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); | |
d97e63b6 | 44 | |
df24a2b9 | 45 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 46 | { |
e2c89907 | 47 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
48 | } |
49 | ||
b4ce94de CM |
50 | /* |
51 | * set all locked nodes in the path to blocking locks. This should | |
52 | * be done before scheduling | |
53 | */ | |
54 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
55 | { | |
56 | int i; | |
57 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
58 | if (!p->nodes[i] || !p->locks[i]) |
59 | continue; | |
60 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
61 | if (p->locks[i] == BTRFS_READ_LOCK) | |
62 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
63 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
64 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
65 | } |
66 | } | |
67 | ||
68 | /* | |
69 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
70 | * |
71 | * held is used to keep lockdep happy, when lockdep is enabled | |
72 | * we set held to a blocking lock before we go around and | |
73 | * retake all the spinlocks in the path. You can safely use NULL | |
74 | * for held | |
b4ce94de | 75 | */ |
4008c04a | 76 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 77 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
78 | { |
79 | int i; | |
4008c04a | 80 | |
bd681513 CM |
81 | if (held) { |
82 | btrfs_set_lock_blocking_rw(held, held_rw); | |
83 | if (held_rw == BTRFS_WRITE_LOCK) | |
84 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
85 | else if (held_rw == BTRFS_READ_LOCK) | |
86 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
87 | } | |
4008c04a | 88 | btrfs_set_path_blocking(p); |
4008c04a CM |
89 | |
90 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
91 | if (p->nodes[i] && p->locks[i]) { |
92 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
93 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
94 | p->locks[i] = BTRFS_WRITE_LOCK; | |
95 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
96 | p->locks[i] = BTRFS_READ_LOCK; | |
97 | } | |
b4ce94de | 98 | } |
4008c04a | 99 | |
4008c04a | 100 | if (held) |
bd681513 | 101 | btrfs_clear_lock_blocking_rw(held, held_rw); |
b4ce94de CM |
102 | } |
103 | ||
d352ac68 | 104 | /* this also releases the path */ |
df24a2b9 | 105 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 106 | { |
ff175d57 JJ |
107 | if (!p) |
108 | return; | |
b3b4aa74 | 109 | btrfs_release_path(p); |
df24a2b9 | 110 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
111 | } |
112 | ||
d352ac68 CM |
113 | /* |
114 | * path release drops references on the extent buffers in the path | |
115 | * and it drops any locks held by this path | |
116 | * | |
117 | * It is safe to call this on paths that no locks or extent buffers held. | |
118 | */ | |
b3b4aa74 | 119 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
120 | { |
121 | int i; | |
a2135011 | 122 | |
234b63a0 | 123 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 124 | p->slots[i] = 0; |
eb60ceac | 125 | if (!p->nodes[i]) |
925baedd CM |
126 | continue; |
127 | if (p->locks[i]) { | |
bd681513 | 128 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
129 | p->locks[i] = 0; |
130 | } | |
5f39d397 | 131 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 132 | p->nodes[i] = NULL; |
eb60ceac CM |
133 | } |
134 | } | |
135 | ||
d352ac68 CM |
136 | /* |
137 | * safely gets a reference on the root node of a tree. A lock | |
138 | * is not taken, so a concurrent writer may put a different node | |
139 | * at the root of the tree. See btrfs_lock_root_node for the | |
140 | * looping required. | |
141 | * | |
142 | * The extent buffer returned by this has a reference taken, so | |
143 | * it won't disappear. It may stop being the root of the tree | |
144 | * at any time because there are no locks held. | |
145 | */ | |
925baedd CM |
146 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
147 | { | |
148 | struct extent_buffer *eb; | |
240f62c8 | 149 | |
3083ee2e JB |
150 | while (1) { |
151 | rcu_read_lock(); | |
152 | eb = rcu_dereference(root->node); | |
153 | ||
154 | /* | |
155 | * RCU really hurts here, we could free up the root node because | |
01327610 | 156 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
157 | * the inc_not_zero dance and if it doesn't work then |
158 | * synchronize_rcu and try again. | |
159 | */ | |
160 | if (atomic_inc_not_zero(&eb->refs)) { | |
161 | rcu_read_unlock(); | |
162 | break; | |
163 | } | |
164 | rcu_read_unlock(); | |
165 | synchronize_rcu(); | |
166 | } | |
925baedd CM |
167 | return eb; |
168 | } | |
169 | ||
d352ac68 CM |
170 | /* loop around taking references on and locking the root node of the |
171 | * tree until you end up with a lock on the root. A locked buffer | |
172 | * is returned, with a reference held. | |
173 | */ | |
925baedd CM |
174 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
175 | { | |
176 | struct extent_buffer *eb; | |
177 | ||
d397712b | 178 | while (1) { |
925baedd CM |
179 | eb = btrfs_root_node(root); |
180 | btrfs_tree_lock(eb); | |
240f62c8 | 181 | if (eb == root->node) |
925baedd | 182 | break; |
925baedd CM |
183 | btrfs_tree_unlock(eb); |
184 | free_extent_buffer(eb); | |
185 | } | |
186 | return eb; | |
187 | } | |
188 | ||
bd681513 CM |
189 | /* loop around taking references on and locking the root node of the |
190 | * tree until you end up with a lock on the root. A locked buffer | |
191 | * is returned, with a reference held. | |
192 | */ | |
84f7d8e6 | 193 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
194 | { |
195 | struct extent_buffer *eb; | |
196 | ||
197 | while (1) { | |
198 | eb = btrfs_root_node(root); | |
199 | btrfs_tree_read_lock(eb); | |
200 | if (eb == root->node) | |
201 | break; | |
202 | btrfs_tree_read_unlock(eb); | |
203 | free_extent_buffer(eb); | |
204 | } | |
205 | return eb; | |
206 | } | |
207 | ||
d352ac68 CM |
208 | /* cowonly root (everything not a reference counted cow subvolume), just get |
209 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
210 | * get properly updated on disk. | |
211 | */ | |
0b86a832 CM |
212 | static void add_root_to_dirty_list(struct btrfs_root *root) |
213 | { | |
0b246afa JM |
214 | struct btrfs_fs_info *fs_info = root->fs_info; |
215 | ||
e7070be1 JB |
216 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
217 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
218 | return; | |
219 | ||
0b246afa | 220 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
221 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
222 | /* Want the extent tree to be the last on the list */ | |
223 | if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
224 | list_move_tail(&root->dirty_list, | |
0b246afa | 225 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
226 | else |
227 | list_move(&root->dirty_list, | |
0b246afa | 228 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 229 | } |
0b246afa | 230 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
231 | } |
232 | ||
d352ac68 CM |
233 | /* |
234 | * used by snapshot creation to make a copy of a root for a tree with | |
235 | * a given objectid. The buffer with the new root node is returned in | |
236 | * cow_ret, and this func returns zero on success or a negative error code. | |
237 | */ | |
be20aa9d CM |
238 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
239 | struct btrfs_root *root, | |
240 | struct extent_buffer *buf, | |
241 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
242 | { | |
0b246afa | 243 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 244 | struct extent_buffer *cow; |
be20aa9d CM |
245 | int ret = 0; |
246 | int level; | |
5d4f98a2 | 247 | struct btrfs_disk_key disk_key; |
be20aa9d | 248 | |
27cdeb70 | 249 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 250 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
251 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
252 | trans->transid != root->last_trans); | |
be20aa9d CM |
253 | |
254 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
255 | if (level == 0) |
256 | btrfs_item_key(buf, &disk_key, 0); | |
257 | else | |
258 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 259 | |
4d75f8a9 DS |
260 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
261 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 262 | if (IS_ERR(cow)) |
be20aa9d CM |
263 | return PTR_ERR(cow); |
264 | ||
58e8012c | 265 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
266 | btrfs_set_header_bytenr(cow, cow->start); |
267 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
268 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
269 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
270 | BTRFS_HEADER_FLAG_RELOC); | |
271 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
272 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
273 | else | |
274 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 275 | |
0b246afa | 276 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 277 | |
be20aa9d | 278 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 279 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 280 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 281 | else |
e339a6b0 | 282 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 283 | |
be20aa9d CM |
284 | if (ret) |
285 | return ret; | |
286 | ||
287 | btrfs_mark_buffer_dirty(cow); | |
288 | *cow_ret = cow; | |
289 | return 0; | |
290 | } | |
291 | ||
bd989ba3 JS |
292 | enum mod_log_op { |
293 | MOD_LOG_KEY_REPLACE, | |
294 | MOD_LOG_KEY_ADD, | |
295 | MOD_LOG_KEY_REMOVE, | |
296 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
297 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
298 | MOD_LOG_MOVE_KEYS, | |
299 | MOD_LOG_ROOT_REPLACE, | |
300 | }; | |
301 | ||
bd989ba3 JS |
302 | struct tree_mod_root { |
303 | u64 logical; | |
304 | u8 level; | |
305 | }; | |
306 | ||
307 | struct tree_mod_elem { | |
308 | struct rb_node node; | |
298cfd36 | 309 | u64 logical; |
097b8a7c | 310 | u64 seq; |
bd989ba3 JS |
311 | enum mod_log_op op; |
312 | ||
313 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
314 | int slot; | |
315 | ||
316 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
317 | u64 generation; | |
318 | ||
319 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
320 | struct btrfs_disk_key key; | |
321 | u64 blockptr; | |
322 | ||
323 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
324 | struct { |
325 | int dst_slot; | |
326 | int nr_items; | |
327 | } move; | |
bd989ba3 JS |
328 | |
329 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
330 | struct tree_mod_root old_root; | |
331 | }; | |
332 | ||
097b8a7c | 333 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 334 | { |
097b8a7c | 335 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
336 | } |
337 | ||
097b8a7c JS |
338 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
339 | { | |
340 | read_unlock(&fs_info->tree_mod_log_lock); | |
341 | } | |
342 | ||
343 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
344 | { | |
345 | write_lock(&fs_info->tree_mod_log_lock); | |
346 | } | |
347 | ||
348 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
349 | { | |
350 | write_unlock(&fs_info->tree_mod_log_lock); | |
351 | } | |
352 | ||
fc36ed7e | 353 | /* |
fcebe456 | 354 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 355 | */ |
fcebe456 | 356 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
357 | { |
358 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
359 | } | |
360 | ||
097b8a7c JS |
361 | /* |
362 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
363 | * passed does not already have a sequence number set. So when a caller expects | |
364 | * to record tree modifications, it should ensure to set elem->seq to zero | |
365 | * before calling btrfs_get_tree_mod_seq. | |
366 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
367 | * blocker was added. | |
368 | */ | |
369 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
370 | struct seq_list *elem) | |
bd989ba3 | 371 | { |
097b8a7c | 372 | tree_mod_log_write_lock(fs_info); |
bd989ba3 | 373 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 374 | if (!elem->seq) { |
fcebe456 | 375 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
376 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
377 | } | |
bd989ba3 | 378 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
379 | tree_mod_log_write_unlock(fs_info); |
380 | ||
fcebe456 | 381 | return elem->seq; |
bd989ba3 JS |
382 | } |
383 | ||
384 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
385 | struct seq_list *elem) | |
386 | { | |
387 | struct rb_root *tm_root; | |
388 | struct rb_node *node; | |
389 | struct rb_node *next; | |
390 | struct seq_list *cur_elem; | |
391 | struct tree_mod_elem *tm; | |
392 | u64 min_seq = (u64)-1; | |
393 | u64 seq_putting = elem->seq; | |
394 | ||
395 | if (!seq_putting) | |
396 | return; | |
397 | ||
bd989ba3 JS |
398 | spin_lock(&fs_info->tree_mod_seq_lock); |
399 | list_del(&elem->list); | |
097b8a7c | 400 | elem->seq = 0; |
bd989ba3 JS |
401 | |
402 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 403 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
404 | if (seq_putting > cur_elem->seq) { |
405 | /* | |
406 | * blocker with lower sequence number exists, we | |
407 | * cannot remove anything from the log | |
408 | */ | |
097b8a7c JS |
409 | spin_unlock(&fs_info->tree_mod_seq_lock); |
410 | return; | |
bd989ba3 JS |
411 | } |
412 | min_seq = cur_elem->seq; | |
413 | } | |
414 | } | |
097b8a7c JS |
415 | spin_unlock(&fs_info->tree_mod_seq_lock); |
416 | ||
bd989ba3 JS |
417 | /* |
418 | * anything that's lower than the lowest existing (read: blocked) | |
419 | * sequence number can be removed from the tree. | |
420 | */ | |
097b8a7c | 421 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
422 | tm_root = &fs_info->tree_mod_log; |
423 | for (node = rb_first(tm_root); node; node = next) { | |
424 | next = rb_next(node); | |
6b4df8b6 | 425 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 426 | if (tm->seq > min_seq) |
bd989ba3 JS |
427 | continue; |
428 | rb_erase(node, tm_root); | |
bd989ba3 JS |
429 | kfree(tm); |
430 | } | |
097b8a7c | 431 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
432 | } |
433 | ||
434 | /* | |
435 | * key order of the log: | |
298cfd36 | 436 | * node/leaf start address -> sequence |
bd989ba3 | 437 | * |
298cfd36 CR |
438 | * The 'start address' is the logical address of the *new* root node |
439 | * for root replace operations, or the logical address of the affected | |
440 | * block for all other operations. | |
5de865ee FDBM |
441 | * |
442 | * Note: must be called with write lock (tree_mod_log_write_lock). | |
bd989ba3 JS |
443 | */ |
444 | static noinline int | |
445 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
446 | { | |
447 | struct rb_root *tm_root; | |
448 | struct rb_node **new; | |
449 | struct rb_node *parent = NULL; | |
450 | struct tree_mod_elem *cur; | |
c8cc6341 | 451 | |
fcebe456 | 452 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 453 | |
bd989ba3 JS |
454 | tm_root = &fs_info->tree_mod_log; |
455 | new = &tm_root->rb_node; | |
456 | while (*new) { | |
6b4df8b6 | 457 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 458 | parent = *new; |
298cfd36 | 459 | if (cur->logical < tm->logical) |
bd989ba3 | 460 | new = &((*new)->rb_left); |
298cfd36 | 461 | else if (cur->logical > tm->logical) |
bd989ba3 | 462 | new = &((*new)->rb_right); |
097b8a7c | 463 | else if (cur->seq < tm->seq) |
bd989ba3 | 464 | new = &((*new)->rb_left); |
097b8a7c | 465 | else if (cur->seq > tm->seq) |
bd989ba3 | 466 | new = &((*new)->rb_right); |
5de865ee FDBM |
467 | else |
468 | return -EEXIST; | |
bd989ba3 JS |
469 | } |
470 | ||
471 | rb_link_node(&tm->node, parent, new); | |
472 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 473 | return 0; |
bd989ba3 JS |
474 | } |
475 | ||
097b8a7c JS |
476 | /* |
477 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
478 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
479 | * this until all tree mod log insertions are recorded in the rb tree and then | |
480 | * call tree_mod_log_write_unlock() to release. | |
481 | */ | |
e9b7fd4d JS |
482 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
483 | struct extent_buffer *eb) { | |
484 | smp_mb(); | |
485 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
486 | return 1; | |
097b8a7c JS |
487 | if (eb && btrfs_header_level(eb) == 0) |
488 | return 1; | |
5de865ee FDBM |
489 | |
490 | tree_mod_log_write_lock(fs_info); | |
491 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { | |
492 | tree_mod_log_write_unlock(fs_info); | |
493 | return 1; | |
494 | } | |
495 | ||
e9b7fd4d JS |
496 | return 0; |
497 | } | |
498 | ||
5de865ee FDBM |
499 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
500 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
501 | struct extent_buffer *eb) | |
502 | { | |
503 | smp_mb(); | |
504 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
505 | return 0; | |
506 | if (eb && btrfs_header_level(eb) == 0) | |
507 | return 0; | |
508 | ||
509 | return 1; | |
510 | } | |
511 | ||
512 | static struct tree_mod_elem * | |
513 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
514 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 515 | { |
097b8a7c | 516 | struct tree_mod_elem *tm; |
bd989ba3 | 517 | |
c8cc6341 JB |
518 | tm = kzalloc(sizeof(*tm), flags); |
519 | if (!tm) | |
5de865ee | 520 | return NULL; |
bd989ba3 | 521 | |
298cfd36 | 522 | tm->logical = eb->start; |
bd989ba3 JS |
523 | if (op != MOD_LOG_KEY_ADD) { |
524 | btrfs_node_key(eb, &tm->key, slot); | |
525 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
526 | } | |
527 | tm->op = op; | |
528 | tm->slot = slot; | |
529 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 530 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 531 | |
5de865ee | 532 | return tm; |
097b8a7c JS |
533 | } |
534 | ||
e09c2efe DS |
535 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
536 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 537 | { |
5de865ee FDBM |
538 | struct tree_mod_elem *tm; |
539 | int ret; | |
540 | ||
e09c2efe | 541 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
542 | return 0; |
543 | ||
544 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
545 | if (!tm) | |
546 | return -ENOMEM; | |
547 | ||
e09c2efe | 548 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 549 | kfree(tm); |
097b8a7c | 550 | return 0; |
5de865ee FDBM |
551 | } |
552 | ||
e09c2efe DS |
553 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
554 | tree_mod_log_write_unlock(eb->fs_info); | |
5de865ee FDBM |
555 | if (ret) |
556 | kfree(tm); | |
097b8a7c | 557 | |
5de865ee | 558 | return ret; |
097b8a7c JS |
559 | } |
560 | ||
6074d45f DS |
561 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
562 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 563 | { |
5de865ee FDBM |
564 | struct tree_mod_elem *tm = NULL; |
565 | struct tree_mod_elem **tm_list = NULL; | |
566 | int ret = 0; | |
bd989ba3 | 567 | int i; |
5de865ee | 568 | int locked = 0; |
bd989ba3 | 569 | |
6074d45f | 570 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 571 | return 0; |
bd989ba3 | 572 | |
176ef8f5 | 573 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
574 | if (!tm_list) |
575 | return -ENOMEM; | |
576 | ||
176ef8f5 | 577 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
578 | if (!tm) { |
579 | ret = -ENOMEM; | |
580 | goto free_tms; | |
581 | } | |
582 | ||
298cfd36 | 583 | tm->logical = eb->start; |
5de865ee FDBM |
584 | tm->slot = src_slot; |
585 | tm->move.dst_slot = dst_slot; | |
586 | tm->move.nr_items = nr_items; | |
587 | tm->op = MOD_LOG_MOVE_KEYS; | |
588 | ||
589 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
590 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 591 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
592 | if (!tm_list[i]) { |
593 | ret = -ENOMEM; | |
594 | goto free_tms; | |
595 | } | |
596 | } | |
597 | ||
6074d45f | 598 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
599 | goto free_tms; |
600 | locked = 1; | |
601 | ||
01763a2e JS |
602 | /* |
603 | * When we override something during the move, we log these removals. | |
604 | * This can only happen when we move towards the beginning of the | |
605 | * buffer, i.e. dst_slot < src_slot. | |
606 | */ | |
bd989ba3 | 607 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 608 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
609 | if (ret) |
610 | goto free_tms; | |
bd989ba3 JS |
611 | } |
612 | ||
6074d45f | 613 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
614 | if (ret) |
615 | goto free_tms; | |
6074d45f | 616 | tree_mod_log_write_unlock(eb->fs_info); |
5de865ee | 617 | kfree(tm_list); |
f395694c | 618 | |
5de865ee FDBM |
619 | return 0; |
620 | free_tms: | |
621 | for (i = 0; i < nr_items; i++) { | |
622 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 623 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
624 | kfree(tm_list[i]); |
625 | } | |
626 | if (locked) | |
6074d45f | 627 | tree_mod_log_write_unlock(eb->fs_info); |
5de865ee FDBM |
628 | kfree(tm_list); |
629 | kfree(tm); | |
bd989ba3 | 630 | |
5de865ee | 631 | return ret; |
bd989ba3 JS |
632 | } |
633 | ||
5de865ee FDBM |
634 | static inline int |
635 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
636 | struct tree_mod_elem **tm_list, | |
637 | int nritems) | |
097b8a7c | 638 | { |
5de865ee | 639 | int i, j; |
097b8a7c JS |
640 | int ret; |
641 | ||
097b8a7c | 642 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
643 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
644 | if (ret) { | |
645 | for (j = nritems - 1; j > i; j--) | |
646 | rb_erase(&tm_list[j]->node, | |
647 | &fs_info->tree_mod_log); | |
648 | return ret; | |
649 | } | |
097b8a7c | 650 | } |
5de865ee FDBM |
651 | |
652 | return 0; | |
097b8a7c JS |
653 | } |
654 | ||
95b757c1 DS |
655 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
656 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 657 | { |
95b757c1 | 658 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
659 | struct tree_mod_elem *tm = NULL; |
660 | struct tree_mod_elem **tm_list = NULL; | |
661 | int nritems = 0; | |
662 | int ret = 0; | |
663 | int i; | |
bd989ba3 | 664 | |
5de865ee | 665 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
666 | return 0; |
667 | ||
5de865ee FDBM |
668 | if (log_removal && btrfs_header_level(old_root) > 0) { |
669 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 670 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 671 | GFP_NOFS); |
5de865ee FDBM |
672 | if (!tm_list) { |
673 | ret = -ENOMEM; | |
674 | goto free_tms; | |
675 | } | |
676 | for (i = 0; i < nritems; i++) { | |
677 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 678 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
679 | if (!tm_list[i]) { |
680 | ret = -ENOMEM; | |
681 | goto free_tms; | |
682 | } | |
683 | } | |
684 | } | |
d9abbf1c | 685 | |
bcc8e07f | 686 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
687 | if (!tm) { |
688 | ret = -ENOMEM; | |
689 | goto free_tms; | |
690 | } | |
bd989ba3 | 691 | |
298cfd36 | 692 | tm->logical = new_root->start; |
bd989ba3 JS |
693 | tm->old_root.logical = old_root->start; |
694 | tm->old_root.level = btrfs_header_level(old_root); | |
695 | tm->generation = btrfs_header_generation(old_root); | |
696 | tm->op = MOD_LOG_ROOT_REPLACE; | |
697 | ||
5de865ee FDBM |
698 | if (tree_mod_dont_log(fs_info, NULL)) |
699 | goto free_tms; | |
700 | ||
701 | if (tm_list) | |
702 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
703 | if (!ret) | |
704 | ret = __tree_mod_log_insert(fs_info, tm); | |
705 | ||
706 | tree_mod_log_write_unlock(fs_info); | |
707 | if (ret) | |
708 | goto free_tms; | |
709 | kfree(tm_list); | |
710 | ||
711 | return ret; | |
712 | ||
713 | free_tms: | |
714 | if (tm_list) { | |
715 | for (i = 0; i < nritems; i++) | |
716 | kfree(tm_list[i]); | |
717 | kfree(tm_list); | |
718 | } | |
719 | kfree(tm); | |
720 | ||
721 | return ret; | |
bd989ba3 JS |
722 | } |
723 | ||
724 | static struct tree_mod_elem * | |
725 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
726 | int smallest) | |
727 | { | |
728 | struct rb_root *tm_root; | |
729 | struct rb_node *node; | |
730 | struct tree_mod_elem *cur = NULL; | |
731 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 732 | |
097b8a7c | 733 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
734 | tm_root = &fs_info->tree_mod_log; |
735 | node = tm_root->rb_node; | |
736 | while (node) { | |
6b4df8b6 | 737 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 738 | if (cur->logical < start) { |
bd989ba3 | 739 | node = node->rb_left; |
298cfd36 | 740 | } else if (cur->logical > start) { |
bd989ba3 | 741 | node = node->rb_right; |
097b8a7c | 742 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
743 | node = node->rb_left; |
744 | } else if (!smallest) { | |
745 | /* we want the node with the highest seq */ | |
746 | if (found) | |
097b8a7c | 747 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
748 | found = cur; |
749 | node = node->rb_left; | |
097b8a7c | 750 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
751 | /* we want the node with the smallest seq */ |
752 | if (found) | |
097b8a7c | 753 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
754 | found = cur; |
755 | node = node->rb_right; | |
756 | } else { | |
757 | found = cur; | |
758 | break; | |
759 | } | |
760 | } | |
097b8a7c | 761 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
762 | |
763 | return found; | |
764 | } | |
765 | ||
766 | /* | |
767 | * this returns the element from the log with the smallest time sequence | |
768 | * value that's in the log (the oldest log item). any element with a time | |
769 | * sequence lower than min_seq will be ignored. | |
770 | */ | |
771 | static struct tree_mod_elem * | |
772 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
773 | u64 min_seq) | |
774 | { | |
775 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
776 | } | |
777 | ||
778 | /* | |
779 | * this returns the element from the log with the largest time sequence | |
780 | * value that's in the log (the most recent log item). any element with | |
781 | * a time sequence lower than min_seq will be ignored. | |
782 | */ | |
783 | static struct tree_mod_elem * | |
784 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
785 | { | |
786 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
787 | } | |
788 | ||
5de865ee | 789 | static noinline int |
bd989ba3 JS |
790 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
791 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 792 | unsigned long src_offset, int nr_items) |
bd989ba3 | 793 | { |
5de865ee FDBM |
794 | int ret = 0; |
795 | struct tree_mod_elem **tm_list = NULL; | |
796 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 797 | int i; |
5de865ee | 798 | int locked = 0; |
bd989ba3 | 799 | |
5de865ee FDBM |
800 | if (!tree_mod_need_log(fs_info, NULL)) |
801 | return 0; | |
bd989ba3 | 802 | |
c8cc6341 | 803 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
804 | return 0; |
805 | ||
31e818fe | 806 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
807 | GFP_NOFS); |
808 | if (!tm_list) | |
809 | return -ENOMEM; | |
bd989ba3 | 810 | |
5de865ee FDBM |
811 | tm_list_add = tm_list; |
812 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 813 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
814 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
815 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
816 | if (!tm_list_rem[i]) { | |
817 | ret = -ENOMEM; | |
818 | goto free_tms; | |
819 | } | |
820 | ||
821 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
