 | 3717 | clean_tree_block(fs_info, left); |
f0486c68 | 3718 | |
5f39d397 | 3719 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3720 | |
5f39d397 CM |
3721 | btrfs_item_key(right, &disk_key, 0); |
3722 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3723 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3724 | |
00ec4c51 | 3725 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3726 | if (path->slots[0] >= left_nritems) { |
3727 | path->slots[0] -= left_nritems; | |
925baedd | 3728 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
7c302b49 | 3729 | clean_tree_block(fs_info, path->nodes[0]); |
925baedd | 3730 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3731 | free_extent_buffer(path->nodes[0]); |
3732 | path->nodes[0] = right; | |
00ec4c51 CM |
3733 | path->slots[1] += 1; |
3734 | } else { | |
925baedd | 3735 | btrfs_tree_unlock(right); |
5f39d397 | 3736 | free_extent_buffer(right); |
00ec4c51 CM |
3737 | } |
3738 | return 0; | |
925baedd CM |
3739 | |
3740 | out_unlock: | |
3741 | btrfs_tree_unlock(right); | |
3742 | free_extent_buffer(right); | |
3743 | return 1; | |
00ec4c51 | 3744 | } |
925baedd | 3745 | |
44871b1b CM |
3746 | /* |
3747 | * push some data in the path leaf to the right, trying to free up at | |
3748 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3749 | * | |
3750 | * returns 1 if the push failed because the other node didn't have enough | |
3751 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3752 | * |
3753 | * this will push starting from min_slot to the end of the leaf. It won't | |
3754 | * push any slot lower than min_slot | |
44871b1b CM |
3755 | */ |
3756 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3757 | *root, struct btrfs_path *path, |
3758 | int min_data_size, int data_size, | |
3759 | int empty, u32 min_slot) | |
44871b1b | 3760 | { |
2ff7e61e | 3761 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3762 | struct extent_buffer *left = path->nodes[0]; |
3763 | struct extent_buffer *right; | |
3764 | struct extent_buffer *upper; | |
3765 | int slot; | |
3766 | int free_space; | |
3767 | u32 left_nritems; | |
3768 | int ret; | |
3769 | ||
3770 | if (!path->nodes[1]) | |
3771 | return 1; | |
3772 | ||
3773 | slot = path->slots[1]; | |
3774 | upper = path->nodes[1]; | |
3775 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3776 | return 1; | |
3777 | ||
3778 | btrfs_assert_tree_locked(path->nodes[1]); | |
3779 | ||
2ff7e61e | 3780 | right = read_node_slot(fs_info, upper, slot + 1); |
fb770ae4 LB |
3781 | /* |
3782 | * slot + 1 is not valid or we fail to read the right node, | |
3783 | * no big deal, just return. | |
3784 | */ | |
3785 | if (IS_ERR(right)) | |
91ca338d TI |
3786 | return 1; |
3787 | ||
44871b1b CM |
3788 | btrfs_tree_lock(right); |
3789 | btrfs_set_lock_blocking(right); | |
3790 | ||
2ff7e61e | 3791 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3792 | if (free_space < data_size) |
3793 | goto out_unlock; | |
3794 | ||
3795 | /* cow and double check */ | |
3796 | ret = btrfs_cow_block(trans, root, right, upper, | |
3797 | slot + 1, &right); | |
3798 | if (ret) | |
3799 | goto out_unlock; | |
3800 | ||
2ff7e61e | 3801 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3802 | if (free_space < data_size) |
3803 | goto out_unlock; | |
3804 | ||
3805 | left_nritems = btrfs_header_nritems(left); | |
3806 | if (left_nritems == 0) | |
3807 | goto out_unlock; | |
3808 | ||
2ef1fed2 FDBM |
3809 | if (path->slots[0] == left_nritems && !empty) { |
3810 | /* Key greater than all keys in the leaf, right neighbor has | |
3811 | * enough room for it and we're not emptying our leaf to delete | |
3812 | * it, therefore use right neighbor to insert the new item and | |
3813 | * no need to touch/dirty our left leaft. */ | |
3814 | btrfs_tree_unlock(left); | |
3815 | free_extent_buffer(left); | |
3816 | path->nodes[0] = right; | |
3817 | path->slots[0] = 0; | |
3818 | path->slots[1]++; | |
3819 | return 0; | |
3820 | } | |
3821 | ||
1e47eef2 | 3822 | return __push_leaf_right(fs_info, path, min_data_size, empty, |
99d8f83c | 3823 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3824 | out_unlock: |
3825 | btrfs_tree_unlock(right); | |
3826 | free_extent_buffer(right); | |
3827 | return 1; | |
3828 | } | |
3829 | ||
74123bd7 CM |
3830 | /* |
3831 | * push some data in the path leaf to the left, trying to free up at | |
3832 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3833 | * |
3834 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3835 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3836 | * items | |
74123bd7 | 3837 | */ |
66cb7ddb | 3838 | static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3839 | struct btrfs_path *path, int data_size, |
3840 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3841 | int free_space, u32 right_nritems, |
3842 | u32 max_slot) | |
be0e5c09 | 3843 | { |
5f39d397 CM |
3844 | struct btrfs_disk_key disk_key; |
3845 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3846 | int i; |
be0e5c09 CM |
3847 | int push_space = 0; |
3848 | int push_items = 0; | |
0783fcfc | 3849 | struct btrfs_item *item; |
7518a238 | 3850 | u32 old_left_nritems; |
34a38218 | 3851 | u32 nr; |
aa5d6bed | 3852 | int ret = 0; |
db94535d CM |
3853 | u32 this_item_size; |
3854 | u32 old_left_item_size; | |
cfed81a0 CM |
3855 | struct btrfs_map_token token; |
3856 | ||
3857 | btrfs_init_map_token(&token); | |
be0e5c09 | 3858 | |
34a38218 | 3859 | if (empty) |
99d8f83c | 3860 | nr = min(right_nritems, max_slot); |
34a38218 | 3861 | else |
99d8f83c | 3862 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3863 | |
3864 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3865 | item = btrfs_item_nr(i); |
db94535d | 3866 | |
31840ae1 ZY |
3867 | if (!empty && push_items > 0) { |
3868 | if (path->slots[0] < i) | |
3869 | break; | |
3870 | if (path->slots[0] == i) { | |
2ff7e61e | 3871 | int space = btrfs_leaf_free_space(fs_info, right); |
31840ae1 ZY |
3872 | if (space + push_space * 2 > free_space) |
3873 | break; | |
3874 | } | |
3875 | } | |
3876 | ||
be0e5c09 | 3877 | if (path->slots[0] == i) |
87b29b20 | 3878 | push_space += data_size; |
db94535d CM |
3879 | |
3880 | this_item_size = btrfs_item_size(right, item); | |
3881 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3882 | break; |
db94535d | 3883 | |
be0e5c09 | 3884 | push_items++; |
db94535d CM |
3885 | push_space += this_item_size + sizeof(*item); |
3886 | } | |
3887 | ||
be0e5c09 | 3888 | if (push_items == 0) { |
925baedd CM |
3889 | ret = 1; |
3890 | goto out; | |
be0e5c09 | 3891 | } |
fae7f21c | 3892 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3893 | |
be0e5c09 | 3894 | /* push data from right to left */ |
5f39d397 CM |
3895 | copy_extent_buffer(left, right, |
3896 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3897 | btrfs_item_nr_offset(0), | |
3898 | push_items * sizeof(struct btrfs_item)); | |
3899 | ||
0b246afa | 3900 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3901 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3902 | |
3d9ec8c4 | 3903 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3904 | leaf_data_end(fs_info, left) - push_space, |
3d9ec8c4 | 3905 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3906 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3907 | push_space); |
5f39d397 | 3908 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3909 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3910 | |
db94535d | 3911 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3912 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3913 | u32 ioff; |
db94535d | 3914 | |
dd3cc16b | 3915 | item = btrfs_item_nr(i); |
db94535d | 3916 | |
cfed81a0 CM |
3917 | ioff = btrfs_token_item_offset(left, item, &token); |
3918 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3919 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3920 | &token); |
be0e5c09 | 3921 | } |
5f39d397 | 3922 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3923 | |
3924 | /* fixup right node */ | |
31b1a2bd JL |
3925 | if (push_items > right_nritems) |
3926 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3927 | right_nritems); |
34a38218 CM |
3928 | |
3929 | if (push_items < right_nritems) { | |
3930 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2ff7e61e | 3931 | leaf_data_end(fs_info, right); |
3d9ec8c4 | 3932 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3933 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3934 | BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3935 | leaf_data_end(fs_info, right), push_space); |
34a38218 CM |
3936 | |
3937 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3938 | btrfs_item_nr_offset(push_items), |
3939 | (btrfs_header_nritems(right) - push_items) * | |
3940 | sizeof(struct btrfs_item)); | |
34a38218 | 3941 | } |
eef1c494 Y |
3942 | right_nritems -= push_items; |
3943 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3944 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3945 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3946 | item = btrfs_item_nr(i); |
db94535d | 3947 | |
cfed81a0 CM |
3948 | push_space = push_space - btrfs_token_item_size(right, |
3949 | item, &token); | |
3950 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3951 | } |
eb60ceac | 3952 | |
5f39d397 | 3953 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3954 | if (right_nritems) |
3955 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3956 | else |
7c302b49 | 3957 | clean_tree_block(fs_info, right); |
098f59c2 | 3958 | |
5f39d397 | 3959 | btrfs_item_key(right, &disk_key, 0); |
0b246afa | 3960 | fixup_low_keys(fs_info, path, &disk_key, 1); |
be0e5c09 CM |
3961 | |
3962 | /* then fixup the leaf pointer in the path */ | |
3963 | if (path->slots[0] < push_items) { | |
3964 | path->slots[0] += old_left_nritems; | |
925baedd | 3965 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3966 | free_extent_buffer(path->nodes[0]); |
3967 | path->nodes[0] = left; | |
be0e5c09 CM |
3968 | path->slots[1] -= 1; |
3969 | } else { | |
925baedd | 3970 | btrfs_tree_unlock(left); |
5f39d397 | 3971 | free_extent_buffer(left); |
be0e5c09 CM |
3972 | path->slots[0] -= push_items; |
3973 | } | |
eb60ceac | 3974 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3975 | return ret; |
925baedd CM |
3976 | out: |
3977 | btrfs_tree_unlock(left); | |
3978 | free_extent_buffer(left); | |
3979 | return ret; | |
be0e5c09 CM |
3980 | } |
3981 | ||
44871b1b CM |
3982 | /* |
3983 | * push some data in the path leaf to the left, trying to free up at | |
3984 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3985 | * |
