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 | ||
fc36ed7e | 333 | /* |
fcebe456 | 334 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 335 | */ |
fcebe456 | 336 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
337 | { |
338 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
339 | } | |
340 | ||
097b8a7c JS |
341 | /* |
342 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
343 | * passed does not already have a sequence number set. So when a caller expects | |
344 | * to record tree modifications, it should ensure to set elem->seq to zero | |
345 | * before calling btrfs_get_tree_mod_seq. | |
346 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
347 | * blocker was added. | |
348 | */ | |
349 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
350 | struct seq_list *elem) | |
bd989ba3 | 351 | { |
b1a09f1e | 352 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 353 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 354 | if (!elem->seq) { |
fcebe456 | 355 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
356 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
357 | } | |
bd989ba3 | 358 | spin_unlock(&fs_info->tree_mod_seq_lock); |
b1a09f1e | 359 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 360 | |
fcebe456 | 361 | return elem->seq; |
bd989ba3 JS |
362 | } |
363 | ||
364 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
365 | struct seq_list *elem) | |
366 | { | |
367 | struct rb_root *tm_root; | |
368 | struct rb_node *node; | |
369 | struct rb_node *next; | |
370 | struct seq_list *cur_elem; | |
371 | struct tree_mod_elem *tm; | |
372 | u64 min_seq = (u64)-1; | |
373 | u64 seq_putting = elem->seq; | |
374 | ||
375 | if (!seq_putting) | |
376 | return; | |
377 | ||
bd989ba3 JS |
378 | spin_lock(&fs_info->tree_mod_seq_lock); |
379 | list_del(&elem->list); | |
097b8a7c | 380 | elem->seq = 0; |
bd989ba3 JS |
381 | |
382 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 383 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
384 | if (seq_putting > cur_elem->seq) { |
385 | /* | |
386 | * blocker with lower sequence number exists, we | |
387 | * cannot remove anything from the log | |
388 | */ | |
097b8a7c JS |
389 | spin_unlock(&fs_info->tree_mod_seq_lock); |
390 | return; | |
bd989ba3 JS |
391 | } |
392 | min_seq = cur_elem->seq; | |
393 | } | |
394 | } | |
097b8a7c JS |
395 | spin_unlock(&fs_info->tree_mod_seq_lock); |
396 | ||
bd989ba3 JS |
397 | /* |
398 | * anything that's lower than the lowest existing (read: blocked) | |
399 | * sequence number can be removed from the tree. | |
400 | */ | |
b1a09f1e | 401 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
402 | tm_root = &fs_info->tree_mod_log; |
403 | for (node = rb_first(tm_root); node; node = next) { | |
404 | next = rb_next(node); | |
6b4df8b6 | 405 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 406 | if (tm->seq > min_seq) |
bd989ba3 JS |
407 | continue; |
408 | rb_erase(node, tm_root); | |
bd989ba3 JS |
409 | kfree(tm); |
410 | } | |
b1a09f1e | 411 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
412 | } |
413 | ||
414 | /* | |
415 | * key order of the log: | |
298cfd36 | 416 | * node/leaf start address -> sequence |
bd989ba3 | 417 | * |
298cfd36 CR |
418 | * The 'start address' is the logical address of the *new* root node |
419 | * for root replace operations, or the logical address of the affected | |
420 | * block for all other operations. | |
5de865ee | 421 | * |
b1a09f1e | 422 | * Note: must be called with write lock for fs_info::tree_mod_log_lock. |
bd989ba3 JS |
423 | */ |
424 | static noinline int | |
425 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
426 | { | |
427 | struct rb_root *tm_root; | |
428 | struct rb_node **new; | |
429 | struct rb_node *parent = NULL; | |
430 | struct tree_mod_elem *cur; | |
c8cc6341 | 431 | |
fcebe456 | 432 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 433 | |
bd989ba3 JS |
434 | tm_root = &fs_info->tree_mod_log; |
435 | new = &tm_root->rb_node; | |
436 | while (*new) { | |
6b4df8b6 | 437 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 438 | parent = *new; |
298cfd36 | 439 | if (cur->logical < tm->logical) |
bd989ba3 | 440 | new = &((*new)->rb_left); |
298cfd36 | 441 | else if (cur->logical > tm->logical) |
bd989ba3 | 442 | new = &((*new)->rb_right); |
097b8a7c | 443 | else if (cur->seq < tm->seq) |
bd989ba3 | 444 | new = &((*new)->rb_left); |
097b8a7c | 445 | else if (cur->seq > tm->seq) |
bd989ba3 | 446 | new = &((*new)->rb_right); |
5de865ee FDBM |
447 | else |
448 | return -EEXIST; | |
bd989ba3 JS |
449 | } |
450 | ||
451 | rb_link_node(&tm->node, parent, new); | |
452 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 453 | return 0; |
bd989ba3 JS |
454 | } |
455 | ||
097b8a7c JS |
456 | /* |
457 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
458 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
459 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 460 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 461 | */ |
e9b7fd4d JS |
462 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
463 | struct extent_buffer *eb) { | |
464 | smp_mb(); | |
465 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
466 | return 1; | |
097b8a7c JS |
467 | if (eb && btrfs_header_level(eb) == 0) |
468 | return 1; | |
5de865ee | 469 | |
b1a09f1e | 470 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 471 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 472 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
473 | return 1; |
474 | } | |
475 | ||
e9b7fd4d JS |
476 | return 0; |
477 | } | |
478 | ||
5de865ee FDBM |
479 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
480 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
481 | struct extent_buffer *eb) | |
482 | { | |
483 | smp_mb(); | |
484 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
485 | return 0; | |
486 | if (eb && btrfs_header_level(eb) == 0) | |
487 | return 0; | |
488 | ||
489 | return 1; | |
490 | } | |
491 | ||
492 | static struct tree_mod_elem * | |
493 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
494 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 495 | { |
097b8a7c | 496 | struct tree_mod_elem *tm; |
bd989ba3 | 497 | |
c8cc6341 JB |
498 | tm = kzalloc(sizeof(*tm), flags); |
499 | if (!tm) | |
5de865ee | 500 | return NULL; |
bd989ba3 | 501 | |
298cfd36 | 502 | tm->logical = eb->start; |
bd989ba3 JS |
503 | if (op != MOD_LOG_KEY_ADD) { |
504 | btrfs_node_key(eb, &tm->key, slot); | |
505 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
506 | } | |
507 | tm->op = op; | |
508 | tm->slot = slot; | |
509 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 510 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 511 | |
5de865ee | 512 | return tm; |
097b8a7c JS |
513 | } |
514 | ||
e09c2efe DS |
515 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
516 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 517 | { |
5de865ee FDBM |
518 | struct tree_mod_elem *tm; |
519 | int ret; | |
520 | ||
e09c2efe | 521 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
522 | return 0; |
523 | ||
524 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
525 | if (!tm) | |
526 | return -ENOMEM; | |
527 | ||
e09c2efe | 528 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 529 | kfree(tm); |
097b8a7c | 530 | return 0; |
5de865ee FDBM |
531 | } |
532 | ||
e09c2efe | 533 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 534 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
535 | if (ret) |
536 | kfree(tm); | |
097b8a7c | 537 | |
5de865ee | 538 | return ret; |
097b8a7c JS |
539 | } |
540 | ||
6074d45f DS |
541 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
542 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 543 | { |
5de865ee FDBM |
544 | struct tree_mod_elem *tm = NULL; |
545 | struct tree_mod_elem **tm_list = NULL; | |
546 | int ret = 0; | |
bd989ba3 | 547 | int i; |
5de865ee | 548 | int locked = 0; |
bd989ba3 | 549 | |
6074d45f | 550 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 551 | return 0; |
bd989ba3 | 552 | |
176ef8f5 | 553 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
554 | if (!tm_list) |
555 | return -ENOMEM; | |
556 | ||
176ef8f5 | 557 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
558 | if (!tm) { |
559 | ret = -ENOMEM; | |
560 | goto free_tms; | |
561 | } | |
562 | ||
298cfd36 | 563 | tm->logical = eb->start; |
5de865ee FDBM |
564 | tm->slot = src_slot; |
565 | tm->move.dst_slot = dst_slot; | |
566 | tm->move.nr_items = nr_items; | |
567 | tm->op = MOD_LOG_MOVE_KEYS; | |
568 | ||
569 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
570 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 571 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
572 | if (!tm_list[i]) { |
573 | ret = -ENOMEM; | |
574 | goto free_tms; | |
575 | } | |
576 | } | |
577 | ||
6074d45f | 578 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
579 | goto free_tms; |
580 | locked = 1; | |
581 | ||
01763a2e JS |
582 | /* |
583 | * When we override something during the move, we log these removals. | |
584 | * This can only happen when we move towards the beginning of the | |
585 | * buffer, i.e. dst_slot < src_slot. | |
586 | */ | |
bd989ba3 | 587 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 588 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
589 | if (ret) |
590 | goto free_tms; | |
bd989ba3 JS |
591 | } |
592 | ||
6074d45f | 593 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
594 | if (ret) |
595 | goto free_tms; | |
b1a09f1e | 596 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 597 | kfree(tm_list); |
f395694c | 598 | |
5de865ee FDBM |
599 | return 0; |
600 | free_tms: | |
601 | for (i = 0; i < nr_items; i++) { | |
602 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 603 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
604 | kfree(tm_list[i]); |
605 | } | |
606 | if (locked) | |
b1a09f1e | 607 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
608 | kfree(tm_list); |
609 | kfree(tm); | |
bd989ba3 | 610 | |
5de865ee | 611 | return ret; |
bd989ba3 JS |
612 | } |
613 | ||
5de865ee FDBM |
614 | static inline int |
615 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
616 | struct tree_mod_elem **tm_list, | |
617 | int nritems) | |
097b8a7c | 618 | { |
5de865ee | 619 | int i, j; |
097b8a7c JS |
620 | int ret; |
621 | ||
097b8a7c | 622 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
623 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
624 | if (ret) { | |
625 | for (j = nritems - 1; j > i; j--) | |
626 | rb_erase(&tm_list[j]->node, | |
627 | &fs_info->tree_mod_log); | |
628 | return ret; | |
629 | } | |
097b8a7c | 630 | } |
5de865ee FDBM |
631 | |
632 | return 0; | |
097b8a7c JS |
633 | } |
634 | ||
95b757c1 DS |
635 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
636 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 637 | { |
95b757c1 | 638 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
639 | struct tree_mod_elem *tm = NULL; |
640 | struct tree_mod_elem **tm_list = NULL; | |
641 | int nritems = 0; | |
642 | int ret = 0; | |
643 | int i; | |
bd989ba3 | 644 | |
5de865ee | 645 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
646 | return 0; |
647 | ||
5de865ee FDBM |
648 | if (log_removal && btrfs_header_level(old_root) > 0) { |
649 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 650 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 651 | GFP_NOFS); |
5de865ee FDBM |
652 | if (!tm_list) { |
653 | ret = -ENOMEM; | |
654 | goto free_tms; | |
655 | } | |
656 | for (i = 0; i < nritems; i++) { | |
657 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 658 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
659 | if (!tm_list[i]) { |
660 | ret = -ENOMEM; | |
661 | goto free_tms; | |
662 | } | |
663 | } | |
664 | } | |
d9abbf1c | 665 | |
bcc8e07f | 666 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
667 | if (!tm) { |
668 | ret = -ENOMEM; | |
669 | goto free_tms; | |
670 | } | |
bd989ba3 | 671 | |
298cfd36 | 672 | tm->logical = new_root->start; |
bd989ba3 JS |
673 | tm->old_root.logical = old_root->start; |
674 | tm->old_root.level = btrfs_header_level(old_root); | |
675 | tm->generation = btrfs_header_generation(old_root); | |
676 | tm->op = MOD_LOG_ROOT_REPLACE; | |
677 | ||
5de865ee FDBM |
678 | if (tree_mod_dont_log(fs_info, NULL)) |
679 | goto free_tms; | |
680 | ||
681 | if (tm_list) | |
682 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
683 | if (!ret) | |
684 | ret = __tree_mod_log_insert(fs_info, tm); | |
685 | ||
b1a09f1e | 686 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
687 | if (ret) |
688 | goto free_tms; | |
689 | kfree(tm_list); | |
690 | ||
691 | return ret; | |
692 | ||
693 | free_tms: | |
694 | if (tm_list) { | |
695 | for (i = 0; i < nritems; i++) | |
696 | kfree(tm_list[i]); | |
697 | kfree(tm_list); | |
698 | } | |
699 | kfree(tm); | |
700 | ||
701 | return ret; | |
bd989ba3 JS |
702 | } |
703 | ||
704 | static struct tree_mod_elem * | |
705 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
706 | int smallest) | |
707 | { | |
708 | struct rb_root *tm_root; | |
709 | struct rb_node *node; | |
710 | struct tree_mod_elem *cur = NULL; | |
711 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 712 | |
b1a09f1e | 713 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
714 | tm_root = &fs_info->tree_mod_log; |
715 | node = tm_root->rb_node; | |
716 | while (node) { | |
6b4df8b6 | 717 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 718 | if (cur->logical < start) { |
bd989ba3 | 719 | node = node->rb_left; |
298cfd36 | 720 | } else if (cur->logical > start) { |
bd989ba3 | 721 | node = node->rb_right; |
097b8a7c | 722 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
723 | node = node->rb_left; |
724 | } else if (!smallest) { | |
725 | /* we want the node with the highest seq */ | |
726 | if (found) | |
097b8a7c | 727 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
728 | found = cur; |
729 | node = node->rb_left; | |
097b8a7c | 730 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
731 | /* we want the node with the smallest seq */ |
732 | if (found) | |
097b8a7c | 733 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
734 | found = cur; |
735 | node = node->rb_right; | |
736 | } else { | |
737 | found = cur; | |
738 | break; | |
739 | } | |
740 | } | |
b1a09f1e | 741 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
742 | |
743 | return found; | |
744 | } | |
745 | ||
746 | /* | |
747 | * this returns the element from the log with the smallest time sequence | |
748 | * value that's in the log (the oldest log item). any element with a time | |
749 | * sequence lower than min_seq will be ignored. | |
750 | */ | |
751 | static struct tree_mod_elem * | |
752 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
753 | u64 min_seq) | |
754 | { | |
755 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
756 | } | |
757 | ||
758 | /* | |
759 | * this returns the element from the log with the largest time sequence | |
760 | * value that's in the log (the most recent log item). any element with | |
761 | * a time sequence lower than min_seq will be ignored. | |
762 | */ | |
763 | static struct tree_mod_elem * | |
764 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
765 | { | |
766 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
767 | } | |
768 | ||
5de865ee | 769 | static noinline int |
bd989ba3 JS |
770 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
771 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 772 | unsigned long src_offset, int nr_items) |
bd989ba3 | 773 | { |
5de865ee FDBM |
774 | int ret = 0; |
775 | struct tree_mod_elem **tm_list = NULL; | |
776 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 777 | int i; |
5de865ee | 778 | int locked = 0; |
bd989ba3 | 779 | |
5de865ee FDBM |
780 | if (!tree_mod_need_log(fs_info, NULL)) |
781 | return 0; | |
bd989ba3 | 782 | |
c8cc6341 | 783 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
784 | return 0; |
785 | ||
31e818fe | 786 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
787 | GFP_NOFS); |
788 | if (!tm_list) | |
789 | return -ENOMEM; | |
bd989ba3 | 790 | |
5de865ee FDBM |
791 | tm_list_add = tm_list; |
792 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 793 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
794 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
795 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
796 | if (!tm_list_rem[i]) { | |
797 | ret = -ENOMEM; | |
798 | goto free_tms; | |
799 | } | |
800 | ||
801 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
802 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
803 | if (!tm_list_add[i]) { | |
804 | ret = -ENOMEM; | |
805 | goto free_tms; | |
806 | } | |
807 | } | |
808 | ||
809 | if (tree_mod_dont_log(fs_info, NULL)) | |
810 | goto free_tms; | |
811 | locked = 1; | |
812 | ||
813 | for (i = 0; i < nr_items; i++) { | |
814 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
815 | if (ret) | |
816 | goto free_tms; | |
817 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
818 | if (ret) | |
819 | goto free_tms; | |
bd989ba3 | 820 | } |
5de865ee | 821 | |
b1a09f1e | 822 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
823 | kfree(tm_list); |
824 | ||
825 | return 0; | |
826 | ||
827 | free_tms: | |
828 | for (i = 0; i < nr_items * 2; i++) { | |
829 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
830 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
831 | kfree(tm_list[i]); | |
832 | } | |
833 | if (locked) | |
b1a09f1e | 834 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
835 | kfree(tm_list); |
836 | ||
837 | return ret; | |
bd989ba3 JS |
838 | } |
839 | ||
db7279a2 | 840 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 841 | { |
5de865ee FDBM |
842 | struct tree_mod_elem **tm_list = NULL; |
843 | int nritems = 0; | |
844 | int i; | |
845 | int ret = 0; | |
846 | ||
847 | if (btrfs_header_level(eb) == 0) | |
848 | return 0; | |
849 | ||
db7279a2 | 850 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
851 | return 0; |
852 | ||
853 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 854 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
855 | if (!tm_list) |
856 | return -ENOMEM; | |
857 | ||
858 | for (i = 0; i < nritems; i++) { | |
859 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
860 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
861 | if (!tm_list[i]) { | |
862 | ret = -ENOMEM; | |
863 | goto free_tms; | |
864 | } | |
865 | } | |
866 | ||
db7279a2 | 867 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
868 | goto free_tms; |
869 | ||
db7279a2 | 870 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 871 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
872 | if (ret) |
873 | goto free_tms; | |
874 | kfree(tm_list); | |
875 | ||
876 | return 0; | |
877 | ||
878 | free_tms: | |
879 | for (i = 0; i < nritems; i++) | |
880 | kfree(tm_list[i]); | |
881 | kfree(tm_list); | |
882 | ||
883 | return ret; | |
bd989ba3 JS |
884 | } |
885 | ||
5d4f98a2 YZ |
886 | /* |
887 | * check if the tree block can be shared by multiple trees | |
888 | */ | |
889 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
890 | struct extent_buffer *buf) | |
891 | { | |
892 | /* | |
01327610 | 893 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
894 | * are never shared. If a block was allocated after the last |
895 | * snapshot and the block was not allocated by tree relocation, | |
896 | * we know the block is not shared. | |
897 | */ | |
27cdeb70 | 898 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
899 | buf != root->node && buf != root->commit_root && |
900 | (btrfs_header_generation(buf) <= | |
901 | btrfs_root_last_snapshot(&root->root_item) || | |
902 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
903 | return 1; | |
904 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 905 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
906 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
907 | return 1; | |
908 | #endif | |
909 | return 0; | |
910 | } | |
911 | ||
912 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
913 | struct btrfs_root *root, | |
914 | struct extent_buffer *buf, | |
f0486c68 YZ |
915 | struct extent_buffer *cow, |
916 | int *last_ref) | |
5d4f98a2 | 917 | { |
0b246afa | 918 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
919 | u64 refs; |
920 | u64 owner; | |
921 | u64 flags; | |
922 | u64 new_flags = 0; | |
923 | int ret; | |
924 | ||
925 | /* | |
926 | * Backrefs update rules: | |
927 | * | |
928 | * Always use full backrefs for extent pointers in tree block | |
929 | * allocated by tree relocation. | |
930 | * | |
931 | * If a shared tree block is no longer referenced by its owner | |
932 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
933 | * use full backrefs for extent pointers in tree block. | |
934 | * | |
935 | * If a tree block is been relocating | |
936 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
937 | * use full backrefs for extent pointers in tree block. | |
938 | * The reason for this is some operations (such as drop tree) | |
939 | * are only allowed for blocks use full backrefs. | |
940 | */ | |
941 | ||
942 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 943 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
944 | btrfs_header_level(buf), 1, |
945 | &refs, &flags); | |
be1a5564 MF |
946 | if (ret) |
947 | return ret; | |
e5df9573 MF |
948 | if (refs == 0) { |
949 | ret = -EROFS; | |
0b246afa | 950 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
951 | return ret; |
952 | } | |
5d4f98a2 YZ |
953 | } else { |
954 | refs = 1; | |
955 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
956 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
957 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
958 | else | |
959 | flags = 0; | |
960 | } | |
961 | ||
962 | owner = btrfs_header_owner(buf); | |
963 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
964 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
965 | ||
966 | if (refs > 1) { | |
967 | if ((owner == root->root_key.objectid || | |
968 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
969 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 970 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
971 | if (ret) |
972 | return ret; | |
5d4f98a2 YZ |
973 | |
974 | if (root->root_key.objectid == | |
975 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 976 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
977 | if (ret) |
978 | return ret; | |
e339a6b0 | 979 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
980 | if (ret) |
981 | return ret; | |
5d4f98a2 YZ |
982 | } |
983 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
984 | } else { | |
985 | ||
986 | if (root->root_key.objectid == | |
987 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 988 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 989 | else |
e339a6b0 | 990 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
991 | if (ret) |
992 | return ret; | |