822 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
823 | if (!tm_list_add[i]) { | |
824 | ret = -ENOMEM; | |
825 | goto free_tms; | |
826 | } | |
827 | } | |
828 | ||
829 | if (tree_mod_dont_log(fs_info, NULL)) | |
830 | goto free_tms; | |
831 | locked = 1; | |
832 | ||
833 | for (i = 0; i < nr_items; i++) { | |
834 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
835 | if (ret) | |
836 | goto free_tms; | |
837 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
838 | if (ret) | |
839 | goto free_tms; | |
bd989ba3 | 840 | } |
5de865ee FDBM |
841 | |
842 | tree_mod_log_write_unlock(fs_info); | |
843 | kfree(tm_list); | |
844 | ||
845 | return 0; | |
846 | ||
847 | free_tms: | |
848 | for (i = 0; i < nr_items * 2; i++) { | |
849 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
850 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
851 | kfree(tm_list[i]); | |
852 | } | |
853 | if (locked) | |
854 | tree_mod_log_write_unlock(fs_info); | |
855 | kfree(tm_list); | |
856 | ||
857 | return ret; | |
bd989ba3 JS |
858 | } |
859 | ||
a446a979 | 860 | static inline void tree_mod_log_eb_move(struct extent_buffer *dst, |
bd989ba3 JS |
861 | int dst_offset, int src_offset, int nr_items) |
862 | { | |
863 | int ret; | |
6074d45f | 864 | ret = tree_mod_log_insert_move(dst, dst_offset, src_offset, nr_items); |
bd989ba3 JS |
865 | BUG_ON(ret < 0); |
866 | } | |
867 | ||
3ac6de1a DS |
868 | static noinline void tree_mod_log_set_node_key(struct extent_buffer *eb, |
869 | int slot, int atomic) | |
bd989ba3 JS |
870 | { |
871 | int ret; | |
872 | ||
e09c2efe | 873 | ret = tree_mod_log_insert_key(eb, slot, MOD_LOG_KEY_REPLACE, |
c8cc6341 | 874 | atomic ? GFP_ATOMIC : GFP_NOFS); |
bd989ba3 JS |
875 | BUG_ON(ret < 0); |
876 | } | |
877 | ||
db7279a2 | 878 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 879 | { |
5de865ee FDBM |
880 | struct tree_mod_elem **tm_list = NULL; |
881 | int nritems = 0; | |
882 | int i; | |
883 | int ret = 0; | |
884 | ||
885 | if (btrfs_header_level(eb) == 0) | |
886 | return 0; | |
887 | ||
db7279a2 | 888 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
889 | return 0; |
890 | ||
891 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 892 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
893 | if (!tm_list) |
894 | return -ENOMEM; | |
895 | ||
896 | for (i = 0; i < nritems; i++) { | |
897 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
898 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
899 | if (!tm_list[i]) { | |
900 | ret = -ENOMEM; | |
901 | goto free_tms; | |
902 | } | |
903 | } | |
904 | ||
db7279a2 | 905 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
906 | goto free_tms; |
907 | ||
db7279a2 DS |
908 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
909 | tree_mod_log_write_unlock(eb->fs_info); | |
5de865ee FDBM |
910 | if (ret) |
911 | goto free_tms; | |
912 | kfree(tm_list); | |
913 | ||
914 | return 0; | |
915 | ||
916 | free_tms: | |
917 | for (i = 0; i < nritems; i++) | |
918 | kfree(tm_list[i]); | |
919 | kfree(tm_list); | |
920 | ||
921 | return ret; | |
bd989ba3 JS |
922 | } |
923 | ||
097b8a7c | 924 | static noinline void |
bd989ba3 | 925 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
926 | struct extent_buffer *new_root_node, |
927 | int log_removal) | |
bd989ba3 JS |
928 | { |
929 | int ret; | |
95b757c1 | 930 | ret = tree_mod_log_insert_root(root->node, new_root_node, log_removal); |
bd989ba3 JS |
931 | BUG_ON(ret < 0); |
932 | } | |
933 | ||
5d4f98a2 YZ |
934 | /* |
935 | * check if the tree block can be shared by multiple trees | |
936 | */ | |
937 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
938 | struct extent_buffer *buf) | |
939 | { | |
940 | /* | |
01327610 | 941 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
942 | * are never shared. If a block was allocated after the last |
943 | * snapshot and the block was not allocated by tree relocation, | |
944 | * we know the block is not shared. | |
945 | */ | |
27cdeb70 | 946 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
947 | buf != root->node && buf != root->commit_root && |
948 | (btrfs_header_generation(buf) <= | |
949 | btrfs_root_last_snapshot(&root->root_item) || | |
950 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
951 | return 1; | |
952 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 953 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
954 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
955 | return 1; | |
956 | #endif | |
957 | return 0; | |
958 | } | |
959 | ||
960 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
961 | struct btrfs_root *root, | |
962 | struct extent_buffer *buf, | |
f0486c68 YZ |
963 | struct extent_buffer *cow, |
964 | int *last_ref) | |
5d4f98a2 | 965 | { |
0b246afa | 966 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
967 | u64 refs; |
968 | u64 owner; | |
969 | u64 flags; | |
970 | u64 new_flags = 0; | |
971 | int ret; | |
972 | ||
973 | /* | |
974 | * Backrefs update rules: | |
975 | * | |
976 | * Always use full backrefs for extent pointers in tree block | |
977 | * allocated by tree relocation. | |
978 | * | |
979 | * If a shared tree block is no longer referenced by its owner | |
980 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
981 | * use full backrefs for extent pointers in tree block. | |
982 | * | |
983 | * If a tree block is been relocating | |
984 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
985 | * use full backrefs for extent pointers in tree block. | |
986 | * The reason for this is some operations (such as drop tree) | |
987 | * are only allowed for blocks use full backrefs. | |
988 | */ | |
989 | ||
990 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 991 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
992 | btrfs_header_level(buf), 1, |
993 | &refs, &flags); | |
be1a5564 MF |
994 | if (ret) |
995 | return ret; | |
e5df9573 MF |
996 | if (refs == 0) { |
997 | ret = -EROFS; | |
0b246afa | 998 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
999 | return ret; |
1000 | } | |
5d4f98a2 YZ |
1001 | } else { |
1002 | refs = 1; | |
1003 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
1004 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1005 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1006 | else | |
1007 | flags = 0; | |
1008 | } | |
1009 | ||
1010 | owner = btrfs_header_owner(buf); | |
1011 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
1012 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
1013 | ||
1014 | if (refs > 1) { | |
1015 | if ((owner == root->root_key.objectid || | |
1016 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
1017 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 1018 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
1019 | if (ret) |
1020 | return ret; | |
5d4f98a2 YZ |
1021 | |
1022 | if (root->root_key.objectid == | |
1023 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 1024 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
1025 | if (ret) |
1026 | return ret; | |
e339a6b0 | 1027 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
1028 | if (ret) |
1029 | return ret; | |
5d4f98a2 YZ |
1030 | } |
1031 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1032 | } else { | |
1033 | ||
1034 | if (root->root_key.objectid == | |
1035 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1036 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1037 | else |
e339a6b0 | 1038 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
1039 | if (ret) |
1040 | return ret; | |
5d4f98a2 YZ |
1041 | } |
1042 | if (new_flags != 0) { | |
b1c79e09 JB |
1043 | int level = btrfs_header_level(buf); |
1044 | ||
2ff7e61e | 1045 | ret = btrfs_set_disk_extent_flags(trans, fs_info, |
5d4f98a2 YZ |
1046 | buf->start, |
1047 | buf->len, | |
b1c79e09 | 1048 | new_flags, level, 0); |
be1a5564 MF |
1049 | if (ret) |
1050 | return ret; | |
5d4f98a2 YZ |
1051 | } |
1052 | } else { | |
1053 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
1054 | if (root->root_key.objectid == | |
1055 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1056 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1057 | else |
e339a6b0 | 1058 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
1059 | if (ret) |
1060 | return ret; | |
e339a6b0 | 1061 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
1062 | if (ret) |
1063 | return ret; | |
5d4f98a2 | 1064 | } |
7c302b49 | 1065 | clean_tree_block(fs_info, buf); |
f0486c68 | 1066 | *last_ref = 1; |
5d4f98a2 YZ |
1067 | } |
1068 | return 0; | |
1069 | } | |
1070 | ||
d352ac68 | 1071 | /* |
d397712b CM |
1072 | * does the dirty work in cow of a single block. The parent block (if |
1073 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1074 | * dirty and returned locked. If you modify the block it needs to be marked | |
1075 | * dirty again. | |
d352ac68 CM |
1076 | * |
1077 | * search_start -- an allocation hint for the new block | |
1078 | * | |
d397712b CM |
1079 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1080 | * bytes the allocator should try to find free next to the block it returns. | |
1081 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1082 | */ |
d397712b | 1083 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1084 | struct btrfs_root *root, |
1085 | struct extent_buffer *buf, | |
1086 | struct extent_buffer *parent, int parent_slot, | |
1087 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1088 | u64 search_start, u64 empty_size) |
02217ed2 | 1089 | { |
0b246afa | 1090 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1091 | struct btrfs_disk_key disk_key; |
5f39d397 | 1092 | struct extent_buffer *cow; |
be1a5564 | 1093 | int level, ret; |
f0486c68 | 1094 | int last_ref = 0; |
925baedd | 1095 | int unlock_orig = 0; |
0f5053eb | 1096 | u64 parent_start = 0; |
7bb86316 | 1097 | |
925baedd CM |
1098 | if (*cow_ret == buf) |
1099 | unlock_orig = 1; | |
1100 | ||
b9447ef8 | 1101 | btrfs_assert_tree_locked(buf); |
925baedd | 1102 | |
27cdeb70 | 1103 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1104 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1105 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1106 | trans->transid != root->last_trans); | |
5f39d397 | 1107 | |
7bb86316 | 1108 | level = btrfs_header_level(buf); |
31840ae1 | 1109 | |
5d4f98a2 YZ |
1110 | if (level == 0) |
1111 | btrfs_item_key(buf, &disk_key, 0); | |
1112 | else | |
1113 | btrfs_node_key(buf, &disk_key, 0); | |
1114 | ||
0f5053eb GR |
1115 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1116 | parent_start = parent->start; | |
5d4f98a2 | 1117 | |
4d75f8a9 DS |
1118 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1119 | root->root_key.objectid, &disk_key, level, | |
1120 | search_start, empty_size); | |
54aa1f4d CM |
1121 | if (IS_ERR(cow)) |
1122 | return PTR_ERR(cow); | |
6702ed49 | 1123 | |
b4ce94de CM |
1124 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1125 | ||
58e8012c | 1126 | copy_extent_buffer_full(cow, buf); |
db94535d | 1127 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1128 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1129 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1130 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1131 | BTRFS_HEADER_FLAG_RELOC); | |
1132 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1133 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1134 | else | |
1135 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1136 | |
0b246afa | 1137 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 1138 | |
be1a5564 | 1139 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1140 | if (ret) { |
66642832 | 1141 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1142 | return ret; |
1143 | } | |
1a40e23b | 1144 | |
27cdeb70 | 1145 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1146 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1147 | if (ret) { |
66642832 | 1148 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1149 | return ret; |
93314e3b | 1150 | } |
83d4cfd4 | 1151 | } |
3fd0a558 | 1152 | |
02217ed2 | 1153 | if (buf == root->node) { |
925baedd | 1154 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1155 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1156 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1157 | parent_start = buf->start; | |
925baedd | 1158 | |
5f39d397 | 1159 | extent_buffer_get(cow); |
90f8d62e | 1160 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1161 | rcu_assign_pointer(root->node, cow); |
925baedd | 1162 | |
f0486c68 | 1163 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1164 | last_ref); |
5f39d397 | 1165 | free_extent_buffer(buf); |
0b86a832 | 1166 | add_root_to_dirty_list(root); |
02217ed2 | 1167 | } else { |
5d4f98a2 | 1168 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1169 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1170 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1171 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1172 | cow->start); |
74493f7a CM |
1173 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1174 | trans->transid); | |
d6025579 | 1175 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1176 | if (last_ref) { |
db7279a2 | 1177 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1178 | if (ret) { |
66642832 | 1179 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1180 | return ret; |
1181 | } | |
1182 | } | |
f0486c68 | 1183 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1184 | last_ref); |
02217ed2 | 1185 | } |
925baedd CM |
1186 | if (unlock_orig) |
1187 | btrfs_tree_unlock(buf); | |
3083ee2e | 1188 | free_extent_buffer_stale(buf); |
ccd467d6 | 1189 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1190 | *cow_ret = cow; |
02217ed2 CM |
1191 | return 0; |
1192 | } | |
1193 | ||
5d9e75c4 JS |
1194 | /* |
1195 | * returns the logical address of the oldest predecessor of the given root. | |
1196 | * entries older than time_seq are ignored. | |
1197 | */ | |
bcd24dab DS |
1198 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1199 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1200 | { |
1201 | struct tree_mod_elem *tm; | |
1202 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1203 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1204 | int looped = 0; |
1205 | ||
1206 | if (!time_seq) | |
35a3621b | 1207 | return NULL; |
5d9e75c4 JS |
1208 | |
1209 | /* | |
298cfd36 CR |
1210 | * the very last operation that's logged for a root is the |
1211 | * replacement operation (if it is replaced at all). this has | |
1212 | * the logical address of the *new* root, making it the very | |
1213 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1214 | */ |
1215 | while (1) { | |
bcd24dab | 1216 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1217 | time_seq); |
1218 | if (!looped && !tm) | |
35a3621b | 1219 | return NULL; |
5d9e75c4 | 1220 | /* |
28da9fb4 JS |
1221 | * if there are no tree operation for the oldest root, we simply |
1222 | * return it. this should only happen if that (old) root is at | |
1223 | * level 0. | |
5d9e75c4 | 1224 | */ |
28da9fb4 JS |
1225 | if (!tm) |
1226 | break; | |
5d9e75c4 | 1227 | |
28da9fb4 JS |
1228 | /* |
1229 | * if there's an operation that's not a root replacement, we | |
1230 | * found the oldest version of our root. normally, we'll find a | |
1231 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1232 | */ | |
5d9e75c4 JS |
1233 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1234 | break; | |
1235 | ||
1236 | found = tm; | |
1237 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1238 | looped = 1; |
1239 | } | |
1240 | ||
a95236d9 JS |
1241 | /* if there's no old root to return, return what we found instead */ |
1242 | if (!found) | |
1243 | found = tm; | |
1244 | ||
5d9e75c4 JS |
1245 | return found; |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1250 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1251 | * time_seq). |
1252 | */ | |
1253 | static void | |
f1ca7e98 JB |
1254 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1255 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1256 | { |
1257 | u32 n; | |
1258 | struct rb_node *next; | |
1259 | struct tree_mod_elem *tm = first_tm; | |
1260 | unsigned long o_dst; | |
1261 | unsigned long o_src; | |
1262 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1263 | ||
1264 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1265 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1266 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1267 | /* |
1268 | * all the operations are recorded with the operator used for | |
1269 | * the modification. as we're going backwards, we do the | |
1270 | * opposite of each operation here. | |
1271 | */ | |
1272 | switch (tm->op) { | |
1273 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1274 | BUG_ON(tm->slot < n); | |
1c697d4a | 1275 | /* Fallthrough */ |
95c80bb1 | 1276 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1277 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1278 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1279 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1280 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1281 | tm->generation); | |
4c3e6969 | 1282 | n++; |
5d9e75c4 JS |
1283 | break; |
1284 | case MOD_LOG_KEY_REPLACE: | |
1285 | BUG_ON(tm->slot >= n); | |
1286 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1287 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1288 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1289 | tm->generation); | |
1290 | break; | |
1291 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1292 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1293 | n--; |
1294 | break; | |
1295 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1296 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1297 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1298 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1299 | tm->move.nr_items * p_size); |
1300 | break; | |
1301 | case MOD_LOG_ROOT_REPLACE: | |
1302 | /* | |
1303 | * this operation is special. for roots, this must be | |
1304 | * handled explicitly before rewinding. | |
1305 | * for non-roots, this operation may exist if the node | |
1306 | * was a root: root A -> child B; then A gets empty and | |
1307 | * B is promoted to the new root. in the mod log, we'll | |
1308 | * have a root-replace operation for B, a tree block | |
1309 | * that is no root. we simply ignore that operation. | |
1310 | */ | |
1311 | break; | |
1312 | } | |
1313 | next = rb_next(&tm->node); | |
1314 | if (!next) | |
1315 | break; | |
6b4df8b6 | 1316 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1317 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1318 | break; |
1319 | } | |
f1ca7e98 | 1320 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1321 | btrfs_set_header_nritems(eb, n); |
1322 | } | |
1323 | ||
47fb091f | 1324 | /* |
01327610 | 1325 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1326 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1327 | * returned buffer is always read-locked. If the returned buffer is not the | |
1328 | * input buffer, the lock on the input buffer is released and the input buffer | |
1329 | * is freed (its refcount is decremented). | |
1330 | */ | |
5d9e75c4 | 1331 | static struct extent_buffer * |
9ec72677 JB |
1332 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1333 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1334 | { |
1335 | struct extent_buffer *eb_rewin; | |
1336 | struct tree_mod_elem *tm; | |
1337 | ||
1338 | if (!time_seq) | |
1339 | return eb; | |
1340 | ||
1341 | if (btrfs_header_level(eb) == 0) | |
1342 | return eb; | |
1343 | ||
1344 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1345 | if (!tm) | |
1346 | return eb; | |
1347 | ||
9ec72677 JB |
1348 | btrfs_set_path_blocking(path); |
1349 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1350 | ||
5d9e75c4 JS |
1351 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1352 | BUG_ON(tm->slot != 0); | |
da17066c | 1353 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1354 | if (!eb_rewin) { |
9ec72677 | 1355 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1356 | free_extent_buffer(eb); |
1357 | return NULL; | |
1358 | } | |
5d9e75c4 JS |
1359 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1360 | btrfs_set_header_backref_rev(eb_rewin, | |
1361 | btrfs_header_backref_rev(eb)); | |
1362 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1363 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1364 | } else { |
1365 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1366 | if (!eb_rewin) { |
9ec72677 | 1367 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1368 | free_extent_buffer(eb); |
1369 | return NULL; | |
1370 | } | |
5d9e75c4 JS |
1371 | } |
1372 | ||
9ec72677 JB |
1373 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1374 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1375 | free_extent_buffer(eb); |
1376 | ||
47fb091f JS |
1377 | extent_buffer_get(eb_rewin); |
1378 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1379 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1380 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1381 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1382 | |
1383 | return eb_rewin; | |
1384 | } | |
1385 | ||
8ba97a15 JS |
1386 | /* |
1387 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1388 | * value. If there are no changes, the current root->root_node is returned. If | |
1389 | * anything changed in between, there's a fresh buffer allocated on which the | |
1390 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1391 | * Returns NULL on error (with no locks held). | |
1392 | */ | |
5d9e75c4 JS |
1393 | static inline struct extent_buffer * |
1394 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1395 | { | |
0b246afa | 1396 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1397 | struct tree_mod_elem *tm; |
30b0463a JS |
1398 | struct extent_buffer *eb = NULL; |
1399 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1400 | struct extent_buffer *old; |
a95236d9 | 1401 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1402 | u64 old_generation = 0; |
a95236d9 | 1403 | u64 logical; |
5d9e75c4 | 1404 | |
30b0463a | 1405 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1406 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1407 | if (!tm) |
30b0463a | 1408 | return eb_root; |
5d9e75c4 | 1409 | |
a95236d9 JS |
1410 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1411 | old_root = &tm->old_root; | |
1412 | old_generation = tm->generation; | |
1413 | logical = old_root->logical; | |
1414 | } else { | |
30b0463a | 1415 | logical = eb_root->start; |
a95236d9 | 1416 | } |
5d9e75c4 | 1417 | |
0b246afa | 1418 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1419 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1420 | btrfs_tree_read_unlock(eb_root); |
1421 | free_extent_buffer(eb_root); | |
2ff7e61e | 1422 | old = read_tree_block(fs_info, logical, 0); |
64c043de LB |
1423 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1424 | if (!IS_ERR(old)) | |
1425 | free_extent_buffer(old); | |
0b246afa JM |
1426 | btrfs_warn(fs_info, |
1427 | "failed to read tree block %llu from get_old_root", | |
1428 | logical); | |
834328a8 | 1429 | } else { |
7bfdcf7f LB |
1430 | eb = btrfs_clone_extent_buffer(old); |
1431 | free_extent_buffer(old); | |
834328a8 JS |
1432 | } |
1433 | } else if (old_root) { | |
30b0463a JS |
1434 | btrfs_tree_read_unlock(eb_root); |
1435 | free_extent_buffer(eb_root); | |
0b246afa | 1436 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1437 | } else { |
9ec72677 | 1438 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1439 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1440 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1441 | free_extent_buffer(eb_root); |
834328a8 JS |
1442 | } |
1443 | ||
8ba97a15 JS |
1444 | if (!eb) |
1445 | return NULL; | |
d6381084 | 1446 | extent_buffer_get(eb); |
8ba97a15 | 1447 | btrfs_tree_read_lock(eb); |
a95236d9 | 1448 | if (old_root) { |
5d9e75c4 JS |
1449 | btrfs_set_header_bytenr(eb, eb->start); |
1450 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1451 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1452 | btrfs_set_header_level(eb, old_root->level); |
1453 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1454 | } |
28da9fb4 | 1455 | if (tm) |
0b246afa | 1456 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1457 | else |
1458 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1459 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1460 | |
1461 | return eb; | |
1462 | } | |
1463 | ||
5b6602e7 JS |
1464 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1465 | { | |
1466 | struct tree_mod_elem *tm; | |
1467 | int level; | |
30b0463a | 1468 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1469 | |
bcd24dab | 1470 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1471 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1472 | level = tm->old_root.level; | |
1473 | } else { | |
30b0463a | 1474 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1475 | } |
30b0463a | 1476 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1477 | |
1478 | return level; | |
1479 | } | |
1480 | ||
5d4f98a2 YZ |
1481 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1482 | struct btrfs_root *root, | |
1483 | struct extent_buffer *buf) | |
1484 | { | |
f5ee5c9a | 1485 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1486 | return 0; |
fccb84c9 | 1487 | |
f1ebcc74 LB |
1488 | /* ensure we can see the force_cow */ |
1489 | smp_rmb(); | |
1490 | ||
1491 | /* | |
1492 | * We do not need to cow a block if | |
1493 | * 1) this block is not created or changed in this transaction; | |
1494 | * 2) this block does not belong to TREE_RELOC tree; | |
1495 | * 3) the root is not forced COW. | |
1496 | * | |
1497 | * What is forced COW: | |
01327610 | 1498 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1499 | * after we've finished coping src root, we must COW the shared |
1500 | * block to ensure the metadata consistency. | |
1501 | */ | |
5d4f98a2 YZ |
1502 | if (btrfs_header_generation(buf) == trans->transid && |
1503 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1504 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1505 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1506 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1507 | return 0; |
1508 | return 1; | |
1509 | } | |
1510 | ||
d352ac68 CM |
1511 | /* |
1512 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1513 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1514 | * once per transaction, as long as it hasn't been written yet |
1515 | */ | |
d397712b | 1516 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1517 | struct btrfs_root *root, struct extent_buffer *buf, |
1518 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1519 | struct extent_buffer **cow_ret) |
6702ed49 | 1520 | { |
0b246afa | 1521 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1522 | u64 search_start; |
f510cfec | 1523 | int ret; |
dc17ff8f | 1524 | |
0b246afa | 1525 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1526 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1527 | trans->transid, |
0b246afa | 1528 | fs_info->running_transaction->transid); |
31b1a2bd | 1529 | |
0b246afa | 1530 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1531 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1532 | trans->transid, fs_info->generation); |
dc17ff8f | 1533 | |
5d4f98a2 | 1534 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1535 | trans->dirty = true; |
6702ed49 CM |
1536 | *cow_ret = buf; |
1537 | return 0; | |
1538 | } | |
c487685d | 1539 | |
ee22184b | 1540 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1541 | |