3986 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3987 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3988 | * items | |
44871b1b CM |
3989 | */ |
3990 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3991 | *root, struct btrfs_path *path, int min_data_size, |
3992 | int data_size, int empty, u32 max_slot) | |
44871b1b | 3993 | { |
2ff7e61e | 3994 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3995 | struct extent_buffer *right = path->nodes[0]; |
3996 | struct extent_buffer *left; | |
3997 | int slot; | |
3998 | int free_space; | |
3999 | u32 right_nritems; | |
4000 | int ret = 0; | |
4001 | ||
4002 | slot = path->slots[1]; | |
4003 | if (slot == 0) | |
4004 | return 1; | |
4005 | if (!path->nodes[1]) | |
4006 | return 1; | |
4007 | ||
4008 | right_nritems = btrfs_header_nritems(right); | |
4009 | if (right_nritems == 0) | |
4010 | return 1; | |
4011 | ||
4012 | btrfs_assert_tree_locked(path->nodes[1]); | |
4013 | ||
2ff7e61e | 4014 | left = read_node_slot(fs_info, path->nodes[1], slot - 1); |
fb770ae4 LB |
4015 | /* |
4016 | * slot - 1 is not valid or we fail to read the left node, | |
4017 | * no big deal, just return. | |
4018 | */ | |
4019 | if (IS_ERR(left)) | |
91ca338d TI |
4020 | return 1; |
4021 | ||
44871b1b CM |
4022 | btrfs_tree_lock(left); |
4023 | btrfs_set_lock_blocking(left); | |
4024 | ||
2ff7e61e | 4025 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4026 | if (free_space < data_size) { |
4027 | ret = 1; | |
4028 | goto out; | |
4029 | } | |
4030 | ||
4031 | /* cow and double check */ | |
4032 | ret = btrfs_cow_block(trans, root, left, | |
4033 | path->nodes[1], slot - 1, &left); | |
4034 | if (ret) { | |
4035 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4036 | if (ret == -ENOSPC) |
4037 | ret = 1; | |
44871b1b CM |
4038 | goto out; |
4039 | } | |
4040 | ||
2ff7e61e | 4041 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4042 | if (free_space < data_size) { |
4043 | ret = 1; | |
4044 | goto out; | |
4045 | } | |
4046 | ||
66cb7ddb | 4047 | return __push_leaf_left(fs_info, path, min_data_size, |
99d8f83c CM |
4048 | empty, left, free_space, right_nritems, |
4049 | max_slot); | |
44871b1b CM |
4050 | out: |
4051 | btrfs_tree_unlock(left); | |
4052 | free_extent_buffer(left); | |
4053 | return ret; | |
4054 | } | |
4055 | ||
4056 | /* | |
4057 | * split the path's leaf in two, making sure there is at least data_size | |
4058 | * available for the resulting leaf level of the path. | |
44871b1b | 4059 | */ |
143bede5 | 4060 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2ff7e61e | 4061 | struct btrfs_fs_info *fs_info, |
143bede5 JM |
4062 | struct btrfs_path *path, |
4063 | struct extent_buffer *l, | |
4064 | struct extent_buffer *right, | |
4065 | int slot, int mid, int nritems) | |
44871b1b CM |
4066 | { |
4067 | int data_copy_size; | |
4068 | int rt_data_off; | |
4069 | int i; | |
44871b1b | 4070 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4071 | struct btrfs_map_token token; |
4072 | ||
4073 | btrfs_init_map_token(&token); | |
44871b1b CM |
4074 | |
4075 | nritems = nritems - mid; | |
4076 | btrfs_set_header_nritems(right, nritems); | |
2ff7e61e | 4077 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l); |
44871b1b CM |
4078 | |
4079 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4080 | btrfs_item_nr_offset(mid), | |
4081 | nritems * sizeof(struct btrfs_item)); | |
4082 | ||
4083 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4084 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4085 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
2ff7e61e | 4086 | leaf_data_end(fs_info, l), data_copy_size); |
44871b1b | 4087 | |
0b246afa | 4088 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4089 | |
4090 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4091 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4092 | u32 ioff; |
4093 | ||
cfed81a0 CM |
4094 | ioff = btrfs_token_item_offset(right, item, &token); |
4095 | btrfs_set_token_item_offset(right, item, | |
4096 | ioff + rt_data_off, &token); | |
44871b1b CM |
4097 | } |
4098 | ||
44871b1b | 4099 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4100 | btrfs_item_key(right, &disk_key, 0); |
2ff7e61e | 4101 | insert_ptr(trans, fs_info, path, &disk_key, right->start, |
c3e06965 | 4102 | path->slots[1] + 1, 1); |
44871b1b CM |
4103 | |
4104 | btrfs_mark_buffer_dirty(right); | |
4105 | btrfs_mark_buffer_dirty(l); | |
4106 | BUG_ON(path->slots[0] != slot); | |
4107 | ||
44871b1b CM |
4108 | if (mid <= slot) { |
4109 | btrfs_tree_unlock(path->nodes[0]); | |
4110 | free_extent_buffer(path->nodes[0]); | |
4111 | path->nodes[0] = right; | |
4112 | path->slots[0] -= mid; | |
4113 | path->slots[1] += 1; | |
4114 | } else { | |
4115 | btrfs_tree_unlock(right); | |
4116 | free_extent_buffer(right); | |
4117 | } | |
4118 | ||
4119 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4120 | } |
4121 | ||
99d8f83c CM |
4122 | /* |
4123 | * double splits happen when we need to insert a big item in the middle | |
4124 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4125 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4126 | * A B C | |
4127 | * | |
4128 | * We avoid this by trying to push the items on either side of our target | |
4129 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4130 | * completely. | |
4131 | */ | |
4132 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4133 | struct btrfs_root *root, | |
4134 | struct btrfs_path *path, | |
4135 | int data_size) | |
4136 | { | |
2ff7e61e | 4137 | struct btrfs_fs_info *fs_info = root->fs_info; |
99d8f83c CM |
4138 | int ret; |
4139 | int progress = 0; | |
4140 | int slot; | |
4141 | u32 nritems; | |
5a4267ca | 4142 | int space_needed = data_size; |
99d8f83c CM |
4143 | |
4144 | slot = path->slots[0]; | |
5a4267ca | 4145 | if (slot < btrfs_header_nritems(path->nodes[0])) |
2ff7e61e | 4146 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); |
99d8f83c CM |
4147 | |
4148 | /* | |
4149 | * try to push all the items after our slot into the | |
4150 | * right leaf | |
4151 | */ | |
5a4267ca | 4152 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4153 | if (ret < 0) |
4154 | return ret; | |
4155 | ||
4156 | if (ret == 0) | |
4157 | progress++; | |
4158 | ||
4159 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4160 | /* | |
4161 | * our goal is to get our slot at the start or end of a leaf. If | |
4162 | * we've done so we're done | |
4163 | */ | |
4164 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4165 | return 0; | |
4166 | ||
2ff7e61e | 4167 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4168 | return 0; |
4169 | ||
4170 | /* try to push all the items before our slot into the next leaf */ | |
4171 | slot = path->slots[0]; | |
263d3995 FM |
4172 | space_needed = data_size; |
4173 | if (slot > 0) | |
4174 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); | |
5a4267ca | 4175 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4176 | if (ret < 0) |
4177 | return ret; | |
4178 | ||
4179 | if (ret == 0) | |
4180 | progress++; | |
4181 | ||
4182 | if (progress) | |
4183 | return 0; | |
4184 | return 1; | |
4185 | } | |
4186 | ||
74123bd7 CM |
4187 | /* |
4188 | * split the path's leaf in two, making sure there is at least data_size | |
4189 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4190 | * |
4191 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4192 | */ |
e02119d5 CM |
4193 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4194 | struct btrfs_root *root, | |
310712b2 | 4195 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4196 | struct btrfs_path *path, int data_size, |
4197 | int extend) | |
be0e5c09 | 4198 | { |
5d4f98a2 | 4199 | struct btrfs_disk_key disk_key; |
5f39d397 | 4200 | struct extent_buffer *l; |
7518a238 | 4201 | u32 nritems; |
eb60ceac CM |
4202 | int mid; |
4203 | int slot; | |
5f39d397 | 4204 | struct extent_buffer *right; |
b7a0365e | 4205 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4206 | int ret = 0; |
aa5d6bed | 4207 | int wret; |
5d4f98a2 | 4208 | int split; |
cc0c5538 | 4209 | int num_doubles = 0; |
99d8f83c | 4210 | int tried_avoid_double = 0; |
aa5d6bed | 4211 | |
a5719521 YZ |
4212 | l = path->nodes[0]; |
4213 | slot = path->slots[0]; | |
4214 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4215 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4216 | return -EOVERFLOW; |
4217 | ||
40689478 | 4218 | /* first try to make some room by pushing left and right */ |
33157e05 | 4219 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4220 | int space_needed = data_size; |
4221 | ||
4222 | if (slot < btrfs_header_nritems(l)) | |
2ff7e61e | 4223 | space_needed -= btrfs_leaf_free_space(fs_info, l); |
5a4267ca FDBM |
4224 | |
4225 | wret = push_leaf_right(trans, root, path, space_needed, | |
4226 | space_needed, 0, 0); | |
d397712b | 4227 | if (wret < 0) |
eaee50e8 | 4228 | return wret; |
3685f791 | 4229 | if (wret) { |
263d3995 FM |
4230 | space_needed = data_size; |
4231 | if (slot > 0) | |
4232 | space_needed -= btrfs_leaf_free_space(fs_info, | |
4233 | l); | |
5a4267ca FDBM |
4234 | wret = push_leaf_left(trans, root, path, space_needed, |
4235 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4236 | if (wret < 0) |
4237 | return wret; | |
4238 | } | |
4239 | l = path->nodes[0]; | |
aa5d6bed | 4240 | |
3685f791 | 4241 | /* did the pushes work? */ |
2ff7e61e | 4242 | if (btrfs_leaf_free_space(fs_info, l) >= data_size) |
3685f791 | 4243 | return 0; |
3326d1b0 | 4244 | } |
aa5d6bed | 4245 | |
5c680ed6 | 4246 | if (!path->nodes[1]) { |
fdd99c72 | 4247 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4248 | if (ret) |
4249 | return ret; | |
4250 | } | |
cc0c5538 | 4251 | again: |
5d4f98a2 | 4252 | split = 1; |
cc0c5538 | 4253 | l = path->nodes[0]; |
eb60ceac | 4254 | slot = path->slots[0]; |
5f39d397 | 4255 | nritems = btrfs_header_nritems(l); |
d397712b | 4256 | mid = (nritems + 1) / 2; |
54aa1f4d | 4257 | |
5d4f98a2 YZ |
4258 | if (mid <= slot) { |
4259 | if (nritems == 1 || | |
4260 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4261 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4262 | if (slot >= nritems) { |
4263 | split = 0; | |
4264 | } else { | |
4265 | mid = slot; | |
4266 | if (mid != nritems && | |
4267 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4268 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4269 | if (data_size && !tried_avoid_double) |
4270 | goto push_for_double; | |
5d4f98a2 YZ |
4271 | split = 2; |
4272 | } | |
4273 | } | |
4274 | } | |
4275 | } else { | |
4276 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4277 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4278 | if (!extend && data_size && slot == 0) { |