5d4f98a2 YZ |
993 | } |
994 | if (new_flags != 0) { | |
b1c79e09 JB |
995 | int level = btrfs_header_level(buf); |
996 | ||
2ff7e61e | 997 | ret = btrfs_set_disk_extent_flags(trans, fs_info, |
5d4f98a2 YZ |
998 | buf->start, |
999 | buf->len, | |
b1c79e09 | 1000 | new_flags, level, 0); |
be1a5564 MF |
1001 | if (ret) |
1002 | return ret; | |
5d4f98a2 YZ |
1003 | } |
1004 | } else { | |
1005 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
1006 | if (root->root_key.objectid == | |
1007 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1008 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1009 | else |
e339a6b0 | 1010 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
1011 | if (ret) |
1012 | return ret; | |
e339a6b0 | 1013 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
1014 | if (ret) |
1015 | return ret; | |
5d4f98a2 | 1016 | } |
7c302b49 | 1017 | clean_tree_block(fs_info, buf); |
f0486c68 | 1018 | *last_ref = 1; |
5d4f98a2 YZ |
1019 | } |
1020 | return 0; | |
1021 | } | |
1022 | ||
d352ac68 | 1023 | /* |
d397712b CM |
1024 | * does the dirty work in cow of a single block. The parent block (if |
1025 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1026 | * dirty and returned locked. If you modify the block it needs to be marked | |
1027 | * dirty again. | |
d352ac68 CM |
1028 | * |
1029 | * search_start -- an allocation hint for the new block | |
1030 | * | |
d397712b CM |
1031 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1032 | * bytes the allocator should try to find free next to the block it returns. | |
1033 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1034 | */ |
d397712b | 1035 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1036 | struct btrfs_root *root, |
1037 | struct extent_buffer *buf, | |
1038 | struct extent_buffer *parent, int parent_slot, | |
1039 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1040 | u64 search_start, u64 empty_size) |
02217ed2 | 1041 | { |
0b246afa | 1042 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1043 | struct btrfs_disk_key disk_key; |
5f39d397 | 1044 | struct extent_buffer *cow; |
be1a5564 | 1045 | int level, ret; |
f0486c68 | 1046 | int last_ref = 0; |
925baedd | 1047 | int unlock_orig = 0; |
0f5053eb | 1048 | u64 parent_start = 0; |
7bb86316 | 1049 | |
925baedd CM |
1050 | if (*cow_ret == buf) |
1051 | unlock_orig = 1; | |
1052 | ||
b9447ef8 | 1053 | btrfs_assert_tree_locked(buf); |
925baedd | 1054 | |
27cdeb70 | 1055 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1056 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1057 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1058 | trans->transid != root->last_trans); | |
5f39d397 | 1059 | |
7bb86316 | 1060 | level = btrfs_header_level(buf); |
31840ae1 | 1061 | |
5d4f98a2 YZ |
1062 | if (level == 0) |
1063 | btrfs_item_key(buf, &disk_key, 0); | |
1064 | else | |
1065 | btrfs_node_key(buf, &disk_key, 0); | |
1066 | ||
0f5053eb GR |
1067 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1068 | parent_start = parent->start; | |
5d4f98a2 | 1069 | |
4d75f8a9 DS |
1070 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1071 | root->root_key.objectid, &disk_key, level, | |
1072 | search_start, empty_size); | |
54aa1f4d CM |
1073 | if (IS_ERR(cow)) |
1074 | return PTR_ERR(cow); | |
6702ed49 | 1075 | |
b4ce94de CM |
1076 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1077 | ||
58e8012c | 1078 | copy_extent_buffer_full(cow, buf); |
db94535d | 1079 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1080 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1081 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1082 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1083 | BTRFS_HEADER_FLAG_RELOC); | |
1084 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1085 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1086 | else | |
1087 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1088 | |
0b246afa | 1089 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 1090 | |
be1a5564 | 1091 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1092 | if (ret) { |
66642832 | 1093 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1094 | return ret; |
1095 | } | |
1a40e23b | 1096 | |
27cdeb70 | 1097 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1098 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1099 | if (ret) { |
66642832 | 1100 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1101 | return ret; |
93314e3b | 1102 | } |
83d4cfd4 | 1103 | } |
3fd0a558 | 1104 | |
02217ed2 | 1105 | if (buf == root->node) { |
925baedd | 1106 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1107 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1108 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1109 | parent_start = buf->start; | |
925baedd | 1110 | |
5f39d397 | 1111 | extent_buffer_get(cow); |
d9d19a01 DS |
1112 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1113 | BUG_ON(ret < 0); | |
240f62c8 | 1114 | rcu_assign_pointer(root->node, cow); |
925baedd | 1115 | |
f0486c68 | 1116 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1117 | last_ref); |
5f39d397 | 1118 | free_extent_buffer(buf); |
0b86a832 | 1119 | add_root_to_dirty_list(root); |
02217ed2 | 1120 | } else { |
5d4f98a2 | 1121 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1122 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1123 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1124 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1125 | cow->start); |
74493f7a CM |
1126 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1127 | trans->transid); | |
d6025579 | 1128 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1129 | if (last_ref) { |
db7279a2 | 1130 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1131 | if (ret) { |
66642832 | 1132 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1133 | return ret; |
1134 | } | |
1135 | } | |
f0486c68 | 1136 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1137 | last_ref); |
02217ed2 | 1138 | } |
925baedd CM |
1139 | if (unlock_orig) |
1140 | btrfs_tree_unlock(buf); | |
3083ee2e | 1141 | free_extent_buffer_stale(buf); |
ccd467d6 | 1142 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1143 | *cow_ret = cow; |
02217ed2 CM |
1144 | return 0; |
1145 | } | |
1146 | ||
5d9e75c4 JS |
1147 | /* |
1148 | * returns the logical address of the oldest predecessor of the given root. | |
1149 | * entries older than time_seq are ignored. | |
1150 | */ | |
bcd24dab DS |
1151 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1152 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1153 | { |
1154 | struct tree_mod_elem *tm; | |
1155 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1156 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1157 | int looped = 0; |
1158 | ||
1159 | if (!time_seq) | |
35a3621b | 1160 | return NULL; |
5d9e75c4 JS |
1161 | |
1162 | /* | |
298cfd36 CR |
1163 | * the very last operation that's logged for a root is the |
1164 | * replacement operation (if it is replaced at all). this has | |
1165 | * the logical address of the *new* root, making it the very | |
1166 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1167 | */ |
1168 | while (1) { | |
bcd24dab | 1169 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1170 | time_seq); |
1171 | if (!looped && !tm) | |
35a3621b | 1172 | return NULL; |
5d9e75c4 | 1173 | /* |
28da9fb4 JS |
1174 | * if there are no tree operation for the oldest root, we simply |
1175 | * return it. this should only happen if that (old) root is at | |
1176 | * level 0. | |
5d9e75c4 | 1177 | */ |
28da9fb4 JS |
1178 | if (!tm) |
1179 | break; | |
5d9e75c4 | 1180 | |
28da9fb4 JS |
1181 | /* |
1182 | * if there's an operation that's not a root replacement, we | |
1183 | * found the oldest version of our root. normally, we'll find a | |
1184 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1185 | */ | |
5d9e75c4 JS |
1186 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1187 | break; | |
1188 | ||
1189 | found = tm; | |
1190 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1191 | looped = 1; |
1192 | } | |
1193 | ||
a95236d9 JS |
1194 | /* if there's no old root to return, return what we found instead */ |
1195 | if (!found) | |
1196 | found = tm; | |
1197 | ||
5d9e75c4 JS |
1198 | return found; |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1203 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1204 | * time_seq). |
1205 | */ | |
1206 | static void | |
f1ca7e98 JB |
1207 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1208 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1209 | { |
1210 | u32 n; | |
1211 | struct rb_node *next; | |
1212 | struct tree_mod_elem *tm = first_tm; | |
1213 | unsigned long o_dst; | |
1214 | unsigned long o_src; | |
1215 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1216 | ||
1217 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1218 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1219 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1220 | /* |
1221 | * all the operations are recorded with the operator used for | |
1222 | * the modification. as we're going backwards, we do the | |
1223 | * opposite of each operation here. | |
1224 | */ | |
1225 | switch (tm->op) { | |
1226 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1227 | BUG_ON(tm->slot < n); | |
1c697d4a | 1228 | /* Fallthrough */ |
95c80bb1 | 1229 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1230 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1231 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1232 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1233 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1234 | tm->generation); | |
4c3e6969 | 1235 | n++; |
5d9e75c4 JS |
1236 | break; |
1237 | case MOD_LOG_KEY_REPLACE: | |
1238 | BUG_ON(tm->slot >= n); | |
1239 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1240 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1241 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1242 | tm->generation); | |
1243 | break; | |
1244 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1245 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1246 | n--; |
1247 | break; | |
1248 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1249 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1250 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1251 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1252 | tm->move.nr_items * p_size); |
1253 | break; | |
1254 | case MOD_LOG_ROOT_REPLACE: | |
1255 | /* | |
1256 | * this operation is special. for roots, this must be | |
1257 | * handled explicitly before rewinding. | |
1258 | * for non-roots, this operation may exist if the node | |
1259 | * was a root: root A -> child B; then A gets empty and | |
1260 | * B is promoted to the new root. in the mod log, we'll | |
1261 | * have a root-replace operation for B, a tree block | |
1262 | * that is no root. we simply ignore that operation. | |
1263 | */ | |
1264 | break; | |
1265 | } | |
1266 | next = rb_next(&tm->node); | |
1267 | if (!next) | |
1268 | break; | |
6b4df8b6 | 1269 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1270 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1271 | break; |
1272 | } | |
b1a09f1e | 1273 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1274 | btrfs_set_header_nritems(eb, n); |
1275 | } | |
1276 | ||
47fb091f | 1277 | /* |
01327610 | 1278 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1279 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1280 | * returned buffer is always read-locked. If the returned buffer is not the | |
1281 | * input buffer, the lock on the input buffer is released and the input buffer | |
1282 | * is freed (its refcount is decremented). | |
1283 | */ | |
5d9e75c4 | 1284 | static struct extent_buffer * |
9ec72677 JB |
1285 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1286 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1287 | { |
1288 | struct extent_buffer *eb_rewin; | |
1289 | struct tree_mod_elem *tm; | |
1290 | ||
1291 | if (!time_seq) | |
1292 | return eb; | |
1293 | ||
1294 | if (btrfs_header_level(eb) == 0) | |
1295 | return eb; | |
1296 | ||
1297 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1298 | if (!tm) | |
1299 | return eb; | |
1300 | ||
9ec72677 JB |
1301 | btrfs_set_path_blocking(path); |
1302 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1303 | ||
5d9e75c4 JS |
1304 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1305 | BUG_ON(tm->slot != 0); | |
da17066c | 1306 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1307 | if (!eb_rewin) { |
9ec72677 | 1308 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1309 | free_extent_buffer(eb); |
1310 | return NULL; | |
1311 | } | |
5d9e75c4 JS |
1312 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1313 | btrfs_set_header_backref_rev(eb_rewin, | |
1314 | btrfs_header_backref_rev(eb)); | |
1315 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1316 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1317 | } else { |
1318 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1319 | if (!eb_rewin) { |
9ec72677 | 1320 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1321 | free_extent_buffer(eb); |
1322 | return NULL; | |
1323 | } | |
5d9e75c4 JS |
1324 | } |
1325 | ||
9ec72677 JB |
1326 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1327 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1328 | free_extent_buffer(eb); |
1329 | ||
47fb091f JS |
1330 | extent_buffer_get(eb_rewin); |
1331 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1332 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1333 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1334 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1335 | |
1336 | return eb_rewin; | |
1337 | } | |
1338 | ||
8ba97a15 JS |
1339 | /* |
1340 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1341 | * value. If there are no changes, the current root->root_node is returned. If | |
1342 | * anything changed in between, there's a fresh buffer allocated on which the | |
1343 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1344 | * Returns NULL on error (with no locks held). | |
1345 | */ | |
5d9e75c4 JS |
1346 | static inline struct extent_buffer * |
1347 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1348 | { | |
0b246afa | 1349 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1350 | struct tree_mod_elem *tm; |
30b0463a JS |
1351 | struct extent_buffer *eb = NULL; |
1352 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1353 | struct extent_buffer *old; |
a95236d9 | 1354 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1355 | u64 old_generation = 0; |
a95236d9 | 1356 | u64 logical; |
5d9e75c4 | 1357 | |
30b0463a | 1358 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1359 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1360 | if (!tm) |
30b0463a | 1361 | return eb_root; |
5d9e75c4 | 1362 | |
a95236d9 JS |
1363 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1364 | old_root = &tm->old_root; | |
1365 | old_generation = tm->generation; | |
1366 | logical = old_root->logical; | |
1367 | } else { | |
30b0463a | 1368 | logical = eb_root->start; |
a95236d9 | 1369 | } |
5d9e75c4 | 1370 | |
0b246afa | 1371 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1372 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1373 | btrfs_tree_read_unlock(eb_root); |
1374 | free_extent_buffer(eb_root); | |
2ff7e61e | 1375 | old = read_tree_block(fs_info, logical, 0); |
64c043de LB |
1376 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1377 | if (!IS_ERR(old)) | |
1378 | free_extent_buffer(old); | |
0b246afa JM |
1379 | btrfs_warn(fs_info, |
1380 | "failed to read tree block %llu from get_old_root", | |
1381 | logical); | |
834328a8 | 1382 | } else { |
7bfdcf7f LB |
1383 | eb = btrfs_clone_extent_buffer(old); |
1384 | free_extent_buffer(old); | |
834328a8 JS |
1385 | } |
1386 | } else if (old_root) { | |
30b0463a JS |
1387 | btrfs_tree_read_unlock(eb_root); |
1388 | free_extent_buffer(eb_root); | |
0b246afa | 1389 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1390 | } else { |
9ec72677 | 1391 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1392 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1393 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1394 | free_extent_buffer(eb_root); |
834328a8 JS |
1395 | } |
1396 | ||
8ba97a15 JS |
1397 | if (!eb) |
1398 | return NULL; | |
d6381084 | 1399 | extent_buffer_get(eb); |
8ba97a15 | 1400 | btrfs_tree_read_lock(eb); |
a95236d9 | 1401 | if (old_root) { |
5d9e75c4 JS |
1402 | btrfs_set_header_bytenr(eb, eb->start); |
1403 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1404 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1405 | btrfs_set_header_level(eb, old_root->level); |
1406 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1407 | } |
28da9fb4 | 1408 | if (tm) |
0b246afa | 1409 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1410 | else |
1411 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1412 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1413 | |
1414 | return eb; | |
1415 | } | |
1416 | ||
5b6602e7 JS |
1417 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1418 | { | |
1419 | struct tree_mod_elem *tm; | |
1420 | int level; | |
30b0463a | 1421 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1422 | |
bcd24dab | 1423 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1424 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1425 | level = tm->old_root.level; | |
1426 | } else { | |
30b0463a | 1427 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1428 | } |
30b0463a | 1429 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1430 | |
1431 | return level; | |
1432 | } | |
1433 | ||
5d4f98a2 YZ |
1434 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1435 | struct btrfs_root *root, | |
1436 | struct extent_buffer *buf) | |
1437 | { | |
f5ee5c9a | 1438 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1439 | return 0; |
fccb84c9 | 1440 | |
f1ebcc74 LB |
1441 | /* ensure we can see the force_cow */ |
1442 | smp_rmb(); | |
1443 | ||
1444 | /* | |
1445 | * We do not need to cow a block if | |
1446 | * 1) this block is not created or changed in this transaction; | |
1447 | * 2) this block does not belong to TREE_RELOC tree; | |
1448 | * 3) the root is not forced COW. | |
1449 | * | |
1450 | * What is forced COW: | |
01327610 | 1451 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1452 | * after we've finished coping src root, we must COW the shared |
1453 | * block to ensure the metadata consistency. | |
1454 | */ | |
5d4f98a2 YZ |
1455 | if (btrfs_header_generation(buf) == trans->transid && |
1456 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1457 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1458 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1459 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1460 | return 0; |
1461 | return 1; | |
1462 | } | |
1463 | ||
d352ac68 CM |
1464 | /* |
1465 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1466 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1467 | * once per transaction, as long as it hasn't been written yet |
1468 | */ | |
d397712b | 1469 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1470 | struct btrfs_root *root, struct extent_buffer *buf, |
1471 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1472 | struct extent_buffer **cow_ret) |
6702ed49 | 1473 | { |
0b246afa | 1474 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1475 | u64 search_start; |
f510cfec | 1476 | int ret; |
dc17ff8f | 1477 | |
0b246afa | 1478 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1479 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1480 | trans->transid, |
0b246afa | 1481 | fs_info->running_transaction->transid); |
31b1a2bd | 1482 | |
0b246afa | 1483 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1484 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1485 | trans->transid, fs_info->generation); |
dc17ff8f | 1486 | |
5d4f98a2 | 1487 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1488 | trans->dirty = true; |
6702ed49 CM |
1489 | *cow_ret = buf; |
1490 | return 0; | |
1491 | } | |
c487685d | 1492 | |
ee22184b | 1493 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1494 | |
1495 | if (parent) | |
1496 | btrfs_set_lock_blocking(parent); | |
1497 | btrfs_set_lock_blocking(buf); | |
1498 | ||
f510cfec | 1499 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1500 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1501 | |
1502 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1503 | ||
f510cfec | 1504 | return ret; |
6702ed49 CM |
1505 | } |
1506 | ||
d352ac68 CM |
1507 | /* |
1508 | * helper function for defrag to decide if two blocks pointed to by a | |
1509 | * node are actually close by | |
1510 | */ | |
6b80053d | 1511 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1512 | { |
6b80053d | 1513 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1514 | return 1; |
6b80053d | 1515 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1516 | return 1; |
1517 | return 0; | |
1518 | } | |
1519 | ||
081e9573 CM |
1520 | /* |
1521 | * compare two keys in a memcmp fashion | |
1522 | */ | |
310712b2 OS |
1523 | static int comp_keys(const struct btrfs_disk_key *disk, |
1524 | const struct btrfs_key *k2) | |
081e9573 CM |
1525 | { |
1526 | struct btrfs_key k1; | |
1527 | ||
1528 | btrfs_disk_key_to_cpu(&k1, disk); | |
1529 | ||
20736aba | 1530 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1531 | } |
1532 | ||
f3465ca4 JB |
1533 | /* |
1534 | * same as comp_keys only with two btrfs_key's | |
1535 | */ | |
310712b2 | 1536 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1537 | { |
1538 | if (k1->objectid > k2->objectid) | |
1539 | return 1; | |
1540 | if (k1->objectid < k2->objectid) | |
1541 | return -1; | |
1542 | if (k1->type > k2->type) | |
1543 | return 1; | |
1544 | if (k1->type < k2->type) | |
1545 | return -1; | |
1546 | if (k1->offset > k2->offset) | |
1547 | return 1; | |
1548 | if (k1->offset < k2->offset) | |
1549 | return -1; | |
1550 | return 0; | |
1551 | } | |
081e9573 | 1552 | |
d352ac68 CM |
1553 | /* |
1554 | * this is used by the defrag code to go through all the | |
1555 | * leaves pointed to by a node and reallocate them so that | |
1556 | * disk order is close to key order | |
1557 | */ | |
6702ed49 | 1558 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1559 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1560 | int start_slot, u64 *last_ret, |
a6b6e75e | 1561 | struct btrfs_key *progress) |
6702ed49 | 1562 | { |
0b246afa | 1563 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1564 | struct extent_buffer *cur; |
6702ed49 | 1565 | u64 blocknr; |
ca7a79ad | 1566 | u64 gen; |
e9d0b13b CM |
1567 | u64 search_start = *last_ret; |
1568 | u64 last_block = 0; | |
6702ed49 CM |
1569 | u64 other; |
1570 | u32 parent_nritems; | |
6702ed49 CM |
1571 | int end_slot; |
1572 | int i; | |
1573 | int err = 0; | |