1542 | if (parent) | |
1543 | btrfs_set_lock_blocking(parent); | |
1544 | btrfs_set_lock_blocking(buf); | |
1545 | ||
f510cfec | 1546 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1547 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1548 | |
1549 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1550 | ||
f510cfec | 1551 | return ret; |
6702ed49 CM |
1552 | } |
1553 | ||
d352ac68 CM |
1554 | /* |
1555 | * helper function for defrag to decide if two blocks pointed to by a | |
1556 | * node are actually close by | |
1557 | */ | |
6b80053d | 1558 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1559 | { |
6b80053d | 1560 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1561 | return 1; |
6b80053d | 1562 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1563 | return 1; |
1564 | return 0; | |
1565 | } | |
1566 | ||
081e9573 CM |
1567 | /* |
1568 | * compare two keys in a memcmp fashion | |
1569 | */ | |
310712b2 OS |
1570 | static int comp_keys(const struct btrfs_disk_key *disk, |
1571 | const struct btrfs_key *k2) | |
081e9573 CM |
1572 | { |
1573 | struct btrfs_key k1; | |
1574 | ||
1575 | btrfs_disk_key_to_cpu(&k1, disk); | |
1576 | ||
20736aba | 1577 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1578 | } |
1579 | ||
f3465ca4 JB |
1580 | /* |
1581 | * same as comp_keys only with two btrfs_key's | |
1582 | */ | |
310712b2 | 1583 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1584 | { |
1585 | if (k1->objectid > k2->objectid) | |
1586 | return 1; | |
1587 | if (k1->objectid < k2->objectid) | |
1588 | return -1; | |
1589 | if (k1->type > k2->type) | |
1590 | return 1; | |
1591 | if (k1->type < k2->type) | |
1592 | return -1; | |
1593 | if (k1->offset > k2->offset) | |
1594 | return 1; | |
1595 | if (k1->offset < k2->offset) | |
1596 | return -1; | |
1597 | return 0; | |
1598 | } | |
081e9573 | 1599 | |
d352ac68 CM |
1600 | /* |
1601 | * this is used by the defrag code to go through all the | |
1602 | * leaves pointed to by a node and reallocate them so that | |
1603 | * disk order is close to key order | |
1604 | */ | |
6702ed49 | 1605 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1606 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1607 | int start_slot, u64 *last_ret, |
a6b6e75e | 1608 | struct btrfs_key *progress) |
6702ed49 | 1609 | { |
0b246afa | 1610 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1611 | struct extent_buffer *cur; |
6702ed49 | 1612 | u64 blocknr; |
ca7a79ad | 1613 | u64 gen; |
e9d0b13b CM |
1614 | u64 search_start = *last_ret; |
1615 | u64 last_block = 0; | |
6702ed49 CM |
1616 | u64 other; |
1617 | u32 parent_nritems; | |
6702ed49 CM |
1618 | int end_slot; |
1619 | int i; | |
1620 | int err = 0; | |
f2183bde | 1621 | int parent_level; |
6b80053d CM |
1622 | int uptodate; |
1623 | u32 blocksize; | |
081e9573 CM |
1624 | int progress_passed = 0; |
1625 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1626 | |
5708b959 | 1627 | parent_level = btrfs_header_level(parent); |
5708b959 | 1628 | |
0b246afa JM |
1629 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1630 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1631 | |
6b80053d | 1632 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1633 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1634 | end_slot = parent_nritems - 1; |
6702ed49 | 1635 | |
5dfe2be7 | 1636 | if (parent_nritems <= 1) |
6702ed49 CM |
1637 | return 0; |
1638 | ||
b4ce94de CM |
1639 | btrfs_set_lock_blocking(parent); |
1640 | ||
5dfe2be7 | 1641 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 1642 | int close = 1; |
a6b6e75e | 1643 | |
081e9573 CM |
1644 | btrfs_node_key(parent, &disk_key, i); |
1645 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1646 | continue; | |
1647 | ||
1648 | progress_passed = 1; | |
6b80053d | 1649 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1650 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1651 | if (last_block == 0) |
1652 | last_block = blocknr; | |
5708b959 | 1653 | |
6702ed49 | 1654 | if (i > 0) { |
6b80053d CM |
1655 | other = btrfs_node_blockptr(parent, i - 1); |
1656 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1657 | } |
5dfe2be7 | 1658 | if (!close && i < end_slot) { |
6b80053d CM |
1659 | other = btrfs_node_blockptr(parent, i + 1); |
1660 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1661 | } |
e9d0b13b CM |
1662 | if (close) { |
1663 | last_block = blocknr; | |
6702ed49 | 1664 | continue; |
e9d0b13b | 1665 | } |
6702ed49 | 1666 | |
0b246afa | 1667 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1668 | if (cur) |
b9fab919 | 1669 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1670 | else |
1671 | uptodate = 0; | |
5708b959 | 1672 | if (!cur || !uptodate) { |
6b80053d | 1673 | if (!cur) { |
2ff7e61e | 1674 | cur = read_tree_block(fs_info, blocknr, gen); |
64c043de LB |
1675 | if (IS_ERR(cur)) { |
1676 | return PTR_ERR(cur); | |
1677 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1678 | free_extent_buffer(cur); |
97d9a8a4 | 1679 | return -EIO; |
416bc658 | 1680 | } |
6b80053d | 1681 | } else if (!uptodate) { |
018642a1 TI |
1682 | err = btrfs_read_buffer(cur, gen); |
1683 | if (err) { | |
1684 | free_extent_buffer(cur); | |
1685 | return err; | |
1686 | } | |
f2183bde | 1687 | } |
6702ed49 | 1688 | } |
e9d0b13b | 1689 | if (search_start == 0) |
6b80053d | 1690 | search_start = last_block; |
e9d0b13b | 1691 | |
e7a84565 | 1692 | btrfs_tree_lock(cur); |
b4ce94de | 1693 | btrfs_set_lock_blocking(cur); |
6b80053d | 1694 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1695 | &cur, search_start, |
6b80053d | 1696 | min(16 * blocksize, |
9fa8cfe7 | 1697 | (end_slot - i) * blocksize)); |
252c38f0 | 1698 | if (err) { |
e7a84565 | 1699 | btrfs_tree_unlock(cur); |
6b80053d | 1700 | free_extent_buffer(cur); |
6702ed49 | 1701 | break; |
252c38f0 | 1702 | } |
e7a84565 CM |
1703 | search_start = cur->start; |
1704 | last_block = cur->start; | |
f2183bde | 1705 | *last_ret = search_start; |
e7a84565 CM |
1706 | btrfs_tree_unlock(cur); |
1707 | free_extent_buffer(cur); | |
6702ed49 CM |
1708 | } |
1709 | return err; | |
1710 | } | |
1711 | ||
74123bd7 | 1712 | /* |
5f39d397 CM |
1713 | * search for key in the extent_buffer. The items start at offset p, |
1714 | * and they are item_size apart. There are 'max' items in p. | |
1715 | * | |
74123bd7 CM |
1716 | * the slot in the array is returned via slot, and it points to |
1717 | * the place where you would insert key if it is not found in | |
1718 | * the array. | |
1719 | * | |
1720 | * slot may point to max if the key is bigger than all of the keys | |
1721 | */ | |
e02119d5 | 1722 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1723 | unsigned long p, int item_size, |
1724 | const struct btrfs_key *key, | |
e02119d5 | 1725 | int max, int *slot) |
be0e5c09 CM |
1726 | { |
1727 | int low = 0; | |
1728 | int high = max; | |
1729 | int mid; | |
1730 | int ret; | |
479965d6 | 1731 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1732 | struct btrfs_disk_key unaligned; |
1733 | unsigned long offset; | |
5f39d397 CM |
1734 | char *kaddr = NULL; |
1735 | unsigned long map_start = 0; | |
1736 | unsigned long map_len = 0; | |
479965d6 | 1737 | int err; |
be0e5c09 | 1738 | |
5e24e9af LB |
1739 | if (low > high) { |
1740 | btrfs_err(eb->fs_info, | |
1741 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1742 | __func__, low, high, eb->start, | |
1743 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1744 | return -EINVAL; | |
1745 | } | |
1746 | ||
d397712b | 1747 | while (low < high) { |
be0e5c09 | 1748 | mid = (low + high) / 2; |
5f39d397 CM |
1749 | offset = p + mid * item_size; |
1750 | ||
a6591715 | 1751 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1752 | (offset + sizeof(struct btrfs_disk_key)) > |
1753 | map_start + map_len) { | |
934d375b CM |
1754 | |
1755 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1756 | sizeof(struct btrfs_disk_key), |
a6591715 | 1757 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1758 | |
1759 | if (!err) { | |
1760 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1761 | map_start); | |
415b35a5 | 1762 | } else if (err == 1) { |
479965d6 CM |
1763 | read_extent_buffer(eb, &unaligned, |
1764 | offset, sizeof(unaligned)); | |
1765 | tmp = &unaligned; | |
415b35a5 LB |
1766 | } else { |
1767 | return err; | |
479965d6 | 1768 | } |
5f39d397 | 1769 | |
5f39d397 CM |
1770 | } else { |
1771 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1772 | map_start); | |
1773 | } | |
be0e5c09 CM |
1774 | ret = comp_keys(tmp, key); |
1775 | ||
1776 | if (ret < 0) | |
1777 | low = mid + 1; | |
1778 | else if (ret > 0) | |
1779 | high = mid; | |
1780 | else { | |
1781 | *slot = mid; | |
1782 | return 0; | |
1783 | } | |
1784 | } | |
1785 | *slot = low; | |
1786 | return 1; | |
1787 | } | |
1788 | ||
97571fd0 CM |
1789 | /* |
1790 | * simple bin_search frontend that does the right thing for | |
1791 | * leaves vs nodes | |
1792 | */ | |
a74b35ec NB |
1793 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1794 | int level, int *slot) | |
be0e5c09 | 1795 | { |
f775738f | 1796 | if (level == 0) |
5f39d397 CM |
1797 | return generic_bin_search(eb, |
1798 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1799 | sizeof(struct btrfs_item), |
5f39d397 | 1800 | key, btrfs_header_nritems(eb), |
7518a238 | 1801 | slot); |
f775738f | 1802 | else |
5f39d397 CM |
1803 | return generic_bin_search(eb, |
1804 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1805 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1806 | key, btrfs_header_nritems(eb), |
7518a238 | 1807 | slot); |
be0e5c09 CM |
1808 | } |
1809 | ||
f0486c68 YZ |
1810 | static void root_add_used(struct btrfs_root *root, u32 size) |
1811 | { | |
1812 | spin_lock(&root->accounting_lock); | |
1813 | btrfs_set_root_used(&root->root_item, | |
1814 | btrfs_root_used(&root->root_item) + size); | |
1815 | spin_unlock(&root->accounting_lock); | |
1816 | } | |
1817 | ||
1818 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1819 | { | |
1820 | spin_lock(&root->accounting_lock); | |
1821 | btrfs_set_root_used(&root->root_item, | |
1822 | btrfs_root_used(&root->root_item) - size); | |
1823 | spin_unlock(&root->accounting_lock); | |
1824 | } | |
1825 | ||
d352ac68 CM |
1826 | /* given a node and slot number, this reads the blocks it points to. The |
1827 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1828 | */ |
2ff7e61e JM |
1829 | static noinline struct extent_buffer * |
1830 | read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent, | |
1831 | int slot) | |
bb803951 | 1832 | { |
ca7a79ad | 1833 | int level = btrfs_header_level(parent); |
416bc658 JB |
1834 | struct extent_buffer *eb; |
1835 | ||
fb770ae4 LB |
1836 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1837 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1838 | |
1839 | BUG_ON(level == 0); | |
1840 | ||
2ff7e61e | 1841 | eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot), |
416bc658 | 1842 | btrfs_node_ptr_generation(parent, slot)); |
fb770ae4 LB |
1843 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1844 | free_extent_buffer(eb); | |
1845 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1846 | } |
1847 | ||
1848 | return eb; | |
bb803951 CM |
1849 | } |
1850 | ||
d352ac68 CM |
1851 | /* |
1852 | * node level balancing, used to make sure nodes are in proper order for | |
1853 | * item deletion. We balance from the top down, so we have to make sure | |
1854 | * that a deletion won't leave an node completely empty later on. | |
1855 | */ | |
e02119d5 | 1856 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1857 | struct btrfs_root *root, |
1858 | struct btrfs_path *path, int level) | |
bb803951 | 1859 | { |
0b246afa | 1860 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1861 | struct extent_buffer *right = NULL; |
1862 | struct extent_buffer *mid; | |
1863 | struct extent_buffer *left = NULL; | |
1864 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1865 | int ret = 0; |
1866 | int wret; | |
1867 | int pslot; | |
bb803951 | 1868 | int orig_slot = path->slots[level]; |
79f95c82 | 1869 | u64 orig_ptr; |
bb803951 CM |
1870 | |
1871 | if (level == 0) | |
1872 | return 0; | |
1873 | ||
5f39d397 | 1874 | mid = path->nodes[level]; |
b4ce94de | 1875 | |
bd681513 CM |
1876 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1877 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1878 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1879 | ||
1d4f8a0c | 1880 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1881 | |
a05a9bb1 | 1882 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1883 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1884 | pslot = path->slots[level + 1]; |
1885 | } | |
bb803951 | 1886 | |
40689478 CM |
1887 | /* |
1888 | * deal with the case where there is only one pointer in the root | |
1889 | * by promoting the node below to a root | |
1890 | */ | |
5f39d397 CM |
1891 | if (!parent) { |
1892 | struct extent_buffer *child; | |
bb803951 | 1893 | |
5f39d397 | 1894 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1895 | return 0; |
1896 | ||
1897 | /* promote the child to a root */ | |
2ff7e61e | 1898 | child = read_node_slot(fs_info, mid, 0); |
fb770ae4 LB |
1899 | if (IS_ERR(child)) { |
1900 | ret = PTR_ERR(child); | |
0b246afa | 1901 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1902 | goto enospc; |
1903 | } | |
1904 | ||
925baedd | 1905 | btrfs_tree_lock(child); |
b4ce94de | 1906 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1907 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1908 | if (ret) { |
1909 | btrfs_tree_unlock(child); | |
1910 | free_extent_buffer(child); | |
1911 | goto enospc; | |
1912 | } | |
2f375ab9 | 1913 | |
90f8d62e | 1914 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1915 | rcu_assign_pointer(root->node, child); |
925baedd | 1916 | |
0b86a832 | 1917 | add_root_to_dirty_list(root); |
925baedd | 1918 | btrfs_tree_unlock(child); |
b4ce94de | 1919 | |
925baedd | 1920 | path->locks[level] = 0; |
bb803951 | 1921 | path->nodes[level] = NULL; |
7c302b49 | 1922 | clean_tree_block(fs_info, mid); |
925baedd | 1923 | btrfs_tree_unlock(mid); |
bb803951 | 1924 | /* once for the path */ |
5f39d397 | 1925 | free_extent_buffer(mid); |
f0486c68 YZ |
1926 | |
1927 | root_sub_used(root, mid->len); | |
5581a51a | 1928 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1929 | /* once for the root ptr */ |
3083ee2e | 1930 | free_extent_buffer_stale(mid); |
f0486c68 | 1931 | return 0; |
bb803951 | 1932 | } |
5f39d397 | 1933 | if (btrfs_header_nritems(mid) > |
0b246afa | 1934 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1935 | return 0; |
1936 | ||
2ff7e61e | 1937 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
1938 | if (IS_ERR(left)) |
1939 | left = NULL; | |
1940 | ||
5f39d397 | 1941 | if (left) { |
925baedd | 1942 | btrfs_tree_lock(left); |
b4ce94de | 1943 | btrfs_set_lock_blocking(left); |
5f39d397 | 1944 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1945 | parent, pslot - 1, &left); |
54aa1f4d CM |
1946 | if (wret) { |
1947 | ret = wret; | |
1948 | goto enospc; | |
1949 | } | |
2cc58cf2 | 1950 | } |
fb770ae4 | 1951 | |
2ff7e61e | 1952 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
1953 | if (IS_ERR(right)) |
1954 | right = NULL; | |
1955 | ||
5f39d397 | 1956 | if (right) { |
925baedd | 1957 | btrfs_tree_lock(right); |
b4ce94de | 1958 | btrfs_set_lock_blocking(right); |
5f39d397 | 1959 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1960 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1961 | if (wret) { |
1962 | ret = wret; | |
1963 | goto enospc; | |
1964 | } | |
1965 | } | |
1966 | ||
1967 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1968 | if (left) { |
1969 | orig_slot += btrfs_header_nritems(left); | |
2ff7e61e | 1970 | wret = push_node_left(trans, fs_info, left, mid, 1); |
79f95c82 CM |
1971 | if (wret < 0) |
1972 | ret = wret; | |
bb803951 | 1973 | } |
79f95c82 CM |
1974 | |
1975 | /* | |
1976 | * then try to empty the right most buffer into the middle | |
1977 | */ | |
5f39d397 | 1978 | if (right) { |
2ff7e61e | 1979 | wret = push_node_left(trans, fs_info, mid, right, 1); |
54aa1f4d | 1980 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1981 | ret = wret; |
5f39d397 | 1982 | if (btrfs_header_nritems(right) == 0) { |
7c302b49 | 1983 | clean_tree_block(fs_info, right); |
925baedd | 1984 | btrfs_tree_unlock(right); |
afe5fea7 | 1985 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1986 | root_sub_used(root, right->len); |
5581a51a | 1987 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1988 | free_extent_buffer_stale(right); |
f0486c68 | 1989 | right = NULL; |
bb803951 | 1990 | } else { |
5f39d397 CM |
1991 | struct btrfs_disk_key right_key; |
1992 | btrfs_node_key(right, &right_key, 0); | |
3ac6de1a | 1993 | tree_mod_log_set_node_key(parent, pslot + 1, 0); |
5f39d397 CM |
1994 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1995 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1996 | } |
1997 | } | |
5f39d397 | 1998 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1999 | /* |
2000 | * we're not allowed to leave a node with one item in the | |
2001 | * tree during a delete. A deletion from lower in the tree | |
2002 | * could try to delete the only pointer in this node. | |
2003 | * So, pull some keys from the left. | |
2004 | * There has to be a left pointer at this point because | |
2005 | * otherwise we would have pulled some pointers from the | |
2006 | * right | |
2007 | */ | |
305a26af MF |
2008 | if (!left) { |
2009 | ret = -EROFS; | |
0b246afa | 2010 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
2011 | goto enospc; |
2012 | } | |
2ff7e61e | 2013 | wret = balance_node_right(trans, fs_info, mid, left); |
54aa1f4d | 2014 | if (wret < 0) { |
79f95c82 | 2015 | ret = wret; |
54aa1f4d CM |
2016 | goto enospc; |
2017 | } | |
bce4eae9 | 2018 | if (wret == 1) { |
2ff7e61e | 2019 | wret = push_node_left(trans, fs_info, left, mid, 1); |
bce4eae9 CM |
2020 | if (wret < 0) |
2021 | ret = wret; | |
2022 | } | |
79f95c82 CM |
2023 | BUG_ON(wret == 1); |
2024 | } | |
5f39d397 | 2025 | if (btrfs_header_nritems(mid) == 0) { |
7c302b49 | 2026 | clean_tree_block(fs_info, mid); |
925baedd | 2027 | btrfs_tree_unlock(mid); |
afe5fea7 | 2028 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2029 | root_sub_used(root, mid->len); |
5581a51a | 2030 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2031 | free_extent_buffer_stale(mid); |
f0486c68 | 2032 | mid = NULL; |
79f95c82 CM |
2033 | } else { |
2034 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2035 | struct btrfs_disk_key mid_key; |
2036 | btrfs_node_key(mid, &mid_key, 0); | |
3ac6de1a | 2037 | tree_mod_log_set_node_key(parent, pslot, 0); |
5f39d397 CM |
2038 | btrfs_set_node_key(parent, &mid_key, pslot); |
2039 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2040 | } |
bb803951 | 2041 | |
79f95c82 | 2042 | /* update the path */ |
5f39d397 CM |
2043 | if (left) { |
2044 | if (btrfs_header_nritems(left) > orig_slot) { | |
2045 | extent_buffer_get(left); | |
925baedd | 2046 | /* left was locked after cow */ |
5f39d397 | 2047 | path->nodes[level] = left; |
bb803951 CM |
2048 | path->slots[level + 1] -= 1; |
2049 | path->slots[level] = orig_slot; | |
925baedd CM |
2050 | if (mid) { |
2051 | btrfs_tree_unlock(mid); | |
5f39d397 | 2052 | free_extent_buffer(mid); |
925baedd | 2053 | } |
bb803951 | 2054 | } else { |
5f39d397 | 2055 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2056 | path->slots[level] = orig_slot; |
2057 | } | |
2058 | } | |
79f95c82 | 2059 | /* double check we haven't messed things up */ |
e20d96d6 | 2060 | if (orig_ptr != |
5f39d397 | 2061 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2062 | BUG(); |
54aa1f4d | 2063 | enospc: |
925baedd CM |
2064 | if (right) { |
2065 | btrfs_tree_unlock(right); | |
5f39d397 | 2066 | free_extent_buffer(right); |
925baedd CM |
2067 | } |
2068 | if (left) { | |
2069 | if (path->nodes[level] != left) | |
2070 | btrfs_tree_unlock(left); | |
5f39d397 | 2071 | free_extent_buffer(left); |
925baedd | 2072 | } |
bb803951 CM |
2073 | return ret; |
2074 | } | |
2075 | ||
d352ac68 CM |
2076 | /* Node balancing for insertion. Here we only split or push nodes around |
2077 | * when they are completely full. This is also done top down, so we | |
2078 | * have to be pessimistic. | |
2079 | */ | |
d397712b | 2080 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2081 | struct btrfs_root *root, |
2082 | struct btrfs_path *path, int level) | |
e66f709b | 2083 | { |
0b246afa | 2084 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2085 | struct extent_buffer *right = NULL; |
2086 | struct extent_buffer *mid; | |
2087 | struct extent_buffer *left = NULL; | |
2088 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2089 | int ret = 0; |
2090 | int wret; | |
2091 | int pslot; | |
2092 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2093 | |
2094 | if (level == 0) | |
2095 | return 1; | |
2096 | ||
5f39d397 | 2097 | mid = path->nodes[level]; |
7bb86316 | 2098 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2099 | |
a05a9bb1 | 2100 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2101 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2102 | pslot = path->slots[level + 1]; |
2103 | } | |
e66f709b | 2104 | |
5f39d397 | 2105 | if (!parent) |
e66f709b | 2106 | return 1; |
e66f709b | 2107 | |
2ff7e61e | 2108 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
2109 | if (IS_ERR(left)) |
2110 | left = NULL; | |
e66f709b CM |
2111 | |
2112 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2113 | if (left) { |
e66f709b | 2114 | u32 left_nr; |
925baedd CM |
2115 | |
2116 | btrfs_tree_lock(left); | |
b4ce94de CM |
2117 | btrfs_set_lock_blocking(left); |
2118 | ||
5f39d397 | 2119 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2120 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2121 | wret = 1; |
2122 | } else { | |
5f39d397 | 2123 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2124 | pslot - 1, &left); |
54aa1f4d CM |
2125 | if (ret) |
2126 | wret = 1; | |
2127 | else { | |
2ff7e61e | 2128 | wret = push_node_left(trans, fs_info, |
971a1f66 | 2129 | left, mid, 0); |
54aa1f4d | 2130 | } |
33ade1f8 | 2131 | } |
e66f709b CM |
2132 | if (wret < 0) |
2133 | ret = wret; | |
2134 | if (wret == 0) { | |
5f39d397 | 2135 | struct btrfs_disk_key disk_key; |
e66f709b | 2136 | orig_slot += left_nr; |
5f39d397 | 2137 | btrfs_node_key(mid, &disk_key, 0); |
3ac6de1a | 2138 | tree_mod_log_set_node_key(parent, pslot, 0); |
5f39d397 CM |
2139 | btrfs_set_node_key(parent, &disk_key, pslot); |
2140 | btrfs_mark_buffer_dirty(parent); | |
2141 | if (btrfs_header_nritems(left) > orig_slot) { | |
2142 | path->nodes[level] = left; | |
e66f709b CM |
2143 | path->slots[level + 1] -= 1; |
2144 | path->slots[level] = orig_slot; | |
925baedd | 2145 | btrfs_tree_unlock(mid); |
5f39d397 | 2146 | free_extent_buffer(mid); |
e66f709b CM |
2147 | } else { |
2148 | orig_slot -= | |
5f39d397 | 2149 | btrfs_header_nritems(left); |
e66f709b | 2150 | path->slots[level] = orig_slot; |
925baedd | 2151 | btrfs_tree_unlock(left); |
5f39d397 | 2152 | free_extent_buffer(left); |
e66f709b | 2153 | } |
e66f709b CM |
2154 | return 0; |
2155 | } | |
925baedd | 2156 | btrfs_tree_unlock(left); |
5f39d397 | 2157 | free_extent_buffer(left); |
e66f709b | 2158 | } |
2ff7e61e | 2159 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
2160 | if (IS_ERR(right)) |
2161 | right = NULL; | |
e66f709b CM |
2162 | |
2163 | /* | |
2164 | * then try to empty the right most buffer into the middle | |
2165 | */ | |
5f39d397 | 2166 | if (right) { |
33ade1f8 | 2167 | u32 right_nr; |
b4ce94de | 2168 | |
925baedd | 2169 | btrfs_tree_lock(right); |
b4ce94de CM |
2170 | btrfs_set_lock_blocking(right); |
2171 | ||
5f39d397 | 2172 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2173 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2174 | wret = 1; |
2175 | } else { | |
5f39d397 CM |
2176 | ret = btrfs_cow_block(trans, root, right, |
2177 | parent, pslot + 1, | |
9fa8cfe7 | 2178 | &right); |
54aa1f4d CM |
2179 | if (ret) |
2180 | wret = 1; | |
2181 | else { | |
2ff7e61e | 2182 | wret = balance_node_right(trans, fs_info, |
5f39d397 | 2183 | right, mid); |
54aa1f4d | 2184 | } |
33ade1f8 | 2185 | } |
e66f709b CM |
2186 | if (wret < 0) |
2187 | ret = wret; | |
2188 | if (wret == 0) { | |
5f39d397 CM |
2189 | struct btrfs_disk_key disk_key; |
2190 | ||
2191 | btrfs_node_key(right, &disk_key, 0); | |
3ac6de1a | 2192 | tree_mod_log_set_node_key(parent, pslot + 1, 0); |
5f39d397 CM |
2193 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2194 | btrfs_mark_buffer_dirty(parent); | |
2195 | ||
2196 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2197 | path->nodes[level] = right; | |
e66f709b CM |
2198 | path->slots[level + 1] += 1; |
2199 | path->slots[level] = orig_slot - | |
5f39d397 | 2200 | btrfs_header_nritems(mid); |
925baedd | 2201 | btrfs_tree_unlock(mid); |
5f39d397 | 2202 | free_extent_buffer(mid); |
e66f709b | 2203 | } else { |
925baedd | 2204 | btrfs_tree_unlock(right); |
5f39d397 | 2205 | free_extent_buffer(right); |
e66f709b | 2206 | } |
e66f709b CM |
2207 | return 0; |
2208 | } | |
925baedd | 2209 | btrfs_tree_unlock(right); |
5f39d397 | 2210 | free_extent_buffer(right); |
e66f709b | 2211 | } |
e66f709b CM |
2212 | return 1; |
2213 | } | |
2214 | ||
3c69faec | 2215 | /* |
d352ac68 CM |
2216 | * readahead one full node of leaves, finding things that are close |
2217 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2218 | */ |
2ff7e61e | 2219 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2220 | struct btrfs_path *path, |
2221 | int level, int slot, u64 objectid) | |
3c69faec | 2222 | { |
5f39d397 | 2223 | struct extent_buffer *node; |
01f46658 | 2224 | struct btrfs_disk_key disk_key; |
3c69faec | 2225 | u32 nritems; |
3c69faec | 2226 | u64 search; |
a7175319 | 2227 | u64 target; |
6b80053d | 2228 | u64 nread = 0; |
5f39d397 | 2229 | struct extent_buffer *eb; |
6b80053d CM |
2230 | u32 nr; |
2231 | u32 blocksize; | |
2232 | u32 nscan = 0; | |
db94535d | 2233 | |
a6b6e75e | 2234 | if (level != 1) |
6702ed49 CM |
2235 | return; |
2236 | ||
2237 | if (!path->nodes[level]) | |
3c69faec CM |
2238 | return; |
2239 | ||
5f39d397 | 2240 | node = path->nodes[level]; |
925baedd | 2241 | |
3c69faec | 2242 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2243 | blocksize = fs_info->nodesize; |
2244 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2245 | if (eb) { |
2246 | free_extent_buffer(eb); | |
3c69faec CM |
2247 | return; |
2248 | } | |
2249 | ||
a7175319 | 2250 | target = search; |
6b80053d | 2251 | |
5f39d397 | 2252 | nritems = btrfs_header_nritems(node); |
6b80053d | 2253 | nr = slot; |
25b8b936 | 2254 | |
d397712b | 2255 | while (1) { |
e4058b54 | 2256 | if (path->reada == READA_BACK) { |
6b80053d CM |
2257 | if (nr == 0) |
2258 | break; | |
2259 | nr--; | |
e4058b54 | 2260 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2261 | nr++; |
2262 | if (nr >= nritems) | |
2263 | break; | |
3c69faec | 2264 | } |
e4058b54 | 2265 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2266 | btrfs_node_key(node, &disk_key, nr); |
2267 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2268 | break; | |
2269 | } | |
6b80053d | 2270 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2271 | if ((search <= target && target - search <= 65536) || |
2272 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2273 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2274 | nread += blocksize; |
2275 | } | |
2276 | nscan++; | |
a7175319 | 2277 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2278 | break; |
3c69faec CM |
2279 | } |
2280 | } | |
925baedd | 2281 | |