4279 | split = 0; | |
4280 | } else if ((extend || !data_size) && slot == 0) { | |
4281 | mid = 1; | |
4282 | } else { | |
4283 | mid = slot; | |
4284 | if (mid != nritems && | |
4285 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4286 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4287 | if (data_size && !tried_avoid_double) |
4288 | goto push_for_double; | |
67871254 | 4289 | split = 2; |
5d4f98a2 YZ |
4290 | } |
4291 | } | |
4292 | } | |
4293 | } | |
4294 | ||
4295 | if (split == 0) | |
4296 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4297 | else | |
4298 | btrfs_item_key(l, &disk_key, mid); | |
4299 | ||
4d75f8a9 DS |
4300 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4301 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4302 | if (IS_ERR(right)) |
5f39d397 | 4303 | return PTR_ERR(right); |
f0486c68 | 4304 | |
0b246afa | 4305 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4306 | |
b159fa28 | 4307 | memzero_extent_buffer(right, 0, sizeof(struct btrfs_header)); |
db94535d | 4308 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4309 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4310 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4311 | btrfs_set_header_owner(right, root->root_key.objectid); |
4312 | btrfs_set_header_level(right, 0); | |
d24ee97b DS |
4313 | write_extent_buffer_fsid(right, fs_info->fsid); |
4314 | write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid); | |
44871b1b | 4315 | |
5d4f98a2 YZ |
4316 | if (split == 0) { |
4317 | if (mid <= slot) { | |
4318 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4319 | insert_ptr(trans, fs_info, path, &disk_key, |
4320 | right->start, path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
4321 | btrfs_tree_unlock(path->nodes[0]); |
4322 | free_extent_buffer(path->nodes[0]); | |
4323 | path->nodes[0] = right; | |
4324 | path->slots[0] = 0; | |
4325 | path->slots[1] += 1; | |
4326 | } else { | |
4327 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4328 | insert_ptr(trans, fs_info, path, &disk_key, |
4329 | right->start, path->slots[1], 1); | |
5d4f98a2 YZ |
4330 | btrfs_tree_unlock(path->nodes[0]); |
4331 | free_extent_buffer(path->nodes[0]); | |
4332 | path->nodes[0] = right; | |
4333 | path->slots[0] = 0; | |
143bede5 | 4334 | if (path->slots[1] == 0) |
b7a0365e | 4335 | fixup_low_keys(fs_info, path, &disk_key, 1); |
d4dbff95 | 4336 | } |
196e0249 LB |
4337 | /* |
4338 | * We create a new leaf 'right' for the required ins_len and | |
4339 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4340 | * the content of ins_len to 'right'. | |
4341 | */ | |
5d4f98a2 | 4342 | return ret; |
d4dbff95 | 4343 | } |
74123bd7 | 4344 | |
2ff7e61e | 4345 | copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems); |
31840ae1 | 4346 | |
5d4f98a2 | 4347 | if (split == 2) { |
cc0c5538 CM |
4348 | BUG_ON(num_doubles != 0); |
4349 | num_doubles++; | |
4350 | goto again; | |
a429e513 | 4351 | } |
44871b1b | 4352 | |
143bede5 | 4353 | return 0; |
99d8f83c CM |
4354 | |
4355 | push_for_double: | |
4356 | push_for_double_split(trans, root, path, data_size); | |
4357 | tried_avoid_double = 1; | |
2ff7e61e | 4358 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4359 | return 0; |
4360 | goto again; | |
be0e5c09 CM |
4361 | } |
4362 | ||
ad48fd75 YZ |
4363 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4364 | struct btrfs_root *root, | |
4365 | struct btrfs_path *path, int ins_len) | |
459931ec | 4366 | { |
2ff7e61e | 4367 | struct btrfs_fs_info *fs_info = root->fs_info; |
ad48fd75 | 4368 | struct btrfs_key key; |
459931ec | 4369 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4370 | struct btrfs_file_extent_item *fi; |
4371 | u64 extent_len = 0; | |
4372 | u32 item_size; | |
4373 | int ret; | |
459931ec CM |
4374 | |
4375 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4376 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4377 | ||
4378 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4379 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4380 | ||
2ff7e61e | 4381 | if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len) |
ad48fd75 | 4382 | return 0; |
459931ec CM |
4383 | |
4384 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4385 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4386 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4387 | struct btrfs_file_extent_item); | |
4388 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4389 | } | |
b3b4aa74 | 4390 | btrfs_release_path(path); |
459931ec | 4391 | |
459931ec | 4392 | path->keep_locks = 1; |
ad48fd75 YZ |
4393 | path->search_for_split = 1; |
4394 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4395 | path->search_for_split = 0; |
a8df6fe6 FM |
4396 | if (ret > 0) |
4397 | ret = -EAGAIN; | |
ad48fd75 YZ |
4398 | if (ret < 0) |
4399 | goto err; | |
459931ec | 4400 | |
ad48fd75 YZ |
4401 | ret = -EAGAIN; |
4402 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4403 | /* if our item isn't there, return now */ |
4404 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4405 | goto err; |
4406 | ||
109f6aef | 4407 | /* the leaf has changed, it now has room. return now */ |
2ff7e61e | 4408 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len) |
109f6aef CM |
4409 | goto err; |
4410 | ||
ad48fd75 YZ |
4411 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4412 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4413 | struct btrfs_file_extent_item); | |
4414 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4415 | goto err; | |
459931ec CM |
4416 | } |
4417 | ||
b9473439 | 4418 | btrfs_set_path_blocking(path); |
ad48fd75 | 4419 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4420 | if (ret) |
4421 | goto err; | |
459931ec | 4422 | |
ad48fd75 | 4423 | path->keep_locks = 0; |
b9473439 | 4424 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4425 | return 0; |
4426 | err: | |
4427 | path->keep_locks = 0; | |
4428 | return ret; | |
4429 | } | |
4430 | ||
4961e293 | 4431 | static noinline int split_item(struct btrfs_fs_info *fs_info, |
ad48fd75 | 4432 | struct btrfs_path *path, |
310712b2 | 4433 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4434 | unsigned long split_offset) |
4435 | { | |
4436 | struct extent_buffer *leaf; | |
4437 | struct btrfs_item *item; | |
4438 | struct btrfs_item *new_item; | |
4439 | int slot; | |
4440 | char *buf; | |
4441 | u32 nritems; | |
4442 | u32 item_size; | |
4443 | u32 orig_offset; | |
4444 | struct btrfs_disk_key disk_key; | |
4445 | ||
b9473439 | 4446 | leaf = path->nodes[0]; |
2ff7e61e | 4447 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4448 | |
b4ce94de CM |
4449 | btrfs_set_path_blocking(path); |
4450 | ||
dd3cc16b | 4451 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4452 | orig_offset = btrfs_item_offset(leaf, item); |
4453 | item_size = btrfs_item_size(leaf, item); | |
4454 | ||
459931ec | 4455 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4456 | if (!buf) |
4457 | return -ENOMEM; | |
4458 | ||
459931ec CM |
4459 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4460 | path->slots[0]), item_size); | |
459931ec | 4461 | |
ad48fd75 | 4462 | slot = path->slots[0] + 1; |
459931ec | 4463 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4464 | if (slot != nritems) { |
4465 | /* shift the items */ | |
4466 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4467 | btrfs_item_nr_offset(slot), |
4468 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4469 | } |
4470 | ||
4471 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4472 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4473 | ||
dd3cc16b | 4474 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4475 | |
4476 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4477 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4478 | ||
4479 | btrfs_set_item_offset(leaf, item, | |
4480 | orig_offset + item_size - split_offset); | |
4481 | btrfs_set_item_size(leaf, item, split_offset); | |
4482 | ||
4483 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4484 | ||
4485 | /* write the data for the start of the original item */ | |
4486 | write_extent_buffer(leaf, buf, | |
4487 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4488 | split_offset); | |
4489 | ||
4490 | /* write the data for the new item */ | |
4491 | write_extent_buffer(leaf, buf + split_offset, | |
4492 | btrfs_item_ptr_offset(leaf, slot), | |
4493 | item_size - split_offset); | |
4494 | btrfs_mark_buffer_dirty(leaf); | |
4495 | ||
2ff7e61e | 4496 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0); |
459931ec | 4497 | kfree(buf); |
ad48fd75 YZ |
4498 | return 0; |
4499 | } | |
4500 | ||
4501 | /* | |
4502 | * This function splits a single item into two items, | |
4503 | * giving 'new_key' to the new item and splitting the | |
4504 | * old one at split_offset (from the start of the item). | |
4505 | * | |
4506 | * The path may be released by this operation. After | |
4507 | * the split, the path is pointing to the old item. The | |
4508 | * new item is going to be in the same node as the old one. | |
4509 | * | |
4510 | * Note, the item being split must be smaller enough to live alone on | |
4511 | * a tree block with room for one extra struct btrfs_item | |
4512 | * | |
4513 | * This allows us to split the item in place, keeping a lock on the | |
4514 | * leaf the entire time. | |
4515 | */ | |
4516 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4517 | struct btrfs_root *root, | |
4518 | struct btrfs_path *path, | |
310712b2 | 4519 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4520 | unsigned long split_offset) |
4521 | { | |
4522 | int ret; | |
4523 | ret = setup_leaf_for_split(trans, root, path, | |
4524 | sizeof(struct btrfs_item)); | |
4525 | if (ret) | |
4526 | return ret; | |
4527 | ||
4961e293 | 4528 | ret = split_item(root->fs_info, path, new_key, split_offset); |
459931ec CM |
4529 | return ret; |
4530 | } | |
4531 | ||
ad48fd75 YZ |
4532 | /* |
4533 | * This function duplicate a item, giving 'new_key' to the new item. | |
4534 | * It guarantees both items live in the same tree leaf and the new item | |
4535 | * is contiguous with the original item. | |
4536 | * | |
4537 | * This allows us to split file extent in place, keeping a lock on the | |
4538 | * leaf the entire time. | |
4539 | */ | |
4540 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4541 | struct btrfs_root *root, | |
4542 | struct btrfs_path *path, | |
310712b2 | 4543 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4544 | { |
4545 | struct extent_buffer *leaf; | |
4546 | int ret; | |
4547 | u32 item_size; | |