f2183bde | 1574 | int parent_level; |
6b80053d CM |
1575 | int uptodate; |
1576 | u32 blocksize; | |
081e9573 CM |
1577 | int progress_passed = 0; |
1578 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1579 | |
5708b959 | 1580 | parent_level = btrfs_header_level(parent); |
5708b959 | 1581 | |
0b246afa JM |
1582 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1583 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1584 | |
6b80053d | 1585 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1586 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1587 | end_slot = parent_nritems - 1; |
6702ed49 | 1588 | |
5dfe2be7 | 1589 | if (parent_nritems <= 1) |
6702ed49 CM |
1590 | return 0; |
1591 | ||
b4ce94de CM |
1592 | btrfs_set_lock_blocking(parent); |
1593 | ||
5dfe2be7 | 1594 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 1595 | int close = 1; |
a6b6e75e | 1596 | |
081e9573 CM |
1597 | btrfs_node_key(parent, &disk_key, i); |
1598 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1599 | continue; | |
1600 | ||
1601 | progress_passed = 1; | |
6b80053d | 1602 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1603 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1604 | if (last_block == 0) |
1605 | last_block = blocknr; | |
5708b959 | 1606 | |
6702ed49 | 1607 | if (i > 0) { |
6b80053d CM |
1608 | other = btrfs_node_blockptr(parent, i - 1); |
1609 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1610 | } |
5dfe2be7 | 1611 | if (!close && i < end_slot) { |
6b80053d CM |
1612 | other = btrfs_node_blockptr(parent, i + 1); |
1613 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1614 | } |
e9d0b13b CM |
1615 | if (close) { |
1616 | last_block = blocknr; | |
6702ed49 | 1617 | continue; |
e9d0b13b | 1618 | } |
6702ed49 | 1619 | |
0b246afa | 1620 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1621 | if (cur) |
b9fab919 | 1622 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1623 | else |
1624 | uptodate = 0; | |
5708b959 | 1625 | if (!cur || !uptodate) { |
6b80053d | 1626 | if (!cur) { |
2ff7e61e | 1627 | cur = read_tree_block(fs_info, blocknr, gen); |
64c043de LB |
1628 | if (IS_ERR(cur)) { |
1629 | return PTR_ERR(cur); | |
1630 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1631 | free_extent_buffer(cur); |
97d9a8a4 | 1632 | return -EIO; |
416bc658 | 1633 | } |
6b80053d | 1634 | } else if (!uptodate) { |
018642a1 TI |
1635 | err = btrfs_read_buffer(cur, gen); |
1636 | if (err) { | |
1637 | free_extent_buffer(cur); | |
1638 | return err; | |
1639 | } | |
f2183bde | 1640 | } |
6702ed49 | 1641 | } |
e9d0b13b | 1642 | if (search_start == 0) |
6b80053d | 1643 | search_start = last_block; |
e9d0b13b | 1644 | |
e7a84565 | 1645 | btrfs_tree_lock(cur); |
b4ce94de | 1646 | btrfs_set_lock_blocking(cur); |
6b80053d | 1647 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1648 | &cur, search_start, |
6b80053d | 1649 | min(16 * blocksize, |
9fa8cfe7 | 1650 | (end_slot - i) * blocksize)); |
252c38f0 | 1651 | if (err) { |
e7a84565 | 1652 | btrfs_tree_unlock(cur); |
6b80053d | 1653 | free_extent_buffer(cur); |
6702ed49 | 1654 | break; |
252c38f0 | 1655 | } |
e7a84565 CM |
1656 | search_start = cur->start; |
1657 | last_block = cur->start; | |
f2183bde | 1658 | *last_ret = search_start; |
e7a84565 CM |
1659 | btrfs_tree_unlock(cur); |
1660 | free_extent_buffer(cur); | |
6702ed49 CM |
1661 | } |
1662 | return err; | |
1663 | } | |
1664 | ||
74123bd7 | 1665 | /* |
5f39d397 CM |
1666 | * search for key in the extent_buffer. The items start at offset p, |
1667 | * and they are item_size apart. There are 'max' items in p. | |
1668 | * | |
74123bd7 CM |
1669 | * the slot in the array is returned via slot, and it points to |
1670 | * the place where you would insert key if it is not found in | |
1671 | * the array. | |
1672 | * | |
1673 | * slot may point to max if the key is bigger than all of the keys | |
1674 | */ | |
e02119d5 | 1675 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1676 | unsigned long p, int item_size, |
1677 | const struct btrfs_key *key, | |
e02119d5 | 1678 | int max, int *slot) |
be0e5c09 CM |
1679 | { |
1680 | int low = 0; | |
1681 | int high = max; | |
1682 | int mid; | |
1683 | int ret; | |
479965d6 | 1684 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1685 | struct btrfs_disk_key unaligned; |
1686 | unsigned long offset; | |
5f39d397 CM |
1687 | char *kaddr = NULL; |
1688 | unsigned long map_start = 0; | |
1689 | unsigned long map_len = 0; | |
479965d6 | 1690 | int err; |
be0e5c09 | 1691 | |
5e24e9af LB |
1692 | if (low > high) { |
1693 | btrfs_err(eb->fs_info, | |
1694 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1695 | __func__, low, high, eb->start, | |
1696 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1697 | return -EINVAL; | |
1698 | } | |
1699 | ||
d397712b | 1700 | while (low < high) { |
be0e5c09 | 1701 | mid = (low + high) / 2; |
5f39d397 CM |
1702 | offset = p + mid * item_size; |
1703 | ||
a6591715 | 1704 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1705 | (offset + sizeof(struct btrfs_disk_key)) > |
1706 | map_start + map_len) { | |
934d375b CM |
1707 | |
1708 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1709 | sizeof(struct btrfs_disk_key), |
a6591715 | 1710 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1711 | |
1712 | if (!err) { | |
1713 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1714 | map_start); | |
415b35a5 | 1715 | } else if (err == 1) { |
479965d6 CM |
1716 | read_extent_buffer(eb, &unaligned, |
1717 | offset, sizeof(unaligned)); | |
1718 | tmp = &unaligned; | |
415b35a5 LB |
1719 | } else { |
1720 | return err; | |
479965d6 | 1721 | } |
5f39d397 | 1722 | |
5f39d397 CM |
1723 | } else { |
1724 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1725 | map_start); | |
1726 | } | |
be0e5c09 CM |
1727 | ret = comp_keys(tmp, key); |
1728 | ||
1729 | if (ret < 0) | |
1730 | low = mid + 1; | |
1731 | else if (ret > 0) | |
1732 | high = mid; | |
1733 | else { | |
1734 | *slot = mid; | |
1735 | return 0; | |
1736 | } | |
1737 | } | |
1738 | *slot = low; | |
1739 | return 1; | |
1740 | } | |
1741 | ||
97571fd0 CM |
1742 | /* |
1743 | * simple bin_search frontend that does the right thing for | |
1744 | * leaves vs nodes | |
1745 | */ | |
a74b35ec NB |
1746 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1747 | int level, int *slot) | |
be0e5c09 | 1748 | { |
f775738f | 1749 | if (level == 0) |
5f39d397 CM |
1750 | return generic_bin_search(eb, |
1751 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1752 | sizeof(struct btrfs_item), |
5f39d397 | 1753 | key, btrfs_header_nritems(eb), |
7518a238 | 1754 | slot); |
f775738f | 1755 | else |
5f39d397 CM |
1756 | return generic_bin_search(eb, |
1757 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1758 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1759 | key, btrfs_header_nritems(eb), |
7518a238 | 1760 | slot); |
be0e5c09 CM |
1761 | } |
1762 | ||
f0486c68 YZ |
1763 | static void root_add_used(struct btrfs_root *root, u32 size) |
1764 | { | |
1765 | spin_lock(&root->accounting_lock); | |
1766 | btrfs_set_root_used(&root->root_item, | |
1767 | btrfs_root_used(&root->root_item) + size); | |
1768 | spin_unlock(&root->accounting_lock); | |
1769 | } | |
1770 | ||
1771 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1772 | { | |
1773 | spin_lock(&root->accounting_lock); | |
1774 | btrfs_set_root_used(&root->root_item, | |
1775 | btrfs_root_used(&root->root_item) - size); | |
1776 | spin_unlock(&root->accounting_lock); | |
1777 | } | |
1778 | ||
d352ac68 CM |
1779 | /* given a node and slot number, this reads the blocks it points to. The |
1780 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1781 | */ |
2ff7e61e JM |
1782 | static noinline struct extent_buffer * |
1783 | read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent, | |
1784 | int slot) | |
bb803951 | 1785 | { |
ca7a79ad | 1786 | int level = btrfs_header_level(parent); |
416bc658 JB |
1787 | struct extent_buffer *eb; |
1788 | ||
fb770ae4 LB |
1789 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1790 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1791 | |
1792 | BUG_ON(level == 0); | |
1793 | ||
2ff7e61e | 1794 | eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot), |
416bc658 | 1795 | btrfs_node_ptr_generation(parent, slot)); |
fb770ae4 LB |
1796 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1797 | free_extent_buffer(eb); | |
1798 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1799 | } |
1800 | ||
1801 | return eb; | |
bb803951 CM |
1802 | } |
1803 | ||
d352ac68 CM |
1804 | /* |
1805 | * node level balancing, used to make sure nodes are in proper order for | |
1806 | * item deletion. We balance from the top down, so we have to make sure | |
1807 | * that a deletion won't leave an node completely empty later on. | |
1808 | */ | |
e02119d5 | 1809 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1810 | struct btrfs_root *root, |
1811 | struct btrfs_path *path, int level) | |
bb803951 | 1812 | { |
0b246afa | 1813 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1814 | struct extent_buffer *right = NULL; |
1815 | struct extent_buffer *mid; | |
1816 | struct extent_buffer *left = NULL; | |
1817 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1818 | int ret = 0; |
1819 | int wret; | |
1820 | int pslot; | |
bb803951 | 1821 | int orig_slot = path->slots[level]; |
79f95c82 | 1822 | u64 orig_ptr; |
bb803951 CM |
1823 | |
1824 | if (level == 0) | |
1825 | return 0; | |
1826 | ||
5f39d397 | 1827 | mid = path->nodes[level]; |
b4ce94de | 1828 | |
bd681513 CM |
1829 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1830 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1831 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1832 | ||
1d4f8a0c | 1833 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1834 | |
a05a9bb1 | 1835 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1836 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1837 | pslot = path->slots[level + 1]; |
1838 | } | |
bb803951 | 1839 | |
40689478 CM |
1840 | /* |
1841 | * deal with the case where there is only one pointer in the root | |
1842 | * by promoting the node below to a root | |
1843 | */ | |
5f39d397 CM |
1844 | if (!parent) { |
1845 | struct extent_buffer *child; | |
bb803951 | 1846 | |
5f39d397 | 1847 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1848 | return 0; |
1849 | ||
1850 | /* promote the child to a root */ | |
2ff7e61e | 1851 | child = read_node_slot(fs_info, mid, 0); |
fb770ae4 LB |
1852 | if (IS_ERR(child)) { |
1853 | ret = PTR_ERR(child); | |
0b246afa | 1854 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1855 | goto enospc; |
1856 | } | |
1857 | ||
925baedd | 1858 | btrfs_tree_lock(child); |
b4ce94de | 1859 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1860 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1861 | if (ret) { |
1862 | btrfs_tree_unlock(child); | |
1863 | free_extent_buffer(child); | |
1864 | goto enospc; | |
1865 | } | |
2f375ab9 | 1866 | |
d9d19a01 DS |
1867 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1868 | BUG_ON(ret < 0); | |
240f62c8 | 1869 | rcu_assign_pointer(root->node, child); |
925baedd | 1870 | |
0b86a832 | 1871 | add_root_to_dirty_list(root); |
925baedd | 1872 | btrfs_tree_unlock(child); |
b4ce94de | 1873 | |
925baedd | 1874 | path->locks[level] = 0; |
bb803951 | 1875 | path->nodes[level] = NULL; |
7c302b49 | 1876 | clean_tree_block(fs_info, mid); |
925baedd | 1877 | btrfs_tree_unlock(mid); |
bb803951 | 1878 | /* once for the path */ |
5f39d397 | 1879 | free_extent_buffer(mid); |
f0486c68 YZ |
1880 | |
1881 | root_sub_used(root, mid->len); | |
5581a51a | 1882 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1883 | /* once for the root ptr */ |
3083ee2e | 1884 | free_extent_buffer_stale(mid); |
f0486c68 | 1885 | return 0; |
bb803951 | 1886 | } |
5f39d397 | 1887 | if (btrfs_header_nritems(mid) > |
0b246afa | 1888 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1889 | return 0; |
1890 | ||
2ff7e61e | 1891 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
1892 | if (IS_ERR(left)) |
1893 | left = NULL; | |
1894 | ||
5f39d397 | 1895 | if (left) { |
925baedd | 1896 | btrfs_tree_lock(left); |
b4ce94de | 1897 | btrfs_set_lock_blocking(left); |
5f39d397 | 1898 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1899 | parent, pslot - 1, &left); |
54aa1f4d CM |
1900 | if (wret) { |
1901 | ret = wret; | |
1902 | goto enospc; | |
1903 | } | |
2cc58cf2 | 1904 | } |
fb770ae4 | 1905 | |
2ff7e61e | 1906 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
1907 | if (IS_ERR(right)) |
1908 | right = NULL; | |
1909 | ||
5f39d397 | 1910 | if (right) { |
925baedd | 1911 | btrfs_tree_lock(right); |
b4ce94de | 1912 | btrfs_set_lock_blocking(right); |
5f39d397 | 1913 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1914 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1915 | if (wret) { |
1916 | ret = wret; | |
1917 | goto enospc; | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1922 | if (left) { |
1923 | orig_slot += btrfs_header_nritems(left); | |
2ff7e61e | 1924 | wret = push_node_left(trans, fs_info, left, mid, 1); |
79f95c82 CM |
1925 | if (wret < 0) |
1926 | ret = wret; | |
bb803951 | 1927 | } |
79f95c82 CM |
1928 | |
1929 | /* | |
1930 | * then try to empty the right most buffer into the middle | |
1931 | */ | |
5f39d397 | 1932 | if (right) { |
2ff7e61e | 1933 | wret = push_node_left(trans, fs_info, mid, right, 1); |
54aa1f4d | 1934 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1935 | ret = wret; |
5f39d397 | 1936 | if (btrfs_header_nritems(right) == 0) { |
7c302b49 | 1937 | clean_tree_block(fs_info, right); |
925baedd | 1938 | btrfs_tree_unlock(right); |
afe5fea7 | 1939 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1940 | root_sub_used(root, right->len); |
5581a51a | 1941 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1942 | free_extent_buffer_stale(right); |
f0486c68 | 1943 | right = NULL; |
bb803951 | 1944 | } else { |
5f39d397 CM |
1945 | struct btrfs_disk_key right_key; |
1946 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1947 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1948 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1949 | BUG_ON(ret < 0); | |
5f39d397 CM |
1950 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1951 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1952 | } |
1953 | } | |
5f39d397 | 1954 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1955 | /* |
1956 | * we're not allowed to leave a node with one item in the | |
1957 | * tree during a delete. A deletion from lower in the tree | |
1958 | * could try to delete the only pointer in this node. | |
1959 | * So, pull some keys from the left. | |
1960 | * There has to be a left pointer at this point because | |
1961 | * otherwise we would have pulled some pointers from the | |
1962 | * right | |
1963 | */ | |
305a26af MF |
1964 | if (!left) { |
1965 | ret = -EROFS; | |
0b246afa | 1966 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1967 | goto enospc; |
1968 | } | |
2ff7e61e | 1969 | wret = balance_node_right(trans, fs_info, mid, left); |
54aa1f4d | 1970 | if (wret < 0) { |
79f95c82 | 1971 | ret = wret; |
54aa1f4d CM |
1972 | goto enospc; |
1973 | } | |
bce4eae9 | 1974 | if (wret == 1) { |
2ff7e61e | 1975 | wret = push_node_left(trans, fs_info, left, mid, 1); |
bce4eae9 CM |
1976 | if (wret < 0) |
1977 | ret = wret; | |
1978 | } | |
79f95c82 CM |
1979 | BUG_ON(wret == 1); |
1980 | } | |
5f39d397 | 1981 | if (btrfs_header_nritems(mid) == 0) { |
7c302b49 | 1982 | clean_tree_block(fs_info, mid); |
925baedd | 1983 | btrfs_tree_unlock(mid); |
afe5fea7 | 1984 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1985 | root_sub_used(root, mid->len); |
5581a51a | 1986 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1987 | free_extent_buffer_stale(mid); |
f0486c68 | 1988 | mid = NULL; |
79f95c82 CM |
1989 | } else { |
1990 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1991 | struct btrfs_disk_key mid_key; |
1992 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
1993 | ret = tree_mod_log_insert_key(parent, pslot, |
1994 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1995 | BUG_ON(ret < 0); | |
5f39d397 CM |
1996 | btrfs_set_node_key(parent, &mid_key, pslot); |
1997 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1998 | } |
bb803951 | 1999 | |
79f95c82 | 2000 | /* update the path */ |
5f39d397 CM |
2001 | if (left) { |
2002 | if (btrfs_header_nritems(left) > orig_slot) { | |
2003 | extent_buffer_get(left); | |
925baedd | 2004 | /* left was locked after cow */ |
5f39d397 | 2005 | path->nodes[level] = left; |
bb803951 CM |
2006 | path->slots[level + 1] -= 1; |
2007 | path->slots[level] = orig_slot; | |
925baedd CM |
2008 | if (mid) { |
2009 | btrfs_tree_unlock(mid); | |
5f39d397 | 2010 | free_extent_buffer(mid); |
925baedd | 2011 | } |
bb803951 | 2012 | } else { |
5f39d397 | 2013 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2014 | path->slots[level] = orig_slot; |
2015 | } | |
2016 | } | |
79f95c82 | 2017 | /* double check we haven't messed things up */ |
e20d96d6 | 2018 | if (orig_ptr != |
5f39d397 | 2019 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2020 | BUG(); |
54aa1f4d | 2021 | enospc: |
925baedd CM |
2022 | if (right) { |
2023 | btrfs_tree_unlock(right); | |
5f39d397 | 2024 | free_extent_buffer(right); |
925baedd CM |
2025 | } |
2026 | if (left) { | |
2027 | if (path->nodes[level] != left) | |
2028 | btrfs_tree_unlock(left); | |
5f39d397 | 2029 | free_extent_buffer(left); |
925baedd | 2030 | } |
bb803951 CM |
2031 | return ret; |
2032 | } | |
2033 | ||
d352ac68 CM |
2034 | /* Node balancing for insertion. Here we only split or push nodes around |
2035 | * when they are completely full. This is also done top down, so we | |
2036 | * have to be pessimistic. | |
2037 | */ | |
d397712b | 2038 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2039 | struct btrfs_root *root, |
2040 | struct btrfs_path *path, int level) | |
e66f709b | 2041 | { |
0b246afa | 2042 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2043 | struct extent_buffer *right = NULL; |
2044 | struct extent_buffer *mid; | |
2045 | struct extent_buffer *left = NULL; | |
2046 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2047 | int ret = 0; |
2048 | int wret; | |
2049 | int pslot; | |
2050 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2051 | |
2052 | if (level == 0) | |
2053 | return 1; | |
2054 | ||
5f39d397 | 2055 | mid = path->nodes[level]; |
7bb86316 | 2056 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2057 | |
a05a9bb1 | 2058 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2059 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2060 | pslot = path->slots[level + 1]; |
2061 | } | |
e66f709b | 2062 | |
5f39d397 | 2063 | if (!parent) |
e66f709b | 2064 | return 1; |
e66f709b | 2065 | |
2ff7e61e | 2066 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
2067 | if (IS_ERR(left)) |
2068 | left = NULL; | |
e66f709b CM |
2069 | |
2070 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2071 | if (left) { |
e66f709b | 2072 | u32 left_nr; |
925baedd CM |
2073 | |
2074 | btrfs_tree_lock(left); | |
b4ce94de CM |
2075 | btrfs_set_lock_blocking(left); |
2076 | ||
5f39d397 | 2077 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2078 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2079 | wret = 1; |
2080 | } else { | |
5f39d397 | 2081 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2082 | pslot - 1, &left); |
54aa1f4d CM |
2083 | if (ret) |
2084 | wret = 1; | |
2085 | else { | |
2ff7e61e | 2086 | wret = push_node_left(trans, fs_info, |
971a1f66 | 2087 | left, mid, 0); |
54aa1f4d | 2088 | } |
33ade1f8 | 2089 | } |
e66f709b CM |
2090 | if (wret < 0) |
2091 | ret = wret; | |
2092 | if (wret == 0) { | |
5f39d397 | 2093 | struct btrfs_disk_key disk_key; |
e66f709b | 2094 | orig_slot += left_nr; |
5f39d397 | 2095 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2096 | ret = tree_mod_log_insert_key(parent, pslot, |
2097 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2098 | BUG_ON(ret < 0); | |
5f39d397 CM |
2099 | btrfs_set_node_key(parent, &disk_key, pslot); |
2100 | btrfs_mark_buffer_dirty(parent); | |
2101 | if (btrfs_header_nritems(left) > orig_slot) { | |
2102 | path->nodes[level] = left; | |
e66f709b CM |
2103 | path->slots[level + 1] -= 1; |
2104 | path->slots[level] = orig_slot; | |
925baedd | 2105 | btrfs_tree_unlock(mid); |
5f39d397 | 2106 | free_extent_buffer(mid); |
e66f709b CM |
2107 | } else { |
2108 | orig_slot -= | |
5f39d397 | 2109 | btrfs_header_nritems(left); |
e66f709b | 2110 | path->slots[level] = orig_slot; |
925baedd | 2111 | btrfs_tree_unlock(left); |
5f39d397 | 2112 | free_extent_buffer(left); |
e66f709b | 2113 | } |
e66f709b CM |
2114 | return 0; |
2115 | } | |
925baedd | 2116 | btrfs_tree_unlock(left); |
5f39d397 | 2117 | free_extent_buffer(left); |
e66f709b | 2118 | } |
2ff7e61e | 2119 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
2120 | if (IS_ERR(right)) |
2121 | right = NULL; | |
e66f709b CM |
2122 | |
2123 | /* | |
2124 | * then try to empty the right most buffer into the middle | |
2125 | */ | |
5f39d397 | 2126 | if (right) { |
33ade1f8 | 2127 | u32 right_nr; |
b4ce94de | 2128 | |
925baedd | 2129 | btrfs_tree_lock(right); |
b4ce94de CM |
2130 | btrfs_set_lock_blocking(right); |
2131 | ||
5f39d397 | 2132 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2133 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2134 | wret = 1; |
2135 | } else { | |
5f39d397 CM |
2136 | ret = btrfs_cow_block(trans, root, right, |
2137 | parent, pslot + 1, | |
9fa8cfe7 | 2138 | &right); |
54aa1f4d CM |
2139 | if (ret) |
2140 | wret = 1; | |
2141 | else { | |
2ff7e61e | 2142 | wret = balance_node_right(trans, fs_info, |
5f39d397 | 2143 | right, mid); |
54aa1f4d | 2144 | } |
33ade1f8 | 2145 | } |
e66f709b CM |
2146 | if (wret < 0) |
2147 | ret = wret; | |
2148 | if (wret == 0) { | |
5f39d397 CM |
2149 | struct btrfs_disk_key disk_key; |
2150 | ||
2151 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2152 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2153 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2154 | BUG_ON(ret < 0); | |
5f39d397 CM |
2155 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2156 | btrfs_mark_buffer_dirty(parent); | |
2157 | ||
2158 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2159 | path->nodes[level] = right; | |
e66f709b CM |
2160 | path->slots[level + 1] += 1; |
2161 | path->slots[level] = orig_slot - | |
5f39d397 | 2162 | btrfs_header_nritems(mid); |
925baedd | 2163 | btrfs_tree_unlock(mid); |
5f39d397 | 2164 | free_extent_buffer(mid); |
e66f709b | 2165 | } else { |
925baedd | 2166 | btrfs_tree_unlock(right); |
5f39d397 | 2167 | free_extent_buffer(right); |
e66f709b | 2168 | } |
e66f709b CM |
2169 | return 0; |
2170 | } | |
925baedd | 2171 | btrfs_tree_unlock(right); |
5f39d397 | 2172 | free_extent_buffer(right); |
e66f709b | 2173 | } |
e66f709b CM |
2174 | return 1; |
2175 | } | |
2176 | ||
3c69faec | 2177 | /* |