2ff7e61e | 2282 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2283 | struct btrfs_path *path, int level) |
b4ce94de CM |
2284 | { |
2285 | int slot; | |
2286 | int nritems; | |
2287 | struct extent_buffer *parent; | |
2288 | struct extent_buffer *eb; | |
2289 | u64 gen; | |
2290 | u64 block1 = 0; | |
2291 | u64 block2 = 0; | |
b4ce94de | 2292 | |
8c594ea8 | 2293 | parent = path->nodes[level + 1]; |
b4ce94de | 2294 | if (!parent) |
0b08851f | 2295 | return; |
b4ce94de CM |
2296 | |
2297 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2298 | slot = path->slots[level + 1]; |
b4ce94de CM |
2299 | |
2300 | if (slot > 0) { | |
2301 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2302 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2303 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2304 | /* |
2305 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2306 | * don't want to return eagain here. That will loop | |
2307 | * forever | |
2308 | */ | |
2309 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2310 | block1 = 0; |
2311 | free_extent_buffer(eb); | |
2312 | } | |
8c594ea8 | 2313 | if (slot + 1 < nritems) { |
b4ce94de CM |
2314 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2315 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2316 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2317 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2318 | block2 = 0; |
2319 | free_extent_buffer(eb); | |
2320 | } | |
8c594ea8 | 2321 | |
0b08851f | 2322 | if (block1) |
2ff7e61e | 2323 | readahead_tree_block(fs_info, block1); |
0b08851f | 2324 | if (block2) |
2ff7e61e | 2325 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2326 | } |
2327 | ||
2328 | ||
d352ac68 | 2329 | /* |
d397712b CM |
2330 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2331 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2332 | * operations on the tree might require changing key pointers higher up in the | |
2333 | * tree. | |
d352ac68 | 2334 | * |
d397712b CM |
2335 | * callers might also have set path->keep_locks, which tells this code to keep |
2336 | * the lock if the path points to the last slot in the block. This is part of | |
2337 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2338 | * |
d397712b CM |
2339 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2340 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2341 | */ |
e02119d5 | 2342 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2343 | int lowest_unlock, int min_write_lock_level, |
2344 | int *write_lock_level) | |
925baedd CM |
2345 | { |
2346 | int i; | |
2347 | int skip_level = level; | |
051e1b9f | 2348 | int no_skips = 0; |
925baedd CM |
2349 | struct extent_buffer *t; |
2350 | ||
2351 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2352 | if (!path->nodes[i]) | |
2353 | break; | |
2354 | if (!path->locks[i]) | |
2355 | break; | |
051e1b9f | 2356 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2357 | skip_level = i + 1; |
2358 | continue; | |
2359 | } | |
051e1b9f | 2360 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2361 | u32 nritems; |
2362 | t = path->nodes[i]; | |
2363 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2364 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2365 | skip_level = i + 1; |
2366 | continue; | |
2367 | } | |
2368 | } | |
051e1b9f CM |
2369 | if (skip_level < i && i >= lowest_unlock) |
2370 | no_skips = 1; | |
2371 | ||
925baedd CM |
2372 | t = path->nodes[i]; |
2373 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2374 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2375 | path->locks[i] = 0; |
f7c79f30 CM |
2376 | if (write_lock_level && |
2377 | i > min_write_lock_level && | |
2378 | i <= *write_lock_level) { | |
2379 | *write_lock_level = i - 1; | |
2380 | } | |
925baedd CM |
2381 | } |
2382 | } | |
2383 | } | |
2384 | ||
b4ce94de CM |
2385 | /* |
2386 | * This releases any locks held in the path starting at level and | |
2387 | * going all the way up to the root. | |
2388 | * | |
2389 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2390 | * corner cases, such as COW of the block at slot zero in the node. This | |
2391 | * ignores those rules, and it should only be called when there are no | |
2392 | * more updates to be done higher up in the tree. | |
2393 | */ | |
2394 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2395 | { | |
2396 | int i; | |
2397 | ||
09a2a8f9 | 2398 | if (path->keep_locks) |
b4ce94de CM |
2399 | return; |
2400 | ||
2401 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2402 | if (!path->nodes[i]) | |
12f4dacc | 2403 | continue; |
b4ce94de | 2404 | if (!path->locks[i]) |
12f4dacc | 2405 | continue; |
bd681513 | 2406 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2407 | path->locks[i] = 0; |
2408 | } | |
2409 | } | |
2410 | ||
c8c42864 CM |
2411 | /* |
2412 | * helper function for btrfs_search_slot. The goal is to find a block | |
2413 | * in cache without setting the path to blocking. If we find the block | |
2414 | * we return zero and the path is unchanged. | |
2415 | * | |
2416 | * If we can't find the block, we set the path blocking and do some | |
2417 | * reada. -EAGAIN is returned and the search must be repeated. | |
2418 | */ | |
2419 | static int | |
d07b8528 LB |
2420 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2421 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2422 | const struct btrfs_key *key) |
c8c42864 | 2423 | { |
0b246afa | 2424 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2425 | u64 blocknr; |
2426 | u64 gen; | |
c8c42864 CM |
2427 | struct extent_buffer *b = *eb_ret; |
2428 | struct extent_buffer *tmp; | |
76a05b35 | 2429 | int ret; |
c8c42864 CM |
2430 | |
2431 | blocknr = btrfs_node_blockptr(b, slot); | |
2432 | gen = btrfs_node_ptr_generation(b, slot); | |
c8c42864 | 2433 | |
0b246afa | 2434 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2435 | if (tmp) { |
b9fab919 | 2436 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2437 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2438 | *eb_ret = tmp; | |
2439 | return 0; | |
2440 | } | |
2441 | ||
2442 | /* the pages were up to date, but we failed | |
2443 | * the generation number check. Do a full | |
2444 | * read for the generation number that is correct. | |
2445 | * We must do this without dropping locks so | |
2446 | * we can trust our generation number | |
2447 | */ | |
2448 | btrfs_set_path_blocking(p); | |
2449 | ||
2450 | /* now we're allowed to do a blocking uptodate check */ | |
2451 | ret = btrfs_read_buffer(tmp, gen); | |
2452 | if (!ret) { | |
2453 | *eb_ret = tmp; | |
2454 | return 0; | |
cb44921a | 2455 | } |
bdf7c00e JB |
2456 | free_extent_buffer(tmp); |
2457 | btrfs_release_path(p); | |
2458 | return -EIO; | |
c8c42864 CM |
2459 | } |
2460 | ||
2461 | /* | |
2462 | * reduce lock contention at high levels | |
2463 | * of the btree by dropping locks before | |
76a05b35 CM |
2464 | * we read. Don't release the lock on the current |
2465 | * level because we need to walk this node to figure | |
2466 | * out which blocks to read. | |
c8c42864 | 2467 | */ |
8c594ea8 CM |
2468 | btrfs_unlock_up_safe(p, level + 1); |
2469 | btrfs_set_path_blocking(p); | |
2470 | ||
cb44921a | 2471 | free_extent_buffer(tmp); |
e4058b54 | 2472 | if (p->reada != READA_NONE) |
2ff7e61e | 2473 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2474 | |
b3b4aa74 | 2475 | btrfs_release_path(p); |
76a05b35 CM |
2476 | |
2477 | ret = -EAGAIN; | |
2ff7e61e | 2478 | tmp = read_tree_block(fs_info, blocknr, 0); |
64c043de | 2479 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2480 | /* |
2481 | * If the read above didn't mark this buffer up to date, | |
2482 | * it will never end up being up to date. Set ret to EIO now | |
2483 | * and give up so that our caller doesn't loop forever | |
2484 | * on our EAGAINs. | |
2485 | */ | |
b9fab919 | 2486 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2487 | ret = -EIO; |
c8c42864 | 2488 | free_extent_buffer(tmp); |
c871b0f2 LB |
2489 | } else { |
2490 | ret = PTR_ERR(tmp); | |
76a05b35 CM |
2491 | } |
2492 | return ret; | |
c8c42864 CM |
2493 | } |
2494 | ||
2495 | /* | |
2496 | * helper function for btrfs_search_slot. This does all of the checks | |
2497 | * for node-level blocks and does any balancing required based on | |
2498 | * the ins_len. | |
2499 | * | |
2500 | * If no extra work was required, zero is returned. If we had to | |
2501 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2502 | * start over | |
2503 | */ | |
2504 | static int | |
2505 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2506 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2507 | struct extent_buffer *b, int level, int ins_len, |
2508 | int *write_lock_level) | |
c8c42864 | 2509 | { |
0b246afa | 2510 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2511 | int ret; |
0b246afa | 2512 | |
c8c42864 | 2513 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2514 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2515 | int sret; |
2516 | ||
bd681513 CM |
2517 | if (*write_lock_level < level + 1) { |
2518 | *write_lock_level = level + 1; | |
2519 | btrfs_release_path(p); | |
2520 | goto again; | |
2521 | } | |
2522 | ||
c8c42864 | 2523 | btrfs_set_path_blocking(p); |
2ff7e61e | 2524 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2525 | sret = split_node(trans, root, p, level); |
bd681513 | 2526 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2527 | |
2528 | BUG_ON(sret > 0); | |
2529 | if (sret) { | |
2530 | ret = sret; | |
2531 | goto done; | |
2532 | } | |
2533 | b = p->nodes[level]; | |
2534 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2535 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2536 | int sret; |
2537 | ||
bd681513 CM |
2538 | if (*write_lock_level < level + 1) { |
2539 | *write_lock_level = level + 1; | |
2540 | btrfs_release_path(p); | |
2541 | goto again; | |
2542 | } | |
2543 | ||
c8c42864 | 2544 | btrfs_set_path_blocking(p); |
2ff7e61e | 2545 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2546 | sret = balance_level(trans, root, p, level); |
bd681513 | 2547 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2548 | |
2549 | if (sret) { | |
2550 | ret = sret; | |
2551 | goto done; | |
2552 | } | |
2553 | b = p->nodes[level]; | |
2554 | if (!b) { | |
b3b4aa74 | 2555 | btrfs_release_path(p); |
c8c42864 CM |
2556 | goto again; |
2557 | } | |
2558 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2559 | } | |
2560 | return 0; | |
2561 | ||
2562 | again: | |
2563 | ret = -EAGAIN; | |
2564 | done: | |
2565 | return ret; | |
2566 | } | |
2567 | ||
d7396f07 | 2568 | static void key_search_validate(struct extent_buffer *b, |
310712b2 | 2569 | const struct btrfs_key *key, |
d7396f07 FDBM |
2570 | int level) |
2571 | { | |
2572 | #ifdef CONFIG_BTRFS_ASSERT | |
2573 | struct btrfs_disk_key disk_key; | |
2574 | ||
2575 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2576 | ||
2577 | if (level == 0) | |
2578 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2579 | offsetof(struct btrfs_leaf, items[0].key), | |
2580 | sizeof(disk_key))); | |
2581 | else | |
2582 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2583 | offsetof(struct btrfs_node, ptrs[0].key), | |
2584 | sizeof(disk_key))); | |
2585 | #endif | |
2586 | } | |
2587 | ||
310712b2 | 2588 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2589 | int level, int *prev_cmp, int *slot) |
2590 | { | |
2591 | if (*prev_cmp != 0) { | |
a74b35ec | 2592 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2593 | return *prev_cmp; |
2594 | } | |
2595 | ||
2596 | key_search_validate(b, key, level); | |
2597 | *slot = 0; | |
2598 | ||
2599 | return 0; | |
2600 | } | |
2601 | ||
381cf658 | 2602 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2603 | u64 iobjectid, u64 ioff, u8 key_type, |
2604 | struct btrfs_key *found_key) | |
2605 | { | |
2606 | int ret; | |
2607 | struct btrfs_key key; | |
2608 | struct extent_buffer *eb; | |
381cf658 DS |
2609 | |
2610 | ASSERT(path); | |
1d4c08e0 | 2611 | ASSERT(found_key); |
e33d5c3d KN |
2612 | |
2613 | key.type = key_type; | |
2614 | key.objectid = iobjectid; | |
2615 | key.offset = ioff; | |
2616 | ||
2617 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2618 | if (ret < 0) |
e33d5c3d KN |
2619 | return ret; |
2620 | ||
2621 | eb = path->nodes[0]; | |
2622 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2623 | ret = btrfs_next_leaf(fs_root, path); | |
2624 | if (ret) | |
2625 | return ret; | |
2626 | eb = path->nodes[0]; | |
2627 | } | |
2628 | ||
2629 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2630 | if (found_key->type != key.type || | |
2631 | found_key->objectid != key.objectid) | |
2632 | return 1; | |
2633 | ||
2634 | return 0; | |
2635 | } | |
2636 | ||
74123bd7 | 2637 | /* |
4271ecea NB |
2638 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2639 | * modifications to preserve tree invariants. | |
74123bd7 | 2640 | * |
4271ecea NB |
2641 | * @trans: Handle of transaction, used when modifying the tree |
2642 | * @p: Holds all btree nodes along the search path | |
2643 | * @root: The root node of the tree | |
2644 | * @key: The key we are looking for | |
2645 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2646 | * deletions it's -1. 0 for plain searches | |
2647 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2648 | * when modifying the tree. | |
97571fd0 | 2649 | * |
4271ecea NB |
2650 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2651 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2652 | * | |
2653 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2654 | * of the path (level 0) | |
2655 | * | |
2656 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2657 | * points to the slot where it should be inserted | |
2658 | * | |
2659 | * If an error is encountered while searching the tree a negative error number | |
2660 | * is returned | |
74123bd7 | 2661 | */ |
310712b2 OS |
2662 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2663 | const struct btrfs_key *key, struct btrfs_path *p, | |
2664 | int ins_len, int cow) | |
be0e5c09 | 2665 | { |
0b246afa | 2666 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2667 | struct extent_buffer *b; |
be0e5c09 CM |
2668 | int slot; |
2669 | int ret; | |
33c66f43 | 2670 | int err; |
be0e5c09 | 2671 | int level; |
925baedd | 2672 | int lowest_unlock = 1; |
bd681513 CM |
2673 | int root_lock; |
2674 | /* everything at write_lock_level or lower must be write locked */ | |
2675 | int write_lock_level = 0; | |
9f3a7427 | 2676 | u8 lowest_level = 0; |
f7c79f30 | 2677 | int min_write_lock_level; |
d7396f07 | 2678 | int prev_cmp; |
9f3a7427 | 2679 | |
6702ed49 | 2680 | lowest_level = p->lowest_level; |
323ac95b | 2681 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2682 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2683 | BUG_ON(!cow && ins_len); |
25179201 | 2684 | |
bd681513 | 2685 | if (ins_len < 0) { |
925baedd | 2686 | lowest_unlock = 2; |
65b51a00 | 2687 | |
bd681513 CM |
2688 | /* when we are removing items, we might have to go up to level |
2689 | * two as we update tree pointers Make sure we keep write | |
2690 | * for those levels as well | |
2691 | */ | |
2692 | write_lock_level = 2; | |
2693 | } else if (ins_len > 0) { | |
2694 | /* | |
2695 | * for inserting items, make sure we have a write lock on | |
2696 | * level 1 so we can update keys | |
2697 | */ | |
2698 | write_lock_level = 1; | |
2699 | } | |
2700 | ||
2701 | if (!cow) | |
2702 | write_lock_level = -1; | |
2703 | ||
09a2a8f9 | 2704 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2705 | write_lock_level = BTRFS_MAX_LEVEL; |
2706 | ||
f7c79f30 CM |
2707 | min_write_lock_level = write_lock_level; |
2708 | ||
bb803951 | 2709 | again: |
d7396f07 | 2710 | prev_cmp = -1; |
bd681513 CM |
2711 | /* |
2712 | * we try very hard to do read locks on the root | |
2713 | */ | |
2714 | root_lock = BTRFS_READ_LOCK; | |
2715 | level = 0; | |
5d4f98a2 | 2716 | if (p->search_commit_root) { |
bd681513 CM |
2717 | /* |
2718 | * the commit roots are read only | |
2719 | * so we always do read locks | |
2720 | */ | |
3f8a18cc | 2721 | if (p->need_commit_sem) |
0b246afa | 2722 | down_read(&fs_info->commit_root_sem); |
5d4f98a2 YZ |
2723 | b = root->commit_root; |
2724 | extent_buffer_get(b); | |
bd681513 | 2725 | level = btrfs_header_level(b); |
3f8a18cc | 2726 | if (p->need_commit_sem) |
0b246afa | 2727 | up_read(&fs_info->commit_root_sem); |
5d4f98a2 | 2728 | if (!p->skip_locking) |
bd681513 | 2729 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2730 | } else { |
bd681513 | 2731 | if (p->skip_locking) { |
5d4f98a2 | 2732 | b = btrfs_root_node(root); |
bd681513 CM |
2733 | level = btrfs_header_level(b); |
2734 | } else { | |
2735 | /* we don't know the level of the root node | |
2736 | * until we actually have it read locked | |
2737 | */ | |
2738 | b = btrfs_read_lock_root_node(root); | |
2739 | level = btrfs_header_level(b); | |
2740 | if (level <= write_lock_level) { | |
2741 | /* whoops, must trade for write lock */ | |
2742 | btrfs_tree_read_unlock(b); | |
2743 | free_extent_buffer(b); | |
2744 | b = btrfs_lock_root_node(root); | |
2745 | root_lock = BTRFS_WRITE_LOCK; | |
2746 | ||
2747 | /* the level might have changed, check again */ | |
2748 | level = btrfs_header_level(b); | |
2749 | } | |
2750 | } | |
5d4f98a2 | 2751 | } |
bd681513 CM |
2752 | p->nodes[level] = b; |
2753 | if (!p->skip_locking) | |
2754 | p->locks[level] = root_lock; | |
925baedd | 2755 | |
eb60ceac | 2756 | while (b) { |
5f39d397 | 2757 | level = btrfs_header_level(b); |
65b51a00 CM |
2758 | |
2759 | /* | |
2760 | * setup the path here so we can release it under lock | |
2761 | * contention with the cow code | |
2762 | */ | |
02217ed2 | 2763 | if (cow) { |
9ea2c7c9 NB |
2764 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2765 | ||
c8c42864 CM |
2766 | /* |
2767 | * if we don't really need to cow this block | |
2768 | * then we don't want to set the path blocking, | |
2769 | * so we test it here | |
2770 | */ | |
64c12921 JM |
2771 | if (!should_cow_block(trans, root, b)) { |
2772 | trans->dirty = true; | |
65b51a00 | 2773 | goto cow_done; |
64c12921 | 2774 | } |
5d4f98a2 | 2775 | |
bd681513 CM |
2776 | /* |
2777 | * must have write locks on this node and the | |
2778 | * parent | |
2779 | */ | |
5124e00e JB |
2780 | if (level > write_lock_level || |
2781 | (level + 1 > write_lock_level && | |
2782 | level + 1 < BTRFS_MAX_LEVEL && | |
2783 | p->nodes[level + 1])) { | |
bd681513 CM |
2784 | write_lock_level = level + 1; |
2785 | btrfs_release_path(p); | |
2786 | goto again; | |
2787 | } | |
2788 | ||
160f4089 | 2789 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2790 | if (last_level) |
2791 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2792 | &b); | |
2793 | else | |
2794 | err = btrfs_cow_block(trans, root, b, | |
2795 | p->nodes[level + 1], | |
2796 | p->slots[level + 1], &b); | |
33c66f43 | 2797 | if (err) { |
33c66f43 | 2798 | ret = err; |
65b51a00 | 2799 | goto done; |
54aa1f4d | 2800 | } |
02217ed2 | 2801 | } |
65b51a00 | 2802 | cow_done: |
eb60ceac | 2803 | p->nodes[level] = b; |
bd681513 | 2804 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2805 | |
2806 | /* | |
2807 | * we have a lock on b and as long as we aren't changing | |
2808 | * the tree, there is no way to for the items in b to change. | |
2809 | * It is safe to drop the lock on our parent before we | |
2810 | * go through the expensive btree search on b. | |
2811 | * | |
eb653de1 FDBM |
2812 | * If we're inserting or deleting (ins_len != 0), then we might |
2813 | * be changing slot zero, which may require changing the parent. | |
2814 | * So, we can't drop the lock until after we know which slot | |
2815 | * we're operating on. | |
b4ce94de | 2816 | */ |
eb653de1 FDBM |
2817 | if (!ins_len && !p->keep_locks) { |
2818 | int u = level + 1; | |
2819 | ||
2820 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2821 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2822 | p->locks[u] = 0; | |
2823 | } | |
2824 | } | |
b4ce94de | 2825 | |
d7396f07 | 2826 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2827 | if (ret < 0) |
2828 | goto done; | |
b4ce94de | 2829 | |
5f39d397 | 2830 | if (level != 0) { |
33c66f43 YZ |
2831 | int dec = 0; |
2832 | if (ret && slot > 0) { | |
2833 | dec = 1; | |
be0e5c09 | 2834 | slot -= 1; |
33c66f43 | 2835 | } |
be0e5c09 | 2836 | p->slots[level] = slot; |
33c66f43 | 2837 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2838 | ins_len, &write_lock_level); |
33c66f43 | 2839 | if (err == -EAGAIN) |
c8c42864 | 2840 | goto again; |
33c66f43 YZ |
2841 | if (err) { |
2842 | ret = err; | |
c8c42864 | 2843 | goto done; |
33c66f43 | 2844 | } |
c8c42864 CM |
2845 | b = p->nodes[level]; |
2846 | slot = p->slots[level]; | |
b4ce94de | 2847 | |
bd681513 CM |
2848 | /* |
2849 | * slot 0 is special, if we change the key | |
2850 | * we have to update the parent pointer | |
2851 | * which means we must have a write lock | |
2852 | * on the parent | |
2853 | */ | |
eb653de1 | 2854 | if (slot == 0 && ins_len && |
bd681513 CM |
2855 | write_lock_level < level + 1) { |
2856 | write_lock_level = level + 1; | |
2857 | btrfs_release_path(p); | |
2858 | goto again; | |
2859 | } | |
2860 | ||
f7c79f30 CM |
2861 | unlock_up(p, level, lowest_unlock, |
2862 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2863 | |
925baedd | 2864 | if (level == lowest_level) { |
33c66f43 YZ |
2865 | if (dec) |
2866 | p->slots[level]++; | |
5b21f2ed | 2867 | goto done; |
925baedd | 2868 | } |
ca7a79ad | 2869 | |
d07b8528 | 2870 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2871 | slot, key); |
33c66f43 | 2872 | if (err == -EAGAIN) |
c8c42864 | 2873 | goto again; |
33c66f43 YZ |
2874 | if (err) { |
2875 | ret = err; | |
76a05b35 | 2876 | goto done; |
33c66f43 | 2877 | } |
76a05b35 | 2878 | |
b4ce94de | 2879 | if (!p->skip_locking) { |
bd681513 CM |
2880 | level = btrfs_header_level(b); |
2881 | if (level <= write_lock_level) { | |
2882 | err = btrfs_try_tree_write_lock(b); | |
2883 | if (!err) { | |
2884 | btrfs_set_path_blocking(p); | |
2885 | btrfs_tree_lock(b); | |
2886 | btrfs_clear_path_blocking(p, b, | |
2887 | BTRFS_WRITE_LOCK); | |
2888 | } | |
2889 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2890 | } else { | |
f82c458a | 2891 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2892 | if (!err) { |
2893 | btrfs_set_path_blocking(p); | |
2894 | btrfs_tree_read_lock(b); | |
2895 | btrfs_clear_path_blocking(p, b, | |
2896 | BTRFS_READ_LOCK); | |
2897 | } | |
2898 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2899 | } |
bd681513 | 2900 | p->nodes[level] = b; |
b4ce94de | 2901 | } |
be0e5c09 CM |
2902 | } else { |
2903 | p->slots[level] = slot; | |
87b29b20 | 2904 | if (ins_len > 0 && |
2ff7e61e | 2905 | btrfs_leaf_free_space(fs_info, b) < ins_len) { |
bd681513 CM |
2906 | if (write_lock_level < 1) { |
2907 | write_lock_level = 1; | |
2908 | btrfs_release_path(p); | |
2909 | goto again; | |
2910 | } | |
2911 | ||
b4ce94de | 2912 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2913 | err = split_leaf(trans, root, key, |
2914 | p, ins_len, ret == 0); | |
bd681513 | 2915 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2916 | |
33c66f43 YZ |
2917 | BUG_ON(err > 0); |
2918 | if (err) { | |
2919 | ret = err; | |
65b51a00 CM |
2920 | goto done; |
2921 | } | |
5c680ed6 | 2922 | } |
459931ec | 2923 | if (!p->search_for_split) |
f7c79f30 CM |
2924 | unlock_up(p, level, lowest_unlock, |
2925 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2926 | goto done; |
be0e5c09 CM |
2927 | } |
2928 | } | |
65b51a00 CM |
2929 | ret = 1; |
2930 | done: | |
b4ce94de CM |
2931 | /* |
2932 | * we don't really know what they plan on doing with the path | |
2933 | * from here on, so for now just mark it as blocking | |
2934 | */ | |
b9473439 CM |
2935 | if (!p->leave_spinning) |
2936 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2937 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2938 | btrfs_release_path(p); |
65b51a00 | 2939 | return ret; |
be0e5c09 CM |
2940 | } |
2941 | ||
5d9e75c4 JS |
2942 | /* |
2943 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2944 | * current state of the tree together with the operations recorded in the tree | |
2945 | * modification log to search for the key in a previous version of this tree, as | |
2946 | * denoted by the time_seq parameter. | |
2947 | * | |
2948 | * Naturally, there is no support for insert, delete or cow operations. | |
2949 | * | |
2950 | * The resulting path and return value will be set up as if we called | |
2951 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2952 | */ | |
310712b2 | 2953 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2954 | struct btrfs_path *p, u64 time_seq) |
2955 | { | |
0b246afa | 2956 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2957 | struct extent_buffer *b; |
2958 | int slot; | |
2959 | int ret; | |
2960 | int err; | |
2961 | int level; | |
2962 | int lowest_unlock = 1; | |
2963 | u8 lowest_level = 0; | |
d4b4087c | 2964 | int prev_cmp = -1; |
5d9e75c4 JS |
2965 | |
2966 | lowest_level = p->lowest_level; | |
2967 | WARN_ON(p->nodes[0] != NULL); | |
2968 | ||
2969 | if (p->search_commit_root) { | |
2970 | BUG_ON(time_seq); | |
2971 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2972 | } | |
2973 | ||
2974 | again: | |
5d9e75c4 | 2975 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2976 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2977 | p->locks[level] = BTRFS_READ_LOCK; |
2978 | ||
2979 | while (b) { | |
2980 | level = btrfs_header_level(b); | |
2981 | p->nodes[level] = b; | |
2982 | btrfs_clear_path_blocking(p, NULL, 0); | |
2983 | ||
2984 | /* | |
2985 | * we have a lock on b and as long as we aren't changing | |
2986 | * the tree, there is no way to for the items in b to change. | |
2987 | * It is safe to drop the lock on our parent before we | |
2988 | * go through the expensive btree search on b. | |
2989 | */ | |
2990 | btrfs_unlock_up_safe(p, level + 1); | |
2991 | ||
d4b4087c | 2992 | /* |
01327610 | 2993 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2994 | * time. |
2995 | */ | |
2996 | prev_cmp = -1; | |
d7396f07 | 2997 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
2998 | |
2999 | if (level != 0) { | |
3000 | int dec = 0; | |
3001 | if (ret && slot > 0) { | |
3002 | dec = 1; | |
3003 | slot -= 1; | |
3004 | } | |
3005 | p->slots[level] = slot; | |
3006 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3007 | ||
3008 | if (level == lowest_level) { | |
3009 | if (dec) | |
3010 | p->slots[level]++; | |
3011 | goto done; | |
3012 | } | |
3013 | ||
d07b8528 | 3014 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 3015 | slot, key); |
5d9e75c4 JS |
3016 | if (err == -EAGAIN) |
3017 | goto again; | |
3018 | if (err) { | |
3019 | ret = err; | |
3020 | goto done; | |
3021 | } | |
3022 | ||
3023 | level = btrfs_header_level(b); | |
f82c458a | 3024 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3025 | if (!err) { |
3026 | btrfs_set_path_blocking(p); | |
3027 | btrfs_tree_read_lock(b); | |
3028 | btrfs_clear_path_blocking(p, b, | |
3029 | BTRFS_READ_LOCK); | |
3030 | } | |
0b246afa | 3031 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3032 | if (!b) { |
3033 | ret = -ENOMEM; | |
3034 | goto done; | |
3035 | } | |
5d9e75c4 JS |
3036 | p->locks[level] = BTRFS_READ_LOCK; |
3037 | p->nodes[level] = b; | |
5d9e75c4 JS |
3038 | } else { |
3039 | p->slots[level] = slot; | |
3040 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3041 | goto done; | |
3042 | } | |
3043 | } | |
3044 | ret = 1; | |
3045 | done: | |
3046 | if (!p->leave_spinning) | |
3047 | btrfs_set_path_blocking(p); | |
3048 | if (ret < 0) | |
3049 | btrfs_release_path(p); | |
3050 | ||
3051 | return ret; | |
3052 | } | |
3053 | ||
2f38b3e1 AJ |
3054 | /* |