4548 | ||
4549 | leaf = path->nodes[0]; | |
4550 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4551 | ret = setup_leaf_for_split(trans, root, path, | |
4552 | item_size + sizeof(struct btrfs_item)); | |
4553 | if (ret) | |
4554 | return ret; | |
4555 | ||
4556 | path->slots[0]++; | |
afe5fea7 | 4557 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4558 | item_size, item_size + |
4559 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4560 | leaf = path->nodes[0]; |
4561 | memcpy_extent_buffer(leaf, | |
4562 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4563 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4564 | item_size); | |
4565 | return 0; | |
4566 | } | |
4567 | ||
d352ac68 CM |
4568 | /* |
4569 | * make the item pointed to by the path smaller. new_size indicates | |
4570 | * how small to make it, and from_end tells us if we just chop bytes | |
4571 | * off the end of the item or if we shift the item to chop bytes off | |
4572 | * the front. | |
4573 | */ | |
2ff7e61e JM |
4574 | void btrfs_truncate_item(struct btrfs_fs_info *fs_info, |
4575 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 4576 | { |
b18c6685 | 4577 | int slot; |
5f39d397 CM |
4578 | struct extent_buffer *leaf; |
4579 | struct btrfs_item *item; | |
b18c6685 CM |
4580 | u32 nritems; |
4581 | unsigned int data_end; | |
4582 | unsigned int old_data_start; | |
4583 | unsigned int old_size; | |
4584 | unsigned int size_diff; | |
4585 | int i; | |
cfed81a0 CM |
4586 | struct btrfs_map_token token; |
4587 | ||
4588 | btrfs_init_map_token(&token); | |
b18c6685 | 4589 | |
5f39d397 | 4590 | leaf = path->nodes[0]; |
179e29e4 CM |
4591 | slot = path->slots[0]; |
4592 | ||
4593 | old_size = btrfs_item_size_nr(leaf, slot); | |
4594 | if (old_size == new_size) | |
143bede5 | 4595 | return; |
b18c6685 | 4596 | |
5f39d397 | 4597 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4598 | data_end = leaf_data_end(fs_info, leaf); |
b18c6685 | 4599 | |
5f39d397 | 4600 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4601 | |
b18c6685 CM |
4602 | size_diff = old_size - new_size; |
4603 | ||
4604 | BUG_ON(slot < 0); | |
4605 | BUG_ON(slot >= nritems); | |
4606 | ||
4607 | /* | |
4608 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4609 | */ | |
4610 | /* first correct the data pointers */ | |
4611 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4612 | u32 ioff; |
dd3cc16b | 4613 | item = btrfs_item_nr(i); |
db94535d | 4614 | |
cfed81a0 CM |
4615 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4616 | btrfs_set_token_item_offset(leaf, item, | |
4617 | ioff + size_diff, &token); | |
b18c6685 | 4618 | } |
db94535d | 4619 | |
b18c6685 | 4620 | /* shift the data */ |
179e29e4 | 4621 | if (from_end) { |
3d9ec8c4 NB |
4622 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4623 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4624 | data_end, old_data_start + new_size - data_end); |
4625 | } else { | |
4626 | struct btrfs_disk_key disk_key; | |
4627 | u64 offset; | |
4628 | ||
4629 | btrfs_item_key(leaf, &disk_key, slot); | |
4630 | ||
4631 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4632 | unsigned long ptr; | |
4633 | struct btrfs_file_extent_item *fi; | |
4634 | ||
4635 | fi = btrfs_item_ptr(leaf, slot, | |
4636 | struct btrfs_file_extent_item); | |
4637 | fi = (struct btrfs_file_extent_item *)( | |
4638 | (unsigned long)fi - size_diff); | |
4639 | ||
4640 | if (btrfs_file_extent_type(leaf, fi) == | |
4641 | BTRFS_FILE_EXTENT_INLINE) { | |
4642 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4643 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4644 | (unsigned long)fi, |
7ec20afb | 4645 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4646 | } |
4647 | } | |
4648 | ||
3d9ec8c4 NB |
4649 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4650 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4651 | data_end, old_data_start - data_end); |
4652 | ||
4653 | offset = btrfs_disk_key_offset(&disk_key); | |
4654 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4655 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4656 | if (slot == 0) | |
0b246afa | 4657 | fixup_low_keys(fs_info, path, &disk_key, 1); |
179e29e4 | 4658 | } |
5f39d397 | 4659 | |
dd3cc16b | 4660 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4661 | btrfs_set_item_size(leaf, item, new_size); |
4662 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4663 | |
2ff7e61e | 4664 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4665 | btrfs_print_leaf(leaf); |
b18c6685 | 4666 | BUG(); |
5f39d397 | 4667 | } |
b18c6685 CM |
4668 | } |
4669 | ||
d352ac68 | 4670 | /* |
8f69dbd2 | 4671 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4672 | */ |
2ff7e61e | 4673 | void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 4674 | u32 data_size) |
6567e837 | 4675 | { |
6567e837 | 4676 | int slot; |
5f39d397 CM |
4677 | struct extent_buffer *leaf; |
4678 | struct btrfs_item *item; | |
6567e837 CM |
4679 | u32 nritems; |
4680 | unsigned int data_end; | |
4681 | unsigned int old_data; | |
4682 | unsigned int old_size; | |
4683 | int i; | |
cfed81a0 CM |
4684 | struct btrfs_map_token token; |
4685 | ||
4686 | btrfs_init_map_token(&token); | |
6567e837 | 4687 | |
5f39d397 | 4688 | leaf = path->nodes[0]; |
6567e837 | 4689 | |
5f39d397 | 4690 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4691 | data_end = leaf_data_end(fs_info, leaf); |
6567e837 | 4692 | |
2ff7e61e | 4693 | if (btrfs_leaf_free_space(fs_info, leaf) < data_size) { |
a4f78750 | 4694 | btrfs_print_leaf(leaf); |
6567e837 | 4695 | BUG(); |
5f39d397 | 4696 | } |
6567e837 | 4697 | slot = path->slots[0]; |
5f39d397 | 4698 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4699 | |
4700 | BUG_ON(slot < 0); | |
3326d1b0 | 4701 | if (slot >= nritems) { |
a4f78750 | 4702 | btrfs_print_leaf(leaf); |
0b246afa JM |
4703 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4704 | slot, nritems); | |
3326d1b0 CM |
4705 | BUG_ON(1); |
4706 | } | |
6567e837 CM |
4707 | |
4708 | /* | |
4709 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4710 | */ | |
4711 | /* first correct the data pointers */ | |
4712 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4713 | u32 ioff; |
dd3cc16b | 4714 | item = btrfs_item_nr(i); |
db94535d | 4715 | |
cfed81a0 CM |
4716 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4717 | btrfs_set_token_item_offset(leaf, item, | |
4718 | ioff - data_size, &token); | |
6567e837 | 4719 | } |
5f39d397 | 4720 | |
6567e837 | 4721 | /* shift the data */ |
3d9ec8c4 NB |
4722 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4723 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4724 | data_end, old_data - data_end); |
5f39d397 | 4725 | |
6567e837 | 4726 | data_end = old_data; |
5f39d397 | 4727 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4728 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4729 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4730 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4731 | |
2ff7e61e | 4732 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4733 | btrfs_print_leaf(leaf); |
6567e837 | 4734 | BUG(); |
5f39d397 | 4735 | } |
6567e837 CM |
4736 | } |
4737 | ||
74123bd7 | 4738 | /* |
44871b1b CM |
4739 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4740 | * to save stack depth by doing the bulk of the work in a function | |
4741 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4742 | */ |
afe5fea7 | 4743 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4744 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4745 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4746 | { |
0b246afa | 4747 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4748 | struct btrfs_item *item; |
9c58309d | 4749 | int i; |
7518a238 | 4750 | u32 nritems; |
be0e5c09 | 4751 | unsigned int data_end; |
e2fa7227 | 4752 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4753 | struct extent_buffer *leaf; |
4754 | int slot; | |
cfed81a0 CM |
4755 | struct btrfs_map_token token; |
4756 | ||
24cdc847 FM |
4757 | if (path->slots[0] == 0) { |
4758 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
0b246afa | 4759 | fixup_low_keys(fs_info, path, &disk_key, 1); |
24cdc847 FM |
4760 | } |
4761 | btrfs_unlock_up_safe(path, 1); | |
4762 | ||
cfed81a0 | 4763 | btrfs_init_map_token(&token); |
e2fa7227 | 4764 | |
5f39d397 | 4765 | leaf = path->nodes[0]; |
44871b1b | 4766 | slot = path->slots[0]; |
74123bd7 | 4767 | |
5f39d397 | 4768 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4769 | data_end = leaf_data_end(fs_info, leaf); |
eb60ceac | 4770 | |
2ff7e61e | 4771 | if (btrfs_leaf_free_space(fs_info, leaf) < total_size) { |
a4f78750 | 4772 | btrfs_print_leaf(leaf); |
0b246afa | 4773 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
2ff7e61e | 4774 | total_size, btrfs_leaf_free_space(fs_info, leaf)); |
be0e5c09 | 4775 | BUG(); |
d4dbff95 | 4776 | } |
5f39d397 | 4777 | |
be0e5c09 | 4778 | if (slot != nritems) { |
5f39d397 | 4779 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4780 | |
5f39d397 | 4781 | if (old_data < data_end) { |
a4f78750 | 4782 | btrfs_print_leaf(leaf); |
0b246afa | 4783 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4784 | slot, old_data, data_end); |
5f39d397 CM |
4785 | BUG_ON(1); |
4786 | } | |
be0e5c09 CM |
4787 | /* |
4788 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4789 | */ | |
4790 | /* first correct the data pointers */ | |
0783fcfc | 4791 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4792 | u32 ioff; |
db94535d | 4793 | |
62e85577 | 4794 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4795 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4796 | btrfs_set_token_item_offset(leaf, item, | |
4797 | ioff - total_data, &token); | |
0783fcfc | 4798 | } |
be0e5c09 | 4799 | /* shift the items */ |
9c58309d | 4800 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4801 | btrfs_item_nr_offset(slot), |
d6025579 | 4802 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4803 | |
4804 | /* shift the data */ | |
3d9ec8c4 NB |
4805 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4806 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4807 | data_end, old_data - data_end); |
be0e5c09 CM |
4808 | data_end = old_data; |
4809 | } | |
5f39d397 | 4810 | |
62e2749e | 4811 | /* setup the item for the new data */ |
9c58309d CM |
4812 | for (i = 0; i < nr; i++) { |
4813 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4814 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4815 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4816 | btrfs_set_token_item_offset(leaf, item, |
4817 | data_end - data_size[i], &token); | |