d352ac68 CM |
2178 | * readahead one full node of leaves, finding things that are close |
2179 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2180 | */ |
2ff7e61e | 2181 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2182 | struct btrfs_path *path, |
2183 | int level, int slot, u64 objectid) | |
3c69faec | 2184 | { |
5f39d397 | 2185 | struct extent_buffer *node; |
01f46658 | 2186 | struct btrfs_disk_key disk_key; |
3c69faec | 2187 | u32 nritems; |
3c69faec | 2188 | u64 search; |
a7175319 | 2189 | u64 target; |
6b80053d | 2190 | u64 nread = 0; |
5f39d397 | 2191 | struct extent_buffer *eb; |
6b80053d CM |
2192 | u32 nr; |
2193 | u32 blocksize; | |
2194 | u32 nscan = 0; | |
db94535d | 2195 | |
a6b6e75e | 2196 | if (level != 1) |
6702ed49 CM |
2197 | return; |
2198 | ||
2199 | if (!path->nodes[level]) | |
3c69faec CM |
2200 | return; |
2201 | ||
5f39d397 | 2202 | node = path->nodes[level]; |
925baedd | 2203 | |
3c69faec | 2204 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2205 | blocksize = fs_info->nodesize; |
2206 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2207 | if (eb) { |
2208 | free_extent_buffer(eb); | |
3c69faec CM |
2209 | return; |
2210 | } | |
2211 | ||
a7175319 | 2212 | target = search; |
6b80053d | 2213 | |
5f39d397 | 2214 | nritems = btrfs_header_nritems(node); |
6b80053d | 2215 | nr = slot; |
25b8b936 | 2216 | |
d397712b | 2217 | while (1) { |
e4058b54 | 2218 | if (path->reada == READA_BACK) { |
6b80053d CM |
2219 | if (nr == 0) |
2220 | break; | |
2221 | nr--; | |
e4058b54 | 2222 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2223 | nr++; |
2224 | if (nr >= nritems) | |
2225 | break; | |
3c69faec | 2226 | } |
e4058b54 | 2227 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2228 | btrfs_node_key(node, &disk_key, nr); |
2229 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2230 | break; | |
2231 | } | |
6b80053d | 2232 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2233 | if ((search <= target && target - search <= 65536) || |
2234 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2235 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2236 | nread += blocksize; |
2237 | } | |
2238 | nscan++; | |
a7175319 | 2239 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2240 | break; |
3c69faec CM |
2241 | } |
2242 | } | |
925baedd | 2243 | |
2ff7e61e | 2244 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2245 | struct btrfs_path *path, int level) |
b4ce94de CM |
2246 | { |
2247 | int slot; | |
2248 | int nritems; | |
2249 | struct extent_buffer *parent; | |
2250 | struct extent_buffer *eb; | |
2251 | u64 gen; | |
2252 | u64 block1 = 0; | |
2253 | u64 block2 = 0; | |
b4ce94de | 2254 | |
8c594ea8 | 2255 | parent = path->nodes[level + 1]; |
b4ce94de | 2256 | if (!parent) |
0b08851f | 2257 | return; |
b4ce94de CM |
2258 | |
2259 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2260 | slot = path->slots[level + 1]; |
b4ce94de CM |
2261 | |
2262 | if (slot > 0) { | |
2263 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2264 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2265 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2266 | /* |
2267 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2268 | * don't want to return eagain here. That will loop | |
2269 | * forever | |
2270 | */ | |
2271 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2272 | block1 = 0; |
2273 | free_extent_buffer(eb); | |
2274 | } | |
8c594ea8 | 2275 | if (slot + 1 < nritems) { |
b4ce94de CM |
2276 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2277 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2278 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2279 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2280 | block2 = 0; |
2281 | free_extent_buffer(eb); | |
2282 | } | |
8c594ea8 | 2283 | |
0b08851f | 2284 | if (block1) |
2ff7e61e | 2285 | readahead_tree_block(fs_info, block1); |
0b08851f | 2286 | if (block2) |
2ff7e61e | 2287 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2288 | } |
2289 | ||
2290 | ||
d352ac68 | 2291 | /* |
d397712b CM |
2292 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2293 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2294 | * operations on the tree might require changing key pointers higher up in the | |
2295 | * tree. | |
d352ac68 | 2296 | * |
d397712b CM |
2297 | * callers might also have set path->keep_locks, which tells this code to keep |
2298 | * the lock if the path points to the last slot in the block. This is part of | |
2299 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2300 | * |
d397712b CM |
2301 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2302 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2303 | */ |
e02119d5 | 2304 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2305 | int lowest_unlock, int min_write_lock_level, |
2306 | int *write_lock_level) | |
925baedd CM |
2307 | { |
2308 | int i; | |
2309 | int skip_level = level; | |
051e1b9f | 2310 | int no_skips = 0; |
925baedd CM |
2311 | struct extent_buffer *t; |
2312 | ||
2313 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2314 | if (!path->nodes[i]) | |
2315 | break; | |
2316 | if (!path->locks[i]) | |
2317 | break; | |
051e1b9f | 2318 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2319 | skip_level = i + 1; |
2320 | continue; | |
2321 | } | |
051e1b9f | 2322 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2323 | u32 nritems; |
2324 | t = path->nodes[i]; | |
2325 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2326 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2327 | skip_level = i + 1; |
2328 | continue; | |
2329 | } | |
2330 | } | |
051e1b9f CM |
2331 | if (skip_level < i && i >= lowest_unlock) |
2332 | no_skips = 1; | |
2333 | ||
925baedd CM |
2334 | t = path->nodes[i]; |
2335 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2336 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2337 | path->locks[i] = 0; |
f7c79f30 CM |
2338 | if (write_lock_level && |
2339 | i > min_write_lock_level && | |
2340 | i <= *write_lock_level) { | |
2341 | *write_lock_level = i - 1; | |
2342 | } | |
925baedd CM |
2343 | } |
2344 | } | |
2345 | } | |
2346 | ||
b4ce94de CM |
2347 | /* |
2348 | * This releases any locks held in the path starting at level and | |
2349 | * going all the way up to the root. | |
2350 | * | |
2351 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2352 | * corner cases, such as COW of the block at slot zero in the node. This | |
2353 | * ignores those rules, and it should only be called when there are no | |
2354 | * more updates to be done higher up in the tree. | |
2355 | */ | |
2356 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2357 | { | |
2358 | int i; | |
2359 | ||
09a2a8f9 | 2360 | if (path->keep_locks) |
b4ce94de CM |
2361 | return; |
2362 | ||
2363 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2364 | if (!path->nodes[i]) | |
12f4dacc | 2365 | continue; |
b4ce94de | 2366 | if (!path->locks[i]) |
12f4dacc | 2367 | continue; |
bd681513 | 2368 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2369 | path->locks[i] = 0; |
2370 | } | |
2371 | } | |
2372 | ||
c8c42864 CM |
2373 | /* |
2374 | * helper function for btrfs_search_slot. The goal is to find a block | |
2375 | * in cache without setting the path to blocking. If we find the block | |
2376 | * we return zero and the path is unchanged. | |
2377 | * | |
2378 | * If we can't find the block, we set the path blocking and do some | |
2379 | * reada. -EAGAIN is returned and the search must be repeated. | |
2380 | */ | |
2381 | static int | |
d07b8528 LB |
2382 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2383 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2384 | const struct btrfs_key *key) |
c8c42864 | 2385 | { |
0b246afa | 2386 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2387 | u64 blocknr; |
2388 | u64 gen; | |
c8c42864 CM |
2389 | struct extent_buffer *b = *eb_ret; |
2390 | struct extent_buffer *tmp; | |
76a05b35 | 2391 | int ret; |
c8c42864 CM |
2392 | |
2393 | blocknr = btrfs_node_blockptr(b, slot); | |
2394 | gen = btrfs_node_ptr_generation(b, slot); | |
c8c42864 | 2395 | |
0b246afa | 2396 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2397 | if (tmp) { |
b9fab919 | 2398 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2399 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2400 | *eb_ret = tmp; | |
2401 | return 0; | |
2402 | } | |
2403 | ||
2404 | /* the pages were up to date, but we failed | |
2405 | * the generation number check. Do a full | |
2406 | * read for the generation number that is correct. | |
2407 | * We must do this without dropping locks so | |
2408 | * we can trust our generation number | |
2409 | */ | |
2410 | btrfs_set_path_blocking(p); | |
2411 | ||
2412 | /* now we're allowed to do a blocking uptodate check */ | |
2413 | ret = btrfs_read_buffer(tmp, gen); | |
2414 | if (!ret) { | |
2415 | *eb_ret = tmp; | |
2416 | return 0; | |
cb44921a | 2417 | } |
bdf7c00e JB |
2418 | free_extent_buffer(tmp); |
2419 | btrfs_release_path(p); | |
2420 | return -EIO; | |
c8c42864 CM |
2421 | } |
2422 | ||
2423 | /* | |
2424 | * reduce lock contention at high levels | |
2425 | * of the btree by dropping locks before | |
76a05b35 CM |
2426 | * we read. Don't release the lock on the current |
2427 | * level because we need to walk this node to figure | |
2428 | * out which blocks to read. | |
c8c42864 | 2429 | */ |
8c594ea8 CM |
2430 | btrfs_unlock_up_safe(p, level + 1); |
2431 | btrfs_set_path_blocking(p); | |
2432 | ||
cb44921a | 2433 | free_extent_buffer(tmp); |
e4058b54 | 2434 | if (p->reada != READA_NONE) |
2ff7e61e | 2435 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2436 | |
b3b4aa74 | 2437 | btrfs_release_path(p); |
76a05b35 CM |
2438 | |
2439 | ret = -EAGAIN; | |
2ff7e61e | 2440 | tmp = read_tree_block(fs_info, blocknr, 0); |
64c043de | 2441 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2442 | /* |
2443 | * If the read above didn't mark this buffer up to date, | |
2444 | * it will never end up being up to date. Set ret to EIO now | |
2445 | * and give up so that our caller doesn't loop forever | |
2446 | * on our EAGAINs. | |
2447 | */ | |
b9fab919 | 2448 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2449 | ret = -EIO; |
c8c42864 | 2450 | free_extent_buffer(tmp); |
c871b0f2 LB |
2451 | } else { |
2452 | ret = PTR_ERR(tmp); | |
76a05b35 CM |
2453 | } |
2454 | return ret; | |
c8c42864 CM |
2455 | } |
2456 | ||
2457 | /* | |
2458 | * helper function for btrfs_search_slot. This does all of the checks | |
2459 | * for node-level blocks and does any balancing required based on | |
2460 | * the ins_len. | |
2461 | * | |
2462 | * If no extra work was required, zero is returned. If we had to | |
2463 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2464 | * start over | |
2465 | */ | |
2466 | static int | |
2467 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2468 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2469 | struct extent_buffer *b, int level, int ins_len, |
2470 | int *write_lock_level) | |
c8c42864 | 2471 | { |
0b246afa | 2472 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2473 | int ret; |
0b246afa | 2474 | |
c8c42864 | 2475 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2476 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2477 | int sret; |
2478 | ||
bd681513 CM |
2479 | if (*write_lock_level < level + 1) { |
2480 | *write_lock_level = level + 1; | |
2481 | btrfs_release_path(p); | |
2482 | goto again; | |
2483 | } | |
2484 | ||
c8c42864 | 2485 | btrfs_set_path_blocking(p); |
2ff7e61e | 2486 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2487 | sret = split_node(trans, root, p, level); |
bd681513 | 2488 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2489 | |
2490 | BUG_ON(sret > 0); | |
2491 | if (sret) { | |
2492 | ret = sret; | |
2493 | goto done; | |
2494 | } | |
2495 | b = p->nodes[level]; | |
2496 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2497 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2498 | int sret; |
2499 | ||
bd681513 CM |
2500 | if (*write_lock_level < level + 1) { |
2501 | *write_lock_level = level + 1; | |
2502 | btrfs_release_path(p); | |
2503 | goto again; | |
2504 | } | |
2505 | ||
c8c42864 | 2506 | btrfs_set_path_blocking(p); |
2ff7e61e | 2507 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2508 | sret = balance_level(trans, root, p, level); |
bd681513 | 2509 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2510 | |
2511 | if (sret) { | |
2512 | ret = sret; | |
2513 | goto done; | |
2514 | } | |
2515 | b = p->nodes[level]; | |
2516 | if (!b) { | |
b3b4aa74 | 2517 | btrfs_release_path(p); |
c8c42864 CM |
2518 | goto again; |
2519 | } | |
2520 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2521 | } | |
2522 | return 0; | |
2523 | ||
2524 | again: | |
2525 | ret = -EAGAIN; | |
2526 | done: | |
2527 | return ret; | |
2528 | } | |
2529 | ||
d7396f07 | 2530 | static void key_search_validate(struct extent_buffer *b, |
310712b2 | 2531 | const struct btrfs_key *key, |
d7396f07 FDBM |
2532 | int level) |
2533 | { | |
2534 | #ifdef CONFIG_BTRFS_ASSERT | |
2535 | struct btrfs_disk_key disk_key; | |
2536 | ||
2537 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2538 | ||
2539 | if (level == 0) | |
2540 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2541 | offsetof(struct btrfs_leaf, items[0].key), | |
2542 | sizeof(disk_key))); | |
2543 | else | |
2544 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2545 | offsetof(struct btrfs_node, ptrs[0].key), | |
2546 | sizeof(disk_key))); | |
2547 | #endif | |
2548 | } | |
2549 | ||
310712b2 | 2550 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2551 | int level, int *prev_cmp, int *slot) |
2552 | { | |
2553 | if (*prev_cmp != 0) { | |
a74b35ec | 2554 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2555 | return *prev_cmp; |
2556 | } | |
2557 | ||
2558 | key_search_validate(b, key, level); | |
2559 | *slot = 0; | |
2560 | ||
2561 | return 0; | |
2562 | } | |
2563 | ||
381cf658 | 2564 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2565 | u64 iobjectid, u64 ioff, u8 key_type, |
2566 | struct btrfs_key *found_key) | |
2567 | { | |
2568 | int ret; | |
2569 | struct btrfs_key key; | |
2570 | struct extent_buffer *eb; | |
381cf658 DS |
2571 | |
2572 | ASSERT(path); | |
1d4c08e0 | 2573 | ASSERT(found_key); |
e33d5c3d KN |
2574 | |
2575 | key.type = key_type; | |
2576 | key.objectid = iobjectid; | |
2577 | key.offset = ioff; | |
2578 | ||
2579 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2580 | if (ret < 0) |
e33d5c3d KN |
2581 | return ret; |
2582 | ||
2583 | eb = path->nodes[0]; | |
2584 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2585 | ret = btrfs_next_leaf(fs_root, path); | |
2586 | if (ret) | |
2587 | return ret; | |
2588 | eb = path->nodes[0]; | |
2589 | } | |
2590 | ||
2591 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2592 | if (found_key->type != key.type || | |
2593 | found_key->objectid != key.objectid) | |
2594 | return 1; | |
2595 | ||
2596 | return 0; | |
2597 | } | |
2598 | ||
74123bd7 | 2599 | /* |
4271ecea NB |
2600 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2601 | * modifications to preserve tree invariants. | |
74123bd7 | 2602 | * |
4271ecea NB |
2603 | * @trans: Handle of transaction, used when modifying the tree |
2604 | * @p: Holds all btree nodes along the search path | |
2605 | * @root: The root node of the tree | |
2606 | * @key: The key we are looking for | |
2607 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2608 | * deletions it's -1. 0 for plain searches | |
2609 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2610 | * when modifying the tree. | |
97571fd0 | 2611 | * |
4271ecea NB |
2612 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2613 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2614 | * | |
2615 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2616 | * of the path (level 0) | |
2617 | * | |
2618 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2619 | * points to the slot where it should be inserted | |
2620 | * | |
2621 | * If an error is encountered while searching the tree a negative error number | |
2622 | * is returned | |
74123bd7 | 2623 | */ |
310712b2 OS |
2624 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2625 | const struct btrfs_key *key, struct btrfs_path *p, | |
2626 | int ins_len, int cow) | |
be0e5c09 | 2627 | { |
0b246afa | 2628 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2629 | struct extent_buffer *b; |
be0e5c09 CM |
2630 | int slot; |
2631 | int ret; | |
33c66f43 | 2632 | int err; |
be0e5c09 | 2633 | int level; |
925baedd | 2634 | int lowest_unlock = 1; |
bd681513 CM |
2635 | int root_lock; |
2636 | /* everything at write_lock_level or lower must be write locked */ | |
2637 | int write_lock_level = 0; | |
9f3a7427 | 2638 | u8 lowest_level = 0; |
f7c79f30 | 2639 | int min_write_lock_level; |
d7396f07 | 2640 | int prev_cmp; |
9f3a7427 | 2641 | |
6702ed49 | 2642 | lowest_level = p->lowest_level; |
323ac95b | 2643 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2644 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2645 | BUG_ON(!cow && ins_len); |
25179201 | 2646 | |
bd681513 | 2647 | if (ins_len < 0) { |
925baedd | 2648 | lowest_unlock = 2; |
65b51a00 | 2649 | |
bd681513 CM |
2650 | /* when we are removing items, we might have to go up to level |
2651 | * two as we update tree pointers Make sure we keep write | |
2652 | * for those levels as well | |
2653 | */ | |
2654 | write_lock_level = 2; | |
2655 | } else if (ins_len > 0) { | |
2656 | /* | |
2657 | * for inserting items, make sure we have a write lock on | |
2658 | * level 1 so we can update keys | |
2659 | */ | |
2660 | write_lock_level = 1; | |
2661 | } | |
2662 | ||
2663 | if (!cow) | |
2664 | write_lock_level = -1; | |
2665 | ||
09a2a8f9 | 2666 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2667 | write_lock_level = BTRFS_MAX_LEVEL; |
2668 | ||
f7c79f30 CM |
2669 | min_write_lock_level = write_lock_level; |
2670 | ||
bb803951 | 2671 | again: |
d7396f07 | 2672 | prev_cmp = -1; |
bd681513 CM |
2673 | /* |
2674 | * we try very hard to do read locks on the root | |
2675 | */ | |
2676 | root_lock = BTRFS_READ_LOCK; | |
2677 | level = 0; | |
5d4f98a2 | 2678 | if (p->search_commit_root) { |
bd681513 CM |
2679 | /* |
2680 | * the commit roots are read only | |
2681 | * so we always do read locks | |
2682 | */ | |
3f8a18cc | 2683 | if (p->need_commit_sem) |
0b246afa | 2684 | down_read(&fs_info->commit_root_sem); |
5d4f98a2 YZ |
2685 | b = root->commit_root; |
2686 | extent_buffer_get(b); | |
bd681513 | 2687 | level = btrfs_header_level(b); |
3f8a18cc | 2688 | if (p->need_commit_sem) |
0b246afa | 2689 | up_read(&fs_info->commit_root_sem); |
5d4f98a2 | 2690 | if (!p->skip_locking) |
bd681513 | 2691 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2692 | } else { |
bd681513 | 2693 | if (p->skip_locking) { |
5d4f98a2 | 2694 | b = btrfs_root_node(root); |
bd681513 CM |
2695 | level = btrfs_header_level(b); |
2696 | } else { | |
2697 | /* we don't know the level of the root node | |
2698 | * until we actually have it read locked | |
2699 | */ | |
2700 | b = btrfs_read_lock_root_node(root); | |
2701 | level = btrfs_header_level(b); | |
2702 | if (level <= write_lock_level) { | |
2703 | /* whoops, must trade for write lock */ | |
2704 | btrfs_tree_read_unlock(b); | |
2705 | free_extent_buffer(b); | |
2706 | b = btrfs_lock_root_node(root); | |
2707 | root_lock = BTRFS_WRITE_LOCK; | |
2708 | ||
2709 | /* the level might have changed, check again */ | |
2710 | level = btrfs_header_level(b); | |
2711 | } | |
2712 | } | |
5d4f98a2 | 2713 | } |
bd681513 CM |
2714 | p->nodes[level] = b; |
2715 | if (!p->skip_locking) | |
2716 | p->locks[level] = root_lock; | |
925baedd | 2717 | |
eb60ceac | 2718 | while (b) { |
5f39d397 | 2719 | level = btrfs_header_level(b); |
65b51a00 CM |
2720 | |
2721 | /* | |
2722 | * setup the path here so we can release it under lock | |
2723 | * contention with the cow code | |
2724 | */ | |
02217ed2 | 2725 | if (cow) { |
9ea2c7c9 NB |
2726 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2727 | ||
c8c42864 CM |
2728 | /* |
2729 | * if we don't really need to cow this block | |
2730 | * then we don't want to set the path blocking, | |
2731 | * so we test it here | |
2732 | */ | |
64c12921 JM |
2733 | if (!should_cow_block(trans, root, b)) { |
2734 | trans->dirty = true; | |
65b51a00 | 2735 | goto cow_done; |
64c12921 | 2736 | } |
5d4f98a2 | 2737 | |
bd681513 CM |
2738 | /* |
2739 | * must have write locks on this node and the | |
2740 | * parent | |
2741 | */ | |
5124e00e JB |
2742 | if (level > write_lock_level || |
2743 | (level + 1 > write_lock_level && | |
2744 | level + 1 < BTRFS_MAX_LEVEL && | |
2745 | p->nodes[level + 1])) { | |
bd681513 CM |
2746 | write_lock_level = level + 1; |
2747 | btrfs_release_path(p); | |
2748 | goto again; | |
2749 | } | |
2750 | ||
160f4089 | 2751 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2752 | if (last_level) |
2753 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2754 | &b); | |
2755 | else | |
2756 | err = btrfs_cow_block(trans, root, b, | |
2757 | p->nodes[level + 1], | |
2758 | p->slots[level + 1], &b); | |
33c66f43 | 2759 | if (err) { |
33c66f43 | 2760 | ret = err; |
65b51a00 | 2761 | goto done; |
54aa1f4d | 2762 | } |
02217ed2 | 2763 | } |
65b51a00 | 2764 | cow_done: |
eb60ceac | 2765 | p->nodes[level] = b; |
bd681513 | 2766 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2767 | |
2768 | /* | |
2769 | * we have a lock on b and as long as we aren't changing | |
2770 | * the tree, there is no way to for the items in b to change. | |
2771 | * It is safe to drop the lock on our parent before we | |
2772 | * go through the expensive btree search on b. | |
2773 | * | |
eb653de1 FDBM |
2774 | * If we're inserting or deleting (ins_len != 0), then we might |
2775 | * be changing slot zero, which may require changing the parent. | |
2776 | * So, we can't drop the lock until after we know which slot | |
2777 | * we're operating on. | |
b4ce94de | 2778 | */ |
eb653de1 FDBM |
2779 | if (!ins_len && !p->keep_locks) { |
2780 | int u = level + 1; | |
2781 | ||
2782 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2783 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2784 | p->locks[u] = 0; | |
2785 | } | |
2786 | } | |
b4ce94de | 2787 | |
d7396f07 | 2788 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2789 | if (ret < 0) |
2790 | goto done; | |
b4ce94de | 2791 | |
5f39d397 | 2792 | if (level != 0) { |
33c66f43 YZ |
2793 | int dec = 0; |
2794 | if (ret && slot > 0) { | |
2795 | dec = 1; | |
be0e5c09 | 2796 | slot -= 1; |
33c66f43 | 2797 | } |
be0e5c09 | 2798 | p->slots[level] = slot; |
33c66f43 | 2799 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2800 | ins_len, &write_lock_level); |
33c66f43 | 2801 | if (err == -EAGAIN) |
c8c42864 | 2802 | goto again; |
33c66f43 YZ |
2803 | if (err) { |
2804 | ret = err; | |
c8c42864 | 2805 | goto done; |
33c66f43 | 2806 | } |