3055 | * helper to use instead of search slot if no exact match is needed but | |
3056 | * instead the next or previous item should be returned. | |
3057 | * When find_higher is true, the next higher item is returned, the next lower | |
3058 | * otherwise. | |
3059 | * When return_any and find_higher are both true, and no higher item is found, | |
3060 | * return the next lower instead. | |
3061 | * When return_any is true and find_higher is false, and no lower item is found, | |
3062 | * return the next higher instead. | |
3063 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3064 | * < 0 on error | |
3065 | */ | |
3066 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3067 | const struct btrfs_key *key, |
3068 | struct btrfs_path *p, int find_higher, | |
3069 | int return_any) | |
2f38b3e1 AJ |
3070 | { |
3071 | int ret; | |
3072 | struct extent_buffer *leaf; | |
3073 | ||
3074 | again: | |
3075 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3076 | if (ret <= 0) | |
3077 | return ret; | |
3078 | /* | |
3079 | * a return value of 1 means the path is at the position where the | |
3080 | * item should be inserted. Normally this is the next bigger item, | |
3081 | * but in case the previous item is the last in a leaf, path points | |
3082 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3083 | * item. | |
3084 | */ | |
3085 | leaf = p->nodes[0]; | |
3086 | ||
3087 | if (find_higher) { | |
3088 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3089 | ret = btrfs_next_leaf(root, p); | |
3090 | if (ret <= 0) | |
3091 | return ret; | |
3092 | if (!return_any) | |
3093 | return 1; | |
3094 | /* | |
3095 | * no higher item found, return the next | |
3096 | * lower instead | |
3097 | */ | |
3098 | return_any = 0; | |
3099 | find_higher = 0; | |
3100 | btrfs_release_path(p); | |
3101 | goto again; | |
3102 | } | |
3103 | } else { | |
e6793769 AJ |
3104 | if (p->slots[0] == 0) { |
3105 | ret = btrfs_prev_leaf(root, p); | |
3106 | if (ret < 0) | |
3107 | return ret; | |
3108 | if (!ret) { | |
23c6bf6a FDBM |
3109 | leaf = p->nodes[0]; |
3110 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3111 | p->slots[0]--; | |
e6793769 | 3112 | return 0; |
2f38b3e1 | 3113 | } |
e6793769 AJ |
3114 | if (!return_any) |
3115 | return 1; | |
3116 | /* | |
3117 | * no lower item found, return the next | |
3118 | * higher instead | |
3119 | */ | |
3120 | return_any = 0; | |
3121 | find_higher = 1; | |
3122 | btrfs_release_path(p); | |
3123 | goto again; | |
3124 | } else { | |
2f38b3e1 AJ |
3125 | --p->slots[0]; |
3126 | } | |
3127 | } | |
3128 | return 0; | |
3129 | } | |
3130 | ||
74123bd7 CM |
3131 | /* |
3132 | * adjust the pointers going up the tree, starting at level | |
3133 | * making sure the right key of each node is points to 'key'. | |
3134 | * This is used after shifting pointers to the left, so it stops | |
3135 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3136 | * higher levels | |
aa5d6bed | 3137 | * |
74123bd7 | 3138 | */ |
b7a0365e DD |
3139 | static void fixup_low_keys(struct btrfs_fs_info *fs_info, |
3140 | struct btrfs_path *path, | |
143bede5 | 3141 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3142 | { |
3143 | int i; | |
5f39d397 CM |
3144 | struct extent_buffer *t; |
3145 | ||
234b63a0 | 3146 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3147 | int tslot = path->slots[i]; |
eb60ceac | 3148 | if (!path->nodes[i]) |
be0e5c09 | 3149 | break; |
5f39d397 | 3150 | t = path->nodes[i]; |
3ac6de1a | 3151 | tree_mod_log_set_node_key(t, tslot, 1); |
5f39d397 | 3152 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3153 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3154 | if (tslot != 0) |
3155 | break; | |
3156 | } | |
3157 | } | |
3158 | ||
31840ae1 ZY |
3159 | /* |
3160 | * update item key. | |
3161 | * | |
3162 | * This function isn't completely safe. It's the caller's responsibility | |
3163 | * that the new key won't break the order | |
3164 | */ | |
b7a0365e DD |
3165 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3166 | struct btrfs_path *path, | |
310712b2 | 3167 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3168 | { |
3169 | struct btrfs_disk_key disk_key; | |
3170 | struct extent_buffer *eb; | |
3171 | int slot; | |
3172 | ||
3173 | eb = path->nodes[0]; | |
3174 | slot = path->slots[0]; | |
3175 | if (slot > 0) { | |
3176 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3177 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3178 | } |
3179 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3180 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3181 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3182 | } |
3183 | ||
3184 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3185 | btrfs_set_item_key(eb, &disk_key, slot); | |
3186 | btrfs_mark_buffer_dirty(eb); | |
3187 | if (slot == 0) | |
b7a0365e | 3188 | fixup_low_keys(fs_info, path, &disk_key, 1); |
31840ae1 ZY |
3189 | } |
3190 | ||
74123bd7 CM |
3191 | /* |
3192 | * try to push data from one node into the next node left in the | |
79f95c82 | 3193 | * tree. |
aa5d6bed CM |
3194 | * |
3195 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3196 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3197 | */ |
98ed5174 | 3198 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
3199 | struct btrfs_fs_info *fs_info, |
3200 | struct extent_buffer *dst, | |
971a1f66 | 3201 | struct extent_buffer *src, int empty) |
be0e5c09 | 3202 | { |
be0e5c09 | 3203 | int push_items = 0; |
bb803951 CM |
3204 | int src_nritems; |
3205 | int dst_nritems; | |
aa5d6bed | 3206 | int ret = 0; |
be0e5c09 | 3207 | |
5f39d397 CM |
3208 | src_nritems = btrfs_header_nritems(src); |
3209 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3210 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3211 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3212 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3213 | |
bce4eae9 | 3214 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3215 | return 1; |
3216 | ||
d397712b | 3217 | if (push_items <= 0) |
be0e5c09 CM |
3218 | return 1; |
3219 | ||
bce4eae9 | 3220 | if (empty) { |
971a1f66 | 3221 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3222 | if (push_items < src_nritems) { |
3223 | /* leave at least 8 pointers in the node if | |
3224 | * we aren't going to empty it | |
3225 | */ | |
3226 | if (src_nritems - push_items < 8) { | |
3227 | if (push_items <= 8) | |
3228 | return 1; | |
3229 | push_items -= 8; | |
3230 | } | |
3231 | } | |
3232 | } else | |
3233 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3234 | |
0b246afa | 3235 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3236 | push_items); |
3237 | if (ret) { | |
66642832 | 3238 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3239 | return ret; |
3240 | } | |
5f39d397 CM |
3241 | copy_extent_buffer(dst, src, |
3242 | btrfs_node_key_ptr_offset(dst_nritems), | |
3243 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3244 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3245 | |
bb803951 | 3246 | if (push_items < src_nritems) { |
57911b8b JS |
3247 | /* |
3248 | * don't call tree_mod_log_eb_move here, key removal was already | |
3249 | * fully logged by tree_mod_log_eb_copy above. | |
3250 | */ | |
5f39d397 CM |
3251 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3252 | btrfs_node_key_ptr_offset(push_items), | |
3253 | (src_nritems - push_items) * | |
3254 | sizeof(struct btrfs_key_ptr)); | |
3255 | } | |
3256 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3257 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3258 | btrfs_mark_buffer_dirty(src); | |
3259 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3260 | |
79f95c82 CM |
3261 | return ret; |
3262 | } | |
3263 | ||
3264 | /* | |
3265 | * try to push data from one node into the next node right in the | |
3266 | * tree. | |
3267 | * | |
3268 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3269 | * error, and > 0 if there was no room in the right hand block. | |
3270 | * | |
3271 | * this will only push up to 1/2 the contents of the left node over | |
3272 | */ | |
5f39d397 | 3273 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 3274 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
3275 | struct extent_buffer *dst, |
3276 | struct extent_buffer *src) | |
79f95c82 | 3277 | { |
79f95c82 CM |
3278 | int push_items = 0; |
3279 | int max_push; | |
3280 | int src_nritems; | |
3281 | int dst_nritems; | |
3282 | int ret = 0; | |
79f95c82 | 3283 | |
7bb86316 CM |
3284 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3285 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3286 | ||
5f39d397 CM |
3287 | src_nritems = btrfs_header_nritems(src); |
3288 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3289 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3290 | if (push_items <= 0) |
79f95c82 | 3291 | return 1; |
bce4eae9 | 3292 | |
d397712b | 3293 | if (src_nritems < 4) |
bce4eae9 | 3294 | return 1; |
79f95c82 CM |
3295 | |
3296 | max_push = src_nritems / 2 + 1; | |
3297 | /* don't try to empty the node */ | |
d397712b | 3298 | if (max_push >= src_nritems) |
79f95c82 | 3299 | return 1; |
252c38f0 | 3300 | |
79f95c82 CM |
3301 | if (max_push < push_items) |
3302 | push_items = max_push; | |
3303 | ||
a446a979 | 3304 | tree_mod_log_eb_move(dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3305 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3306 | btrfs_node_key_ptr_offset(0), | |
3307 | (dst_nritems) * | |
3308 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3309 | |
0b246afa | 3310 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3311 | src_nritems - push_items, push_items); |
3312 | if (ret) { | |
66642832 | 3313 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3314 | return ret; |
3315 | } | |
5f39d397 CM |
3316 | copy_extent_buffer(dst, src, |
3317 | btrfs_node_key_ptr_offset(0), | |
3318 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3319 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3320 | |
5f39d397 CM |
3321 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3322 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3323 | |
5f39d397 CM |
3324 | btrfs_mark_buffer_dirty(src); |
3325 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3326 | |
aa5d6bed | 3327 | return ret; |
be0e5c09 CM |
3328 | } |
3329 | ||
97571fd0 CM |
3330 | /* |
3331 | * helper function to insert a new root level in the tree. | |
3332 | * A new node is allocated, and a single item is inserted to | |
3333 | * point to the existing root | |
aa5d6bed CM |
3334 | * |
3335 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3336 | */ |
d397712b | 3337 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3338 | struct btrfs_root *root, |
fdd99c72 | 3339 | struct btrfs_path *path, int level) |
5c680ed6 | 3340 | { |
0b246afa | 3341 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3342 | u64 lower_gen; |
5f39d397 CM |
3343 | struct extent_buffer *lower; |
3344 | struct extent_buffer *c; | |
925baedd | 3345 | struct extent_buffer *old; |
5f39d397 | 3346 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3347 | |
3348 | BUG_ON(path->nodes[level]); | |
3349 | BUG_ON(path->nodes[level-1] != root->node); | |
3350 | ||
7bb86316 CM |
3351 | lower = path->nodes[level-1]; |
3352 | if (level == 1) | |
3353 | btrfs_item_key(lower, &lower_key, 0); | |
3354 | else | |
3355 | btrfs_node_key(lower, &lower_key, 0); | |
3356 | ||
4d75f8a9 DS |
3357 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3358 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3359 | if (IS_ERR(c)) |
3360 | return PTR_ERR(c); | |
925baedd | 3361 | |
0b246afa | 3362 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3363 | |
b159fa28 | 3364 | memzero_extent_buffer(c, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3365 | btrfs_set_header_nritems(c, 1); |
3366 | btrfs_set_header_level(c, level); | |
db94535d | 3367 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3368 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3369 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3370 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3371 | |
0b246afa JM |
3372 | write_extent_buffer_fsid(c, fs_info->fsid); |
3373 | write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid); | |
e17cade2 | 3374 | |
5f39d397 | 3375 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3376 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3377 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3378 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3379 | |
3380 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3381 | |
5f39d397 | 3382 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3383 | |
925baedd | 3384 | old = root->node; |
fdd99c72 | 3385 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3386 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3387 | |
3388 | /* the super has an extra ref to root->node */ | |
3389 | free_extent_buffer(old); | |
3390 | ||
0b86a832 | 3391 | add_root_to_dirty_list(root); |
5f39d397 CM |
3392 | extent_buffer_get(c); |
3393 | path->nodes[level] = c; | |
95449a16 | 3394 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3395 | path->slots[level] = 0; |
3396 | return 0; | |
3397 | } | |
3398 | ||
74123bd7 CM |
3399 | /* |
3400 | * worker function to insert a single pointer in a node. | |
3401 | * the node should have enough room for the pointer already | |
97571fd0 | 3402 | * |
74123bd7 CM |
3403 | * slot and level indicate where you want the key to go, and |
3404 | * blocknr is the block the key points to. | |
3405 | */ | |
143bede5 | 3406 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2ff7e61e | 3407 | struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 3408 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3409 | int slot, int level) |
74123bd7 | 3410 | { |
5f39d397 | 3411 | struct extent_buffer *lower; |
74123bd7 | 3412 | int nritems; |
f3ea38da | 3413 | int ret; |
5c680ed6 CM |
3414 | |
3415 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3416 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3417 | lower = path->nodes[level]; |
3418 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3419 | BUG_ON(slot > nritems); |
0b246afa | 3420 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3421 | if (slot != nritems) { |
c3e06965 | 3422 | if (level) |
a446a979 DS |
3423 | tree_mod_log_eb_move(lower, slot + 1, slot, |
3424 | nritems - slot); | |
5f39d397 CM |
3425 | memmove_extent_buffer(lower, |
3426 | btrfs_node_key_ptr_offset(slot + 1), | |
3427 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3428 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3429 | } |
c3e06965 | 3430 | if (level) { |
e09c2efe DS |
3431 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3432 | GFP_NOFS); | |
f3ea38da JS |
3433 | BUG_ON(ret < 0); |
3434 | } | |
5f39d397 | 3435 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3436 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3437 | WARN_ON(trans->transid == 0); |
3438 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3439 | btrfs_set_header_nritems(lower, nritems + 1); |
3440 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3441 | } |
3442 | ||
97571fd0 CM |
3443 | /* |
3444 | * split the node at the specified level in path in two. | |
3445 | * The path is corrected to point to the appropriate node after the split | |
3446 | * | |
3447 | * Before splitting this tries to make some room in the node by pushing | |
3448 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3449 | * |
3450 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3451 | */ |
e02119d5 CM |
3452 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3453 | struct btrfs_root *root, | |
3454 | struct btrfs_path *path, int level) | |
be0e5c09 | 3455 | { |
0b246afa | 3456 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3457 | struct extent_buffer *c; |
3458 | struct extent_buffer *split; | |
3459 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3460 | int mid; |
5c680ed6 | 3461 | int ret; |
7518a238 | 3462 | u32 c_nritems; |
eb60ceac | 3463 | |
5f39d397 | 3464 | c = path->nodes[level]; |
7bb86316 | 3465 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3466 | if (c == root->node) { |
d9abbf1c | 3467 | /* |
90f8d62e JS |
3468 | * trying to split the root, lets make a new one |
3469 | * | |
fdd99c72 | 3470 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3471 | * insert_new_root, because that root buffer will be kept as a |
3472 | * normal node. We are going to log removal of half of the | |
3473 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3474 | * tree lock on the buffer, which is why we cannot race with | |
3475 | * other tree_mod_log users. | |
d9abbf1c | 3476 | */ |
fdd99c72 | 3477 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3478 | if (ret) |
3479 | return ret; | |
b3612421 | 3480 | } else { |
e66f709b | 3481 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3482 | c = path->nodes[level]; |
3483 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3484 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3485 | return 0; |
54aa1f4d CM |
3486 | if (ret < 0) |
3487 | return ret; | |
be0e5c09 | 3488 | } |
e66f709b | 3489 | |
5f39d397 | 3490 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3491 | mid = (c_nritems + 1) / 2; |
3492 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3493 | |
4d75f8a9 DS |
3494 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3495 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3496 | if (IS_ERR(split)) |
3497 | return PTR_ERR(split); | |
3498 | ||
0b246afa | 3499 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3500 | |
b159fa28 | 3501 | memzero_extent_buffer(split, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3502 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3503 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3504 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3505 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3506 | btrfs_set_header_owner(split, root->root_key.objectid); |
0b246afa JM |
3507 | write_extent_buffer_fsid(split, fs_info->fsid); |
3508 | write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid); | |
54aa1f4d | 3509 | |
0b246afa | 3510 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3511 | if (ret) { |
66642832 | 3512 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3513 | return ret; |
3514 | } | |
5f39d397 CM |
3515 | copy_extent_buffer(split, c, |
3516 | btrfs_node_key_ptr_offset(0), | |
3517 | btrfs_node_key_ptr_offset(mid), | |
3518 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3519 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3520 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3521 | ret = 0; |
3522 | ||
5f39d397 CM |
3523 | btrfs_mark_buffer_dirty(c); |
3524 | btrfs_mark_buffer_dirty(split); | |
3525 | ||
2ff7e61e | 3526 | insert_ptr(trans, fs_info, path, &disk_key, split->start, |
c3e06965 | 3527 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3528 | |
5de08d7d | 3529 | if (path->slots[level] >= mid) { |
5c680ed6 | 3530 | path->slots[level] -= mid; |
925baedd | 3531 | btrfs_tree_unlock(c); |
5f39d397 CM |
3532 | free_extent_buffer(c); |
3533 | path->nodes[level] = split; | |
5c680ed6 CM |
3534 | path->slots[level + 1] += 1; |
3535 | } else { | |
925baedd | 3536 | btrfs_tree_unlock(split); |
5f39d397 | 3537 | free_extent_buffer(split); |
be0e5c09 | 3538 | } |
aa5d6bed | 3539 | return ret; |
be0e5c09 CM |
3540 | } |
3541 | ||
74123bd7 CM |
3542 | /* |
3543 | * how many bytes are required to store the items in a leaf. start | |
3544 | * and nr indicate which items in the leaf to check. This totals up the | |
3545 | * space used both by the item structs and the item data | |
3546 | */ | |
5f39d397 | 3547 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3548 | { |
41be1f3b JB |
3549 | struct btrfs_item *start_item; |
3550 | struct btrfs_item *end_item; | |
3551 | struct btrfs_map_token token; | |
be0e5c09 | 3552 | int data_len; |
5f39d397 | 3553 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3554 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3555 | |
3556 | if (!nr) | |
3557 | return 0; | |
41be1f3b | 3558 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3559 | start_item = btrfs_item_nr(start); |
3560 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3561 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3562 | btrfs_token_item_size(l, start_item, &token); | |
3563 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3564 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3565 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3566 | return data_len; |
3567 | } | |
3568 | ||
d4dbff95 CM |
3569 | /* |
3570 | * The space between the end of the leaf items and | |
3571 | * the start of the leaf data. IOW, how much room | |
3572 | * the leaf has left for both items and data | |
3573 | */ | |
2ff7e61e | 3574 | noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info, |
e02119d5 | 3575 | struct extent_buffer *leaf) |
d4dbff95 | 3576 | { |
5f39d397 CM |
3577 | int nritems = btrfs_header_nritems(leaf); |
3578 | int ret; | |
0b246afa JM |
3579 | |
3580 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3581 | if (ret < 0) { |
0b246afa JM |
3582 | btrfs_crit(fs_info, |
3583 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3584 | ret, | |
3585 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3586 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3587 | } |
3588 | return ret; | |
d4dbff95 CM |
3589 | } |
3590 | ||
99d8f83c CM |
3591 | /* |
3592 | * min slot controls the lowest index we're willing to push to the | |
3593 | * right. We'll push up to and including min_slot, but no lower | |
3594 | */ | |
1e47eef2 | 3595 | static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3596 | struct btrfs_path *path, |
3597 | int data_size, int empty, | |
3598 | struct extent_buffer *right, | |
99d8f83c CM |
3599 | int free_space, u32 left_nritems, |
3600 | u32 min_slot) | |
00ec4c51 | 3601 | { |
5f39d397 | 3602 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3603 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3604 | struct btrfs_map_token token; |
5f39d397 | 3605 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3606 | int slot; |
34a38218 | 3607 | u32 i; |
00ec4c51 CM |
3608 | int push_space = 0; |
3609 | int push_items = 0; | |
0783fcfc | 3610 | struct btrfs_item *item; |
34a38218 | 3611 | u32 nr; |
7518a238 | 3612 | u32 right_nritems; |
5f39d397 | 3613 | u32 data_end; |
db94535d | 3614 | u32 this_item_size; |
00ec4c51 | 3615 | |
cfed81a0 CM |
3616 | btrfs_init_map_token(&token); |
3617 | ||
34a38218 CM |
3618 | if (empty) |
3619 | nr = 0; | |
3620 | else | |
99d8f83c | 3621 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3622 | |
31840ae1 | 3623 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3624 | push_space += data_size; |
31840ae1 | 3625 | |
44871b1b | 3626 | slot = path->slots[1]; |
34a38218 CM |
3627 | i = left_nritems - 1; |
3628 | while (i >= nr) { | |
dd3cc16b | 3629 | item = btrfs_item_nr(i); |
db94535d | 3630 | |
31840ae1 ZY |
3631 | if (!empty && push_items > 0) { |
3632 | if (path->slots[0] > i) | |
3633 | break; | |
3634 | if (path->slots[0] == i) { | |
2ff7e61e | 3635 | int space = btrfs_leaf_free_space(fs_info, left); |
31840ae1 ZY |
3636 | if (space + push_space * 2 > free_space) |
3637 | break; | |
3638 | } | |
3639 | } | |
3640 | ||
00ec4c51 | 3641 | if (path->slots[0] == i) |
87b29b20 | 3642 | push_space += data_size; |
db94535d | 3643 | |
db94535d CM |
3644 | this_item_size = btrfs_item_size(left, item); |
3645 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3646 | break; |
31840ae1 | 3647 | |
00ec4c51 | 3648 | push_items++; |
db94535d | 3649 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3650 | if (i == 0) |
3651 | break; | |
3652 | i--; | |
db94535d | 3653 | } |
5f39d397 | 3654 | |
925baedd CM |
3655 | if (push_items == 0) |
3656 | goto out_unlock; | |
5f39d397 | 3657 | |
6c1500f2 | 3658 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3659 | |
00ec4c51 | 3660 | /* push left to right */ |
5f39d397 | 3661 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3662 | |
5f39d397 | 3663 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
2ff7e61e | 3664 | push_space -= leaf_data_end(fs_info, left); |
5f39d397 | 3665 | |
00ec4c51 | 3666 | /* make room in the right data area */ |
2ff7e61e | 3667 | data_end = leaf_data_end(fs_info, right); |
5f39d397 | 3668 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3669 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3670 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3671 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3672 | |
00ec4c51 | 3673 | /* copy from the left data area */ |
3d9ec8c4 | 3674 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3675 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3676 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left), |
d6025579 | 3677 | push_space); |
5f39d397 CM |
3678 | |
3679 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3680 | btrfs_item_nr_offset(0), | |
3681 | right_nritems * sizeof(struct btrfs_item)); | |
3682 | ||
00ec4c51 | 3683 | /* copy the items from left to right */ |
5f39d397 CM |
3684 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3685 | btrfs_item_nr_offset(left_nritems - push_items), | |
3686 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3687 | |
3688 | /* update the item pointers */ | |
7518a238 | 3689 | right_nritems += push_items; |
5f39d397 | 3690 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3691 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3692 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3693 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3694 | push_space -= btrfs_token_item_size(right, item, &token); |
3695 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3696 | } |
3697 | ||
7518a238 | 3698 | left_nritems -= push_items; |
5f39d397 | 3699 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3700 | |
34a38218 CM |
3701 | if (left_nritems) |
3702 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3703 | else |
7c302b49 | 3704 | clean_tree_block(fs_info, left); |
f0486c68 | 3705 | |
5f39d397 | 3706 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3707 | |
5f39d397 CM |
3708 | btrfs_item_key(right, &disk_key, 0); |
3709 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3710 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3711 | |
00ec4c51 | 3712 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3713 | if (path->slots[0] >= left_nritems) { |
3714 | path->slots[0] -= left_nritems; | |
925baedd | 3715 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
7c302b49 | 3716 | clean_tree_block(fs_info, path->nodes[0]); |
925baedd | 3717 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3718 | free_extent_buffer(path->nodes[0]); |
3719 | path->nodes[0] = right; | |
00ec4c51 CM |
3720 | path->slots[1] += 1; |
3721 | } else { | |
925baedd | 3722 | btrfs_tree_unlock(right); |
5f39d397 | 3723 | free_extent_buffer(right); |
00ec4c51 CM |
3724 | } |
3725 | return 0; | |
925baedd CM |
3726 | |
3727 | out_unlock: | |
3728 | btrfs_tree_unlock(right); | |
3729 | free_extent_buffer(right); | |
3730 | return 1; | |
00ec4c51 | 3731 | } |
925baedd | 3732 | |
44871b1b CM |
3733 | /* |
3734 | * push some data in the path leaf to the right, trying to free up at | |
3735 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3736 | * | |
3737 | * returns 1 if the push failed because the other node didn't have enough | |
3738 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3739 | * |
3740 | * this will push starting from min_slot to the end of the leaf. It won't | |
3741 | * push any slot lower than min_slot | |
44871b1b CM |
3742 | */ |