9c58309d | 4818 | data_end -= data_size[i]; |
cfed81a0 | 4819 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4820 | } |
44871b1b | 4821 | |
9c58309d | 4822 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4823 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4824 | |
2ff7e61e | 4825 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4826 | btrfs_print_leaf(leaf); |
be0e5c09 | 4827 | BUG(); |
5f39d397 | 4828 | } |
44871b1b CM |
4829 | } |
4830 | ||
4831 | /* | |
4832 | * Given a key and some data, insert items into the tree. | |
4833 | * This does all the path init required, making room in the tree if needed. | |
4834 | */ | |
4835 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4836 | struct btrfs_root *root, | |
4837 | struct btrfs_path *path, | |
310712b2 | 4838 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4839 | int nr) |
4840 | { | |
44871b1b CM |
4841 | int ret = 0; |
4842 | int slot; | |
4843 | int i; | |
4844 | u32 total_size = 0; | |
4845 | u32 total_data = 0; | |
4846 | ||
4847 | for (i = 0; i < nr; i++) | |
4848 | total_data += data_size[i]; | |
4849 | ||
4850 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4851 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4852 | if (ret == 0) | |
4853 | return -EEXIST; | |
4854 | if (ret < 0) | |
143bede5 | 4855 | return ret; |
44871b1b | 4856 | |
44871b1b CM |
4857 | slot = path->slots[0]; |
4858 | BUG_ON(slot < 0); | |
4859 | ||
afe5fea7 | 4860 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4861 | total_data, total_size, nr); |
143bede5 | 4862 | return 0; |
62e2749e CM |
4863 | } |
4864 | ||
4865 | /* | |
4866 | * Given a key and some data, insert an item into the tree. | |
4867 | * This does all the path init required, making room in the tree if needed. | |
4868 | */ | |
310712b2 OS |
4869 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4870 | const struct btrfs_key *cpu_key, void *data, | |
4871 | u32 data_size) | |
62e2749e CM |
4872 | { |
4873 | int ret = 0; | |
2c90e5d6 | 4874 | struct btrfs_path *path; |
5f39d397 CM |
4875 | struct extent_buffer *leaf; |
4876 | unsigned long ptr; | |
62e2749e | 4877 | |
2c90e5d6 | 4878 | path = btrfs_alloc_path(); |
db5b493a TI |
4879 | if (!path) |
4880 | return -ENOMEM; | |
2c90e5d6 | 4881 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4882 | if (!ret) { |
5f39d397 CM |
4883 | leaf = path->nodes[0]; |
4884 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4885 | write_extent_buffer(leaf, data, ptr, data_size); | |
4886 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4887 | } |
2c90e5d6 | 4888 | btrfs_free_path(path); |
aa5d6bed | 4889 | return ret; |
be0e5c09 CM |
4890 | } |
4891 | ||
74123bd7 | 4892 | /* |
5de08d7d | 4893 | * delete the pointer from a given node. |
74123bd7 | 4894 | * |
d352ac68 CM |
4895 | * the tree should have been previously balanced so the deletion does not |
4896 | * empty a node. | |
74123bd7 | 4897 | */ |
afe5fea7 TI |
4898 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4899 | int level, int slot) | |
be0e5c09 | 4900 | { |
0b246afa | 4901 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4902 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4903 | u32 nritems; |
f3ea38da | 4904 | int ret; |
be0e5c09 | 4905 | |
5f39d397 | 4906 | nritems = btrfs_header_nritems(parent); |
d397712b | 4907 | if (slot != nritems - 1) { |
0e411ece | 4908 | if (level) |
0b246afa | 4909 | tree_mod_log_eb_move(fs_info, parent, slot, |
f3ea38da | 4910 | slot + 1, nritems - slot - 1); |
5f39d397 CM |
4911 | memmove_extent_buffer(parent, |
4912 | btrfs_node_key_ptr_offset(slot), | |
4913 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4914 | sizeof(struct btrfs_key_ptr) * |
4915 | (nritems - slot - 1)); | |
57ba86c0 | 4916 | } else if (level) { |
0b246afa | 4917 | ret = tree_mod_log_insert_key(fs_info, parent, slot, |
c8cc6341 | 4918 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4919 | BUG_ON(ret < 0); |
bb803951 | 4920 | } |
f3ea38da | 4921 | |
7518a238 | 4922 | nritems--; |
5f39d397 | 4923 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4924 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4925 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4926 | /* just turn the root into a leaf and break */ |
5f39d397 | 4927 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4928 | } else if (slot == 0) { |
5f39d397 CM |
4929 | struct btrfs_disk_key disk_key; |
4930 | ||
4931 | btrfs_node_key(parent, &disk_key, 0); | |
0b246afa | 4932 | fixup_low_keys(fs_info, path, &disk_key, level + 1); |
be0e5c09 | 4933 | } |
d6025579 | 4934 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4935 | } |
4936 | ||
323ac95b CM |
4937 | /* |
4938 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4939 | * path->nodes[1]. |
323ac95b CM |
4940 | * |
4941 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4942 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4943 | * | |
4944 | * The path must have already been setup for deleting the leaf, including | |
4945 | * all the proper balancing. path->nodes[1] must be locked. | |
4946 | */ | |
143bede5 JM |
4947 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4948 | struct btrfs_root *root, | |
4949 | struct btrfs_path *path, | |
4950 | struct extent_buffer *leaf) | |
323ac95b | 4951 | { |
5d4f98a2 | 4952 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4953 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4954 | |
4d081c41 CM |
4955 | /* |
4956 | * btrfs_free_extent is expensive, we want to make sure we | |
4957 | * aren't holding any locks when we call it | |
4958 | */ | |
4959 | btrfs_unlock_up_safe(path, 0); | |
4960 | ||
f0486c68 YZ |
4961 | root_sub_used(root, leaf->len); |
4962 | ||
3083ee2e | 4963 | extent_buffer_get(leaf); |
5581a51a | 4964 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4965 | free_extent_buffer_stale(leaf); |
323ac95b | 4966 | } |
74123bd7 CM |
4967 | /* |
4968 | * delete the item at the leaf level in path. If that empties | |
4969 | * the leaf, remove it from the tree | |
4970 | */ | |
85e21bac CM |
4971 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4972 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4973 | { |
0b246afa | 4974 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4975 | struct extent_buffer *leaf; |
4976 | struct btrfs_item *item; | |
ce0eac2a AM |
4977 | u32 last_off; |
4978 | u32 dsize = 0; | |
aa5d6bed CM |
4979 | int ret = 0; |
4980 | int wret; | |
85e21bac | 4981 | int i; |
7518a238 | 4982 | u32 nritems; |
cfed81a0 CM |
4983 | struct btrfs_map_token token; |
4984 | ||
4985 | btrfs_init_map_token(&token); | |
be0e5c09 | 4986 | |
5f39d397 | 4987 | leaf = path->nodes[0]; |
85e21bac CM |
4988 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4989 | ||
4990 | for (i = 0; i < nr; i++) | |
4991 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4992 | ||
5f39d397 | 4993 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4994 | |
85e21bac | 4995 | if (slot + nr != nritems) { |
2ff7e61e | 4996 | int data_end = leaf_data_end(fs_info, leaf); |
5f39d397 | 4997 | |
3d9ec8c4 | 4998 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4999 | data_end + dsize, |
3d9ec8c4 | 5000 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 5001 | last_off - data_end); |
5f39d397 | 5002 | |
85e21bac | 5003 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 5004 | u32 ioff; |
db94535d | 5005 | |
dd3cc16b | 5006 | item = btrfs_item_nr(i); |
cfed81a0 CM |
5007 | ioff = btrfs_token_item_offset(leaf, item, &token); |
5008 | btrfs_set_token_item_offset(leaf, item, | |
5009 | ioff + dsize, &token); | |
0783fcfc | 5010 | } |
db94535d | 5011 | |
5f39d397 | 5012 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5013 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5014 | sizeof(struct btrfs_item) * |
85e21bac | 5015 | (nritems - slot - nr)); |
be0e5c09 | 5016 | } |
85e21bac CM |
5017 | btrfs_set_header_nritems(leaf, nritems - nr); |
5018 | nritems -= nr; | |
5f39d397 | 5019 | |
74123bd7 | 5020 | /* delete the leaf if we've emptied it */ |
7518a238 | 5021 | if (nritems == 0) { |
5f39d397 CM |
5022 | if (leaf == root->node) { |
5023 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5024 | } else { |
f0486c68 | 5025 | btrfs_set_path_blocking(path); |
7c302b49 | 5026 | clean_tree_block(fs_info, leaf); |
143bede5 | 5027 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5028 | } |
be0e5c09 | 5029 | } else { |
7518a238 | 5030 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5031 | if (slot == 0) { |
5f39d397 CM |
5032 | struct btrfs_disk_key disk_key; |
5033 | ||
5034 | btrfs_item_key(leaf, &disk_key, 0); | |
0b246afa | 5035 | fixup_low_keys(fs_info, path, &disk_key, 1); |
aa5d6bed | 5036 | } |
aa5d6bed | 5037 | |
74123bd7 | 5038 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5039 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5040 | /* push_leaf_left fixes the path. |
5041 | * make sure the path still points to our leaf | |
5042 | * for possible call to del_ptr below | |
5043 | */ | |
4920c9ac | 5044 | slot = path->slots[1]; |
5f39d397 CM |
5045 | extent_buffer_get(leaf); |
5046 | ||
b9473439 | 5047 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5048 | wret = push_leaf_left(trans, root, path, 1, 1, |
5049 | 1, (u32)-1); | |
54aa1f4d | 5050 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5051 | ret = wret; |
5f39d397 CM |
5052 | |
5053 | if (path->nodes[0] == leaf && | |
5054 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5055 | wret = push_leaf_right(trans, root, path, 1, |
5056 | 1, 1, 0); | |
54aa1f4d | 5057 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5058 | ret = wret; |
5059 | } | |
5f39d397 CM |
5060 | |
5061 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5062 | path->slots[1] = slot; |
143bede5 | 5063 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5064 | free_extent_buffer(leaf); |
143bede5 | 5065 | ret = 0; |
5de08d7d | 5066 | } else { |
925baedd CM |
5067 | /* if we're still in the path, make sure |
5068 | * we're dirty. Otherwise, one of the | |
5069 | * push_leaf functions must have already | |
5070 | * dirtied this buffer | |
5071 | */ | |
5072 | if (path->nodes[0] == leaf) | |
5073 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5074 | free_extent_buffer(leaf); |
be0e5c09 | 5075 | } |
d5719762 | 5076 | } else { |
5f39d397 | 5077 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5078 | } |
5079 | } | |
aa5d6bed | 5080 | return ret; |
be0e5c09 CM |
5081 | } |
5082 | ||
7bb86316 | 5083 | /* |
925baedd | 5084 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5085 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5086 | * returns < 0 on io errors. | |