c8c42864 CM |
2807 | b = p->nodes[level]; |
2808 | slot = p->slots[level]; | |
b4ce94de | 2809 | |
bd681513 CM |
2810 | /* |
2811 | * slot 0 is special, if we change the key | |
2812 | * we have to update the parent pointer | |
2813 | * which means we must have a write lock | |
2814 | * on the parent | |
2815 | */ | |
eb653de1 | 2816 | if (slot == 0 && ins_len && |
bd681513 CM |
2817 | write_lock_level < level + 1) { |
2818 | write_lock_level = level + 1; | |
2819 | btrfs_release_path(p); | |
2820 | goto again; | |
2821 | } | |
2822 | ||
f7c79f30 CM |
2823 | unlock_up(p, level, lowest_unlock, |
2824 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2825 | |
925baedd | 2826 | if (level == lowest_level) { |
33c66f43 YZ |
2827 | if (dec) |
2828 | p->slots[level]++; | |
5b21f2ed | 2829 | goto done; |
925baedd | 2830 | } |
ca7a79ad | 2831 | |
d07b8528 | 2832 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2833 | slot, key); |
33c66f43 | 2834 | if (err == -EAGAIN) |
c8c42864 | 2835 | goto again; |
33c66f43 YZ |
2836 | if (err) { |
2837 | ret = err; | |
76a05b35 | 2838 | goto done; |
33c66f43 | 2839 | } |
76a05b35 | 2840 | |
b4ce94de | 2841 | if (!p->skip_locking) { |
bd681513 CM |
2842 | level = btrfs_header_level(b); |
2843 | if (level <= write_lock_level) { | |
2844 | err = btrfs_try_tree_write_lock(b); | |
2845 | if (!err) { | |
2846 | btrfs_set_path_blocking(p); | |
2847 | btrfs_tree_lock(b); | |
2848 | btrfs_clear_path_blocking(p, b, | |
2849 | BTRFS_WRITE_LOCK); | |
2850 | } | |
2851 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2852 | } else { | |
f82c458a | 2853 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2854 | if (!err) { |
2855 | btrfs_set_path_blocking(p); | |
2856 | btrfs_tree_read_lock(b); | |
2857 | btrfs_clear_path_blocking(p, b, | |
2858 | BTRFS_READ_LOCK); | |
2859 | } | |
2860 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2861 | } |
bd681513 | 2862 | p->nodes[level] = b; |
b4ce94de | 2863 | } |
be0e5c09 CM |
2864 | } else { |
2865 | p->slots[level] = slot; | |
87b29b20 | 2866 | if (ins_len > 0 && |
2ff7e61e | 2867 | btrfs_leaf_free_space(fs_info, b) < ins_len) { |
bd681513 CM |
2868 | if (write_lock_level < 1) { |
2869 | write_lock_level = 1; | |
2870 | btrfs_release_path(p); | |
2871 | goto again; | |
2872 | } | |
2873 | ||
b4ce94de | 2874 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2875 | err = split_leaf(trans, root, key, |
2876 | p, ins_len, ret == 0); | |
bd681513 | 2877 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2878 | |
33c66f43 YZ |
2879 | BUG_ON(err > 0); |
2880 | if (err) { | |
2881 | ret = err; | |
65b51a00 CM |
2882 | goto done; |
2883 | } | |
5c680ed6 | 2884 | } |
459931ec | 2885 | if (!p->search_for_split) |
f7c79f30 CM |
2886 | unlock_up(p, level, lowest_unlock, |
2887 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2888 | goto done; |
be0e5c09 CM |
2889 | } |
2890 | } | |
65b51a00 CM |
2891 | ret = 1; |
2892 | done: | |
b4ce94de CM |
2893 | /* |
2894 | * we don't really know what they plan on doing with the path | |
2895 | * from here on, so for now just mark it as blocking | |
2896 | */ | |
b9473439 CM |
2897 | if (!p->leave_spinning) |
2898 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2899 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2900 | btrfs_release_path(p); |
65b51a00 | 2901 | return ret; |
be0e5c09 CM |
2902 | } |
2903 | ||
5d9e75c4 JS |
2904 | /* |
2905 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2906 | * current state of the tree together with the operations recorded in the tree | |
2907 | * modification log to search for the key in a previous version of this tree, as | |
2908 | * denoted by the time_seq parameter. | |
2909 | * | |
2910 | * Naturally, there is no support for insert, delete or cow operations. | |
2911 | * | |
2912 | * The resulting path and return value will be set up as if we called | |
2913 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2914 | */ | |
310712b2 | 2915 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2916 | struct btrfs_path *p, u64 time_seq) |
2917 | { | |
0b246afa | 2918 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2919 | struct extent_buffer *b; |
2920 | int slot; | |
2921 | int ret; | |
2922 | int err; | |
2923 | int level; | |
2924 | int lowest_unlock = 1; | |
2925 | u8 lowest_level = 0; | |
d4b4087c | 2926 | int prev_cmp = -1; |
5d9e75c4 JS |
2927 | |
2928 | lowest_level = p->lowest_level; | |
2929 | WARN_ON(p->nodes[0] != NULL); | |
2930 | ||
2931 | if (p->search_commit_root) { | |
2932 | BUG_ON(time_seq); | |
2933 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2934 | } | |
2935 | ||
2936 | again: | |
5d9e75c4 | 2937 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2938 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2939 | p->locks[level] = BTRFS_READ_LOCK; |
2940 | ||
2941 | while (b) { | |
2942 | level = btrfs_header_level(b); | |
2943 | p->nodes[level] = b; | |
2944 | btrfs_clear_path_blocking(p, NULL, 0); | |
2945 | ||
2946 | /* | |
2947 | * we have a lock on b and as long as we aren't changing | |
2948 | * the tree, there is no way to for the items in b to change. | |
2949 | * It is safe to drop the lock on our parent before we | |
2950 | * go through the expensive btree search on b. | |
2951 | */ | |
2952 | btrfs_unlock_up_safe(p, level + 1); | |
2953 | ||
d4b4087c | 2954 | /* |
01327610 | 2955 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2956 | * time. |
2957 | */ | |
2958 | prev_cmp = -1; | |
d7396f07 | 2959 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
2960 | |
2961 | if (level != 0) { | |
2962 | int dec = 0; | |
2963 | if (ret && slot > 0) { | |
2964 | dec = 1; | |
2965 | slot -= 1; | |
2966 | } | |
2967 | p->slots[level] = slot; | |
2968 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2969 | ||
2970 | if (level == lowest_level) { | |
2971 | if (dec) | |
2972 | p->slots[level]++; | |
2973 | goto done; | |
2974 | } | |
2975 | ||
d07b8528 | 2976 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2977 | slot, key); |
5d9e75c4 JS |
2978 | if (err == -EAGAIN) |
2979 | goto again; | |
2980 | if (err) { | |
2981 | ret = err; | |
2982 | goto done; | |
2983 | } | |
2984 | ||
2985 | level = btrfs_header_level(b); | |
f82c458a | 2986 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
2987 | if (!err) { |
2988 | btrfs_set_path_blocking(p); | |
2989 | btrfs_tree_read_lock(b); | |
2990 | btrfs_clear_path_blocking(p, b, | |
2991 | BTRFS_READ_LOCK); | |
2992 | } | |
0b246afa | 2993 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
2994 | if (!b) { |
2995 | ret = -ENOMEM; | |
2996 | goto done; | |
2997 | } | |
5d9e75c4 JS |
2998 | p->locks[level] = BTRFS_READ_LOCK; |
2999 | p->nodes[level] = b; | |
5d9e75c4 JS |
3000 | } else { |
3001 | p->slots[level] = slot; | |
3002 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3003 | goto done; | |
3004 | } | |
3005 | } | |
3006 | ret = 1; | |
3007 | done: | |
3008 | if (!p->leave_spinning) | |
3009 | btrfs_set_path_blocking(p); | |
3010 | if (ret < 0) | |
3011 | btrfs_release_path(p); | |
3012 | ||
3013 | return ret; | |
3014 | } | |
3015 | ||
2f38b3e1 AJ |
3016 | /* |
3017 | * helper to use instead of search slot if no exact match is needed but | |
3018 | * instead the next or previous item should be returned. | |
3019 | * When find_higher is true, the next higher item is returned, the next lower | |
3020 | * otherwise. | |
3021 | * When return_any and find_higher are both true, and no higher item is found, | |
3022 | * return the next lower instead. | |
3023 | * When return_any is true and find_higher is false, and no lower item is found, | |
3024 | * return the next higher instead. | |
3025 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3026 | * < 0 on error | |
3027 | */ | |
3028 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3029 | const struct btrfs_key *key, |
3030 | struct btrfs_path *p, int find_higher, | |
3031 | int return_any) | |
2f38b3e1 AJ |
3032 | { |
3033 | int ret; | |
3034 | struct extent_buffer *leaf; | |
3035 | ||
3036 | again: | |
3037 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3038 | if (ret <= 0) | |
3039 | return ret; | |
3040 | /* | |
3041 | * a return value of 1 means the path is at the position where the | |
3042 | * item should be inserted. Normally this is the next bigger item, | |
3043 | * but in case the previous item is the last in a leaf, path points | |
3044 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3045 | * item. | |
3046 | */ | |
3047 | leaf = p->nodes[0]; | |
3048 | ||
3049 | if (find_higher) { | |
3050 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3051 | ret = btrfs_next_leaf(root, p); | |
3052 | if (ret <= 0) | |
3053 | return ret; | |
3054 | if (!return_any) | |
3055 | return 1; | |
3056 | /* | |
3057 | * no higher item found, return the next | |
3058 | * lower instead | |
3059 | */ | |
3060 | return_any = 0; | |
3061 | find_higher = 0; | |
3062 | btrfs_release_path(p); | |
3063 | goto again; | |
3064 | } | |
3065 | } else { | |
e6793769 AJ |
3066 | if (p->slots[0] == 0) { |
3067 | ret = btrfs_prev_leaf(root, p); | |
3068 | if (ret < 0) | |
3069 | return ret; | |
3070 | if (!ret) { | |
23c6bf6a FDBM |
3071 | leaf = p->nodes[0]; |
3072 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3073 | p->slots[0]--; | |
e6793769 | 3074 | return 0; |
2f38b3e1 | 3075 | } |
e6793769 AJ |
3076 | if (!return_any) |
3077 | return 1; | |
3078 | /* | |
3079 | * no lower item found, return the next | |
3080 | * higher instead | |
3081 | */ | |
3082 | return_any = 0; | |
3083 | find_higher = 1; | |
3084 | btrfs_release_path(p); | |
3085 | goto again; | |
3086 | } else { | |
2f38b3e1 AJ |
3087 | --p->slots[0]; |
3088 | } | |
3089 | } | |
3090 | return 0; | |
3091 | } | |
3092 | ||
74123bd7 CM |
3093 | /* |
3094 | * adjust the pointers going up the tree, starting at level | |
3095 | * making sure the right key of each node is points to 'key'. | |
3096 | * This is used after shifting pointers to the left, so it stops | |
3097 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3098 | * higher levels | |
aa5d6bed | 3099 | * |
74123bd7 | 3100 | */ |
b7a0365e DD |
3101 | static void fixup_low_keys(struct btrfs_fs_info *fs_info, |
3102 | struct btrfs_path *path, | |
143bede5 | 3103 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3104 | { |
3105 | int i; | |
5f39d397 | 3106 | struct extent_buffer *t; |
0e82bcfe | 3107 | int ret; |
5f39d397 | 3108 | |
234b63a0 | 3109 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3110 | int tslot = path->slots[i]; |
0e82bcfe | 3111 | |
eb60ceac | 3112 | if (!path->nodes[i]) |
be0e5c09 | 3113 | break; |
5f39d397 | 3114 | t = path->nodes[i]; |
0e82bcfe DS |
3115 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3116 | GFP_ATOMIC); | |
3117 | BUG_ON(ret < 0); | |
5f39d397 | 3118 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3119 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3120 | if (tslot != 0) |
3121 | break; | |
3122 | } | |
3123 | } | |
3124 | ||
31840ae1 ZY |
3125 | /* |
3126 | * update item key. | |
3127 | * | |
3128 | * This function isn't completely safe. It's the caller's responsibility | |
3129 | * that the new key won't break the order | |
3130 | */ | |
b7a0365e DD |
3131 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3132 | struct btrfs_path *path, | |
310712b2 | 3133 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3134 | { |
3135 | struct btrfs_disk_key disk_key; | |
3136 | struct extent_buffer *eb; | |
3137 | int slot; | |
3138 | ||
3139 | eb = path->nodes[0]; | |
3140 | slot = path->slots[0]; | |
3141 | if (slot > 0) { | |
3142 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3143 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3144 | } |
3145 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3146 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3147 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3148 | } |
3149 | ||
3150 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3151 | btrfs_set_item_key(eb, &disk_key, slot); | |
3152 | btrfs_mark_buffer_dirty(eb); | |
3153 | if (slot == 0) | |
b7a0365e | 3154 | fixup_low_keys(fs_info, path, &disk_key, 1); |
31840ae1 ZY |
3155 | } |
3156 | ||
74123bd7 CM |
3157 | /* |
3158 | * try to push data from one node into the next node left in the | |
79f95c82 | 3159 | * tree. |
aa5d6bed CM |
3160 | * |
3161 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3162 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3163 | */ |
98ed5174 | 3164 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
3165 | struct btrfs_fs_info *fs_info, |
3166 | struct extent_buffer *dst, | |
971a1f66 | 3167 | struct extent_buffer *src, int empty) |
be0e5c09 | 3168 | { |
be0e5c09 | 3169 | int push_items = 0; |
bb803951 CM |
3170 | int src_nritems; |
3171 | int dst_nritems; | |
aa5d6bed | 3172 | int ret = 0; |
be0e5c09 | 3173 | |
5f39d397 CM |
3174 | src_nritems = btrfs_header_nritems(src); |
3175 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3176 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3177 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3178 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3179 | |
bce4eae9 | 3180 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3181 | return 1; |
3182 | ||
d397712b | 3183 | if (push_items <= 0) |
be0e5c09 CM |
3184 | return 1; |
3185 | ||
bce4eae9 | 3186 | if (empty) { |
971a1f66 | 3187 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3188 | if (push_items < src_nritems) { |
3189 | /* leave at least 8 pointers in the node if | |
3190 | * we aren't going to empty it | |
3191 | */ | |
3192 | if (src_nritems - push_items < 8) { | |
3193 | if (push_items <= 8) | |
3194 | return 1; | |
3195 | push_items -= 8; | |
3196 | } | |
3197 | } | |
3198 | } else | |
3199 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3200 | |
0b246afa | 3201 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3202 | push_items); |
3203 | if (ret) { | |
66642832 | 3204 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3205 | return ret; |
3206 | } | |
5f39d397 CM |
3207 | copy_extent_buffer(dst, src, |
3208 | btrfs_node_key_ptr_offset(dst_nritems), | |
3209 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3210 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3211 | |
bb803951 | 3212 | if (push_items < src_nritems) { |
57911b8b | 3213 | /* |
bf1d3425 DS |
3214 | * Don't call tree_mod_log_insert_move here, key removal was |
3215 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3216 | */ |
5f39d397 CM |
3217 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3218 | btrfs_node_key_ptr_offset(push_items), | |
3219 | (src_nritems - push_items) * | |
3220 | sizeof(struct btrfs_key_ptr)); | |
3221 | } | |
3222 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3223 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3224 | btrfs_mark_buffer_dirty(src); | |
3225 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3226 | |
79f95c82 CM |
3227 | return ret; |
3228 | } | |
3229 | ||
3230 | /* | |
3231 | * try to push data from one node into the next node right in the | |
3232 | * tree. | |
3233 | * | |
3234 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3235 | * error, and > 0 if there was no room in the right hand block. | |
3236 | * | |
3237 | * this will only push up to 1/2 the contents of the left node over | |
3238 | */ | |
5f39d397 | 3239 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 3240 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
3241 | struct extent_buffer *dst, |
3242 | struct extent_buffer *src) | |
79f95c82 | 3243 | { |
79f95c82 CM |
3244 | int push_items = 0; |
3245 | int max_push; | |
3246 | int src_nritems; | |
3247 | int dst_nritems; | |
3248 | int ret = 0; | |
79f95c82 | 3249 | |
7bb86316 CM |
3250 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3251 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3252 | ||
5f39d397 CM |
3253 | src_nritems = btrfs_header_nritems(src); |
3254 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3255 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3256 | if (push_items <= 0) |
79f95c82 | 3257 | return 1; |
bce4eae9 | 3258 | |
d397712b | 3259 | if (src_nritems < 4) |
bce4eae9 | 3260 | return 1; |
79f95c82 CM |
3261 | |
3262 | max_push = src_nritems / 2 + 1; | |
3263 | /* don't try to empty the node */ | |
d397712b | 3264 | if (max_push >= src_nritems) |
79f95c82 | 3265 | return 1; |
252c38f0 | 3266 | |
79f95c82 CM |
3267 | if (max_push < push_items) |
3268 | push_items = max_push; | |
3269 | ||
bf1d3425 DS |
3270 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3271 | BUG_ON(ret < 0); | |
5f39d397 CM |
3272 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3273 | btrfs_node_key_ptr_offset(0), | |
3274 | (dst_nritems) * | |
3275 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3276 | |
0b246afa | 3277 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3278 | src_nritems - push_items, push_items); |
3279 | if (ret) { | |
66642832 | 3280 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3281 | return ret; |
3282 | } | |
5f39d397 CM |
3283 | copy_extent_buffer(dst, src, |
3284 | btrfs_node_key_ptr_offset(0), | |
3285 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3286 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3287 | |
5f39d397 CM |
3288 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3289 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3290 | |
5f39d397 CM |
3291 | btrfs_mark_buffer_dirty(src); |
3292 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3293 | |
aa5d6bed | 3294 | return ret; |
be0e5c09 CM |
3295 | } |
3296 | ||
97571fd0 CM |
3297 | /* |
3298 | * helper function to insert a new root level in the tree. | |
3299 | * A new node is allocated, and a single item is inserted to | |
3300 | * point to the existing root | |
aa5d6bed CM |
3301 | * |
3302 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3303 | */ |
d397712b | 3304 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3305 | struct btrfs_root *root, |
fdd99c72 | 3306 | struct btrfs_path *path, int level) |
5c680ed6 | 3307 | { |
0b246afa | 3308 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3309 | u64 lower_gen; |
5f39d397 CM |
3310 | struct extent_buffer *lower; |
3311 | struct extent_buffer *c; | |
925baedd | 3312 | struct extent_buffer *old; |
5f39d397 | 3313 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3314 | int ret; |
5c680ed6 CM |
3315 | |
3316 | BUG_ON(path->nodes[level]); | |
3317 | BUG_ON(path->nodes[level-1] != root->node); | |
3318 | ||
7bb86316 CM |
3319 | lower = path->nodes[level-1]; |
3320 | if (level == 1) | |
3321 | btrfs_item_key(lower, &lower_key, 0); | |
3322 | else | |
3323 | btrfs_node_key(lower, &lower_key, 0); | |
3324 | ||
4d75f8a9 DS |
3325 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3326 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3327 | if (IS_ERR(c)) |
3328 | return PTR_ERR(c); | |
925baedd | 3329 | |
0b246afa | 3330 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3331 | |
b159fa28 | 3332 | memzero_extent_buffer(c, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3333 | btrfs_set_header_nritems(c, 1); |
3334 | btrfs_set_header_level(c, level); | |
db94535d | 3335 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3336 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3337 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3338 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3339 | |
0b246afa JM |
3340 | write_extent_buffer_fsid(c, fs_info->fsid); |
3341 | write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid); | |
e17cade2 | 3342 | |
5f39d397 | 3343 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3344 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3345 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3346 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3347 | |
3348 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3349 | |
5f39d397 | 3350 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3351 | |
925baedd | 3352 | old = root->node; |
d9d19a01 DS |
3353 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3354 | BUG_ON(ret < 0); | |
240f62c8 | 3355 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3356 | |
3357 | /* the super has an extra ref to root->node */ | |
3358 | free_extent_buffer(old); | |
3359 | ||
0b86a832 | 3360 | add_root_to_dirty_list(root); |
5f39d397 CM |
3361 | extent_buffer_get(c); |
3362 | path->nodes[level] = c; | |
95449a16 | 3363 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3364 | path->slots[level] = 0; |
3365 | return 0; | |
3366 | } | |
3367 | ||
74123bd7 CM |
3368 | /* |
3369 | * worker function to insert a single pointer in a node. | |
3370 | * the node should have enough room for the pointer already | |
97571fd0 | 3371 | * |
74123bd7 CM |
3372 | * slot and level indicate where you want the key to go, and |
3373 | * blocknr is the block the key points to. | |
3374 | */ | |
143bede5 | 3375 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2ff7e61e | 3376 | struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 3377 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3378 | int slot, int level) |
74123bd7 | 3379 | { |
5f39d397 | 3380 | struct extent_buffer *lower; |
74123bd7 | 3381 | int nritems; |
f3ea38da | 3382 | int ret; |
5c680ed6 CM |
3383 | |
3384 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3385 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3386 | lower = path->nodes[level]; |
3387 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3388 | BUG_ON(slot > nritems); |
0b246afa | 3389 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3390 | if (slot != nritems) { |
bf1d3425 DS |
3391 | if (level) { |
3392 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3393 | nritems - slot); |
bf1d3425 DS |
3394 | BUG_ON(ret < 0); |
3395 | } | |
5f39d397 CM |
3396 | memmove_extent_buffer(lower, |
3397 | btrfs_node_key_ptr_offset(slot + 1), | |
3398 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3399 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3400 | } |
c3e06965 | 3401 | if (level) { |
e09c2efe DS |
3402 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3403 | GFP_NOFS); | |
f3ea38da JS |
3404 | BUG_ON(ret < 0); |
3405 | } | |
5f39d397 | 3406 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3407 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3408 | WARN_ON(trans->transid == 0); |
3409 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3410 | btrfs_set_header_nritems(lower, nritems + 1); |
3411 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3412 | } |
3413 | ||
97571fd0 CM |
3414 | /* |
3415 | * split the node at the specified level in path in two. | |
3416 | * The path is corrected to point to the appropriate node after the split | |
3417 | * | |
3418 | * Before splitting this tries to make some room in the node by pushing | |
3419 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3420 | * |
3421 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3422 | */ |
e02119d5 CM |