3743 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3744 | *root, struct btrfs_path *path, |
3745 | int min_data_size, int data_size, | |
3746 | int empty, u32 min_slot) | |
44871b1b | 3747 | { |
2ff7e61e | 3748 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3749 | struct extent_buffer *left = path->nodes[0]; |
3750 | struct extent_buffer *right; | |
3751 | struct extent_buffer *upper; | |
3752 | int slot; | |
3753 | int free_space; | |
3754 | u32 left_nritems; | |
3755 | int ret; | |
3756 | ||
3757 | if (!path->nodes[1]) | |
3758 | return 1; | |
3759 | ||
3760 | slot = path->slots[1]; | |
3761 | upper = path->nodes[1]; | |
3762 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3763 | return 1; | |
3764 | ||
3765 | btrfs_assert_tree_locked(path->nodes[1]); | |
3766 | ||
2ff7e61e | 3767 | right = read_node_slot(fs_info, upper, slot + 1); |
fb770ae4 LB |
3768 | /* |
3769 | * slot + 1 is not valid or we fail to read the right node, | |
3770 | * no big deal, just return. | |
3771 | */ | |
3772 | if (IS_ERR(right)) | |
91ca338d TI |
3773 | return 1; |
3774 | ||
44871b1b CM |
3775 | btrfs_tree_lock(right); |
3776 | btrfs_set_lock_blocking(right); | |
3777 | ||
2ff7e61e | 3778 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3779 | if (free_space < data_size) |
3780 | goto out_unlock; | |
3781 | ||
3782 | /* cow and double check */ | |
3783 | ret = btrfs_cow_block(trans, root, right, upper, | |
3784 | slot + 1, &right); | |
3785 | if (ret) | |
3786 | goto out_unlock; | |
3787 | ||
2ff7e61e | 3788 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3789 | if (free_space < data_size) |
3790 | goto out_unlock; | |
3791 | ||
3792 | left_nritems = btrfs_header_nritems(left); | |
3793 | if (left_nritems == 0) | |
3794 | goto out_unlock; | |
3795 | ||
2ef1fed2 FDBM |
3796 | if (path->slots[0] == left_nritems && !empty) { |
3797 | /* Key greater than all keys in the leaf, right neighbor has | |
3798 | * enough room for it and we're not emptying our leaf to delete | |
3799 | * it, therefore use right neighbor to insert the new item and | |
3800 | * no need to touch/dirty our left leaft. */ | |
3801 | btrfs_tree_unlock(left); | |
3802 | free_extent_buffer(left); | |
3803 | path->nodes[0] = right; | |
3804 | path->slots[0] = 0; | |
3805 | path->slots[1]++; | |
3806 | return 0; | |
3807 | } | |
3808 | ||
1e47eef2 | 3809 | return __push_leaf_right(fs_info, path, min_data_size, empty, |
99d8f83c | 3810 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3811 | out_unlock: |
3812 | btrfs_tree_unlock(right); | |
3813 | free_extent_buffer(right); | |
3814 | return 1; | |
3815 | } | |
3816 | ||
74123bd7 CM |
3817 | /* |
3818 | * push some data in the path leaf to the left, trying to free up at | |
3819 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3820 | * |
3821 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3822 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3823 | * items | |
74123bd7 | 3824 | */ |
66cb7ddb | 3825 | static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3826 | struct btrfs_path *path, int data_size, |
3827 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3828 | int free_space, u32 right_nritems, |
3829 | u32 max_slot) | |
be0e5c09 | 3830 | { |
5f39d397 CM |
3831 | struct btrfs_disk_key disk_key; |
3832 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3833 | int i; |
be0e5c09 CM |
3834 | int push_space = 0; |
3835 | int push_items = 0; | |
0783fcfc | 3836 | struct btrfs_item *item; |
7518a238 | 3837 | u32 old_left_nritems; |
34a38218 | 3838 | u32 nr; |
aa5d6bed | 3839 | int ret = 0; |
db94535d CM |
3840 | u32 this_item_size; |
3841 | u32 old_left_item_size; | |
cfed81a0 CM |
3842 | struct btrfs_map_token token; |
3843 | ||
3844 | btrfs_init_map_token(&token); | |
be0e5c09 | 3845 | |
34a38218 | 3846 | if (empty) |
99d8f83c | 3847 | nr = min(right_nritems, max_slot); |
34a38218 | 3848 | else |
99d8f83c | 3849 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3850 | |
3851 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3852 | item = btrfs_item_nr(i); |
db94535d | 3853 | |
31840ae1 ZY |
3854 | if (!empty && push_items > 0) { |
3855 | if (path->slots[0] < i) | |
3856 | break; | |
3857 | if (path->slots[0] == i) { | |
2ff7e61e | 3858 | int space = btrfs_leaf_free_space(fs_info, right); |
31840ae1 ZY |
3859 | if (space + push_space * 2 > free_space) |
3860 | break; | |
3861 | } | |
3862 | } | |
3863 | ||
be0e5c09 | 3864 | if (path->slots[0] == i) |
87b29b20 | 3865 | push_space += data_size; |
db94535d CM |
3866 | |
3867 | this_item_size = btrfs_item_size(right, item); | |
3868 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3869 | break; |
db94535d | 3870 | |
be0e5c09 | 3871 | push_items++; |
db94535d CM |
3872 | push_space += this_item_size + sizeof(*item); |
3873 | } | |
3874 | ||
be0e5c09 | 3875 | if (push_items == 0) { |
925baedd CM |
3876 | ret = 1; |
3877 | goto out; | |
be0e5c09 | 3878 | } |
fae7f21c | 3879 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3880 | |
be0e5c09 | 3881 | /* push data from right to left */ |
5f39d397 CM |
3882 | copy_extent_buffer(left, right, |
3883 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3884 | btrfs_item_nr_offset(0), | |
3885 | push_items * sizeof(struct btrfs_item)); | |
3886 | ||
0b246afa | 3887 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3888 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3889 | |
3d9ec8c4 | 3890 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3891 | leaf_data_end(fs_info, left) - push_space, |
3d9ec8c4 | 3892 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3893 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3894 | push_space); |
5f39d397 | 3895 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3896 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3897 | |
db94535d | 3898 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3899 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3900 | u32 ioff; |
db94535d | 3901 | |
dd3cc16b | 3902 | item = btrfs_item_nr(i); |
db94535d | 3903 | |
cfed81a0 CM |
3904 | ioff = btrfs_token_item_offset(left, item, &token); |
3905 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3906 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3907 | &token); |
be0e5c09 | 3908 | } |
5f39d397 | 3909 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3910 | |
3911 | /* fixup right node */ | |
31b1a2bd JL |
3912 | if (push_items > right_nritems) |
3913 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3914 | right_nritems); |
34a38218 CM |
3915 | |
3916 | if (push_items < right_nritems) { | |
3917 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2ff7e61e | 3918 | leaf_data_end(fs_info, right); |
3d9ec8c4 | 3919 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3920 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3921 | BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3922 | leaf_data_end(fs_info, right), push_space); |
34a38218 CM |
3923 | |
3924 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3925 | btrfs_item_nr_offset(push_items), |
3926 | (btrfs_header_nritems(right) - push_items) * | |
3927 | sizeof(struct btrfs_item)); | |
34a38218 | 3928 | } |
eef1c494 Y |
3929 | right_nritems -= push_items; |
3930 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3931 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3932 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3933 | item = btrfs_item_nr(i); |
db94535d | 3934 | |
cfed81a0 CM |
3935 | push_space = push_space - btrfs_token_item_size(right, |
3936 | item, &token); | |
3937 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3938 | } |
eb60ceac | 3939 | |
5f39d397 | 3940 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3941 | if (right_nritems) |
3942 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3943 | else |
7c302b49 | 3944 | clean_tree_block(fs_info, right); |
098f59c2 | 3945 | |
5f39d397 | 3946 | btrfs_item_key(right, &disk_key, 0); |
0b246afa | 3947 | fixup_low_keys(fs_info, path, &disk_key, 1); |
be0e5c09 CM |
3948 | |
3949 | /* then fixup the leaf pointer in the path */ | |
3950 | if (path->slots[0] < push_items) { | |
3951 | path->slots[0] += old_left_nritems; | |
925baedd | 3952 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3953 | free_extent_buffer(path->nodes[0]); |
3954 | path->nodes[0] = left; | |
be0e5c09 CM |
3955 | path->slots[1] -= 1; |
3956 | } else { | |
925baedd | 3957 | btrfs_tree_unlock(left); |
5f39d397 | 3958 | free_extent_buffer(left); |
be0e5c09 CM |
3959 | path->slots[0] -= push_items; |
3960 | } | |
eb60ceac | 3961 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3962 | return ret; |
925baedd CM |
3963 | out: |
3964 | btrfs_tree_unlock(left); | |
3965 | free_extent_buffer(left); | |
3966 | return ret; | |
be0e5c09 CM |
3967 | } |
3968 | ||
44871b1b CM |
3969 | /* |
3970 | * push some data in the path leaf to the left, trying to free up at | |
3971 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3972 | * |
3973 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3974 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3975 | * items | |
44871b1b CM |
3976 | */ |
3977 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3978 | *root, struct btrfs_path *path, int min_data_size, |
3979 | int data_size, int empty, u32 max_slot) | |
44871b1b | 3980 | { |
2ff7e61e | 3981 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3982 | struct extent_buffer *right = path->nodes[0]; |
3983 | struct extent_buffer *left; | |
3984 | int slot; | |
3985 | int free_space; | |
3986 | u32 right_nritems; | |
3987 | int ret = 0; | |
3988 | ||
3989 | slot = path->slots[1]; | |
3990 | if (slot == 0) | |
3991 | return 1; | |
3992 | if (!path->nodes[1]) | |
3993 | return 1; | |
3994 | ||
3995 | right_nritems = btrfs_header_nritems(right); | |
3996 | if (right_nritems == 0) | |
3997 | return 1; | |
3998 | ||
3999 | btrfs_assert_tree_locked(path->nodes[1]); | |
4000 | ||
2ff7e61e | 4001 | left = read_node_slot(fs_info, path->nodes[1], slot - 1); |
fb770ae4 LB |
4002 | /* |
4003 | * slot - 1 is not valid or we fail to read the left node, | |
4004 | * no big deal, just return. | |
4005 | */ | |
4006 | if (IS_ERR(left)) | |
91ca338d TI |
4007 | return 1; |
4008 | ||
44871b1b CM |
4009 | btrfs_tree_lock(left); |
4010 | btrfs_set_lock_blocking(left); | |
4011 | ||
2ff7e61e | 4012 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4013 | if (free_space < data_size) { |
4014 | ret = 1; | |
4015 | goto out; | |
4016 | } | |
4017 | ||
4018 | /* cow and double check */ | |
4019 | ret = btrfs_cow_block(trans, root, left, | |
4020 | path->nodes[1], slot - 1, &left); | |
4021 | if (ret) { | |
4022 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4023 | if (ret == -ENOSPC) |
4024 | ret = 1; | |
44871b1b CM |
4025 | goto out; |
4026 | } | |
4027 | ||
2ff7e61e | 4028 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4029 | if (free_space < data_size) { |
4030 | ret = 1; | |
4031 | goto out; | |
4032 | } | |
4033 | ||
66cb7ddb | 4034 | return __push_leaf_left(fs_info, path, min_data_size, |
99d8f83c CM |
4035 | empty, left, free_space, right_nritems, |
4036 | max_slot); | |
44871b1b CM |
4037 | out: |
4038 | btrfs_tree_unlock(left); | |
4039 | free_extent_buffer(left); | |
4040 | return ret; | |
4041 | } | |
4042 | ||
4043 | /* | |
4044 | * split the path's leaf in two, making sure there is at least data_size | |
4045 | * available for the resulting leaf level of the path. | |
44871b1b | 4046 | */ |
143bede5 | 4047 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2ff7e61e | 4048 | struct btrfs_fs_info *fs_info, |
143bede5 JM |
4049 | struct btrfs_path *path, |
4050 | struct extent_buffer *l, | |
4051 | struct extent_buffer *right, | |
4052 | int slot, int mid, int nritems) | |
44871b1b CM |
4053 | { |
4054 | int data_copy_size; | |
4055 | int rt_data_off; | |
4056 | int i; | |
44871b1b | 4057 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4058 | struct btrfs_map_token token; |
4059 | ||
4060 | btrfs_init_map_token(&token); | |
44871b1b CM |
4061 | |
4062 | nritems = nritems - mid; | |
4063 | btrfs_set_header_nritems(right, nritems); | |
2ff7e61e | 4064 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l); |
44871b1b CM |
4065 | |
4066 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4067 | btrfs_item_nr_offset(mid), | |
4068 | nritems * sizeof(struct btrfs_item)); | |
4069 | ||
4070 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4071 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4072 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
2ff7e61e | 4073 | leaf_data_end(fs_info, l), data_copy_size); |
44871b1b | 4074 | |
0b246afa | 4075 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4076 | |
4077 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4078 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4079 | u32 ioff; |
4080 | ||
cfed81a0 CM |
4081 | ioff = btrfs_token_item_offset(right, item, &token); |
4082 | btrfs_set_token_item_offset(right, item, | |
4083 | ioff + rt_data_off, &token); | |
44871b1b CM |
4084 | } |
4085 | ||
44871b1b | 4086 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4087 | btrfs_item_key(right, &disk_key, 0); |
2ff7e61e | 4088 | insert_ptr(trans, fs_info, path, &disk_key, right->start, |
c3e06965 | 4089 | path->slots[1] + 1, 1); |
44871b1b CM |
4090 | |
4091 | btrfs_mark_buffer_dirty(right); | |
4092 | btrfs_mark_buffer_dirty(l); | |
4093 | BUG_ON(path->slots[0] != slot); | |
4094 | ||
44871b1b CM |
4095 | if (mid <= slot) { |
4096 | btrfs_tree_unlock(path->nodes[0]); | |
4097 | free_extent_buffer(path->nodes[0]); | |
4098 | path->nodes[0] = right; | |
4099 | path->slots[0] -= mid; | |
4100 | path->slots[1] += 1; | |
4101 | } else { | |
4102 | btrfs_tree_unlock(right); | |
4103 | free_extent_buffer(right); | |
4104 | } | |
4105 | ||
4106 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4107 | } |
4108 | ||
99d8f83c CM |
4109 | /* |
4110 | * double splits happen when we need to insert a big item in the middle | |
4111 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4112 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4113 | * A B C | |
4114 | * | |
4115 | * We avoid this by trying to push the items on either side of our target | |
4116 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4117 | * completely. | |
4118 | */ | |
4119 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4120 | struct btrfs_root *root, | |
4121 | struct btrfs_path *path, | |
4122 | int data_size) | |
4123 | { | |
2ff7e61e | 4124 | struct btrfs_fs_info *fs_info = root->fs_info; |
99d8f83c CM |
4125 | int ret; |
4126 | int progress = 0; | |
4127 | int slot; | |
4128 | u32 nritems; | |
5a4267ca | 4129 | int space_needed = data_size; |
99d8f83c CM |
4130 | |
4131 | slot = path->slots[0]; | |
5a4267ca | 4132 | if (slot < btrfs_header_nritems(path->nodes[0])) |
2ff7e61e | 4133 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); |
99d8f83c CM |
4134 | |
4135 | /* | |
4136 | * try to push all the items after our slot into the | |
4137 | * right leaf | |
4138 | */ | |
5a4267ca | 4139 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4140 | if (ret < 0) |
4141 | return ret; | |
4142 | ||
4143 | if (ret == 0) | |
4144 | progress++; | |
4145 | ||
4146 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4147 | /* | |
4148 | * our goal is to get our slot at the start or end of a leaf. If | |
4149 | * we've done so we're done | |
4150 | */ | |
4151 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4152 | return 0; | |
4153 | ||
2ff7e61e | 4154 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4155 | return 0; |
4156 | ||
4157 | /* try to push all the items before our slot into the next leaf */ | |
4158 | slot = path->slots[0]; | |
263d3995 FM |
4159 | space_needed = data_size; |
4160 | if (slot > 0) | |
4161 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); | |
5a4267ca | 4162 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4163 | if (ret < 0) |
4164 | return ret; | |
4165 | ||
4166 | if (ret == 0) | |
4167 | progress++; | |
4168 | ||
4169 | if (progress) | |
4170 | return 0; | |
4171 | return 1; | |
4172 | } | |
4173 | ||
74123bd7 CM |
4174 | /* |
4175 | * split the path's leaf in two, making sure there is at least data_size | |
4176 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4177 | * |
4178 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4179 | */ |
e02119d5 CM |
4180 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4181 | struct btrfs_root *root, | |
310712b2 | 4182 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4183 | struct btrfs_path *path, int data_size, |
4184 | int extend) | |
be0e5c09 | 4185 | { |
5d4f98a2 | 4186 | struct btrfs_disk_key disk_key; |
5f39d397 | 4187 | struct extent_buffer *l; |
7518a238 | 4188 | u32 nritems; |
eb60ceac CM |
4189 | int mid; |
4190 | int slot; | |
5f39d397 | 4191 | struct extent_buffer *right; |
b7a0365e | 4192 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4193 | int ret = 0; |
aa5d6bed | 4194 | int wret; |
5d4f98a2 | 4195 | int split; |
cc0c5538 | 4196 | int num_doubles = 0; |
99d8f83c | 4197 | int tried_avoid_double = 0; |
aa5d6bed | 4198 | |
a5719521 YZ |
4199 | l = path->nodes[0]; |
4200 | slot = path->slots[0]; | |
4201 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4202 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4203 | return -EOVERFLOW; |
4204 | ||
40689478 | 4205 | /* first try to make some room by pushing left and right */ |
33157e05 | 4206 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4207 | int space_needed = data_size; |
4208 | ||
4209 | if (slot < btrfs_header_nritems(l)) | |
2ff7e61e | 4210 | space_needed -= btrfs_leaf_free_space(fs_info, l); |
5a4267ca FDBM |
4211 | |
4212 | wret = push_leaf_right(trans, root, path, space_needed, | |
4213 | space_needed, 0, 0); | |
d397712b | 4214 | if (wret < 0) |
eaee50e8 | 4215 | return wret; |
3685f791 | 4216 | if (wret) { |
263d3995 FM |
4217 | space_needed = data_size; |
4218 | if (slot > 0) | |
4219 | space_needed -= btrfs_leaf_free_space(fs_info, | |
4220 | l); | |
5a4267ca FDBM |
4221 | wret = push_leaf_left(trans, root, path, space_needed, |
4222 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4223 | if (wret < 0) |
4224 | return wret; | |
4225 | } | |
4226 | l = path->nodes[0]; | |
aa5d6bed | 4227 | |
3685f791 | 4228 | /* did the pushes work? */ |
2ff7e61e | 4229 | if (btrfs_leaf_free_space(fs_info, l) >= data_size) |
3685f791 | 4230 | return 0; |
3326d1b0 | 4231 | } |
aa5d6bed | 4232 | |
5c680ed6 | 4233 | if (!path->nodes[1]) { |
fdd99c72 | 4234 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4235 | if (ret) |
4236 | return ret; | |
4237 | } | |
cc0c5538 | 4238 | again: |
5d4f98a2 | 4239 | split = 1; |
cc0c5538 | 4240 | l = path->nodes[0]; |
eb60ceac | 4241 | slot = path->slots[0]; |
5f39d397 | 4242 | nritems = btrfs_header_nritems(l); |
d397712b | 4243 | mid = (nritems + 1) / 2; |
54aa1f4d | 4244 | |
5d4f98a2 YZ |
4245 | if (mid <= slot) { |
4246 | if (nritems == 1 || | |
4247 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4248 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4249 | if (slot >= nritems) { |
4250 | split = 0; | |
4251 | } else { | |
4252 | mid = slot; | |
4253 | if (mid != nritems && | |
4254 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4255 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4256 | if (data_size && !tried_avoid_double) |
4257 | goto push_for_double; | |
5d4f98a2 YZ |
4258 | split = 2; |
4259 | } | |
4260 | } | |
4261 | } | |
4262 | } else { | |
4263 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4264 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4265 | if (!extend && data_size && slot == 0) { |
4266 | split = 0; | |
4267 | } else if ((extend || !data_size) && slot == 0) { | |
4268 | mid = 1; | |
4269 | } else { | |
4270 | mid = slot; | |
4271 | if (mid != nritems && | |
4272 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4273 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4274 | if (data_size && !tried_avoid_double) |
4275 | goto push_for_double; | |
67871254 | 4276 | split = 2; |
5d4f98a2 YZ |
4277 | } |
4278 | } | |
4279 | } | |
4280 | } | |
4281 | ||
4282 | if (split == 0) | |
4283 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4284 | else | |
4285 | btrfs_item_key(l, &disk_key, mid); | |
4286 | ||
4d75f8a9 DS |
4287 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4288 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4289 | if (IS_ERR(right)) |
5f39d397 | 4290 | return PTR_ERR(right); |
f0486c68 | 4291 | |
0b246afa | 4292 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4293 | |
b159fa28 | 4294 | memzero_extent_buffer(right, 0, sizeof(struct btrfs_header)); |
db94535d | 4295 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4296 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4297 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4298 | btrfs_set_header_owner(right, root->root_key.objectid); |
4299 | btrfs_set_header_level(right, 0); | |
d24ee97b DS |
4300 | write_extent_buffer_fsid(right, fs_info->fsid); |
4301 | write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid); | |
44871b1b | 4302 | |
5d4f98a2 YZ |
4303 | if (split == 0) { |
4304 | if (mid <= slot) { | |
4305 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4306 | insert_ptr(trans, fs_info, path, &disk_key, |
4307 | right->start, path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
4308 | btrfs_tree_unlock(path->nodes[0]); |
4309 | free_extent_buffer(path->nodes[0]); | |
4310 | path->nodes[0] = right; | |
4311 | path->slots[0] = 0; | |
4312 | path->slots[1] += 1; | |
4313 | } else { | |
4314 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4315 | insert_ptr(trans, fs_info, path, &disk_key, |
4316 | right->start, path->slots[1], 1); | |
5d4f98a2 YZ |
4317 | btrfs_tree_unlock(path->nodes[0]); |
4318 | free_extent_buffer(path->nodes[0]); | |
4319 | path->nodes[0] = right; | |
4320 | path->slots[0] = 0; | |
143bede5 | 4321 | if (path->slots[1] == 0) |
b7a0365e | 4322 | fixup_low_keys(fs_info, path, &disk_key, 1); |
d4dbff95 | 4323 | } |
196e0249 LB |
4324 | /* |
4325 | * We create a new leaf 'right' for the required ins_len and | |
4326 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4327 | * the content of ins_len to 'right'. | |
4328 | */ | |
5d4f98a2 | 4329 | return ret; |
d4dbff95 | 4330 | } |
74123bd7 | 4331 | |
2ff7e61e | 4332 | copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems); |
31840ae1 | 4333 | |
5d4f98a2 | 4334 | if (split == 2) { |
cc0c5538 CM |
4335 | BUG_ON(num_doubles != 0); |
4336 | num_doubles++; | |
4337 | goto again; | |
a429e513 | 4338 | } |
44871b1b | 4339 | |
143bede5 | 4340 | return 0; |
99d8f83c CM |
4341 | |
4342 | push_for_double: | |
4343 | push_for_double_split(trans, root, path, data_size); | |
4344 | tried_avoid_double = 1; | |
2ff7e61e | 4345 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4346 | return 0; |
4347 | goto again; | |
be0e5c09 CM |
4348 | } |
4349 | ||
ad48fd75 YZ |
4350 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4351 | struct btrfs_root *root, | |
4352 | struct btrfs_path *path, int ins_len) | |
459931ec | 4353 | { |
2ff7e61e | 4354 | struct btrfs_fs_info *fs_info = root->fs_info; |
ad48fd75 | 4355 | struct btrfs_key key; |
459931ec | 4356 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4357 | struct btrfs_file_extent_item *fi; |
4358 | u64 extent_len = 0; | |
4359 | u32 item_size; | |
4360 | int ret; | |
459931ec CM |
4361 | |
4362 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4363 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4364 | ||
4365 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4366 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4367 | ||
2ff7e61e | 4368 | if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len) |
ad48fd75 | 4369 | return 0; |
459931ec CM |
4370 | |
4371 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4372 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4373 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4374 | struct btrfs_file_extent_item); | |
4375 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4376 | } | |
b3b4aa74 | 4377 | btrfs_release_path(path); |
459931ec | 4378 | |
459931ec | 4379 | path->keep_locks = 1; |
ad48fd75 YZ |
4380 | path->search_for_split = 1; |
4381 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4382 | path->search_for_split = 0; |
a8df6fe6 FM |
4383 | if (ret > 0) |
4384 | ret = -EAGAIN; | |
ad48fd75 YZ |
4385 | if (ret < 0) |
4386 | goto err; | |
459931ec | 4387 | |
ad48fd75 YZ |
4388 | ret = -EAGAIN; |
4389 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4390 | /* if our item isn't there, return now */ |
4391 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4392 | goto err; |
4393 | ||
109f6aef | 4394 | /* the leaf has changed, it now has room. return now */ |
2ff7e61e | 4395 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len) |
109f6aef CM |
4396 | goto err; |
4397 | ||
ad48fd75 YZ |
4398 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4399 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4400 | struct btrfs_file_extent_item); | |
4401 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4402 | goto err; | |
459931ec CM |
4403 | } |
4404 | ||
b9473439 | 4405 | btrfs_set_path_blocking(path); |
ad48fd75 | 4406 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4407 | if (ret) |
4408 | goto err; | |
459931ec | 4409 | |
ad48fd75 | 4410 | path->keep_locks = 0; |
b9473439 | 4411 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4412 | return 0; |
4413 | err: | |
4414 | path->keep_locks = 0; | |
4415 | return ret; | |
4416 | } | |
4417 | ||
4961e293 | 4418 | static noinline int split_item(struct btrfs_fs_info *fs_info, |
ad48fd75 | 4419 | struct btrfs_path *path, |
310712b2 | 4420 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4421 | unsigned long split_offset) |
4422 | { | |
4423 | struct extent_buffer *leaf; | |
4424 | struct btrfs_item *item; | |
4425 | struct btrfs_item *new_item; | |
4426 | int slot; | |
4427 | char *buf; | |
4428 | u32 nritems; | |
4429 | u32 item_size; | |
4430 | u32 orig_offset; | |
4431 | struct btrfs_disk_key disk_key; | |
4432 | ||
b9473439 | 4433 | leaf = path->nodes[0]; |
2ff7e61e | 4434 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4435 | |
b4ce94de CM |
4436 | btrfs_set_path_blocking(path); |
4437 | ||
dd3cc16b | 4438 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4439 | orig_offset = btrfs_item_offset(leaf, item); |
4440 | item_size = btrfs_item_size(leaf, item); | |
4441 | ||
459931ec | 4442 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4443 | if (!buf) |
4444 | return -ENOMEM; | |
4445 | ||
459931ec CM |
4446 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4447 | path->slots[0]), item_size); | |
459931ec | 4448 | |
ad48fd75 | 4449 | slot = path->slots[0] + 1; |
459931ec | 4450 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4451 | if (slot != nritems) { |
4452 | /* shift the items */ | |
4453 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4454 | btrfs_item_nr_offset(slot), |
4455 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4456 | } |
4457 | ||
4458 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4459 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4460 | ||
dd3cc16b | 4461 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4462 | |
4463 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4464 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4465 | ||
4466 | btrfs_set_item_offset(leaf, item, | |
4467 | orig_offset + item_size - split_offset); | |
4468 | btrfs_set_item_size(leaf, item, split_offset); | |
4469 | ||
4470 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4471 | ||
4472 | /* write the data for the start of the original item */ | |
4473 | write_extent_buffer(leaf, buf, | |