d352ac68 CM |
5087 | * |
5088 | * This may release the path, and so you may lose any locks held at the | |
5089 | * time you call it. | |
7bb86316 | 5090 | */ |
16e7549f | 5091 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5092 | { |
925baedd CM |
5093 | struct btrfs_key key; |
5094 | struct btrfs_disk_key found_key; | |
5095 | int ret; | |
7bb86316 | 5096 | |
925baedd | 5097 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5098 | |
e8b0d724 | 5099 | if (key.offset > 0) { |
925baedd | 5100 | key.offset--; |
e8b0d724 | 5101 | } else if (key.type > 0) { |
925baedd | 5102 | key.type--; |
e8b0d724 FDBM |
5103 | key.offset = (u64)-1; |
5104 | } else if (key.objectid > 0) { | |
925baedd | 5105 | key.objectid--; |
e8b0d724 FDBM |
5106 | key.type = (u8)-1; |
5107 | key.offset = (u64)-1; | |
5108 | } else { | |
925baedd | 5109 | return 1; |
e8b0d724 | 5110 | } |
7bb86316 | 5111 | |
b3b4aa74 | 5112 | btrfs_release_path(path); |
925baedd CM |
5113 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5114 | if (ret < 0) | |
5115 | return ret; | |
5116 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5117 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5118 | /* |
5119 | * We might have had an item with the previous key in the tree right | |
5120 | * before we released our path. And after we released our path, that | |
5121 | * item might have been pushed to the first slot (0) of the leaf we | |
5122 | * were holding due to a tree balance. Alternatively, an item with the | |
5123 | * previous key can exist as the only element of a leaf (big fat item). | |
5124 | * Therefore account for these 2 cases, so that our callers (like | |
5125 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5126 | * the previous key we computed above. | |
5127 | */ | |
5128 | if (ret <= 0) | |
925baedd CM |
5129 | return 0; |
5130 | return 1; | |
7bb86316 CM |
5131 | } |
5132 | ||
3f157a2f CM |
5133 | /* |
5134 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5135 | * for nodes or leaves that are have a minimum transaction id. |
5136 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5137 | * |
5138 | * This does not cow, but it does stuff the starting key it finds back | |
5139 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5140 | * key and get a writable path. | |
5141 | * | |
3f157a2f CM |
5142 | * This honors path->lowest_level to prevent descent past a given level |
5143 | * of the tree. | |
5144 | * | |
d352ac68 CM |
5145 | * min_trans indicates the oldest transaction that you are interested |
5146 | * in walking through. Any nodes or leaves older than min_trans are | |
5147 | * skipped over (without reading them). | |
5148 | * | |
3f157a2f CM |
5149 | * returns zero if something useful was found, < 0 on error and 1 if there |
5150 | * was nothing in the tree that matched the search criteria. | |
5151 | */ | |
5152 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5153 | struct btrfs_path *path, |
3f157a2f CM |
5154 | u64 min_trans) |
5155 | { | |
2ff7e61e | 5156 | struct btrfs_fs_info *fs_info = root->fs_info; |
3f157a2f CM |
5157 | struct extent_buffer *cur; |
5158 | struct btrfs_key found_key; | |
5159 | int slot; | |
9652480b | 5160 | int sret; |
3f157a2f CM |
5161 | u32 nritems; |
5162 | int level; | |
5163 | int ret = 1; | |
f98de9b9 | 5164 | int keep_locks = path->keep_locks; |
3f157a2f | 5165 | |
f98de9b9 | 5166 | path->keep_locks = 1; |
3f157a2f | 5167 | again: |
bd681513 | 5168 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5169 | level = btrfs_header_level(cur); |
e02119d5 | 5170 | WARN_ON(path->nodes[level]); |
3f157a2f | 5171 | path->nodes[level] = cur; |
bd681513 | 5172 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5173 | |
5174 | if (btrfs_header_generation(cur) < min_trans) { | |
5175 | ret = 1; | |
5176 | goto out; | |
5177 | } | |
d397712b | 5178 | while (1) { |
3f157a2f CM |
5179 | nritems = btrfs_header_nritems(cur); |
5180 | level = btrfs_header_level(cur); | |
a74b35ec | 5181 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
3f157a2f | 5182 | |
323ac95b CM |
5183 | /* at the lowest level, we're done, setup the path and exit */ |
5184 | if (level == path->lowest_level) { | |
e02119d5 CM |
5185 | if (slot >= nritems) |
5186 | goto find_next_key; | |
3f157a2f CM |
5187 | ret = 0; |
5188 | path->slots[level] = slot; | |
5189 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5190 | goto out; | |
5191 | } | |
9652480b Y |
5192 | if (sret && slot > 0) |
5193 | slot--; | |
3f157a2f | 5194 | /* |
de78b51a ES |
5195 | * check this node pointer against the min_trans parameters. |
5196 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5197 | */ |
d397712b | 5198 | while (slot < nritems) { |
3f157a2f | 5199 | u64 gen; |
e02119d5 | 5200 | |
3f157a2f CM |
5201 | gen = btrfs_node_ptr_generation(cur, slot); |
5202 | if (gen < min_trans) { | |
5203 | slot++; | |
5204 | continue; | |
5205 | } | |
de78b51a | 5206 | break; |
3f157a2f | 5207 | } |
e02119d5 | 5208 | find_next_key: |
3f157a2f CM |
5209 | /* |
5210 | * we didn't find a candidate key in this node, walk forward | |
5211 | * and find another one | |
5212 | */ | |
5213 | if (slot >= nritems) { | |
e02119d5 | 5214 | path->slots[level] = slot; |
b4ce94de | 5215 | btrfs_set_path_blocking(path); |
e02119d5 | 5216 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5217 | min_trans); |
e02119d5 | 5218 | if (sret == 0) { |
b3b4aa74 | 5219 | btrfs_release_path(path); |
3f157a2f CM |
5220 | goto again; |
5221 | } else { | |
5222 | goto out; | |
5223 | } | |
5224 | } | |
5225 | /* save our key for returning back */ | |
5226 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5227 | path->slots[level] = slot; | |
5228 | if (level == path->lowest_level) { | |
5229 | ret = 0; | |
3f157a2f CM |
5230 | goto out; |
5231 | } | |
b4ce94de | 5232 | btrfs_set_path_blocking(path); |
2ff7e61e | 5233 | cur = read_node_slot(fs_info, cur, slot); |
fb770ae4 LB |
5234 | if (IS_ERR(cur)) { |
5235 | ret = PTR_ERR(cur); | |
5236 | goto out; | |
5237 | } | |
3f157a2f | 5238 | |
bd681513 | 5239 | btrfs_tree_read_lock(cur); |
b4ce94de | 5240 | |
bd681513 | 5241 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5242 | path->nodes[level - 1] = cur; |
f7c79f30 | 5243 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5244 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5245 | } |
5246 | out: | |
f98de9b9 FM |
5247 | path->keep_locks = keep_locks; |
5248 | if (ret == 0) { | |
5249 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5250 | btrfs_set_path_blocking(path); | |
3f157a2f | 5251 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5252 | } |
3f157a2f CM |
5253 | return ret; |
5254 | } | |
5255 | ||
2ff7e61e | 5256 | static int tree_move_down(struct btrfs_fs_info *fs_info, |
7069830a | 5257 | struct btrfs_path *path, |
ab6a43e1 | 5258 | int *level) |
7069830a | 5259 | { |
fb770ae4 LB |
5260 | struct extent_buffer *eb; |
5261 | ||
74dd17fb | 5262 | BUG_ON(*level == 0); |
2ff7e61e | 5263 | eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]); |
fb770ae4 LB |
5264 | if (IS_ERR(eb)) |
5265 | return PTR_ERR(eb); | |
5266 | ||
5267 | path->nodes[*level - 1] = eb; | |
7069830a AB |
5268 | path->slots[*level - 1] = 0; |
5269 | (*level)--; | |
fb770ae4 | 5270 | return 0; |
7069830a AB |
5271 | } |
5272 | ||
f1e30261 | 5273 | static int tree_move_next_or_upnext(struct btrfs_path *path, |
7069830a AB |
5274 | int *level, int root_level) |
5275 | { | |
5276 | int ret = 0; | |
5277 | int nritems; | |
5278 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5279 | ||
5280 | path->slots[*level]++; | |
5281 | ||
74dd17fb | 5282 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5283 | if (*level == root_level) |
5284 | return -1; | |
5285 | ||
5286 | /* move upnext */ | |
5287 | path->slots[*level] = 0; | |
5288 | free_extent_buffer(path->nodes[*level]); | |
5289 | path->nodes[*level] = NULL; | |
5290 | (*level)++; | |
5291 | path->slots[*level]++; | |
5292 | ||
5293 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5294 | ret = 1; | |
5295 | } | |
5296 | return ret; | |
5297 | } | |
5298 | ||
5299 | /* | |
5300 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5301 | * or down. | |
5302 | */ | |
2ff7e61e | 5303 | static int tree_advance(struct btrfs_fs_info *fs_info, |
7069830a AB |
5304 | struct btrfs_path *path, |
5305 | int *level, int root_level, | |
5306 | int allow_down, | |
5307 | struct btrfs_key *key) | |
5308 | { | |
5309 | int ret; | |
5310 | ||
5311 | if (*level == 0 || !allow_down) { | |
f1e30261 | 5312 | ret = tree_move_next_or_upnext(path, level, root_level); |
7069830a | 5313 | } else { |
ab6a43e1 | 5314 | ret = tree_move_down(fs_info, path, level); |
7069830a AB |
5315 | } |
5316 | if (ret >= 0) { | |
5317 | if (*level == 0) | |
5318 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5319 | path->slots[*level]); | |
5320 | else | |
5321 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5322 | path->slots[*level]); | |
5323 | } | |
5324 | return ret; | |
5325 | } | |
5326 | ||
2ff7e61e | 5327 | static int tree_compare_item(struct btrfs_path *left_path, |
7069830a AB |
5328 | struct btrfs_path *right_path, |
5329 | char *tmp_buf) | |
5330 | { | |
5331 | int cmp; | |
5332 | int len1, len2; | |
5333 | unsigned long off1, off2; | |
5334 | ||
5335 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5336 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5337 | if (len1 != len2) | |
5338 | return 1; | |
5339 | ||
5340 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5341 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5342 | right_path->slots[0]); | |
5343 | ||
5344 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5345 | ||
5346 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5347 | if (cmp) | |
5348 | return 1; | |
5349 | return 0; | |
5350 | } | |
5351 | ||
5352 | #define ADVANCE 1 | |
5353 | #define ADVANCE_ONLY_NEXT -1 | |
5354 | ||
5355 | /* | |
5356 | * This function compares two trees and calls the provided callback for | |
5357 | * every changed/new/deleted item it finds. | |
5358 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5359 | * the compare pretty fast on snapshotted subvolumes. | |
5360 | * | |
5361 | * This currently works on commit roots only. As commit roots are read only, | |
5362 | * we don't do any locking. The commit roots are protected with transactions. | |
5363 | * Transactions are ended and rejoined when a commit is tried in between. | |
5364 | * | |
5365 | * This function checks for modifications done to the trees while comparing. | |
5366 | * If it detects a change, it aborts immediately. | |
5367 | */ | |