3423 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3424 | struct btrfs_root *root, | |
3425 | struct btrfs_path *path, int level) | |
be0e5c09 | 3426 | { |
0b246afa | 3427 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3428 | struct extent_buffer *c; |
3429 | struct extent_buffer *split; | |
3430 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3431 | int mid; |
5c680ed6 | 3432 | int ret; |
7518a238 | 3433 | u32 c_nritems; |
eb60ceac | 3434 | |
5f39d397 | 3435 | c = path->nodes[level]; |
7bb86316 | 3436 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3437 | if (c == root->node) { |
d9abbf1c | 3438 | /* |
90f8d62e JS |
3439 | * trying to split the root, lets make a new one |
3440 | * | |
fdd99c72 | 3441 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3442 | * insert_new_root, because that root buffer will be kept as a |
3443 | * normal node. We are going to log removal of half of the | |
3444 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3445 | * tree lock on the buffer, which is why we cannot race with | |
3446 | * other tree_mod_log users. | |
d9abbf1c | 3447 | */ |
fdd99c72 | 3448 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3449 | if (ret) |
3450 | return ret; | |
b3612421 | 3451 | } else { |
e66f709b | 3452 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3453 | c = path->nodes[level]; |
3454 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3455 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3456 | return 0; |
54aa1f4d CM |
3457 | if (ret < 0) |
3458 | return ret; | |
be0e5c09 | 3459 | } |
e66f709b | 3460 | |
5f39d397 | 3461 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3462 | mid = (c_nritems + 1) / 2; |
3463 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3464 | |
4d75f8a9 DS |
3465 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3466 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3467 | if (IS_ERR(split)) |
3468 | return PTR_ERR(split); | |
3469 | ||
0b246afa | 3470 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3471 | |
b159fa28 | 3472 | memzero_extent_buffer(split, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3473 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3474 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3475 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3476 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3477 | btrfs_set_header_owner(split, root->root_key.objectid); |
0b246afa JM |
3478 | write_extent_buffer_fsid(split, fs_info->fsid); |
3479 | write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid); | |
54aa1f4d | 3480 | |
0b246afa | 3481 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3482 | if (ret) { |
66642832 | 3483 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3484 | return ret; |
3485 | } | |
5f39d397 CM |
3486 | copy_extent_buffer(split, c, |
3487 | btrfs_node_key_ptr_offset(0), | |
3488 | btrfs_node_key_ptr_offset(mid), | |
3489 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3490 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3491 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3492 | ret = 0; |
3493 | ||
5f39d397 CM |
3494 | btrfs_mark_buffer_dirty(c); |
3495 | btrfs_mark_buffer_dirty(split); | |
3496 | ||
2ff7e61e | 3497 | insert_ptr(trans, fs_info, path, &disk_key, split->start, |
c3e06965 | 3498 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3499 | |
5de08d7d | 3500 | if (path->slots[level] >= mid) { |
5c680ed6 | 3501 | path->slots[level] -= mid; |
925baedd | 3502 | btrfs_tree_unlock(c); |
5f39d397 CM |
3503 | free_extent_buffer(c); |
3504 | path->nodes[level] = split; | |
5c680ed6 CM |
3505 | path->slots[level + 1] += 1; |
3506 | } else { | |
925baedd | 3507 | btrfs_tree_unlock(split); |
5f39d397 | 3508 | free_extent_buffer(split); |
be0e5c09 | 3509 | } |
aa5d6bed | 3510 | return ret; |
be0e5c09 CM |
3511 | } |
3512 | ||
74123bd7 CM |
3513 | /* |
3514 | * how many bytes are required to store the items in a leaf. start | |
3515 | * and nr indicate which items in the leaf to check. This totals up the | |
3516 | * space used both by the item structs and the item data | |
3517 | */ | |
5f39d397 | 3518 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3519 | { |
41be1f3b JB |
3520 | struct btrfs_item *start_item; |
3521 | struct btrfs_item *end_item; | |
3522 | struct btrfs_map_token token; | |
be0e5c09 | 3523 | int data_len; |
5f39d397 | 3524 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3525 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3526 | |
3527 | if (!nr) | |
3528 | return 0; | |
41be1f3b | 3529 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3530 | start_item = btrfs_item_nr(start); |
3531 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3532 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3533 | btrfs_token_item_size(l, start_item, &token); | |
3534 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3535 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3536 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3537 | return data_len; |
3538 | } | |
3539 | ||
d4dbff95 CM |
3540 | /* |
3541 | * The space between the end of the leaf items and | |
3542 | * the start of the leaf data. IOW, how much room | |
3543 | * the leaf has left for both items and data | |
3544 | */ | |
2ff7e61e | 3545 | noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info, |
e02119d5 | 3546 | struct extent_buffer *leaf) |
d4dbff95 | 3547 | { |
5f39d397 CM |
3548 | int nritems = btrfs_header_nritems(leaf); |
3549 | int ret; | |
0b246afa JM |
3550 | |
3551 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3552 | if (ret < 0) { |
0b246afa JM |
3553 | btrfs_crit(fs_info, |
3554 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3555 | ret, | |
3556 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3557 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3558 | } |
3559 | return ret; | |
d4dbff95 CM |
3560 | } |
3561 | ||
99d8f83c CM |
3562 | /* |
3563 | * min slot controls the lowest index we're willing to push to the | |
3564 | * right. We'll push up to and including min_slot, but no lower | |
3565 | */ | |
1e47eef2 | 3566 | static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3567 | struct btrfs_path *path, |
3568 | int data_size, int empty, | |
3569 | struct extent_buffer *right, | |
99d8f83c CM |
3570 | int free_space, u32 left_nritems, |
3571 | u32 min_slot) | |
00ec4c51 | 3572 | { |
5f39d397 | 3573 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3574 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3575 | struct btrfs_map_token token; |
5f39d397 | 3576 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3577 | int slot; |
34a38218 | 3578 | u32 i; |
00ec4c51 CM |
3579 | int push_space = 0; |
3580 | int push_items = 0; | |
0783fcfc | 3581 | struct btrfs_item *item; |
34a38218 | 3582 | u32 nr; |
7518a238 | 3583 | u32 right_nritems; |
5f39d397 | 3584 | u32 data_end; |
db94535d | 3585 | u32 this_item_size; |
00ec4c51 | 3586 | |
cfed81a0 CM |
3587 | btrfs_init_map_token(&token); |
3588 | ||
34a38218 CM |
3589 | if (empty) |
3590 | nr = 0; | |
3591 | else | |
99d8f83c | 3592 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3593 | |
31840ae1 | 3594 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3595 | push_space += data_size; |
31840ae1 | 3596 | |
44871b1b | 3597 | slot = path->slots[1]; |
34a38218 CM |
3598 | i = left_nritems - 1; |
3599 | while (i >= nr) { | |
dd3cc16b | 3600 | item = btrfs_item_nr(i); |
db94535d | 3601 | |
31840ae1 ZY |
3602 | if (!empty && push_items > 0) { |
3603 | if (path->slots[0] > i) | |
3604 | break; | |
3605 | if (path->slots[0] == i) { | |
2ff7e61e | 3606 | int space = btrfs_leaf_free_space(fs_info, left); |
31840ae1 ZY |
3607 | if (space + push_space * 2 > free_space) |
3608 | break; | |
3609 | } | |
3610 | } | |
3611 | ||
00ec4c51 | 3612 | if (path->slots[0] == i) |
87b29b20 | 3613 | push_space += data_size; |
db94535d | 3614 | |
db94535d CM |
3615 | this_item_size = btrfs_item_size(left, item); |
3616 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3617 | break; |
31840ae1 | 3618 | |
00ec4c51 | 3619 | push_items++; |
db94535d | 3620 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3621 | if (i == 0) |
3622 | break; | |
3623 | i--; | |
db94535d | 3624 | } |
5f39d397 | 3625 | |
925baedd CM |
3626 | if (push_items == 0) |
3627 | goto out_unlock; | |
5f39d397 | 3628 | |
6c1500f2 | 3629 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3630 | |
00ec4c51 | 3631 | /* push left to right */ |
5f39d397 | 3632 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3633 | |
5f39d397 | 3634 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
2ff7e61e | 3635 | push_space -= leaf_data_end(fs_info, left); |
5f39d397 | 3636 | |
00ec4c51 | 3637 | /* make room in the right data area */ |
2ff7e61e | 3638 | data_end = leaf_data_end(fs_info, right); |
5f39d397 | 3639 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3640 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3641 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3642 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3643 | |
00ec4c51 | 3644 | /* copy from the left data area */ |
3d9ec8c4 | 3645 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3646 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3647 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left), |
d6025579 | 3648 | push_space); |
5f39d397 CM |
3649 | |
3650 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3651 | btrfs_item_nr_offset(0), | |
3652 | right_nritems * sizeof(struct btrfs_item)); | |
3653 | ||
00ec4c51 | 3654 | /* copy the items from left to right */ |
5f39d397 CM |
3655 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3656 | btrfs_item_nr_offset(left_nritems - push_items), | |
3657 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3658 | |
3659 | /* update the item pointers */ | |
7518a238 | 3660 | right_nritems += push_items; |
5f39d397 | 3661 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3662 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3663 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3664 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3665 | push_space -= btrfs_token_item_size(right, item, &token); |
3666 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3667 | } |
3668 | ||
7518a238 | 3669 | left_nritems -= push_items; |
5f39d397 | 3670 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3671 | |
34a38218 CM |
3672 | if (left_nritems) |
3673 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3674 | else |
7c302b49 | 3675 | clean_tree_block(fs_info, left); |
f0486c68 | 3676 | |
5f39d397 | 3677 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3678 | |
5f39d397 CM |
3679 | btrfs_item_key(right, &disk_key, 0); |
3680 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3681 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3682 | |
00ec4c51 | 3683 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3684 | if (path->slots[0] >= left_nritems) { |
3685 | path->slots[0] -= left_nritems; | |
925baedd | 3686 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
7c302b49 | 3687 | clean_tree_block(fs_info, path->nodes[0]); |
925baedd | 3688 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3689 | free_extent_buffer(path->nodes[0]); |
3690 | path->nodes[0] = right; | |
00ec4c51 CM |
3691 | path->slots[1] += 1; |
3692 | } else { | |
925baedd | 3693 | btrfs_tree_unlock(right); |
5f39d397 | 3694 | free_extent_buffer(right); |
00ec4c51 CM |
3695 | } |
3696 | return 0; | |
925baedd CM |
3697 | |
3698 | out_unlock: | |
3699 | btrfs_tree_unlock(right); | |
3700 | free_extent_buffer(right); | |
3701 | return 1; | |
00ec4c51 | 3702 | } |
925baedd | 3703 | |
44871b1b CM |
3704 | /* |
3705 | * push some data in the path leaf to the right, trying to free up at | |
3706 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3707 | * | |
3708 | * returns 1 if the push failed because the other node didn't have enough | |
3709 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3710 | * |
3711 | * this will push starting from min_slot to the end of the leaf. It won't | |
3712 | * push any slot lower than min_slot | |
44871b1b CM |
3713 | */ |
3714 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3715 | *root, struct btrfs_path *path, |
3716 | int min_data_size, int data_size, | |
3717 | int empty, u32 min_slot) | |
44871b1b | 3718 | { |
2ff7e61e | 3719 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3720 | struct extent_buffer *left = path->nodes[0]; |
3721 | struct extent_buffer *right; | |
3722 | struct extent_buffer *upper; | |
3723 | int slot; | |
3724 | int free_space; | |
3725 | u32 left_nritems; | |
3726 | int ret; | |
3727 | ||
3728 | if (!path->nodes[1]) | |
3729 | return 1; | |
3730 | ||
3731 | slot = path->slots[1]; | |
3732 | upper = path->nodes[1]; | |
3733 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3734 | return 1; | |
3735 | ||
3736 | btrfs_assert_tree_locked(path->nodes[1]); | |
3737 | ||
2ff7e61e | 3738 | right = read_node_slot(fs_info, upper, slot + 1); |
fb770ae4 LB |
3739 | /* |
3740 | * slot + 1 is not valid or we fail to read the right node, | |
3741 | * no big deal, just return. | |
3742 | */ | |
3743 | if (IS_ERR(right)) | |
91ca338d TI |
3744 | return 1; |
3745 | ||
44871b1b CM |
3746 | btrfs_tree_lock(right); |
3747 | btrfs_set_lock_blocking(right); | |
3748 | ||
2ff7e61e | 3749 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3750 | if (free_space < data_size) |
3751 | goto out_unlock; | |
3752 | ||
3753 | /* cow and double check */ | |
3754 | ret = btrfs_cow_block(trans, root, right, upper, | |
3755 | slot + 1, &right); | |
3756 | if (ret) | |
3757 | goto out_unlock; | |
3758 | ||
2ff7e61e | 3759 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3760 | if (free_space < data_size) |
3761 | goto out_unlock; | |
3762 | ||
3763 | left_nritems = btrfs_header_nritems(left); | |
3764 | if (left_nritems == 0) | |
3765 | goto out_unlock; | |
3766 | ||
2ef1fed2 FDBM |
3767 | if (path->slots[0] == left_nritems && !empty) { |
3768 | /* Key greater than all keys in the leaf, right neighbor has | |
3769 | * enough room for it and we're not emptying our leaf to delete | |
3770 | * it, therefore use right neighbor to insert the new item and | |
3771 | * no need to touch/dirty our left leaft. */ | |
3772 | btrfs_tree_unlock(left); | |
3773 | free_extent_buffer(left); | |
3774 | path->nodes[0] = right; | |
3775 | path->slots[0] = 0; | |
3776 | path->slots[1]++; | |
3777 | return 0; | |
3778 | } | |
3779 | ||
1e47eef2 | 3780 | return __push_leaf_right(fs_info, path, min_data_size, empty, |
99d8f83c | 3781 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3782 | out_unlock: |
3783 | btrfs_tree_unlock(right); | |
3784 | free_extent_buffer(right); | |
3785 | return 1; | |
3786 | } | |
3787 | ||
74123bd7 CM |
3788 | /* |
3789 | * push some data in the path leaf to the left, trying to free up at | |
3790 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3791 | * |
3792 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3793 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3794 | * items | |
74123bd7 | 3795 | */ |
66cb7ddb | 3796 | static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3797 | struct btrfs_path *path, int data_size, |
3798 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3799 | int free_space, u32 right_nritems, |
3800 | u32 max_slot) | |
be0e5c09 | 3801 | { |
5f39d397 CM |
3802 | struct btrfs_disk_key disk_key; |
3803 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3804 | int i; |
be0e5c09 CM |
3805 | int push_space = 0; |
3806 | int push_items = 0; | |
0783fcfc | 3807 | struct btrfs_item *item; |
7518a238 | 3808 | u32 old_left_nritems; |
34a38218 | 3809 | u32 nr; |
aa5d6bed | 3810 | int ret = 0; |
db94535d CM |
3811 | u32 this_item_size; |
3812 | u32 old_left_item_size; | |
cfed81a0 CM |
3813 | struct btrfs_map_token token; |
3814 | ||
3815 | btrfs_init_map_token(&token); | |
be0e5c09 | 3816 | |
34a38218 | 3817 | if (empty) |
99d8f83c | 3818 | nr = min(right_nritems, max_slot); |
34a38218 | 3819 | else |
99d8f83c | 3820 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3821 | |
3822 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3823 | item = btrfs_item_nr(i); |
db94535d | 3824 | |
31840ae1 ZY |
3825 | if (!empty && push_items > 0) { |
3826 | if (path->slots[0] < i) | |
3827 | break; | |
3828 | if (path->slots[0] == i) { | |
2ff7e61e | 3829 | int space = btrfs_leaf_free_space(fs_info, right); |
31840ae1 ZY |
3830 | if (space + push_space * 2 > free_space) |
3831 | break; | |
3832 | } | |
3833 | } | |
3834 | ||
be0e5c09 | 3835 | if (path->slots[0] == i) |
87b29b20 | 3836 | push_space += data_size; |
db94535d CM |
3837 | |
3838 | this_item_size = btrfs_item_size(right, item); | |
3839 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3840 | break; |
db94535d | 3841 | |
be0e5c09 | 3842 | push_items++; |
db94535d CM |
3843 | push_space += this_item_size + sizeof(*item); |
3844 | } | |
3845 | ||
be0e5c09 | 3846 | if (push_items == 0) { |
925baedd CM |
3847 | ret = 1; |
3848 | goto out; | |
be0e5c09 | 3849 | } |
fae7f21c | 3850 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3851 | |
be0e5c09 | 3852 | /* push data from right to left */ |
5f39d397 CM |
3853 | copy_extent_buffer(left, right, |
3854 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3855 | btrfs_item_nr_offset(0), | |
3856 | push_items * sizeof(struct btrfs_item)); | |
3857 | ||
0b246afa | 3858 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3859 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3860 | |
3d9ec8c4 | 3861 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3862 | leaf_data_end(fs_info, left) - push_space, |
3d9ec8c4 | 3863 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3864 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3865 | push_space); |
5f39d397 | 3866 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3867 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3868 | |
db94535d | 3869 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3870 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3871 | u32 ioff; |
db94535d | 3872 | |
dd3cc16b | 3873 | item = btrfs_item_nr(i); |
db94535d | 3874 | |
cfed81a0 CM |
3875 | ioff = btrfs_token_item_offset(left, item, &token); |
3876 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3877 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3878 | &token); |
be0e5c09 | 3879 | } |
5f39d397 | 3880 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3881 | |
3882 | /* fixup right node */ | |
31b1a2bd JL |
3883 | if (push_items > right_nritems) |
3884 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3885 | right_nritems); |
34a38218 CM |
3886 | |
3887 | if (push_items < right_nritems) { | |
3888 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2ff7e61e | 3889 | leaf_data_end(fs_info, right); |
3d9ec8c4 | 3890 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3891 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3892 | BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3893 | leaf_data_end(fs_info, right), push_space); |
34a38218 CM |
3894 | |
3895 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3896 | btrfs_item_nr_offset(push_items), |
3897 | (btrfs_header_nritems(right) - push_items) * | |
3898 | sizeof(struct btrfs_item)); | |
34a38218 | 3899 | } |
eef1c494 Y |
3900 | right_nritems -= push_items; |
3901 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3902 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3903 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3904 | item = btrfs_item_nr(i); |
db94535d | 3905 | |
cfed81a0 CM |
3906 | push_space = push_space - btrfs_token_item_size(right, |
3907 | item, &token); | |
3908 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3909 | } |
eb60ceac | 3910 | |
5f39d397 | 3911 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3912 | if (right_nritems) |
3913 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3914 | else |
7c302b49 | 3915 | clean_tree_block(fs_info, right); |
098f59c2 | 3916 | |
5f39d397 | 3917 | btrfs_item_key(right, &disk_key, 0); |
0b246afa | 3918 | fixup_low_keys(fs_info, path, &disk_key, 1); |
be0e5c09 CM |
3919 | |
3920 | /* then fixup the leaf pointer in the path */ | |
3921 | if (path->slots[0] < push_items) { | |
3922 | path->slots[0] += old_left_nritems; | |
925baedd | 3923 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3924 | free_extent_buffer(path->nodes[0]); |
3925 | path->nodes[0] = left; | |
be0e5c09 CM |
3926 | path->slots[1] -= 1; |
3927 | } else { | |
925baedd | 3928 | btrfs_tree_unlock(left); |
5f39d397 | 3929 | free_extent_buffer(left); |
be0e5c09 CM |
3930 | path->slots[0] -= push_items; |
3931 | } | |
eb60ceac | 3932 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3933 | return ret; |
925baedd CM |
3934 | out: |
3935 | btrfs_tree_unlock(left); | |
3936 | free_extent_buffer(left); | |
3937 | return ret; | |
be0e5c09 CM |
3938 | } |
3939 | ||
44871b1b CM |
3940 | /* |
3941 | * push some data in the path leaf to the left, trying to free up at | |
3942 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3943 | * |
3944 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3945 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3946 | * items | |
44871b1b CM |
3947 | */ |
3948 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3949 | *root, struct btrfs_path *path, int min_data_size, |
3950 | int data_size, int empty, u32 max_slot) | |
44871b1b | 3951 | { |
2ff7e61e | 3952 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3953 | struct extent_buffer *right = path->nodes[0]; |
3954 | struct extent_buffer *left; | |
3955 | int slot; | |
3956 | int free_space; | |
3957 | u32 right_nritems; | |
3958 | int ret = 0; | |
3959 | ||
3960 | slot = path->slots[1]; | |
3961 | if (slot == 0) | |
3962 | return 1; | |
3963 | if (!path->nodes[1]) | |
3964 | return 1; | |
3965 | ||
3966 | right_nritems = btrfs_header_nritems(right); | |
3967 | if (right_nritems == 0) | |
3968 | return 1; | |
3969 | ||
3970 | btrfs_assert_tree_locked(path->nodes[1]); | |
3971 | ||
2ff7e61e | 3972 | left = read_node_slot(fs_info, path->nodes[1], slot - 1); |
fb770ae4 LB |
3973 | /* |
3974 | * slot - 1 is not valid or we fail to read the left node, | |
3975 | * no big deal, just return. | |
3976 | */ | |
3977 | if (IS_ERR(left)) | |
91ca338d TI |
3978 | return 1; |
3979 | ||
44871b1b CM |
3980 | btrfs_tree_lock(left); |
3981 | btrfs_set_lock_blocking(left); | |
3982 | ||
2ff7e61e | 3983 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
3984 | if (free_space < data_size) { |
3985 | ret = 1; | |
3986 | goto out; | |
3987 | } | |
3988 | ||
3989 | /* cow and double check */ | |
3990 | ret = btrfs_cow_block(trans, root, left, | |
3991 | path->nodes[1], slot - 1, &left); | |
3992 | if (ret) { | |
3993 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3994 | if (ret == -ENOSPC) |
3995 | ret = 1; | |
44871b1b CM |
3996 | goto out; |
3997 | } | |
3998 | ||
2ff7e61e | 3999 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4000 | if (free_space < data_size) { |