4474 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4475 | split_offset); | |
4476 | ||
4477 | /* write the data for the new item */ | |
4478 | write_extent_buffer(leaf, buf + split_offset, | |
4479 | btrfs_item_ptr_offset(leaf, slot), | |
4480 | item_size - split_offset); | |
4481 | btrfs_mark_buffer_dirty(leaf); | |
4482 | ||
2ff7e61e | 4483 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0); |
459931ec | 4484 | kfree(buf); |
ad48fd75 YZ |
4485 | return 0; |
4486 | } | |
4487 | ||
4488 | /* | |
4489 | * This function splits a single item into two items, | |
4490 | * giving 'new_key' to the new item and splitting the | |
4491 | * old one at split_offset (from the start of the item). | |
4492 | * | |
4493 | * The path may be released by this operation. After | |
4494 | * the split, the path is pointing to the old item. The | |
4495 | * new item is going to be in the same node as the old one. | |
4496 | * | |
4497 | * Note, the item being split must be smaller enough to live alone on | |
4498 | * a tree block with room for one extra struct btrfs_item | |
4499 | * | |
4500 | * This allows us to split the item in place, keeping a lock on the | |
4501 | * leaf the entire time. | |
4502 | */ | |
4503 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4504 | struct btrfs_root *root, | |
4505 | struct btrfs_path *path, | |
310712b2 | 4506 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4507 | unsigned long split_offset) |
4508 | { | |
4509 | int ret; | |
4510 | ret = setup_leaf_for_split(trans, root, path, | |
4511 | sizeof(struct btrfs_item)); | |
4512 | if (ret) | |
4513 | return ret; | |
4514 | ||
4961e293 | 4515 | ret = split_item(root->fs_info, path, new_key, split_offset); |
459931ec CM |
4516 | return ret; |
4517 | } | |
4518 | ||
ad48fd75 YZ |
4519 | /* |
4520 | * This function duplicate a item, giving 'new_key' to the new item. | |
4521 | * It guarantees both items live in the same tree leaf and the new item | |
4522 | * is contiguous with the original item. | |
4523 | * | |
4524 | * This allows us to split file extent in place, keeping a lock on the | |
4525 | * leaf the entire time. | |
4526 | */ | |
4527 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4528 | struct btrfs_root *root, | |
4529 | struct btrfs_path *path, | |
310712b2 | 4530 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4531 | { |
4532 | struct extent_buffer *leaf; | |
4533 | int ret; | |
4534 | u32 item_size; | |
4535 | ||
4536 | leaf = path->nodes[0]; | |
4537 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4538 | ret = setup_leaf_for_split(trans, root, path, | |
4539 | item_size + sizeof(struct btrfs_item)); | |
4540 | if (ret) | |
4541 | return ret; | |
4542 | ||
4543 | path->slots[0]++; | |
afe5fea7 | 4544 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4545 | item_size, item_size + |
4546 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4547 | leaf = path->nodes[0]; |
4548 | memcpy_extent_buffer(leaf, | |
4549 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4550 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4551 | item_size); | |
4552 | return 0; | |
4553 | } | |
4554 | ||
d352ac68 CM |
4555 | /* |
4556 | * make the item pointed to by the path smaller. new_size indicates | |
4557 | * how small to make it, and from_end tells us if we just chop bytes | |
4558 | * off the end of the item or if we shift the item to chop bytes off | |
4559 | * the front. | |
4560 | */ | |
2ff7e61e JM |
4561 | void btrfs_truncate_item(struct btrfs_fs_info *fs_info, |
4562 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 4563 | { |
b18c6685 | 4564 | int slot; |
5f39d397 CM |
4565 | struct extent_buffer *leaf; |
4566 | struct btrfs_item *item; | |
b18c6685 CM |
4567 | u32 nritems; |
4568 | unsigned int data_end; | |
4569 | unsigned int old_data_start; | |
4570 | unsigned int old_size; | |
4571 | unsigned int size_diff; | |
4572 | int i; | |
cfed81a0 CM |
4573 | struct btrfs_map_token token; |
4574 | ||
4575 | btrfs_init_map_token(&token); | |
b18c6685 | 4576 | |
5f39d397 | 4577 | leaf = path->nodes[0]; |
179e29e4 CM |
4578 | slot = path->slots[0]; |
4579 | ||
4580 | old_size = btrfs_item_size_nr(leaf, slot); | |
4581 | if (old_size == new_size) | |
143bede5 | 4582 | return; |
b18c6685 | 4583 | |
5f39d397 | 4584 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4585 | data_end = leaf_data_end(fs_info, leaf); |
b18c6685 | 4586 | |
5f39d397 | 4587 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4588 | |
b18c6685 CM |
4589 | size_diff = old_size - new_size; |
4590 | ||
4591 | BUG_ON(slot < 0); | |
4592 | BUG_ON(slot >= nritems); | |
4593 | ||
4594 | /* | |
4595 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4596 | */ | |
4597 | /* first correct the data pointers */ | |
4598 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4599 | u32 ioff; |
dd3cc16b | 4600 | item = btrfs_item_nr(i); |
db94535d | 4601 | |
cfed81a0 CM |
4602 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4603 | btrfs_set_token_item_offset(leaf, item, | |
4604 | ioff + size_diff, &token); | |
b18c6685 | 4605 | } |
db94535d | 4606 | |
b18c6685 | 4607 | /* shift the data */ |
179e29e4 | 4608 | if (from_end) { |
3d9ec8c4 NB |
4609 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4610 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4611 | data_end, old_data_start + new_size - data_end); |
4612 | } else { | |
4613 | struct btrfs_disk_key disk_key; | |
4614 | u64 offset; | |
4615 | ||
4616 | btrfs_item_key(leaf, &disk_key, slot); | |
4617 | ||
4618 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4619 | unsigned long ptr; | |
4620 | struct btrfs_file_extent_item *fi; | |
4621 | ||
4622 | fi = btrfs_item_ptr(leaf, slot, | |
4623 | struct btrfs_file_extent_item); | |
4624 | fi = (struct btrfs_file_extent_item *)( | |
4625 | (unsigned long)fi - size_diff); | |
4626 | ||
4627 | if (btrfs_file_extent_type(leaf, fi) == | |
4628 | BTRFS_FILE_EXTENT_INLINE) { | |
4629 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4630 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4631 | (unsigned long)fi, |
7ec20afb | 4632 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4633 | } |
4634 | } | |
4635 | ||
3d9ec8c4 NB |
4636 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4637 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4638 | data_end, old_data_start - data_end); |
4639 | ||
4640 | offset = btrfs_disk_key_offset(&disk_key); | |
4641 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4642 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4643 | if (slot == 0) | |
0b246afa | 4644 | fixup_low_keys(fs_info, path, &disk_key, 1); |
179e29e4 | 4645 | } |
5f39d397 | 4646 | |
dd3cc16b | 4647 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4648 | btrfs_set_item_size(leaf, item, new_size); |
4649 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4650 | |
2ff7e61e | 4651 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4652 | btrfs_print_leaf(leaf); |
b18c6685 | 4653 | BUG(); |
5f39d397 | 4654 | } |
b18c6685 CM |
4655 | } |
4656 | ||
d352ac68 | 4657 | /* |
8f69dbd2 | 4658 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4659 | */ |
2ff7e61e | 4660 | void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 4661 | u32 data_size) |
6567e837 | 4662 | { |
6567e837 | 4663 | int slot; |
5f39d397 CM |
4664 | struct extent_buffer *leaf; |
4665 | struct btrfs_item *item; | |
6567e837 CM |
4666 | u32 nritems; |
4667 | unsigned int data_end; | |
4668 | unsigned int old_data; | |
4669 | unsigned int old_size; | |
4670 | int i; | |
cfed81a0 CM |
4671 | struct btrfs_map_token token; |
4672 | ||
4673 | btrfs_init_map_token(&token); | |
6567e837 | 4674 | |
5f39d397 | 4675 | leaf = path->nodes[0]; |
6567e837 | 4676 | |
5f39d397 | 4677 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4678 | data_end = leaf_data_end(fs_info, leaf); |
6567e837 | 4679 | |
2ff7e61e | 4680 | if (btrfs_leaf_free_space(fs_info, leaf) < data_size) { |
a4f78750 | 4681 | btrfs_print_leaf(leaf); |
6567e837 | 4682 | BUG(); |
5f39d397 | 4683 | } |
6567e837 | 4684 | slot = path->slots[0]; |
5f39d397 | 4685 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4686 | |
4687 | BUG_ON(slot < 0); | |
3326d1b0 | 4688 | if (slot >= nritems) { |
a4f78750 | 4689 | btrfs_print_leaf(leaf); |
0b246afa JM |
4690 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4691 | slot, nritems); | |
3326d1b0 CM |
4692 | BUG_ON(1); |
4693 | } | |
6567e837 CM |
4694 | |
4695 | /* | |
4696 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4697 | */ | |
4698 | /* first correct the data pointers */ | |
4699 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4700 | u32 ioff; |
dd3cc16b | 4701 | item = btrfs_item_nr(i); |
db94535d | 4702 | |
cfed81a0 CM |
4703 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4704 | btrfs_set_token_item_offset(leaf, item, | |
4705 | ioff - data_size, &token); | |
6567e837 | 4706 | } |
5f39d397 | 4707 | |
6567e837 | 4708 | /* shift the data */ |
3d9ec8c4 NB |
4709 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4710 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4711 | data_end, old_data - data_end); |
5f39d397 | 4712 | |
6567e837 | 4713 | data_end = old_data; |
5f39d397 | 4714 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4715 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4716 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4717 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4718 | |
2ff7e61e | 4719 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4720 | btrfs_print_leaf(leaf); |
6567e837 | 4721 | BUG(); |
5f39d397 | 4722 | } |
6567e837 CM |
4723 | } |
4724 | ||
74123bd7 | 4725 | /* |
44871b1b CM |
4726 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4727 | * to save stack depth by doing the bulk of the work in a function | |
4728 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4729 | */ |
afe5fea7 | 4730 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4731 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4732 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4733 | { |
0b246afa | 4734 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4735 | struct btrfs_item *item; |
9c58309d | 4736 | int i; |
7518a238 | 4737 | u32 nritems; |
be0e5c09 | 4738 | unsigned int data_end; |
e2fa7227 | 4739 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4740 | struct extent_buffer *leaf; |
4741 | int slot; | |
cfed81a0 CM |
4742 | struct btrfs_map_token token; |
4743 | ||
24cdc847 FM |
4744 | if (path->slots[0] == 0) { |
4745 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
0b246afa | 4746 | fixup_low_keys(fs_info, path, &disk_key, 1); |
24cdc847 FM |
4747 | } |
4748 | btrfs_unlock_up_safe(path, 1); | |
4749 | ||
cfed81a0 | 4750 | btrfs_init_map_token(&token); |
e2fa7227 | 4751 | |
5f39d397 | 4752 | leaf = path->nodes[0]; |
44871b1b | 4753 | slot = path->slots[0]; |
74123bd7 | 4754 | |
5f39d397 | 4755 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4756 | data_end = leaf_data_end(fs_info, leaf); |
eb60ceac | 4757 | |
2ff7e61e | 4758 | if (btrfs_leaf_free_space(fs_info, leaf) < total_size) { |
a4f78750 | 4759 | btrfs_print_leaf(leaf); |
0b246afa | 4760 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
2ff7e61e | 4761 | total_size, btrfs_leaf_free_space(fs_info, leaf)); |
be0e5c09 | 4762 | BUG(); |
d4dbff95 | 4763 | } |
5f39d397 | 4764 | |
be0e5c09 | 4765 | if (slot != nritems) { |
5f39d397 | 4766 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4767 | |
5f39d397 | 4768 | if (old_data < data_end) { |
a4f78750 | 4769 | btrfs_print_leaf(leaf); |
0b246afa | 4770 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4771 | slot, old_data, data_end); |
5f39d397 CM |
4772 | BUG_ON(1); |
4773 | } | |
be0e5c09 CM |
4774 | /* |
4775 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4776 | */ | |
4777 | /* first correct the data pointers */ | |
0783fcfc | 4778 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4779 | u32 ioff; |
db94535d | 4780 | |
62e85577 | 4781 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4782 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4783 | btrfs_set_token_item_offset(leaf, item, | |
4784 | ioff - total_data, &token); | |
0783fcfc | 4785 | } |
be0e5c09 | 4786 | /* shift the items */ |
9c58309d | 4787 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4788 | btrfs_item_nr_offset(slot), |
d6025579 | 4789 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4790 | |
4791 | /* shift the data */ | |
3d9ec8c4 NB |
4792 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4793 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4794 | data_end, old_data - data_end); |
be0e5c09 CM |
4795 | data_end = old_data; |
4796 | } | |
5f39d397 | 4797 | |
62e2749e | 4798 | /* setup the item for the new data */ |
9c58309d CM |
4799 | for (i = 0; i < nr; i++) { |
4800 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4801 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4802 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4803 | btrfs_set_token_item_offset(leaf, item, |
4804 | data_end - data_size[i], &token); | |
9c58309d | 4805 | data_end -= data_size[i]; |
cfed81a0 | 4806 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4807 | } |
44871b1b | 4808 | |
9c58309d | 4809 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4810 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4811 | |
2ff7e61e | 4812 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4813 | btrfs_print_leaf(leaf); |
be0e5c09 | 4814 | BUG(); |
5f39d397 | 4815 | } |
44871b1b CM |
4816 | } |
4817 | ||
4818 | /* | |
4819 | * Given a key and some data, insert items into the tree. | |
4820 | * This does all the path init required, making room in the tree if needed. | |
4821 | */ | |
4822 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4823 | struct btrfs_root *root, | |
4824 | struct btrfs_path *path, | |
310712b2 | 4825 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4826 | int nr) |
4827 | { | |
44871b1b CM |
4828 | int ret = 0; |
4829 | int slot; | |
4830 | int i; | |
4831 | u32 total_size = 0; | |
4832 | u32 total_data = 0; | |
4833 | ||
4834 | for (i = 0; i < nr; i++) | |
4835 | total_data += data_size[i]; | |
4836 | ||
4837 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4838 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4839 | if (ret == 0) | |
4840 | return -EEXIST; | |
4841 | if (ret < 0) | |
143bede5 | 4842 | return ret; |
44871b1b | 4843 | |
44871b1b CM |
4844 | slot = path->slots[0]; |
4845 | BUG_ON(slot < 0); | |
4846 | ||
afe5fea7 | 4847 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4848 | total_data, total_size, nr); |
143bede5 | 4849 | return 0; |
62e2749e CM |
4850 | } |
4851 | ||
4852 | /* | |
4853 | * Given a key and some data, insert an item into the tree. | |
4854 | * This does all the path init required, making room in the tree if needed. | |
4855 | */ | |
310712b2 OS |
4856 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4857 | const struct btrfs_key *cpu_key, void *data, | |
4858 | u32 data_size) | |
62e2749e CM |
4859 | { |
4860 | int ret = 0; | |
2c90e5d6 | 4861 | struct btrfs_path *path; |
5f39d397 CM |
4862 | struct extent_buffer *leaf; |
4863 | unsigned long ptr; | |
62e2749e | 4864 | |
2c90e5d6 | 4865 | path = btrfs_alloc_path(); |
db5b493a TI |
4866 | if (!path) |
4867 | return -ENOMEM; | |
2c90e5d6 | 4868 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4869 | if (!ret) { |
5f39d397 CM |
4870 | leaf = path->nodes[0]; |
4871 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4872 | write_extent_buffer(leaf, data, ptr, data_size); | |
4873 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4874 | } |
2c90e5d6 | 4875 | btrfs_free_path(path); |
aa5d6bed | 4876 | return ret; |
be0e5c09 CM |
4877 | } |
4878 | ||
74123bd7 | 4879 | /* |
5de08d7d | 4880 | * delete the pointer from a given node. |
74123bd7 | 4881 | * |
d352ac68 CM |
4882 | * the tree should have been previously balanced so the deletion does not |
4883 | * empty a node. | |
74123bd7 | 4884 | */ |
afe5fea7 TI |
4885 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4886 | int level, int slot) | |
be0e5c09 | 4887 | { |
0b246afa | 4888 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4889 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4890 | u32 nritems; |
f3ea38da | 4891 | int ret; |
be0e5c09 | 4892 | |
5f39d397 | 4893 | nritems = btrfs_header_nritems(parent); |
d397712b | 4894 | if (slot != nritems - 1) { |
0e411ece | 4895 | if (level) |
a446a979 DS |
4896 | tree_mod_log_eb_move(parent, slot, slot + 1, |
4897 | nritems - slot - 1); | |
5f39d397 CM |
4898 | memmove_extent_buffer(parent, |
4899 | btrfs_node_key_ptr_offset(slot), | |
4900 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4901 | sizeof(struct btrfs_key_ptr) * |
4902 | (nritems - slot - 1)); | |
57ba86c0 | 4903 | } else if (level) { |
e09c2efe DS |
4904 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4905 | GFP_NOFS); | |
57ba86c0 | 4906 | BUG_ON(ret < 0); |
bb803951 | 4907 | } |
f3ea38da | 4908 | |
7518a238 | 4909 | nritems--; |
5f39d397 | 4910 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4911 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4912 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4913 | /* just turn the root into a leaf and break */ |
5f39d397 | 4914 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4915 | } else if (slot == 0) { |
5f39d397 CM |
4916 | struct btrfs_disk_key disk_key; |
4917 | ||
4918 | btrfs_node_key(parent, &disk_key, 0); | |
0b246afa | 4919 | fixup_low_keys(fs_info, path, &disk_key, level + 1); |
be0e5c09 | 4920 | } |
d6025579 | 4921 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4922 | } |
4923 | ||
323ac95b CM |
4924 | /* |
4925 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4926 | * path->nodes[1]. |
323ac95b CM |
4927 | * |
4928 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4929 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4930 | * | |
4931 | * The path must have already been setup for deleting the leaf, including | |
4932 | * all the proper balancing. path->nodes[1] must be locked. | |
4933 | */ | |
143bede5 JM |
4934 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4935 | struct btrfs_root *root, | |
4936 | struct btrfs_path *path, | |
4937 | struct extent_buffer *leaf) | |
323ac95b | 4938 | { |
5d4f98a2 | 4939 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4940 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4941 | |
4d081c41 CM |
4942 | /* |
4943 | * btrfs_free_extent is expensive, we want to make sure we | |
4944 | * aren't holding any locks when we call it | |
4945 | */ | |
4946 | btrfs_unlock_up_safe(path, 0); | |
4947 | ||
f0486c68 YZ |
4948 | root_sub_used(root, leaf->len); |
4949 | ||
3083ee2e | 4950 | extent_buffer_get(leaf); |
5581a51a | 4951 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4952 | free_extent_buffer_stale(leaf); |
323ac95b | 4953 | } |
74123bd7 CM |
4954 | /* |
4955 | * delete the item at the leaf level in path. If that empties | |
4956 | * the leaf, remove it from the tree | |
4957 | */ | |
85e21bac CM |
4958 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4959 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4960 | { |
0b246afa | 4961 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4962 | struct extent_buffer *leaf; |
4963 | struct btrfs_item *item; | |
ce0eac2a AM |
4964 | u32 last_off; |
4965 | u32 dsize = 0; | |
aa5d6bed CM |
4966 | int ret = 0; |
4967 | int wret; | |
85e21bac | 4968 | int i; |
7518a238 | 4969 | u32 nritems; |
cfed81a0 CM |
4970 | struct btrfs_map_token token; |
4971 | ||
4972 | btrfs_init_map_token(&token); | |
be0e5c09 | 4973 | |
5f39d397 | 4974 | leaf = path->nodes[0]; |
85e21bac CM |
4975 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4976 | ||
4977 | for (i = 0; i < nr; i++) | |
4978 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4979 | ||
5f39d397 | 4980 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4981 | |
85e21bac | 4982 | if (slot + nr != nritems) { |
2ff7e61e | 4983 | int data_end = leaf_data_end(fs_info, leaf); |
5f39d397 | 4984 | |
3d9ec8c4 | 4985 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4986 | data_end + dsize, |
3d9ec8c4 | 4987 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4988 | last_off - data_end); |
5f39d397 | 4989 | |
85e21bac | 4990 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4991 | u32 ioff; |
db94535d | 4992 | |
dd3cc16b | 4993 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4994 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4995 | btrfs_set_token_item_offset(leaf, item, | |
4996 | ioff + dsize, &token); | |
0783fcfc | 4997 | } |
db94535d | 4998 | |
5f39d397 | 4999 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5000 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5001 | sizeof(struct btrfs_item) * |
85e21bac | 5002 | (nritems - slot - nr)); |
be0e5c09 | 5003 | } |
85e21bac CM |
5004 | btrfs_set_header_nritems(leaf, nritems - nr); |
5005 | nritems -= nr; | |
5f39d397 | 5006 | |
74123bd7 | 5007 | /* delete the leaf if we've emptied it */ |
7518a238 | 5008 | if (nritems == 0) { |
5f39d397 CM |
5009 | if (leaf == root->node) { |
5010 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5011 | } else { |
f0486c68 | 5012 | btrfs_set_path_blocking(path); |
7c302b49 | 5013 | clean_tree_block(fs_info, leaf); |
143bede5 | 5014 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5015 | } |
be0e5c09 | 5016 | } else { |
7518a238 | 5017 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5018 | if (slot == 0) { |
5f39d397 CM |
5019 | struct btrfs_disk_key disk_key; |
5020 | ||
5021 | btrfs_item_key(leaf, &disk_key, 0); | |
0b246afa | 5022 | fixup_low_keys(fs_info, path, &disk_key, 1); |
aa5d6bed | 5023 | } |
aa5d6bed | 5024 | |
74123bd7 | 5025 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5026 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5027 | /* push_leaf_left fixes the path. |
5028 | * make sure the path still points to our leaf | |
5029 | * for possible call to del_ptr below | |
5030 | */ | |
4920c9ac | 5031 | slot = path->slots[1]; |
5f39d397 CM |
5032 | extent_buffer_get(leaf); |
5033 | ||
b9473439 | 5034 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5035 | wret = push_leaf_left(trans, root, path, 1, 1, |
5036 | 1, (u32)-1); | |
54aa1f4d | 5037 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5038 | ret = wret; |
5f39d397 CM |
5039 | |
5040 | if (path->nodes[0] == leaf && | |
5041 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5042 | wret = push_leaf_right(trans, root, path, 1, |
5043 | 1, 1, 0); | |
54aa1f4d | 5044 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5045 | ret = wret; |
5046 | } | |
5f39d397 CM |
5047 | |
5048 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5049 | path->slots[1] = slot; |
143bede5 | 5050 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5051 | free_extent_buffer(leaf); |
143bede5 | 5052 | ret = 0; |
5de08d7d | 5053 | } else { |
925baedd CM |
5054 | /* if we're still in the path, make sure |
5055 | * we're dirty. Otherwise, one of the | |
5056 | * push_leaf functions must have already | |
5057 | * dirtied this buffer | |
5058 | */ | |
5059 | if (path->nodes[0] == leaf) | |
5060 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5061 | free_extent_buffer(leaf); |
be0e5c09 | 5062 | } |
d5719762 | 5063 | } else { |
5f39d397 | 5064 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5065 | } |
5066 | } | |
aa5d6bed | 5067 | return ret; |
be0e5c09 CM |
5068 | } |
5069 | ||
7bb86316 | 5070 | /* |
925baedd | 5071 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5072 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5073 | * returns < 0 on io errors. | |
d352ac68 CM |
5074 | * |
5075 | * This may release the path, and so you may lose any locks held at the | |
5076 | * time you call it. | |
7bb86316 | 5077 | */ |
16e7549f | 5078 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5079 | { |
925baedd CM |
5080 | struct btrfs_key key; |
5081 | struct btrfs_disk_key found_key; | |
5082 | int ret; | |
7bb86316 | 5083 | |
925baedd | 5084 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5085 | |
e8b0d724 | 5086 | if (key.offset > 0) { |
925baedd | 5087 | key.offset--; |
e8b0d724 | 5088 | } else if (key.type > 0) { |
925baedd | 5089 | key.type--; |
e8b0d724 FDBM |
5090 | key.offset = (u64)-1; |
5091 | } else if (key.objectid > 0) { | |
925baedd | 5092 | key.objectid--; |
e8b0d724 FDBM |
5093 | key.type = (u8)-1; |
5094 | key.offset = (u64)-1; | |
5095 | } else { | |
925baedd | 5096 | return 1; |
e8b0d724 | 5097 | } |
7bb86316 | 5098 | |
b3b4aa74 | 5099 | btrfs_release_path(path); |
925baedd CM |
5100 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5101 | if (ret < 0) | |
5102 | return ret; | |
5103 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5104 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5105 | /* |
5106 | * We might have had an item with the previous key in the tree right | |
5107 | * before we released our path. And after we released our path, that | |
5108 | * item might have been pushed to the first slot (0) of the leaf we | |
5109 | * were holding due to a tree balance. Alternatively, an item with the | |
5110 | * previous key can exist as the only element of a leaf (big fat item). | |
5111 | * Therefore account for these 2 cases, so that our callers (like | |
5112 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5113 | * the previous key we computed above. | |
5114 | */ | |
5115 | if (ret <= 0) | |
925baedd CM |
5116 | return 0; |
5117 | return 1; | |
7bb86316 CM |
5118 | } |
5119 | ||
3f157a2f CM |
5120 | /* |
5121 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5122 | * for nodes or leaves that are have a minimum transaction id. |
5123 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5124 | * |
5125 | * This does not cow, but it does stuff the starting key it finds back | |
5126 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5127 | * key and get a writable path. | |
5128 | * | |
3f157a2f CM |
5129 | * This honors path->lowest_level to prevent descent past a given level |
5130 | * of the tree. | |
5131 | * | |
d352ac68 CM |
5132 | * min_trans indicates the oldest transaction that you are interested |
5133 | * in walking through. Any nodes or leaves older than min_trans are | |
5134 | * skipped over (without reading them). | |
5135 | * | |
3f157a2f CM |
5136 | * returns zero if something useful was found, < 0 on error and 1 if there |
5137 | * was nothing in the tree that matched the search criteria. | |
5138 | */ | |
5139 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5140 | struct btrfs_path *path, |
3f157a2f CM |
5141 | u64 min_trans) |
5142 | { | |
2ff7e61e | 5143 | struct btrfs_fs_info *fs_info = root->fs_info; |
3f157a2f CM |
5144 | struct extent_buffer *cur; |
5145 | struct btrfs_key found_key; | |
5146 | int slot; | |
9652480b | 5147 | int sret; |
3f157a2f CM |
5148 | u32 nritems; |
5149 | int level; | |
5150 | int ret = 1; | |
f98de9b9 | 5151 | int keep_locks = path->keep_locks; |
3f157a2f | 5152 | |
f98de9b9 | 5153 | path->keep_locks = 1; |
3f157a2f | 5154 | again: |
bd681513 | 5155 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5156 | level = btrfs_header_level(cur); |
e02119d5 | 5157 | WARN_ON(path->nodes[level]); |
3f157a2f | 5158 | path->nodes[level] = cur; |
bd681513 | 5159 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5160 | |
5161 | if (btrfs_header_generation(cur) < min_trans) { | |
5162 | ret = 1; | |
5163 | goto out; | |
5164 | } | |
d397712b | 5165 | while (1) { |
3f157a2f CM |
5166 | nritems = btrfs_header_nritems(cur); |
5167 | level = btrfs_header_level(cur); | |
a74b35ec | 5168 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
3f157a2f | 5169 | |
323ac95b CM |