5368 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5369 | struct btrfs_root *right_root, | |
5370 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5371 | { | |
0b246afa | 5372 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5373 | int ret; |
5374 | int cmp; | |
7069830a AB |
5375 | struct btrfs_path *left_path = NULL; |
5376 | struct btrfs_path *right_path = NULL; | |
5377 | struct btrfs_key left_key; | |
5378 | struct btrfs_key right_key; | |
5379 | char *tmp_buf = NULL; | |
5380 | int left_root_level; | |
5381 | int right_root_level; | |
5382 | int left_level; | |
5383 | int right_level; | |
5384 | int left_end_reached; | |
5385 | int right_end_reached; | |
5386 | int advance_left; | |
5387 | int advance_right; | |
5388 | u64 left_blockptr; | |
5389 | u64 right_blockptr; | |
6baa4293 FM |
5390 | u64 left_gen; |
5391 | u64 right_gen; | |
7069830a AB |
5392 | |
5393 | left_path = btrfs_alloc_path(); | |
5394 | if (!left_path) { | |
5395 | ret = -ENOMEM; | |
5396 | goto out; | |
5397 | } | |
5398 | right_path = btrfs_alloc_path(); | |
5399 | if (!right_path) { | |
5400 | ret = -ENOMEM; | |
5401 | goto out; | |
5402 | } | |
5403 | ||
752ade68 | 5404 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); |
7069830a | 5405 | if (!tmp_buf) { |
752ade68 MH |
5406 | ret = -ENOMEM; |
5407 | goto out; | |
7069830a AB |
5408 | } |
5409 | ||
5410 | left_path->search_commit_root = 1; | |
5411 | left_path->skip_locking = 1; | |
5412 | right_path->search_commit_root = 1; | |
5413 | right_path->skip_locking = 1; | |
5414 | ||
7069830a AB |
5415 | /* |
5416 | * Strategy: Go to the first items of both trees. Then do | |
5417 | * | |
5418 | * If both trees are at level 0 | |
5419 | * Compare keys of current items | |
5420 | * If left < right treat left item as new, advance left tree | |
5421 | * and repeat | |
5422 | * If left > right treat right item as deleted, advance right tree | |
5423 | * and repeat | |
5424 | * If left == right do deep compare of items, treat as changed if | |
5425 | * needed, advance both trees and repeat | |
5426 | * If both trees are at the same level but not at level 0 | |
5427 | * Compare keys of current nodes/leafs | |
5428 | * If left < right advance left tree and repeat | |
5429 | * If left > right advance right tree and repeat | |
5430 | * If left == right compare blockptrs of the next nodes/leafs | |
5431 | * If they match advance both trees but stay at the same level | |
5432 | * and repeat | |
5433 | * If they don't match advance both trees while allowing to go | |
5434 | * deeper and repeat | |
5435 | * If tree levels are different | |
5436 | * Advance the tree that needs it and repeat | |
5437 | * | |
5438 | * Advancing a tree means: | |
5439 | * If we are at level 0, try to go to the next slot. If that's not | |
5440 | * possible, go one level up and repeat. Stop when we found a level | |
5441 | * where we could go to the next slot. We may at this point be on a | |
5442 | * node or a leaf. | |
5443 | * | |
5444 | * If we are not at level 0 and not on shared tree blocks, go one | |
5445 | * level deeper. | |
5446 | * | |
5447 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5448 | * the right if possible or go up and right. | |
5449 | */ | |
5450 | ||
0b246afa | 5451 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5452 | left_level = btrfs_header_level(left_root->commit_root); |
5453 | left_root_level = left_level; | |
5454 | left_path->nodes[left_level] = left_root->commit_root; | |
5455 | extent_buffer_get(left_path->nodes[left_level]); | |
5456 | ||
5457 | right_level = btrfs_header_level(right_root->commit_root); | |
5458 | right_root_level = right_level; | |
5459 | right_path->nodes[right_level] = right_root->commit_root; | |
5460 | extent_buffer_get(right_path->nodes[right_level]); | |
0b246afa | 5461 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5462 | |
5463 | if (left_level == 0) | |
5464 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5465 | &left_key, left_path->slots[left_level]); | |
5466 | else | |
5467 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5468 | &left_key, left_path->slots[left_level]); | |
5469 | if (right_level == 0) | |
5470 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5471 | &right_key, right_path->slots[right_level]); | |
5472 | else | |
5473 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5474 | &right_key, right_path->slots[right_level]); | |
5475 | ||
5476 | left_end_reached = right_end_reached = 0; | |
5477 | advance_left = advance_right = 0; | |
5478 | ||
5479 | while (1) { | |
7069830a | 5480 | if (advance_left && !left_end_reached) { |
2ff7e61e | 5481 | ret = tree_advance(fs_info, left_path, &left_level, |
7069830a AB |
5482 | left_root_level, |
5483 | advance_left != ADVANCE_ONLY_NEXT, | |
5484 | &left_key); | |
fb770ae4 | 5485 | if (ret == -1) |
7069830a | 5486 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5487 | else if (ret < 0) |
5488 | goto out; | |
7069830a AB |
5489 | advance_left = 0; |
5490 | } | |
5491 | if (advance_right && !right_end_reached) { | |
2ff7e61e | 5492 | ret = tree_advance(fs_info, right_path, &right_level, |
7069830a AB |
5493 | right_root_level, |
5494 | advance_right != ADVANCE_ONLY_NEXT, | |
5495 | &right_key); | |
fb770ae4 | 5496 | if (ret == -1) |
7069830a | 5497 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5498 | else if (ret < 0) |
5499 | goto out; | |
7069830a AB |
5500 | advance_right = 0; |
5501 | } | |
5502 | ||
5503 | if (left_end_reached && right_end_reached) { | |
5504 | ret = 0; | |
5505 | goto out; | |
5506 | } else if (left_end_reached) { | |
5507 | if (right_level == 0) { | |
ee8c494f | 5508 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5509 | &right_key, |
5510 | BTRFS_COMPARE_TREE_DELETED, | |
5511 | ctx); | |
5512 | if (ret < 0) | |
5513 | goto out; | |
5514 | } | |
5515 | advance_right = ADVANCE; | |
5516 | continue; | |
5517 | } else if (right_end_reached) { | |
5518 | if (left_level == 0) { | |
ee8c494f | 5519 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5520 | &left_key, |
5521 | BTRFS_COMPARE_TREE_NEW, | |
5522 | ctx); | |
5523 | if (ret < 0) | |
5524 | goto out; | |
5525 | } | |
5526 | advance_left = ADVANCE; | |
5527 | continue; | |
5528 | } | |
5529 | ||
5530 | if (left_level == 0 && right_level == 0) { | |
5531 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5532 | if (cmp < 0) { | |
ee8c494f | 5533 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5534 | &left_key, |
5535 | BTRFS_COMPARE_TREE_NEW, | |
5536 | ctx); | |
5537 | if (ret < 0) | |
5538 | goto out; | |
5539 | advance_left = ADVANCE; | |
5540 | } else if (cmp > 0) { | |
ee8c494f | 5541 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5542 | &right_key, |
5543 | BTRFS_COMPARE_TREE_DELETED, | |
5544 | ctx); | |
5545 | if (ret < 0) | |
5546 | goto out; | |
5547 | advance_right = ADVANCE; | |
5548 | } else { | |
b99d9a6a | 5549 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5550 | |
74dd17fb | 5551 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
2ff7e61e JM |
5552 | ret = tree_compare_item(left_path, right_path, |
5553 | tmp_buf); | |
ba5e8f2e | 5554 | if (ret) |
b99d9a6a | 5555 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5556 | else |
b99d9a6a | 5557 | result = BTRFS_COMPARE_TREE_SAME; |
ee8c494f | 5558 | ret = changed_cb(left_path, right_path, |
b99d9a6a | 5559 | &left_key, result, ctx); |
ba5e8f2e JB |
5560 | if (ret < 0) |
5561 | goto out; | |
7069830a AB |
5562 | advance_left = ADVANCE; |
5563 | advance_right = ADVANCE; | |
5564 | } | |
5565 | } else if (left_level == right_level) { | |
5566 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5567 | if (cmp < 0) { | |
5568 | advance_left = ADVANCE; | |
5569 | } else if (cmp > 0) { | |
5570 | advance_right = ADVANCE; | |
5571 | } else { | |
5572 | left_blockptr = btrfs_node_blockptr( | |
5573 | left_path->nodes[left_level], | |
5574 | left_path->slots[left_level]); | |
5575 | right_blockptr = btrfs_node_blockptr( | |
5576 | right_path->nodes[right_level], | |
5577 | right_path->slots[right_level]); | |
6baa4293 FM |
5578 | left_gen = btrfs_node_ptr_generation( |
5579 | left_path->nodes[left_level], | |
5580 | left_path->slots[left_level]); | |
5581 | right_gen = btrfs_node_ptr_generation( | |
5582 | right_path->nodes[right_level], | |
5583 | right_path->slots[right_level]); | |
5584 | if (left_blockptr == right_blockptr && | |
5585 | left_gen == right_gen) { | |
7069830a AB |
5586 | /* |
5587 | * As we're on a shared block, don't | |
5588 | * allow to go deeper. | |
5589 | */ | |
5590 | advance_left = ADVANCE_ONLY_NEXT; | |
5591 | advance_right = ADVANCE_ONLY_NEXT; | |
5592 | } else { | |
5593 | advance_left = ADVANCE; | |
5594 | advance_right = ADVANCE; | |
5595 | } | |
5596 | } | |
5597 | } else if (left_level < right_level) { | |
5598 | advance_right = ADVANCE; | |
5599 | } else { | |
5600 | advance_left = ADVANCE; | |
5601 | } | |
5602 | } | |
5603 | ||
5604 | out: | |
5605 | btrfs_free_path(left_path); | |
5606 | btrfs_free_path(right_path); | |
8f282f71 | 5607 | kvfree(tmp_buf); |
7069830a AB |
5608 | return ret; |
5609 | } | |
5610 | ||
3f157a2f CM |
5611 | /* |
5612 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5613 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5614 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5615 | * |
5616 | * 0 is returned if another key is found, < 0 if there are any errors | |
5617 | * and 1 is returned if there are no higher keys in the tree | |
5618 | * | |
5619 | * path->keep_locks should be set to 1 on the search made before | |
5620 | * calling this function. | |
5621 | */ | |
e7a84565 | 5622 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5623 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5624 | { |
e7a84565 CM |
5625 | int slot; |
5626 | struct extent_buffer *c; | |
5627 | ||
934d375b | 5628 | WARN_ON(!path->keep_locks); |
d397712b | 5629 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5630 | if (!path->nodes[level]) |
5631 | return 1; | |
5632 | ||
5633 | slot = path->slots[level] + 1; | |
5634 | c = path->nodes[level]; | |
3f157a2f | 5635 | next: |
e7a84565 | 5636 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5637 | int ret; |
5638 | int orig_lowest; | |
5639 | struct btrfs_key cur_key; | |
5640 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5641 | !path->nodes[level + 1]) | |
e7a84565 | 5642 | return 1; |
33c66f43 YZ |
5643 | |
5644 | if (path->locks[level + 1]) { | |
5645 | level++; | |
5646 | continue; | |
5647 | } | |
5648 | ||
5649 | slot = btrfs_header_nritems(c) - 1; | |
5650 | if (level == 0) | |