4001 | ret = 1; | |
4002 | goto out; | |
4003 | } | |
4004 | ||
66cb7ddb | 4005 | return __push_leaf_left(fs_info, path, min_data_size, |
99d8f83c CM |
4006 | empty, left, free_space, right_nritems, |
4007 | max_slot); | |
44871b1b CM |
4008 | out: |
4009 | btrfs_tree_unlock(left); | |
4010 | free_extent_buffer(left); | |
4011 | return ret; | |
4012 | } | |
4013 | ||
4014 | /* | |
4015 | * split the path's leaf in two, making sure there is at least data_size | |
4016 | * available for the resulting leaf level of the path. | |
44871b1b | 4017 | */ |
143bede5 | 4018 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2ff7e61e | 4019 | struct btrfs_fs_info *fs_info, |
143bede5 JM |
4020 | struct btrfs_path *path, |
4021 | struct extent_buffer *l, | |
4022 | struct extent_buffer *right, | |
4023 | int slot, int mid, int nritems) | |
44871b1b CM |
4024 | { |
4025 | int data_copy_size; | |
4026 | int rt_data_off; | |
4027 | int i; | |
44871b1b | 4028 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4029 | struct btrfs_map_token token; |
4030 | ||
4031 | btrfs_init_map_token(&token); | |
44871b1b CM |
4032 | |
4033 | nritems = nritems - mid; | |
4034 | btrfs_set_header_nritems(right, nritems); | |
2ff7e61e | 4035 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l); |
44871b1b CM |
4036 | |
4037 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4038 | btrfs_item_nr_offset(mid), | |
4039 | nritems * sizeof(struct btrfs_item)); | |
4040 | ||
4041 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4042 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4043 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
2ff7e61e | 4044 | leaf_data_end(fs_info, l), data_copy_size); |
44871b1b | 4045 | |
0b246afa | 4046 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4047 | |
4048 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4049 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4050 | u32 ioff; |
4051 | ||
cfed81a0 CM |
4052 | ioff = btrfs_token_item_offset(right, item, &token); |
4053 | btrfs_set_token_item_offset(right, item, | |
4054 | ioff + rt_data_off, &token); | |
44871b1b CM |
4055 | } |
4056 | ||
44871b1b | 4057 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4058 | btrfs_item_key(right, &disk_key, 0); |
2ff7e61e | 4059 | insert_ptr(trans, fs_info, path, &disk_key, right->start, |
c3e06965 | 4060 | path->slots[1] + 1, 1); |
44871b1b CM |
4061 | |
4062 | btrfs_mark_buffer_dirty(right); | |
4063 | btrfs_mark_buffer_dirty(l); | |
4064 | BUG_ON(path->slots[0] != slot); | |
4065 | ||
44871b1b CM |
4066 | if (mid <= slot) { |
4067 | btrfs_tree_unlock(path->nodes[0]); | |
4068 | free_extent_buffer(path->nodes[0]); | |
4069 | path->nodes[0] = right; | |
4070 | path->slots[0] -= mid; | |
4071 | path->slots[1] += 1; | |
4072 | } else { | |
4073 | btrfs_tree_unlock(right); | |
4074 | free_extent_buffer(right); | |
4075 | } | |
4076 | ||
4077 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4078 | } |
4079 | ||
99d8f83c CM |
4080 | /* |
4081 | * double splits happen when we need to insert a big item in the middle | |
4082 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4083 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4084 | * A B C | |
4085 | * | |
4086 | * We avoid this by trying to push the items on either side of our target | |
4087 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4088 | * completely. | |
4089 | */ | |
4090 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4091 | struct btrfs_root *root, | |
4092 | struct btrfs_path *path, | |
4093 | int data_size) | |
4094 | { | |
2ff7e61e | 4095 | struct btrfs_fs_info *fs_info = root->fs_info; |
99d8f83c CM |
4096 | int ret; |
4097 | int progress = 0; | |
4098 | int slot; | |
4099 | u32 nritems; | |
5a4267ca | 4100 | int space_needed = data_size; |
99d8f83c CM |
4101 | |
4102 | slot = path->slots[0]; | |
5a4267ca | 4103 | if (slot < btrfs_header_nritems(path->nodes[0])) |
2ff7e61e | 4104 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); |
99d8f83c CM |
4105 | |
4106 | /* | |
4107 | * try to push all the items after our slot into the | |
4108 | * right leaf | |
4109 | */ | |
5a4267ca | 4110 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4111 | if (ret < 0) |
4112 | return ret; | |
4113 | ||
4114 | if (ret == 0) | |
4115 | progress++; | |
4116 | ||
4117 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4118 | /* | |
4119 | * our goal is to get our slot at the start or end of a leaf. If | |
4120 | * we've done so we're done | |
4121 | */ | |
4122 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4123 | return 0; | |
4124 | ||
2ff7e61e | 4125 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4126 | return 0; |
4127 | ||
4128 | /* try to push all the items before our slot into the next leaf */ | |
4129 | slot = path->slots[0]; | |
263d3995 FM |
4130 | space_needed = data_size; |
4131 | if (slot > 0) | |
4132 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); | |
5a4267ca | 4133 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4134 | if (ret < 0) |
4135 | return ret; | |
4136 | ||
4137 | if (ret == 0) | |
4138 | progress++; | |
4139 | ||
4140 | if (progress) | |
4141 | return 0; | |
4142 | return 1; | |
4143 | } | |
4144 | ||
74123bd7 CM |
4145 | /* |
4146 | * split the path's leaf in two, making sure there is at least data_size | |
4147 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4148 | * |
4149 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4150 | */ |
e02119d5 CM |
4151 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4152 | struct btrfs_root *root, | |
310712b2 | 4153 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4154 | struct btrfs_path *path, int data_size, |
4155 | int extend) | |
be0e5c09 | 4156 | { |
5d4f98a2 | 4157 | struct btrfs_disk_key disk_key; |
5f39d397 | 4158 | struct extent_buffer *l; |
7518a238 | 4159 | u32 nritems; |
eb60ceac CM |
4160 | int mid; |
4161 | int slot; | |
5f39d397 | 4162 | struct extent_buffer *right; |
b7a0365e | 4163 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4164 | int ret = 0; |
aa5d6bed | 4165 | int wret; |
5d4f98a2 | 4166 | int split; |
cc0c5538 | 4167 | int num_doubles = 0; |
99d8f83c | 4168 | int tried_avoid_double = 0; |
aa5d6bed | 4169 | |
a5719521 YZ |
4170 | l = path->nodes[0]; |
4171 | slot = path->slots[0]; | |
4172 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4173 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4174 | return -EOVERFLOW; |
4175 | ||
40689478 | 4176 | /* first try to make some room by pushing left and right */ |
33157e05 | 4177 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4178 | int space_needed = data_size; |
4179 | ||
4180 | if (slot < btrfs_header_nritems(l)) | |
2ff7e61e | 4181 | space_needed -= btrfs_leaf_free_space(fs_info, l); |
5a4267ca FDBM |
4182 | |
4183 | wret = push_leaf_right(trans, root, path, space_needed, | |
4184 | space_needed, 0, 0); | |
d397712b | 4185 | if (wret < 0) |
eaee50e8 | 4186 | return wret; |
3685f791 | 4187 | if (wret) { |
263d3995 FM |
4188 | space_needed = data_size; |
4189 | if (slot > 0) | |
4190 | space_needed -= btrfs_leaf_free_space(fs_info, | |
4191 | l); | |
5a4267ca FDBM |
4192 | wret = push_leaf_left(trans, root, path, space_needed, |
4193 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4194 | if (wret < 0) |
4195 | return wret; | |
4196 | } | |
4197 | l = path->nodes[0]; | |
aa5d6bed | 4198 | |
3685f791 | 4199 | /* did the pushes work? */ |
2ff7e61e | 4200 | if (btrfs_leaf_free_space(fs_info, l) >= data_size) |
3685f791 | 4201 | return 0; |
3326d1b0 | 4202 | } |
aa5d6bed | 4203 | |
5c680ed6 | 4204 | if (!path->nodes[1]) { |
fdd99c72 | 4205 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4206 | if (ret) |
4207 | return ret; | |
4208 | } | |
cc0c5538 | 4209 | again: |
5d4f98a2 | 4210 | split = 1; |
cc0c5538 | 4211 | l = path->nodes[0]; |
eb60ceac | 4212 | slot = path->slots[0]; |
5f39d397 | 4213 | nritems = btrfs_header_nritems(l); |
d397712b | 4214 | mid = (nritems + 1) / 2; |
54aa1f4d | 4215 | |
5d4f98a2 YZ |
4216 | if (mid <= slot) { |
4217 | if (nritems == 1 || | |
4218 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4219 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4220 | if (slot >= nritems) { |
4221 | split = 0; | |
4222 | } else { | |
4223 | mid = slot; | |
4224 | if (mid != nritems && | |
4225 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4226 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4227 | if (data_size && !tried_avoid_double) |
4228 | goto push_for_double; | |
5d4f98a2 YZ |
4229 | split = 2; |
4230 | } | |
4231 | } | |
4232 | } | |
4233 | } else { | |
4234 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4235 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4236 | if (!extend && data_size && slot == 0) { |
4237 | split = 0; | |
4238 | } else if ((extend || !data_size) && slot == 0) { | |
4239 | mid = 1; | |
4240 | } else { | |
4241 | mid = slot; | |
4242 | if (mid != nritems && | |
4243 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4244 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4245 | if (data_size && !tried_avoid_double) |
4246 | goto push_for_double; | |
67871254 | 4247 | split = 2; |
5d4f98a2 YZ |
4248 | } |
4249 | } | |
4250 | } | |
4251 | } | |
4252 | ||
4253 | if (split == 0) | |
4254 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4255 | else | |
4256 | btrfs_item_key(l, &disk_key, mid); | |
4257 | ||
4d75f8a9 DS |
4258 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4259 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4260 | if (IS_ERR(right)) |
5f39d397 | 4261 | return PTR_ERR(right); |
f0486c68 | 4262 | |
0b246afa | 4263 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4264 | |
b159fa28 | 4265 | memzero_extent_buffer(right, 0, sizeof(struct btrfs_header)); |
db94535d | 4266 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4267 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4268 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4269 | btrfs_set_header_owner(right, root->root_key.objectid); |
4270 | btrfs_set_header_level(right, 0); | |
d24ee97b DS |
4271 | write_extent_buffer_fsid(right, fs_info->fsid); |
4272 | write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid); | |
44871b1b | 4273 | |
5d4f98a2 YZ |
4274 | if (split == 0) { |
4275 | if (mid <= slot) { | |
4276 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4277 | insert_ptr(trans, fs_info, path, &disk_key, |
4278 | right->start, path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
4279 | btrfs_tree_unlock(path->nodes[0]); |
4280 | free_extent_buffer(path->nodes[0]); | |
4281 | path->nodes[0] = right; | |
4282 | path->slots[0] = 0; | |
4283 | path->slots[1] += 1; | |
4284 | } else { | |
4285 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4286 | insert_ptr(trans, fs_info, path, &disk_key, |
4287 | right->start, path->slots[1], 1); | |
5d4f98a2 YZ |
4288 | btrfs_tree_unlock(path->nodes[0]); |
4289 | free_extent_buffer(path->nodes[0]); | |
4290 | path->nodes[0] = right; | |
4291 | path->slots[0] = 0; | |
143bede5 | 4292 | if (path->slots[1] == 0) |
b7a0365e | 4293 | fixup_low_keys(fs_info, path, &disk_key, 1); |
d4dbff95 | 4294 | } |
196e0249 LB |
4295 | /* |
4296 | * We create a new leaf 'right' for the required ins_len and | |
4297 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4298 | * the content of ins_len to 'right'. | |
4299 | */ | |
5d4f98a2 | 4300 | return ret; |
d4dbff95 | 4301 | } |
74123bd7 | 4302 | |
2ff7e61e | 4303 | copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems); |
31840ae1 | 4304 | |
5d4f98a2 | 4305 | if (split == 2) { |
cc0c5538 CM |
4306 | BUG_ON(num_doubles != 0); |
4307 | num_doubles++; | |
4308 | goto again; | |
a429e513 | 4309 | } |
44871b1b | 4310 | |
143bede5 | 4311 | return 0; |
99d8f83c CM |
4312 | |
4313 | push_for_double: | |
4314 | push_for_double_split(trans, root, path, data_size); | |
4315 | tried_avoid_double = 1; | |
2ff7e61e | 4316 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4317 | return 0; |
4318 | goto again; | |
be0e5c09 CM |
4319 | } |
4320 | ||
ad48fd75 YZ |
4321 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4322 | struct btrfs_root *root, | |
4323 | struct btrfs_path *path, int ins_len) | |
459931ec | 4324 | { |
2ff7e61e | 4325 | struct btrfs_fs_info *fs_info = root->fs_info; |
ad48fd75 | 4326 | struct btrfs_key key; |
459931ec | 4327 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4328 | struct btrfs_file_extent_item *fi; |
4329 | u64 extent_len = 0; | |
4330 | u32 item_size; | |
4331 | int ret; | |
459931ec CM |
4332 | |
4333 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4334 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4335 | ||
4336 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4337 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4338 | ||
2ff7e61e | 4339 | if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len) |
ad48fd75 | 4340 | return 0; |
459931ec CM |
4341 | |
4342 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4343 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4344 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4345 | struct btrfs_file_extent_item); | |
4346 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4347 | } | |
b3b4aa74 | 4348 | btrfs_release_path(path); |
459931ec | 4349 | |
459931ec | 4350 | path->keep_locks = 1; |
ad48fd75 YZ |
4351 | path->search_for_split = 1; |
4352 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4353 | path->search_for_split = 0; |
a8df6fe6 FM |
4354 | if (ret > 0) |
4355 | ret = -EAGAIN; | |
ad48fd75 YZ |
4356 | if (ret < 0) |
4357 | goto err; | |
459931ec | 4358 | |
ad48fd75 YZ |
4359 | ret = -EAGAIN; |
4360 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4361 | /* if our item isn't there, return now */ |
4362 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4363 | goto err; |
4364 | ||
109f6aef | 4365 | /* the leaf has changed, it now has room. return now */ |
2ff7e61e | 4366 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len) |
109f6aef CM |
4367 | goto err; |
4368 | ||
ad48fd75 YZ |
4369 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4370 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4371 | struct btrfs_file_extent_item); | |
4372 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4373 | goto err; | |
459931ec CM |
4374 | } |
4375 | ||
b9473439 | 4376 | btrfs_set_path_blocking(path); |
ad48fd75 | 4377 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4378 | if (ret) |
4379 | goto err; | |
459931ec | 4380 | |
ad48fd75 | 4381 | path->keep_locks = 0; |
b9473439 | 4382 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4383 | return 0; |
4384 | err: | |
4385 | path->keep_locks = 0; | |
4386 | return ret; | |
4387 | } | |
4388 | ||
4961e293 | 4389 | static noinline int split_item(struct btrfs_fs_info *fs_info, |
ad48fd75 | 4390 | struct btrfs_path *path, |
310712b2 | 4391 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4392 | unsigned long split_offset) |
4393 | { | |
4394 | struct extent_buffer *leaf; | |
4395 | struct btrfs_item *item; | |
4396 | struct btrfs_item *new_item; | |
4397 | int slot; | |
4398 | char *buf; | |
4399 | u32 nritems; | |
4400 | u32 item_size; | |
4401 | u32 orig_offset; | |
4402 | struct btrfs_disk_key disk_key; | |
4403 | ||
b9473439 | 4404 | leaf = path->nodes[0]; |
2ff7e61e | 4405 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4406 | |
b4ce94de CM |
4407 | btrfs_set_path_blocking(path); |
4408 | ||
dd3cc16b | 4409 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4410 | orig_offset = btrfs_item_offset(leaf, item); |
4411 | item_size = btrfs_item_size(leaf, item); | |
4412 | ||
459931ec | 4413 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4414 | if (!buf) |
4415 | return -ENOMEM; | |
4416 | ||
459931ec CM |
4417 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4418 | path->slots[0]), item_size); | |
459931ec | 4419 | |
ad48fd75 | 4420 | slot = path->slots[0] + 1; |
459931ec | 4421 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4422 | if (slot != nritems) { |
4423 | /* shift the items */ | |
4424 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4425 | btrfs_item_nr_offset(slot), |
4426 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4427 | } |
4428 | ||
4429 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4430 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4431 | ||
dd3cc16b | 4432 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4433 | |
4434 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4435 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4436 | ||
4437 | btrfs_set_item_offset(leaf, item, | |
4438 | orig_offset + item_size - split_offset); | |
4439 | btrfs_set_item_size(leaf, item, split_offset); | |
4440 | ||
4441 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4442 | ||
4443 | /* write the data for the start of the original item */ | |
4444 | write_extent_buffer(leaf, buf, | |
4445 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4446 | split_offset); | |
4447 | ||
4448 | /* write the data for the new item */ | |
4449 | write_extent_buffer(leaf, buf + split_offset, | |
4450 | btrfs_item_ptr_offset(leaf, slot), | |
4451 | item_size - split_offset); | |
4452 | btrfs_mark_buffer_dirty(leaf); | |
4453 | ||
2ff7e61e | 4454 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0); |
459931ec | 4455 | kfree(buf); |
ad48fd75 YZ |
4456 | return 0; |
4457 | } | |
4458 | ||
4459 | /* | |
4460 | * This function splits a single item into two items, | |
4461 | * giving 'new_key' to the new item and splitting the | |
4462 | * old one at split_offset (from the start of the item). | |
4463 | * | |
4464 | * The path may be released by this operation. After | |
4465 | * the split, the path is pointing to the old item. The | |
4466 | * new item is going to be in the same node as the old one. | |
4467 | * | |
4468 | * Note, the item being split must be smaller enough to live alone on | |
4469 | * a tree block with room for one extra struct btrfs_item | |
4470 | * | |
4471 | * This allows us to split the item in place, keeping a lock on the | |
4472 | * leaf the entire time. | |
4473 | */ | |
4474 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4475 | struct btrfs_root *root, | |
4476 | struct btrfs_path *path, | |
310712b2 | 4477 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4478 | unsigned long split_offset) |
4479 | { | |
4480 | int ret; | |
4481 | ret = setup_leaf_for_split(trans, root, path, | |
4482 | sizeof(struct btrfs_item)); | |
4483 | if (ret) | |
4484 | return ret; | |
4485 | ||
4961e293 | 4486 | ret = split_item(root->fs_info, path, new_key, split_offset); |
459931ec CM |
4487 | return ret; |
4488 | } | |
4489 | ||
ad48fd75 YZ |
4490 | /* |
4491 | * This function duplicate a item, giving 'new_key' to the new item. | |
4492 | * It guarantees both items live in the same tree leaf and the new item | |
4493 | * is contiguous with the original item. | |
4494 | * | |
4495 | * This allows us to split file extent in place, keeping a lock on the | |
4496 | * leaf the entire time. | |
4497 | */ | |
4498 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4499 | struct btrfs_root *root, | |
4500 | struct btrfs_path *path, | |
310712b2 | 4501 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4502 | { |
4503 | struct extent_buffer *leaf; | |
4504 | int ret; | |
4505 | u32 item_size; | |
4506 | ||
4507 | leaf = path->nodes[0]; | |
4508 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4509 | ret = setup_leaf_for_split(trans, root, path, | |
4510 | item_size + sizeof(struct btrfs_item)); | |
4511 | if (ret) | |
4512 | return ret; | |
4513 | ||
4514 | path->slots[0]++; | |
afe5fea7 | 4515 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4516 | item_size, item_size + |
4517 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4518 | leaf = path->nodes[0]; |
4519 | memcpy_extent_buffer(leaf, | |
4520 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4521 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4522 | item_size); | |
4523 | return 0; | |
4524 | } | |
4525 | ||
d352ac68 CM |
4526 | /* |
4527 | * make the item pointed to by the path smaller. new_size indicates | |
4528 | * how small to make it, and from_end tells us if we just chop bytes | |
4529 | * off the end of the item or if we shift the item to chop bytes off | |
4530 | * the front. | |
4531 | */ | |
2ff7e61e JM |
4532 | void btrfs_truncate_item(struct btrfs_fs_info *fs_info, |
4533 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 4534 | { |
b18c6685 | 4535 | int slot; |
5f39d397 CM |
4536 | struct extent_buffer *leaf; |
4537 | struct btrfs_item *item; | |
b18c6685 CM |
4538 | u32 nritems; |
4539 | unsigned int data_end; | |
4540 | unsigned int old_data_start; | |
4541 | unsigned int old_size; | |
4542 | unsigned int size_diff; | |
4543 | int i; | |
cfed81a0 CM |
4544 | struct btrfs_map_token token; |
4545 | ||
4546 | btrfs_init_map_token(&token); | |
b18c6685 | 4547 | |
5f39d397 | 4548 | leaf = path->nodes[0]; |
179e29e4 CM |
4549 | slot = path->slots[0]; |
4550 | ||
4551 | old_size = btrfs_item_size_nr(leaf, slot); | |
4552 | if (old_size == new_size) | |
143bede5 | 4553 | return; |
b18c6685 | 4554 | |
5f39d397 | 4555 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4556 | data_end = leaf_data_end(fs_info, leaf); |
b18c6685 | 4557 | |
5f39d397 | 4558 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4559 | |
b18c6685 CM |
4560 | size_diff = old_size - new_size; |
4561 | ||
4562 | BUG_ON(slot < 0); | |
4563 | BUG_ON(slot >= nritems); | |
4564 | ||
4565 | /* | |
4566 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4567 | */ | |
4568 | /* first correct the data pointers */ | |
4569 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4570 | u32 ioff; |
dd3cc16b | 4571 | item = btrfs_item_nr(i); |
db94535d | 4572 | |
cfed81a0 CM |
4573 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4574 | btrfs_set_token_item_offset(leaf, item, | |
4575 | ioff + size_diff, &token); | |
b18c6685 | 4576 | } |
db94535d | 4577 | |
b18c6685 | 4578 | /* shift the data */ |
179e29e4 | 4579 | if (from_end) { |
3d9ec8c4 NB |
4580 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4581 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4582 | data_end, old_data_start + new_size - data_end); |
4583 | } else { | |
4584 | struct btrfs_disk_key disk_key; | |
4585 | u64 offset; | |
4586 | ||
4587 | btrfs_item_key(leaf, &disk_key, slot); | |
4588 | ||
4589 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4590 | unsigned long ptr; | |
4591 | struct btrfs_file_extent_item *fi; | |
4592 | ||
4593 | fi = btrfs_item_ptr(leaf, slot, | |
4594 | struct btrfs_file_extent_item); | |
4595 | fi = (struct btrfs_file_extent_item *)( | |
4596 | (unsigned long)fi - size_diff); | |
4597 | ||
4598 | if (btrfs_file_extent_type(leaf, fi) == | |
4599 | BTRFS_FILE_EXTENT_INLINE) { | |
4600 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4601 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4602 | (unsigned long)fi, |