5170 | /* at the lowest level, we're done, setup the path and exit */ |
5171 | if (level == path->lowest_level) { | |
e02119d5 CM |
5172 | if (slot >= nritems) |
5173 | goto find_next_key; | |
3f157a2f CM |
5174 | ret = 0; |
5175 | path->slots[level] = slot; | |
5176 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5177 | goto out; | |
5178 | } | |
9652480b Y |
5179 | if (sret && slot > 0) |
5180 | slot--; | |
3f157a2f | 5181 | /* |
de78b51a ES |
5182 | * check this node pointer against the min_trans parameters. |
5183 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5184 | */ |
d397712b | 5185 | while (slot < nritems) { |
3f157a2f | 5186 | u64 gen; |
e02119d5 | 5187 | |
3f157a2f CM |
5188 | gen = btrfs_node_ptr_generation(cur, slot); |
5189 | if (gen < min_trans) { | |
5190 | slot++; | |
5191 | continue; | |
5192 | } | |
de78b51a | 5193 | break; |
3f157a2f | 5194 | } |
e02119d5 | 5195 | find_next_key: |
3f157a2f CM |
5196 | /* |
5197 | * we didn't find a candidate key in this node, walk forward | |
5198 | * and find another one | |
5199 | */ | |
5200 | if (slot >= nritems) { | |
e02119d5 | 5201 | path->slots[level] = slot; |
b4ce94de | 5202 | btrfs_set_path_blocking(path); |
e02119d5 | 5203 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5204 | min_trans); |
e02119d5 | 5205 | if (sret == 0) { |
b3b4aa74 | 5206 | btrfs_release_path(path); |
3f157a2f CM |
5207 | goto again; |
5208 | } else { | |
5209 | goto out; | |
5210 | } | |
5211 | } | |
5212 | /* save our key for returning back */ | |
5213 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5214 | path->slots[level] = slot; | |
5215 | if (level == path->lowest_level) { | |
5216 | ret = 0; | |
3f157a2f CM |
5217 | goto out; |
5218 | } | |
b4ce94de | 5219 | btrfs_set_path_blocking(path); |
2ff7e61e | 5220 | cur = read_node_slot(fs_info, cur, slot); |
fb770ae4 LB |
5221 | if (IS_ERR(cur)) { |
5222 | ret = PTR_ERR(cur); | |
5223 | goto out; | |
5224 | } | |
3f157a2f | 5225 | |
bd681513 | 5226 | btrfs_tree_read_lock(cur); |
b4ce94de | 5227 | |
bd681513 | 5228 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5229 | path->nodes[level - 1] = cur; |
f7c79f30 | 5230 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5231 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5232 | } |
5233 | out: | |
f98de9b9 FM |
5234 | path->keep_locks = keep_locks; |
5235 | if (ret == 0) { | |
5236 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5237 | btrfs_set_path_blocking(path); | |
3f157a2f | 5238 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5239 | } |
3f157a2f CM |
5240 | return ret; |
5241 | } | |
5242 | ||
2ff7e61e | 5243 | static int tree_move_down(struct btrfs_fs_info *fs_info, |
7069830a | 5244 | struct btrfs_path *path, |
ab6a43e1 | 5245 | int *level) |
7069830a | 5246 | { |
fb770ae4 LB |
5247 | struct extent_buffer *eb; |
5248 | ||
74dd17fb | 5249 | BUG_ON(*level == 0); |
2ff7e61e | 5250 | eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]); |
fb770ae4 LB |
5251 | if (IS_ERR(eb)) |
5252 | return PTR_ERR(eb); | |
5253 | ||
5254 | path->nodes[*level - 1] = eb; | |
7069830a AB |
5255 | path->slots[*level - 1] = 0; |
5256 | (*level)--; | |
fb770ae4 | 5257 | return 0; |
7069830a AB |
5258 | } |
5259 | ||
f1e30261 | 5260 | static int tree_move_next_or_upnext(struct btrfs_path *path, |
7069830a AB |
5261 | int *level, int root_level) |
5262 | { | |
5263 | int ret = 0; | |
5264 | int nritems; | |
5265 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5266 | ||
5267 | path->slots[*level]++; | |
5268 | ||
74dd17fb | 5269 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5270 | if (*level == root_level) |
5271 | return -1; | |
5272 | ||
5273 | /* move upnext */ | |
5274 | path->slots[*level] = 0; | |
5275 | free_extent_buffer(path->nodes[*level]); | |
5276 | path->nodes[*level] = NULL; | |
5277 | (*level)++; | |
5278 | path->slots[*level]++; | |
5279 | ||
5280 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5281 | ret = 1; | |
5282 | } | |
5283 | return ret; | |
5284 | } | |
5285 | ||
5286 | /* | |
5287 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5288 | * or down. | |
5289 | */ | |
2ff7e61e | 5290 | static int tree_advance(struct btrfs_fs_info *fs_info, |
7069830a AB |
5291 | struct btrfs_path *path, |
5292 | int *level, int root_level, | |
5293 | int allow_down, | |
5294 | struct btrfs_key *key) | |
5295 | { | |
5296 | int ret; | |
5297 | ||
5298 | if (*level == 0 || !allow_down) { | |
f1e30261 | 5299 | ret = tree_move_next_or_upnext(path, level, root_level); |
7069830a | 5300 | } else { |
ab6a43e1 | 5301 | ret = tree_move_down(fs_info, path, level); |
7069830a AB |
5302 | } |
5303 | if (ret >= 0) { | |
5304 | if (*level == 0) | |
5305 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5306 | path->slots[*level]); | |
5307 | else | |
5308 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5309 | path->slots[*level]); | |
5310 | } | |
5311 | return ret; | |
5312 | } | |
5313 | ||
2ff7e61e | 5314 | static int tree_compare_item(struct btrfs_path *left_path, |
7069830a AB |
5315 | struct btrfs_path *right_path, |
5316 | char *tmp_buf) | |
5317 | { | |
5318 | int cmp; | |
5319 | int len1, len2; | |
5320 | unsigned long off1, off2; | |
5321 | ||
5322 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5323 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5324 | if (len1 != len2) | |
5325 | return 1; | |
5326 | ||
5327 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5328 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5329 | right_path->slots[0]); | |
5330 | ||
5331 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5332 | ||
5333 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5334 | if (cmp) | |
5335 | return 1; | |
5336 | return 0; | |
5337 | } | |
5338 | ||
5339 | #define ADVANCE 1 | |
5340 | #define ADVANCE_ONLY_NEXT -1 | |
5341 | ||
5342 | /* | |
5343 | * This function compares two trees and calls the provided callback for | |
5344 | * every changed/new/deleted item it finds. | |
5345 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5346 | * the compare pretty fast on snapshotted subvolumes. | |
5347 | * | |
5348 | * This currently works on commit roots only. As commit roots are read only, | |
5349 | * we don't do any locking. The commit roots are protected with transactions. | |
5350 | * Transactions are ended and rejoined when a commit is tried in between. | |
5351 | * | |
5352 | * This function checks for modifications done to the trees while comparing. | |
5353 | * If it detects a change, it aborts immediately. | |
5354 | */ | |
5355 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5356 | struct btrfs_root *right_root, | |
5357 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5358 | { | |
0b246afa | 5359 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5360 | int ret; |
5361 | int cmp; | |
7069830a AB |
5362 | struct btrfs_path *left_path = NULL; |
5363 | struct btrfs_path *right_path = NULL; | |
5364 | struct btrfs_key left_key; | |
5365 | struct btrfs_key right_key; | |
5366 | char *tmp_buf = NULL; | |
5367 | int left_root_level; | |
5368 | int right_root_level; | |
5369 | int left_level; | |
5370 | int right_level; | |
5371 | int left_end_reached; | |
5372 | int right_end_reached; | |
5373 | int advance_left; | |
5374 | int advance_right; | |
5375 | u64 left_blockptr; | |
5376 | u64 right_blockptr; | |
6baa4293 FM |
5377 | u64 left_gen; |
5378 | u64 right_gen; | |
7069830a AB |
5379 | |
5380 | left_path = btrfs_alloc_path(); | |
5381 | if (!left_path) { | |
5382 | ret = -ENOMEM; | |
5383 | goto out; | |
5384 | } | |
5385 | right_path = btrfs_alloc_path(); | |
5386 | if (!right_path) { | |
5387 | ret = -ENOMEM; | |
5388 | goto out; | |
5389 | } | |
5390 | ||
752ade68 | 5391 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); |
7069830a | 5392 | if (!tmp_buf) { |
752ade68 MH |
5393 | ret = -ENOMEM; |
5394 | goto out; | |
7069830a AB |
5395 | } |
5396 | ||
5397 | left_path->search_commit_root = 1; | |
5398 | left_path->skip_locking = 1; | |
5399 | right_path->search_commit_root = 1; | |
5400 | right_path->skip_locking = 1; | |
5401 | ||
7069830a AB |
5402 | /* |
5403 | * Strategy: Go to the first items of both trees. Then do | |
5404 | * | |
5405 | * If both trees are at level 0 | |
5406 | * Compare keys of current items | |
5407 | * If left < right treat left item as new, advance left tree | |
5408 | * and repeat | |
5409 | * If left > right treat right item as deleted, advance right tree | |
5410 | * and repeat | |
5411 | * If left == right do deep compare of items, treat as changed if | |
5412 | * needed, advance both trees and repeat | |
5413 | * If both trees are at the same level but not at level 0 | |
5414 | * Compare keys of current nodes/leafs | |
5415 | * If left < right advance left tree and repeat | |
5416 | * If left > right advance right tree and repeat | |
5417 | * If left == right compare blockptrs of the next nodes/leafs | |
5418 | * If they match advance both trees but stay at the same level | |
5419 | * and repeat | |
5420 | * If they don't match advance both trees while allowing to go | |
5421 | * deeper and repeat | |
5422 | * If tree levels are different | |
5423 | * Advance the tree that needs it and repeat | |
5424 | * | |
5425 | * Advancing a tree means: | |
5426 | * If we are at level 0, try to go to the next slot. If that's not | |
5427 | * possible, go one level up and repeat. Stop when we found a level | |
5428 | * where we could go to the next slot. We may at this point be on a | |
5429 | * node or a leaf. | |
5430 | * | |
5431 | * If we are not at level 0 and not on shared tree blocks, go one | |
5432 | * level deeper. | |
5433 | * | |
5434 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5435 | * the right if possible or go up and right. | |
5436 | */ | |
5437 | ||
0b246afa | 5438 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5439 | left_level = btrfs_header_level(left_root->commit_root); |
5440 | left_root_level = left_level; | |
5441 | left_path->nodes[left_level] = left_root->commit_root; | |
5442 | extent_buffer_get(left_path->nodes[left_level]); | |
5443 | ||
5444 | right_level = btrfs_header_level(right_root->commit_root); | |
5445 | right_root_level = right_level; | |
5446 | right_path->nodes[right_level] = right_root->commit_root; | |
5447 | extent_buffer_get(right_path->nodes[right_level]); | |
0b246afa | 5448 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5449 | |
5450 | if (left_level == 0) | |
5451 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5452 | &left_key, left_path->slots[left_level]); | |
5453 | else | |
5454 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5455 | &left_key, left_path->slots[left_level]); | |
5456 | if (right_level == 0) | |
5457 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5458 | &right_key, right_path->slots[right_level]); | |
5459 | else | |
5460 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5461 | &right_key, right_path->slots[right_level]); | |
5462 | ||
5463 | left_end_reached = right_end_reached = 0; | |
5464 | advance_left = advance_right = 0; | |
5465 | ||
5466 | while (1) { | |
7069830a | 5467 | if (advance_left && !left_end_reached) { |
2ff7e61e | 5468 | ret = tree_advance(fs_info, left_path, &left_level, |
7069830a AB |
5469 | left_root_level, |
5470 | advance_left != ADVANCE_ONLY_NEXT, | |
5471 | &left_key); | |
fb770ae4 | 5472 | if (ret == -1) |
7069830a | 5473 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5474 | else if (ret < 0) |
5475 | goto out; | |
7069830a AB |
5476 | advance_left = 0; |
5477 | } | |
5478 | if (advance_right && !right_end_reached) { | |
2ff7e61e | 5479 | ret = tree_advance(fs_info, right_path, &right_level, |
7069830a AB |
5480 | right_root_level, |
5481 | advance_right != ADVANCE_ONLY_NEXT, | |
5482 | &right_key); | |
fb770ae4 | 5483 | if (ret == -1) |
7069830a | 5484 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5485 | else if (ret < 0) |
5486 | goto out; | |
7069830a AB |
5487 | advance_right = 0; |
5488 | } | |
5489 | ||
5490 | if (left_end_reached && right_end_reached) { | |
5491 | ret = 0; | |
5492 | goto out; | |
5493 | } else if (left_end_reached) { | |
5494 | if (right_level == 0) { | |
ee8c494f | 5495 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5496 | &right_key, |
5497 | BTRFS_COMPARE_TREE_DELETED, | |
5498 | ctx); | |
5499 | if (ret < 0) | |
5500 | goto out; | |
5501 | } | |
5502 | advance_right = ADVANCE; | |
5503 | continue; | |
5504 | } else if (right_end_reached) { | |
5505 | if (left_level == 0) { | |
ee8c494f | 5506 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5507 | &left_key, |
5508 | BTRFS_COMPARE_TREE_NEW, | |
5509 | ctx); | |
5510 | if (ret < 0) | |
5511 | goto out; | |
5512 | } | |
5513 | advance_left = ADVANCE; | |
5514 | continue; | |
5515 | } | |
5516 | ||
5517 | if (left_level == 0 && right_level == 0) { | |
5518 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5519 | if (cmp < 0) { | |
ee8c494f | 5520 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5521 | &left_key, |
5522 | BTRFS_COMPARE_TREE_NEW, | |
5523 | ctx); | |
5524 | if (ret < 0) | |
5525 | goto out; | |
5526 | advance_left = ADVANCE; | |
5527 | } else if (cmp > 0) { | |
ee8c494f | 5528 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5529 | &right_key, |
5530 | BTRFS_COMPARE_TREE_DELETED, | |
5531 | ctx); | |
5532 | if (ret < 0) | |
5533 | goto out; | |
5534 | advance_right = ADVANCE; | |
5535 | } else { | |
b99d9a6a | 5536 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5537 | |
74dd17fb | 5538 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
2ff7e61e JM |
5539 | ret = tree_compare_item(left_path, right_path, |
5540 | tmp_buf); | |
ba5e8f2e | 5541 | if (ret) |
b99d9a6a | 5542 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5543 | else |
b99d9a6a | 5544 | result = BTRFS_COMPARE_TREE_SAME; |
ee8c494f | 5545 | ret = changed_cb(left_path, right_path, |
b99d9a6a | 5546 | &left_key, result, ctx); |
ba5e8f2e JB |
5547 | if (ret < 0) |
5548 | goto out; | |
7069830a AB |
5549 | advance_left = ADVANCE; |
5550 | advance_right = ADVANCE; | |
5551 | } | |
5552 | } else if (left_level == right_level) { | |
5553 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5554 | if (cmp < 0) { | |
5555 | advance_left = ADVANCE; | |
5556 | } else if (cmp > 0) { | |
5557 | advance_right = ADVANCE; | |
5558 | } else { | |
5559 | left_blockptr = btrfs_node_blockptr( | |
5560 | left_path->nodes[left_level], | |
5561 | left_path->slots[left_level]); | |
5562 | right_blockptr = btrfs_node_blockptr( | |
5563 | right_path->nodes[right_level], | |
5564 | right_path->slots[right_level]); | |
6baa4293 FM |
5565 | left_gen = btrfs_node_ptr_generation( |
5566 | left_path->nodes[left_level], | |
5567 | left_path->slots[left_level]); | |
5568 | right_gen = btrfs_node_ptr_generation( | |
5569 | right_path->nodes[right_level], | |
5570 | right_path->slots[right_level]); | |
5571 | if (left_blockptr == right_blockptr && | |
5572 | left_gen == right_gen) { | |
7069830a AB |
5573 | /* |
5574 | * As we're on a shared block, don't | |
5575 | * allow to go deeper. | |
5576 | */ | |
5577 | advance_left = ADVANCE_ONLY_NEXT; | |
5578 | advance_right = ADVANCE_ONLY_NEXT; | |
5579 | } else { | |
5580 | advance_left = ADVANCE; | |
5581 | advance_right = ADVANCE; | |
5582 | } | |
5583 | } | |
5584 | } else if (left_level < right_level) { | |
5585 | advance_right = ADVANCE; | |
5586 | } else { | |
5587 | advance_left = ADVANCE; | |
5588 | } | |
5589 | } | |
5590 | ||
5591 | out: | |
5592 | btrfs_free_path(left_path); | |
5593 | btrfs_free_path(right_path); | |
8f282f71 | 5594 | kvfree(tmp_buf); |
7069830a AB |
5595 | return ret; |
5596 | } | |
5597 | ||
3f157a2f CM |
5598 | /* |
5599 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5600 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5601 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5602 | * |
5603 | * 0 is returned if another key is found, < 0 if there are any errors | |
5604 | * and 1 is returned if there are no higher keys in the tree | |
5605 | * | |
5606 | * path->keep_locks should be set to 1 on the search made before | |
5607 | * calling this function. | |
5608 | */ | |
e7a84565 | 5609 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5610 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5611 | { |
e7a84565 CM |
5612 | int slot; |
5613 | struct extent_buffer *c; | |
5614 | ||
934d375b | 5615 | WARN_ON(!path->keep_locks); |
d397712b | 5616 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5617 | if (!path->nodes[level]) |
5618 | return 1; | |
5619 | ||
5620 | slot = path->slots[level] + 1; | |
5621 | c = path->nodes[level]; | |
3f157a2f | 5622 | next: |
e7a84565 | 5623 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5624 | int ret; |
5625 | int orig_lowest; | |
5626 | struct btrfs_key cur_key; | |
5627 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5628 | !path->nodes[level + 1]) | |
e7a84565 | 5629 | return 1; |
33c66f43 YZ |
5630 | |
5631 | if (path->locks[level + 1]) { | |
5632 | level++; | |
5633 | continue; | |
5634 | } | |
5635 | ||
5636 | slot = btrfs_header_nritems(c) - 1; | |
5637 | if (level == 0) | |
5638 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5639 | else | |
5640 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5641 | ||
5642 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5643 | btrfs_release_path(path); |
33c66f43 YZ |
5644 | path->lowest_level = level; |
5645 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5646 | 0, 0); | |
5647 | path->lowest_level = orig_lowest; | |
5648 | if (ret < 0) | |
5649 | return ret; | |
5650 | ||
5651 | c = path->nodes[level]; | |
5652 | slot = path->slots[level]; | |
5653 | if (ret == 0) | |
5654 | slot++; | |
5655 | goto next; | |
e7a84565 | 5656 | } |
33c66f43 | 5657 | |
e7a84565 CM |
5658 | if (level == 0) |
5659 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5660 | else { |
3f157a2f CM |
5661 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5662 | ||
3f157a2f CM |
5663 | if (gen < min_trans) { |
5664 | slot++; | |
5665 | goto next; | |
5666 | } | |
e7a84565 | 5667 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5668 | } |
e7a84565 CM |
5669 | return 0; |
5670 | } | |
5671 | return 1; | |
5672 | } | |
5673 | ||
97571fd0 | 5674 | /* |
925baedd | 5675 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5676 | * returns 0 if it found something or 1 if there are no greater leaves. |
5677 | * returns < 0 on io errors. | |
97571fd0 | 5678 | */ |
234b63a0 | 5679 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5680 | { |
5681 | return btrfs_next_old_leaf(root, path, 0); | |
5682 | } | |
5683 | ||
5684 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5685 | u64 time_seq) | |
d97e63b6 CM |
5686 | { |
5687 | int slot; | |
8e73f275 | 5688 | int level; |
5f39d397 | 5689 | struct extent_buffer *c; |
8e73f275 | 5690 | struct extent_buffer *next; |
925baedd CM |
5691 | struct btrfs_key key; |
5692 | u32 nritems; | |
5693 | int ret; | |
8e73f275 | 5694 | int old_spinning = path->leave_spinning; |
bd681513 | 5695 | int next_rw_lock = 0; |
925baedd CM |
5696 | |
5697 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5698 | if (nritems == 0) |
925baedd | 5699 | return 1; |
925baedd | 5700 | |
8e73f275 CM |
5701 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5702 | again: | |
5703 | level = 1; | |
5704 | next = NULL; | |
bd681513 | 5705 | next_rw_lock = 0; |
b3b4aa74 | 5706 | btrfs_release_path(path); |
8e73f275 | 5707 | |
a2135011 | 5708 | path->keep_locks = 1; |
31533fb2 | 5709 | path->leave_spinning = 1; |
8e73f275 | 5710 | |
3d7806ec JS |
5711 | if (time_seq) |
5712 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5713 | else | |
5714 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5715 | path->keep_locks = 0; |
5716 | ||
5717 | if (ret < 0) | |
5718 | return ret; | |
5719 | ||
a2135011 | 5720 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5721 | /* |
5722 | * by releasing the path above we dropped all our locks. A balance | |
5723 | * could have added more items next to the key that used to be | |
5724 | * at the very end of the block. So, check again here and | |
5725 | * advance the path if there are now more items available. | |
5726 | */ | |
a2135011 | 5727 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5728 | if (ret == 0) |
5729 | path->slots[0]++; | |
8e73f275 | 5730 | ret = 0; |
925baedd CM |
5731 | goto done; |
5732 | } | |
0b43e04f LB |
5733 | /* |
5734 | * So the above check misses one case: | |
5735 | * - after releasing the path above, someone has removed the item that | |
5736 | * used to be at the very end of the block, and balance between leafs | |
5737 | * gets another one with bigger key.offset to replace it. | |
5738 | * | |
5739 | * This one should be returned as well, or we can get leaf corruption | |
5740 | * later(esp. in __btrfs_drop_extents()). | |
5741 | * | |
5742 | * And a bit more explanation about this check, | |
5743 | * with ret > 0, the key isn't found, the path points to the slot | |
5744 | * where it should be inserted, so the path->slots[0] item must be the | |
5745 | * bigger one. | |
5746 | */ | |
5747 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5748 | ret = 0; | |
5749 | goto done; | |
5750 | } | |
d97e63b6 | 5751 | |
d397712b | 5752 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5753 | if (!path->nodes[level]) { |
5754 | ret = 1; | |
5755 | goto done; | |
5756 | } | |
5f39d397 | 5757 | |
d97e63b6 CM |
5758 | slot = path->slots[level] + 1; |
5759 | c = path->nodes[level]; | |
5f39d397 | 5760 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5761 | level++; |
8e73f275 CM |
5762 | if (level == BTRFS_MAX_LEVEL) { |
5763 | ret = 1; | |
5764 | goto done; | |
5765 | } | |
d97e63b6 CM |
5766 | continue; |
5767 | } | |
5f39d397 | 5768 | |
925baedd | 5769 | if (next) { |
bd681513 | 5770 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5771 | free_extent_buffer(next); |
925baedd | 5772 | } |
5f39d397 | 5773 | |
8e73f275 | 5774 | next = c; |
bd681513 | 5775 | next_rw_lock = path->locks[level]; |
d07b8528 | 5776 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5777 | slot, &key); |
8e73f275 CM |
5778 | if (ret == -EAGAIN) |
5779 | goto again; | |
5f39d397 | 5780 | |
76a05b35 | 5781 | if (ret < 0) { |
b3b4aa74 | 5782 | btrfs_release_path(path); |
76a05b35 CM |
5783 | goto done; |
5784 | } | |
5785 | ||
5cd57b2c | 5786 | if (!path->skip_locking) { |
bd681513 | 5787 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5788 | if (!ret && time_seq) { |
5789 | /* | |
5790 | * If we don't get the lock, we may be racing | |
5791 | * with push_leaf_left, holding that lock while | |
5792 | * itself waiting for the leaf we've currently | |
5793 | * locked. To solve this situation, we give up | |
5794 | * on our lock and cycle. | |
5795 | */ | |
cf538830 | 5796 | free_extent_buffer(next); |
d42244a0 JS |
5797 | btrfs_release_path(path); |
5798 | cond_resched(); | |
5799 | goto again; | |
5800 | } | |
8e73f275 CM |
5801 | if (!ret) { |
5802 | btrfs_set_path_blocking(path); | |
bd681513 | 5803 | btrfs_tree_read_lock(next); |
31533fb2 | 5804 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5805 | BTRFS_READ_LOCK); |
8e73f275 | 5806 | } |
31533fb2 | 5807 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5808 | } |
d97e63b6 CM |
5809 | break; |
5810 | } | |
5811 | path->slots[level] = slot; | |
d397712b | 5812 | while (1) { |
d97e63b6 CM |
5813 | level--; |
5814 | c = path->nodes[level]; | |
925baedd | 5815 | if (path->locks[level]) |
bd681513 | 5816 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5817 | |
5f39d397 | 5818 | free_extent_buffer(c); |
d97e63b6 CM |
5819 | path->nodes[level] = next; |
5820 | path->slots[level] = 0; | |
a74a4b97 | 5821 | if (!path->skip_locking) |
bd681513 | 5822 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5823 | if (!level) |
5824 | break; | |
b4ce94de | 5825 | |
d07b8528 | 5826 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5827 | 0, &key); |
8e73f275 CM |
5828 | if (ret == -EAGAIN) |
5829 | goto again; | |
5830 | ||
76a05b35 | 5831 | if (ret < 0) { |
b3b4aa74 | 5832 | btrfs_release_path(path); |
76a05b35 CM |
5833 | goto done; |
5834 | } | |
5835 | ||
5cd57b2c | 5836 | if (!path->skip_locking) { |
bd681513 | 5837 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5838 | if (!ret) { |
5839 | btrfs_set_path_blocking(path); | |
bd681513 | 5840 | btrfs_tree_read_lock(next); |
31533fb2 | 5841 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5842 | BTRFS_READ_LOCK); |
5843 | } | |
31533fb2 | 5844 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5845 | } |
d97e63b6 | 5846 | } |
8e73f275 | 5847 | ret = 0; |
925baedd | 5848 | done: |
f7c79f30 | 5849 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5850 | path->leave_spinning = old_spinning; |
5851 | if (!old_spinning) | |
5852 | btrfs_set_path_blocking(path); | |
5853 | ||
5854 | return ret; | |
d97e63b6 | 5855 | } |
0b86a832 | 5856 | |
3f157a2f CM |
5857 | /* |
5858 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5859 | * searching until it gets past min_objectid or finds an item of 'type' | |
5860 | * | |
5861 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5862 | */ | |
0b86a832 CM |
5863 | int btrfs_previous_item(struct btrfs_root *root, |
5864 | struct btrfs_path *path, u64 min_objectid, | |
5865 | int type) | |
5866 | { | |
5867 | struct btrfs_key found_key; | |
5868 | struct extent_buffer *leaf; | |
e02119d5 | 5869 | u32 nritems; |
0b86a832 CM |
5870 | int ret; |
5871 | ||
d397712b | 5872 | while (1) { |
0b86a832 | 5873 | if (path->slots[0] == 0) { |
b4ce94de | 5874 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5875 | ret = btrfs_prev_leaf(root, path); |
5876 | if (ret != 0) | |
5877 | return ret; | |
5878 | } else { | |
5879 | path->slots[0]--; | |
5880 | } | |
5881 | leaf = path->nodes[0]; | |
e02119d5 CM |
5882 | nritems = btrfs_header_nritems(leaf); |
5883 | if (nritems == 0) | |
5884 | return 1; | |
5885 | if (path->slots[0] == nritems) | |
5886 | path->slots[0]--; | |
5887 | ||
0b86a832 | 5888 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5889 | if (found_key.objectid < min_objectid) |
5890 | break; | |
0a4eefbb YZ |
5891 | if (found_key.type == type) |
5892 | return 0; | |
e02119d5 CM |
5893 | if (found_key.objectid == min_objectid && |
5894 | found_key.type < type) | |
5895 | break; | |
0b86a832 CM |
5896 | } |
5897 | return 1; | |
5898 | } | |
ade2e0b3 WS |
5899 | |
5900 | /* | |
5901 | * search in extent tree to find a previous Metadata/Data extent item with | |
5902 | * min objecitd. | |
5903 | * | |
5904 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5905 | */ | |
5906 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5907 | struct btrfs_path *path, u64 min_objectid) | |
5908 | { | |
5909 | struct btrfs_key found_key; | |
5910 | struct extent_buffer *leaf; | |
5911 | u32 nritems; | |
5912 | int ret; | |
5913 | ||
5914 | while (1) { | |
5915 | if (path->slots[0] == 0) { | |
5916 | btrfs_set_path_blocking(path); | |
5917 | ret = btrfs_prev_leaf(root, path); | |
5918 | if (ret != 0) | |
5919 | return ret; | |
5920 | } else { | |
5921 | path->slots[0]--; | |
5922 | } | |
5923 | leaf = path->nodes[0]; | |
5924 | nritems = btrfs_header_nritems(leaf); | |
5925 | if (nritems == 0) | |
5926 | return 1; | |
5927 | if (path->slots[0] == nritems) | |
5928 | path->slots[0]--; | |
5929 | ||
5930 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5931 | if (found_key.objectid < min_objectid) | |
5932 | break; | |
5933 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5934 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5935 | return 0; | |
5936 | if (found_key.objectid == min_objectid && | |
5937 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5938 | break; | |
5939 | } | |
5940 | return 1; | |
5941 | } |