5651 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5652 | else | |
5653 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5654 | ||
5655 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5656 | btrfs_release_path(path); |
33c66f43 YZ |
5657 | path->lowest_level = level; |
5658 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5659 | 0, 0); | |
5660 | path->lowest_level = orig_lowest; | |
5661 | if (ret < 0) | |
5662 | return ret; | |
5663 | ||
5664 | c = path->nodes[level]; | |
5665 | slot = path->slots[level]; | |
5666 | if (ret == 0) | |
5667 | slot++; | |
5668 | goto next; | |
e7a84565 | 5669 | } |
33c66f43 | 5670 | |
e7a84565 CM |
5671 | if (level == 0) |
5672 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5673 | else { |
3f157a2f CM |
5674 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5675 | ||
3f157a2f CM |
5676 | if (gen < min_trans) { |
5677 | slot++; | |
5678 | goto next; | |
5679 | } | |
e7a84565 | 5680 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5681 | } |
e7a84565 CM |
5682 | return 0; |
5683 | } | |
5684 | return 1; | |
5685 | } | |
5686 | ||
97571fd0 | 5687 | /* |
925baedd | 5688 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5689 | * returns 0 if it found something or 1 if there are no greater leaves. |
5690 | * returns < 0 on io errors. | |
97571fd0 | 5691 | */ |
234b63a0 | 5692 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5693 | { |
5694 | return btrfs_next_old_leaf(root, path, 0); | |
5695 | } | |
5696 | ||
5697 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5698 | u64 time_seq) | |
d97e63b6 CM |
5699 | { |
5700 | int slot; | |
8e73f275 | 5701 | int level; |
5f39d397 | 5702 | struct extent_buffer *c; |
8e73f275 | 5703 | struct extent_buffer *next; |
925baedd CM |
5704 | struct btrfs_key key; |
5705 | u32 nritems; | |
5706 | int ret; | |
8e73f275 | 5707 | int old_spinning = path->leave_spinning; |
bd681513 | 5708 | int next_rw_lock = 0; |
925baedd CM |
5709 | |
5710 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5711 | if (nritems == 0) |
925baedd | 5712 | return 1; |
925baedd | 5713 | |
8e73f275 CM |
5714 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5715 | again: | |
5716 | level = 1; | |
5717 | next = NULL; | |
bd681513 | 5718 | next_rw_lock = 0; |
b3b4aa74 | 5719 | btrfs_release_path(path); |
8e73f275 | 5720 | |
a2135011 | 5721 | path->keep_locks = 1; |
31533fb2 | 5722 | path->leave_spinning = 1; |
8e73f275 | 5723 | |
3d7806ec JS |
5724 | if (time_seq) |
5725 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5726 | else | |
5727 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5728 | path->keep_locks = 0; |
5729 | ||
5730 | if (ret < 0) | |
5731 | return ret; | |
5732 | ||
a2135011 | 5733 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5734 | /* |
5735 | * by releasing the path above we dropped all our locks. A balance | |
5736 | * could have added more items next to the key that used to be | |
5737 | * at the very end of the block. So, check again here and | |
5738 | * advance the path if there are now more items available. | |
5739 | */ | |
a2135011 | 5740 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5741 | if (ret == 0) |
5742 | path->slots[0]++; | |
8e73f275 | 5743 | ret = 0; |
925baedd CM |
5744 | goto done; |
5745 | } | |
0b43e04f LB |
5746 | /* |
5747 | * So the above check misses one case: | |
5748 | * - after releasing the path above, someone has removed the item that | |
5749 | * used to be at the very end of the block, and balance between leafs | |
5750 | * gets another one with bigger key.offset to replace it. | |
5751 | * | |
5752 | * This one should be returned as well, or we can get leaf corruption | |
5753 | * later(esp. in __btrfs_drop_extents()). | |
5754 | * | |
5755 | * And a bit more explanation about this check, | |
5756 | * with ret > 0, the key isn't found, the path points to the slot | |
5757 | * where it should be inserted, so the path->slots[0] item must be the | |
5758 | * bigger one. | |
5759 | */ | |
5760 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5761 | ret = 0; | |
5762 | goto done; | |
5763 | } | |
d97e63b6 | 5764 | |
d397712b | 5765 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5766 | if (!path->nodes[level]) { |
5767 | ret = 1; | |
5768 | goto done; | |
5769 | } | |
5f39d397 | 5770 | |
d97e63b6 CM |
5771 | slot = path->slots[level] + 1; |
5772 | c = path->nodes[level]; | |
5f39d397 | 5773 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5774 | level++; |
8e73f275 CM |
5775 | if (level == BTRFS_MAX_LEVEL) { |
5776 | ret = 1; | |
5777 | goto done; | |
5778 | } | |
d97e63b6 CM |
5779 | continue; |
5780 | } | |
5f39d397 | 5781 | |
925baedd | 5782 | if (next) { |
bd681513 | 5783 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5784 | free_extent_buffer(next); |
925baedd | 5785 | } |
5f39d397 | 5786 | |
8e73f275 | 5787 | next = c; |
bd681513 | 5788 | next_rw_lock = path->locks[level]; |
d07b8528 | 5789 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5790 | slot, &key); |
8e73f275 CM |
5791 | if (ret == -EAGAIN) |
5792 | goto again; | |
5f39d397 | 5793 | |
76a05b35 | 5794 | if (ret < 0) { |
b3b4aa74 | 5795 | btrfs_release_path(path); |
76a05b35 CM |
5796 | goto done; |
5797 | } | |
5798 | ||
5cd57b2c | 5799 | if (!path->skip_locking) { |
bd681513 | 5800 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5801 | if (!ret && time_seq) { |
5802 | /* | |
5803 | * If we don't get the lock, we may be racing | |
5804 | * with push_leaf_left, holding that lock while | |
5805 | * itself waiting for the leaf we've currently | |
5806 | * locked. To solve this situation, we give up | |
5807 | * on our lock and cycle. | |
5808 | */ | |
cf538830 | 5809 | free_extent_buffer(next); |
d42244a0 JS |
5810 | btrfs_release_path(path); |
5811 | cond_resched(); | |
5812 | goto again; | |
5813 | } | |
8e73f275 CM |
5814 | if (!ret) { |
5815 | btrfs_set_path_blocking(path); | |
bd681513 | 5816 | btrfs_tree_read_lock(next); |
31533fb2 | 5817 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5818 | BTRFS_READ_LOCK); |
8e73f275 | 5819 | } |
31533fb2 | 5820 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5821 | } |
d97e63b6 CM |
5822 | break; |
5823 | } | |
5824 | path->slots[level] = slot; | |
d397712b | 5825 | while (1) { |
d97e63b6 CM |
5826 | level--; |
5827 | c = path->nodes[level]; | |
925baedd | 5828 | if (path->locks[level]) |
bd681513 | 5829 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5830 | |
5f39d397 | 5831 | free_extent_buffer(c); |
d97e63b6 CM |
5832 | path->nodes[level] = next; |
5833 | path->slots[level] = 0; | |
a74a4b97 | 5834 | if (!path->skip_locking) |
bd681513 | 5835 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5836 | if (!level) |
5837 | break; | |
b4ce94de | 5838 | |
d07b8528 | 5839 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5840 | 0, &key); |
8e73f275 CM |
5841 | if (ret == -EAGAIN) |
5842 | goto again; | |
5843 | ||
76a05b35 | 5844 | if (ret < 0) { |
b3b4aa74 | 5845 | btrfs_release_path(path); |
76a05b35 CM |
5846 | goto done; |
5847 | } | |
5848 | ||
5cd57b2c | 5849 | if (!path->skip_locking) { |
bd681513 | 5850 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5851 | if (!ret) { |
5852 | btrfs_set_path_blocking(path); | |
bd681513 | 5853 | btrfs_tree_read_lock(next); |
31533fb2 | 5854 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5855 | BTRFS_READ_LOCK); |
5856 | } | |
31533fb2 | 5857 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5858 | } |
d97e63b6 | 5859 | } |
8e73f275 | 5860 | ret = 0; |
925baedd | 5861 | done: |
f7c79f30 | 5862 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5863 | path->leave_spinning = old_spinning; |
5864 | if (!old_spinning) | |
5865 | btrfs_set_path_blocking(path); | |
5866 | ||
5867 | return ret; | |
d97e63b6 | 5868 | } |
0b86a832 | 5869 | |
3f157a2f CM |
5870 | /* |
5871 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5872 | * searching until it gets past min_objectid or finds an item of 'type' | |
5873 | * | |
5874 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5875 | */ | |
0b86a832 CM |
5876 | int btrfs_previous_item(struct btrfs_root *root, |
5877 | struct btrfs_path *path, u64 min_objectid, | |
5878 | int type) | |
5879 | { | |
5880 | struct btrfs_key found_key; | |
5881 | struct extent_buffer *leaf; | |
e02119d5 | 5882 | u32 nritems; |
0b86a832 CM |
5883 | int ret; |
5884 | ||
d397712b | 5885 | while (1) { |
0b86a832 | 5886 | if (path->slots[0] == 0) { |
b4ce94de | 5887 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5888 | ret = btrfs_prev_leaf(root, path); |
5889 | if (ret != 0) | |
5890 | return ret; | |
5891 | } else { | |
5892 | path->slots[0]--; | |
5893 | } | |
5894 | leaf = path->nodes[0]; | |
e02119d5 CM |
5895 | nritems = btrfs_header_nritems(leaf); |
5896 | if (nritems == 0) | |
5897 | return 1; | |
5898 | if (path->slots[0] == nritems) | |
5899 | path->slots[0]--; | |
5900 | ||
0b86a832 | 5901 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5902 | if (found_key.objectid < min_objectid) |
5903 | break; | |
0a4eefbb YZ |
5904 | if (found_key.type == type) |
5905 | return 0; | |
e02119d5 CM |
5906 | if (found_key.objectid == min_objectid && |
5907 | found_key.type < type) | |
5908 | break; | |
0b86a832 CM |
5909 | } |
5910 | return 1; | |
5911 | } | |
ade2e0b3 WS |
5912 | |
5913 | /* | |
5914 | * search in extent tree to find a previous Metadata/Data extent item with | |
5915 | * min objecitd. | |
5916 | * | |
5917 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5918 | */ | |
5919 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5920 | struct btrfs_path *path, u64 min_objectid) | |
5921 | { | |
5922 | struct btrfs_key found_key; | |
5923 | struct extent_buffer *leaf; | |
5924 | u32 nritems; | |
5925 | int ret; | |
5926 | ||
5927 | while (1) { | |
5928 | if (path->slots[0] == 0) { | |
5929 | btrfs_set_path_blocking(path); | |
5930 | ret = btrfs_prev_leaf(root, path); | |
5931 | if (ret != 0) | |
5932 | return ret; | |
5933 | } else { | |
5934 | path->slots[0]--; | |
5935 | } | |
5936 | leaf = path->nodes[0]; | |
5937 | nritems = btrfs_header_nritems(leaf); | |
5938 | if (nritems == 0) | |
5939 | return 1; | |
5940 | if (path->slots[0] == nritems) | |
5941 | path->slots[0]--; | |
5942 | ||
5943 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5944 | if (found_key.objectid < min_objectid) | |
5945 | break; | |
5946 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5947 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5948 | return 0; | |
5949 | if (found_key.objectid == min_objectid && | |
5950 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5951 | break; | |
5952 | } | |
5953 | return 1; | |
5954 | } |