7ec20afb | 4603 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4604 | } |
4605 | } | |
4606 | ||
3d9ec8c4 NB |
4607 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4608 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4609 | data_end, old_data_start - data_end); |
4610 | ||
4611 | offset = btrfs_disk_key_offset(&disk_key); | |
4612 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4613 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4614 | if (slot == 0) | |
0b246afa | 4615 | fixup_low_keys(fs_info, path, &disk_key, 1); |
179e29e4 | 4616 | } |
5f39d397 | 4617 | |
dd3cc16b | 4618 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4619 | btrfs_set_item_size(leaf, item, new_size); |
4620 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4621 | |
2ff7e61e | 4622 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4623 | btrfs_print_leaf(leaf); |
b18c6685 | 4624 | BUG(); |
5f39d397 | 4625 | } |
b18c6685 CM |
4626 | } |
4627 | ||
d352ac68 | 4628 | /* |
8f69dbd2 | 4629 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4630 | */ |
2ff7e61e | 4631 | void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 4632 | u32 data_size) |
6567e837 | 4633 | { |
6567e837 | 4634 | int slot; |
5f39d397 CM |
4635 | struct extent_buffer *leaf; |
4636 | struct btrfs_item *item; | |
6567e837 CM |
4637 | u32 nritems; |
4638 | unsigned int data_end; | |
4639 | unsigned int old_data; | |
4640 | unsigned int old_size; | |
4641 | int i; | |
cfed81a0 CM |
4642 | struct btrfs_map_token token; |
4643 | ||
4644 | btrfs_init_map_token(&token); | |
6567e837 | 4645 | |
5f39d397 | 4646 | leaf = path->nodes[0]; |
6567e837 | 4647 | |
5f39d397 | 4648 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4649 | data_end = leaf_data_end(fs_info, leaf); |
6567e837 | 4650 | |
2ff7e61e | 4651 | if (btrfs_leaf_free_space(fs_info, leaf) < data_size) { |
a4f78750 | 4652 | btrfs_print_leaf(leaf); |
6567e837 | 4653 | BUG(); |
5f39d397 | 4654 | } |
6567e837 | 4655 | slot = path->slots[0]; |
5f39d397 | 4656 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4657 | |
4658 | BUG_ON(slot < 0); | |
3326d1b0 | 4659 | if (slot >= nritems) { |
a4f78750 | 4660 | btrfs_print_leaf(leaf); |
0b246afa JM |
4661 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4662 | slot, nritems); | |
3326d1b0 CM |
4663 | BUG_ON(1); |
4664 | } | |
6567e837 CM |
4665 | |
4666 | /* | |
4667 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4668 | */ | |
4669 | /* first correct the data pointers */ | |
4670 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4671 | u32 ioff; |
dd3cc16b | 4672 | item = btrfs_item_nr(i); |
db94535d | 4673 | |
cfed81a0 CM |
4674 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4675 | btrfs_set_token_item_offset(leaf, item, | |
4676 | ioff - data_size, &token); | |
6567e837 | 4677 | } |
5f39d397 | 4678 | |
6567e837 | 4679 | /* shift the data */ |
3d9ec8c4 NB |
4680 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4681 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4682 | data_end, old_data - data_end); |
5f39d397 | 4683 | |
6567e837 | 4684 | data_end = old_data; |
5f39d397 | 4685 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4686 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4687 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4688 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4689 | |
2ff7e61e | 4690 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4691 | btrfs_print_leaf(leaf); |
6567e837 | 4692 | BUG(); |
5f39d397 | 4693 | } |
6567e837 CM |
4694 | } |
4695 | ||
74123bd7 | 4696 | /* |
44871b1b CM |
4697 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4698 | * to save stack depth by doing the bulk of the work in a function | |
4699 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4700 | */ |
afe5fea7 | 4701 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4702 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4703 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4704 | { |
0b246afa | 4705 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4706 | struct btrfs_item *item; |
9c58309d | 4707 | int i; |
7518a238 | 4708 | u32 nritems; |
be0e5c09 | 4709 | unsigned int data_end; |
e2fa7227 | 4710 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4711 | struct extent_buffer *leaf; |
4712 | int slot; | |
cfed81a0 CM |
4713 | struct btrfs_map_token token; |
4714 | ||
24cdc847 FM |
4715 | if (path->slots[0] == 0) { |
4716 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
0b246afa | 4717 | fixup_low_keys(fs_info, path, &disk_key, 1); |
24cdc847 FM |
4718 | } |
4719 | btrfs_unlock_up_safe(path, 1); | |
4720 | ||
cfed81a0 | 4721 | btrfs_init_map_token(&token); |
e2fa7227 | 4722 | |
5f39d397 | 4723 | leaf = path->nodes[0]; |
44871b1b | 4724 | slot = path->slots[0]; |
74123bd7 | 4725 | |
5f39d397 | 4726 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4727 | data_end = leaf_data_end(fs_info, leaf); |
eb60ceac | 4728 | |
2ff7e61e | 4729 | if (btrfs_leaf_free_space(fs_info, leaf) < total_size) { |
a4f78750 | 4730 | btrfs_print_leaf(leaf); |
0b246afa | 4731 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
2ff7e61e | 4732 | total_size, btrfs_leaf_free_space(fs_info, leaf)); |
be0e5c09 | 4733 | BUG(); |
d4dbff95 | 4734 | } |
5f39d397 | 4735 | |
be0e5c09 | 4736 | if (slot != nritems) { |
5f39d397 | 4737 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4738 | |
5f39d397 | 4739 | if (old_data < data_end) { |
a4f78750 | 4740 | btrfs_print_leaf(leaf); |
0b246afa | 4741 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4742 | slot, old_data, data_end); |
5f39d397 CM |
4743 | BUG_ON(1); |
4744 | } | |
be0e5c09 CM |
4745 | /* |
4746 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4747 | */ | |
4748 | /* first correct the data pointers */ | |
0783fcfc | 4749 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4750 | u32 ioff; |
db94535d | 4751 | |
62e85577 | 4752 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4753 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4754 | btrfs_set_token_item_offset(leaf, item, | |
4755 | ioff - total_data, &token); | |
0783fcfc | 4756 | } |
be0e5c09 | 4757 | /* shift the items */ |
9c58309d | 4758 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4759 | btrfs_item_nr_offset(slot), |
d6025579 | 4760 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4761 | |
4762 | /* shift the data */ | |
3d9ec8c4 NB |
4763 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4764 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4765 | data_end, old_data - data_end); |
be0e5c09 CM |
4766 | data_end = old_data; |
4767 | } | |
5f39d397 | 4768 | |
62e2749e | 4769 | /* setup the item for the new data */ |
9c58309d CM |
4770 | for (i = 0; i < nr; i++) { |
4771 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4772 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4773 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4774 | btrfs_set_token_item_offset(leaf, item, |
4775 | data_end - data_size[i], &token); | |
9c58309d | 4776 | data_end -= data_size[i]; |
cfed81a0 | 4777 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4778 | } |
44871b1b | 4779 | |
9c58309d | 4780 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4781 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4782 | |
2ff7e61e | 4783 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4784 | btrfs_print_leaf(leaf); |
be0e5c09 | 4785 | BUG(); |
5f39d397 | 4786 | } |
44871b1b CM |
4787 | } |
4788 | ||
4789 | /* | |
4790 | * Given a key and some data, insert items into the tree. | |
4791 | * This does all the path init required, making room in the tree if needed. | |
4792 | */ | |
4793 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4794 | struct btrfs_root *root, | |
4795 | struct btrfs_path *path, | |
310712b2 | 4796 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4797 | int nr) |
4798 | { | |
44871b1b CM |
4799 | int ret = 0; |
4800 | int slot; | |
4801 | int i; | |
4802 | u32 total_size = 0; | |
4803 | u32 total_data = 0; | |
4804 | ||
4805 | for (i = 0; i < nr; i++) | |
4806 | total_data += data_size[i]; | |
4807 | ||
4808 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4809 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4810 | if (ret == 0) | |
4811 | return -EEXIST; | |
4812 | if (ret < 0) | |
143bede5 | 4813 | return ret; |
44871b1b | 4814 | |
44871b1b CM |
4815 | slot = path->slots[0]; |
4816 | BUG_ON(slot < 0); | |
4817 | ||
afe5fea7 | 4818 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4819 | total_data, total_size, nr); |
143bede5 | 4820 | return 0; |
62e2749e CM |
4821 | } |
4822 | ||
4823 | /* | |
4824 | * Given a key and some data, insert an item into the tree. | |
4825 | * This does all the path init required, making room in the tree if needed. | |
4826 | */ | |
310712b2 OS |
4827 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4828 | const struct btrfs_key *cpu_key, void *data, | |
4829 | u32 data_size) | |
62e2749e CM |
4830 | { |
4831 | int ret = 0; | |
2c90e5d6 | 4832 | struct btrfs_path *path; |
5f39d397 CM |
4833 | struct extent_buffer *leaf; |
4834 | unsigned long ptr; | |
62e2749e | 4835 | |
2c90e5d6 | 4836 | path = btrfs_alloc_path(); |
db5b493a TI |
4837 | if (!path) |
4838 | return -ENOMEM; | |
2c90e5d6 | 4839 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4840 | if (!ret) { |
5f39d397 CM |
4841 | leaf = path->nodes[0]; |
4842 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4843 | write_extent_buffer(leaf, data, ptr, data_size); | |
4844 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4845 | } |
2c90e5d6 | 4846 | btrfs_free_path(path); |
aa5d6bed | 4847 | return ret; |
be0e5c09 CM |
4848 | } |
4849 | ||
74123bd7 | 4850 | /* |
5de08d7d | 4851 | * delete the pointer from a given node. |
74123bd7 | 4852 | * |
d352ac68 CM |
4853 | * the tree should have been previously balanced so the deletion does not |
4854 | * empty a node. | |
74123bd7 | 4855 | */ |
afe5fea7 TI |
4856 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4857 | int level, int slot) | |
be0e5c09 | 4858 | { |
0b246afa | 4859 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4860 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4861 | u32 nritems; |
f3ea38da | 4862 | int ret; |
be0e5c09 | 4863 | |
5f39d397 | 4864 | nritems = btrfs_header_nritems(parent); |
d397712b | 4865 | if (slot != nritems - 1) { |
bf1d3425 DS |
4866 | if (level) { |
4867 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4868 | nritems - slot - 1); |
bf1d3425 DS |
4869 | BUG_ON(ret < 0); |
4870 | } | |
5f39d397 CM |
4871 | memmove_extent_buffer(parent, |
4872 | btrfs_node_key_ptr_offset(slot), | |
4873 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4874 | sizeof(struct btrfs_key_ptr) * |
4875 | (nritems - slot - 1)); | |
57ba86c0 | 4876 | } else if (level) { |
e09c2efe DS |
4877 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4878 | GFP_NOFS); | |
57ba86c0 | 4879 | BUG_ON(ret < 0); |
bb803951 | 4880 | } |
f3ea38da | 4881 | |
7518a238 | 4882 | nritems--; |
5f39d397 | 4883 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4884 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4885 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4886 | /* just turn the root into a leaf and break */ |
5f39d397 | 4887 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4888 | } else if (slot == 0) { |
5f39d397 CM |
4889 | struct btrfs_disk_key disk_key; |
4890 | ||
4891 | btrfs_node_key(parent, &disk_key, 0); | |
0b246afa | 4892 | fixup_low_keys(fs_info, path, &disk_key, level + 1); |
be0e5c09 | 4893 | } |
d6025579 | 4894 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4895 | } |
4896 | ||
323ac95b CM |
4897 | /* |
4898 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4899 | * path->nodes[1]. |
323ac95b CM |
4900 | * |
4901 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4902 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4903 | * | |
4904 | * The path must have already been setup for deleting the leaf, including | |
4905 | * all the proper balancing. path->nodes[1] must be locked. | |
4906 | */ | |
143bede5 JM |
4907 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4908 | struct btrfs_root *root, | |
4909 | struct btrfs_path *path, | |
4910 | struct extent_buffer *leaf) | |
323ac95b | 4911 | { |
5d4f98a2 | 4912 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4913 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4914 | |
4d081c41 CM |
4915 | /* |
4916 | * btrfs_free_extent is expensive, we want to make sure we | |
4917 | * aren't holding any locks when we call it | |
4918 | */ | |
4919 | btrfs_unlock_up_safe(path, 0); | |
4920 | ||
f0486c68 YZ |
4921 | root_sub_used(root, leaf->len); |
4922 | ||
3083ee2e | 4923 | extent_buffer_get(leaf); |
5581a51a | 4924 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4925 | free_extent_buffer_stale(leaf); |
323ac95b | 4926 | } |
74123bd7 CM |
4927 | /* |
4928 | * delete the item at the leaf level in path. If that empties | |
4929 | * the leaf, remove it from the tree | |
4930 | */ | |
85e21bac CM |
4931 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4932 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4933 | { |
0b246afa | 4934 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4935 | struct extent_buffer *leaf; |
4936 | struct btrfs_item *item; | |
ce0eac2a AM |
4937 | u32 last_off; |
4938 | u32 dsize = 0; | |
aa5d6bed CM |
4939 | int ret = 0; |
4940 | int wret; | |
85e21bac | 4941 | int i; |
7518a238 | 4942 | u32 nritems; |
cfed81a0 CM |
4943 | struct btrfs_map_token token; |
4944 | ||
4945 | btrfs_init_map_token(&token); | |
be0e5c09 | 4946 | |
5f39d397 | 4947 | leaf = path->nodes[0]; |
85e21bac CM |
4948 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4949 | ||
4950 | for (i = 0; i < nr; i++) | |
4951 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4952 | ||
5f39d397 | 4953 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4954 | |
85e21bac | 4955 | if (slot + nr != nritems) { |
2ff7e61e | 4956 | int data_end = leaf_data_end(fs_info, leaf); |
5f39d397 | 4957 | |
3d9ec8c4 | 4958 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4959 | data_end + dsize, |
3d9ec8c4 | 4960 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4961 | last_off - data_end); |
5f39d397 | 4962 | |
85e21bac | 4963 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4964 | u32 ioff; |
db94535d | 4965 | |
dd3cc16b | 4966 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4967 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4968 | btrfs_set_token_item_offset(leaf, item, | |
4969 | ioff + dsize, &token); | |
0783fcfc | 4970 | } |
db94535d | 4971 | |
5f39d397 | 4972 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4973 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4974 | sizeof(struct btrfs_item) * |
85e21bac | 4975 | (nritems - slot - nr)); |
be0e5c09 | 4976 | } |
85e21bac CM |
4977 | btrfs_set_header_nritems(leaf, nritems - nr); |
4978 | nritems -= nr; | |
5f39d397 | 4979 | |
74123bd7 | 4980 | /* delete the leaf if we've emptied it */ |
7518a238 | 4981 | if (nritems == 0) { |
5f39d397 CM |
4982 | if (leaf == root->node) { |
4983 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4984 | } else { |
f0486c68 | 4985 | btrfs_set_path_blocking(path); |
7c302b49 | 4986 | clean_tree_block(fs_info, leaf); |
143bede5 | 4987 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4988 | } |
be0e5c09 | 4989 | } else { |
7518a238 | 4990 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4991 | if (slot == 0) { |
5f39d397 CM |
4992 | struct btrfs_disk_key disk_key; |
4993 | ||
4994 | btrfs_item_key(leaf, &disk_key, 0); | |
0b246afa | 4995 | fixup_low_keys(fs_info, path, &disk_key, 1); |
aa5d6bed | 4996 | } |
aa5d6bed | 4997 | |
74123bd7 | 4998 | /* delete the leaf if it is mostly empty */ |
0b246afa | 4999 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5000 | /* push_leaf_left fixes the path. |
5001 | * make sure the path still points to our leaf | |
5002 | * for possible call to del_ptr below | |
5003 | */ | |
4920c9ac | 5004 | slot = path->slots[1]; |
5f39d397 CM |
5005 | extent_buffer_get(leaf); |
5006 | ||
b9473439 | 5007 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5008 | wret = push_leaf_left(trans, root, path, 1, 1, |
5009 | 1, (u32)-1); | |
54aa1f4d | 5010 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5011 | ret = wret; |
5f39d397 CM |
5012 | |
5013 | if (path->nodes[0] == leaf && | |
5014 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5015 | wret = push_leaf_right(trans, root, path, 1, |
5016 | 1, 1, 0); | |
54aa1f4d | 5017 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5018 | ret = wret; |
5019 | } | |
5f39d397 CM |
5020 | |
5021 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5022 | path->slots[1] = slot; |
143bede5 | 5023 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5024 | free_extent_buffer(leaf); |
143bede5 | 5025 | ret = 0; |
5de08d7d | 5026 | } else { |
925baedd CM |
5027 | /* if we're still in the path, make sure |
5028 | * we're dirty. Otherwise, one of the | |
5029 | * push_leaf functions must have already | |
5030 | * dirtied this buffer | |
5031 | */ | |
5032 | if (path->nodes[0] == leaf) | |
5033 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5034 | free_extent_buffer(leaf); |
be0e5c09 | 5035 | } |
d5719762 | 5036 | } else { |
5f39d397 | 5037 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5038 | } |
5039 | } | |
aa5d6bed | 5040 | return ret; |
be0e5c09 CM |
5041 | } |
5042 | ||
7bb86316 | 5043 | /* |
925baedd | 5044 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5045 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5046 | * returns < 0 on io errors. | |
d352ac68 CM |
5047 | * |
5048 | * This may release the path, and so you may lose any locks held at the | |
5049 | * time you call it. | |
7bb86316 | 5050 | */ |
16e7549f | 5051 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5052 | { |
925baedd CM |
5053 | struct btrfs_key key; |
5054 | struct btrfs_disk_key found_key; | |
5055 | int ret; | |
7bb86316 | 5056 | |
925baedd | 5057 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5058 | |
e8b0d724 | 5059 | if (key.offset > 0) { |
925baedd | 5060 | key.offset--; |
e8b0d724 | 5061 | } else if (key.type > 0) { |
925baedd | 5062 | key.type--; |
e8b0d724 FDBM |
5063 | key.offset = (u64)-1; |
5064 | } else if (key.objectid > 0) { | |
925baedd | 5065 | key.objectid--; |
e8b0d724 FDBM |
5066 | key.type = (u8)-1; |
5067 | key.offset = (u64)-1; | |
5068 | } else { | |
925baedd | 5069 | return 1; |
e8b0d724 | 5070 | } |
7bb86316 | 5071 | |
b3b4aa74 | 5072 | btrfs_release_path(path); |
925baedd CM |
5073 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5074 | if (ret < 0) | |
5075 | return ret; | |
5076 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5077 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5078 | /* |
5079 | * We might have had an item with the previous key in the tree right | |
5080 | * before we released our path. And after we released our path, that | |
5081 | * item might have been pushed to the first slot (0) of the leaf we | |
5082 | * were holding due to a tree balance. Alternatively, an item with the | |
5083 | * previous key can exist as the only element of a leaf (big fat item). | |
5084 | * Therefore account for these 2 cases, so that our callers (like | |
5085 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5086 | * the previous key we computed above. | |
5087 | */ | |
5088 | if (ret <= 0) | |
925baedd CM |
5089 | return 0; |
5090 | return 1; | |
7bb86316 CM |
5091 | } |
5092 | ||
3f157a2f CM |
5093 | /* |
5094 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5095 | * for nodes or leaves that are have a minimum transaction id. |
5096 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5097 | * |
5098 | * This does not cow, but it does stuff the starting key it finds back | |
5099 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5100 | * key and get a writable path. | |
5101 | * | |
3f157a2f CM |
5102 | * This honors path->lowest_level to prevent descent past a given level |
5103 | * of the tree. | |
5104 | * | |
d352ac68 CM |
5105 | * min_trans indicates the oldest transaction that you are interested |
5106 | * in walking through. Any nodes or leaves older than min_trans are | |
5107 | * skipped over (without reading them). | |
5108 | * | |
3f157a2f CM |
5109 | * returns zero if something useful was found, < 0 on error and 1 if there |
5110 | * was nothing in the tree that matched the search criteria. | |
5111 | */ | |
5112 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5113 | struct btrfs_path *path, |
3f157a2f CM |
5114 | u64 min_trans) |
5115 | { | |
2ff7e61e | 5116 | struct btrfs_fs_info *fs_info = root->fs_info; |
3f157a2f CM |
5117 | struct extent_buffer *cur; |
5118 | struct btrfs_key found_key; | |
5119 | int slot; | |
9652480b | 5120 | int sret; |
3f157a2f CM |
5121 | u32 nritems; |
5122 | int level; | |
5123 | int ret = 1; | |
f98de9b9 | 5124 | int keep_locks = path->keep_locks; |
3f157a2f | 5125 | |
f98de9b9 | 5126 | path->keep_locks = 1; |
3f157a2f | 5127 | again: |
bd681513 | 5128 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5129 | level = btrfs_header_level(cur); |
e02119d5 | 5130 | WARN_ON(path->nodes[level]); |
3f157a2f | 5131 | path->nodes[level] = cur; |
bd681513 | 5132 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5133 | |
5134 | if (btrfs_header_generation(cur) < min_trans) { | |
5135 | ret = 1; | |
5136 | goto out; | |
5137 | } | |
d397712b | 5138 | while (1) { |
3f157a2f CM |
5139 | nritems = btrfs_header_nritems(cur); |
5140 | level = btrfs_header_level(cur); | |
a74b35ec | 5141 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
3f157a2f | 5142 | |
323ac95b CM |
5143 | /* at the lowest level, we're done, setup the path and exit */ |
5144 | if (level == path->lowest_level) { | |
e02119d5 CM |
5145 | if (slot >= nritems) |
5146 | goto find_next_key; | |
3f157a2f CM |
5147 | ret = 0; |
5148 | path->slots[level] = slot; | |
5149 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5150 | goto out; | |
5151 | } | |
9652480b Y |
5152 | if (sret && slot > 0) |
5153 | slot--; | |
3f157a2f | 5154 | /* |
de78b51a ES |
5155 | * check this node pointer against the min_trans parameters. |
5156 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5157 | */ |
d397712b | 5158 | while (slot < nritems) { |
3f157a2f | 5159 | u64 gen; |
e02119d5 | 5160 | |
3f157a2f CM |
5161 | gen = btrfs_node_ptr_generation(cur, slot); |
5162 | if (gen < min_trans) { | |
5163 | slot++; | |
5164 | continue; | |
5165 | } | |
de78b51a | 5166 | break; |
3f157a2f | 5167 | } |
e02119d5 | 5168 | find_next_key: |
3f157a2f CM |
5169 | /* |
5170 | * we didn't find a candidate key in this node, walk forward | |
5171 | * and find another one | |
5172 | */ | |
5173 | if (slot >= nritems) { | |
e02119d5 | 5174 | path->slots[level] = slot; |
b4ce94de | 5175 | btrfs_set_path_blocking(path); |
e02119d5 | 5176 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5177 | min_trans); |
e02119d5 | 5178 | if (sret == 0) { |
b3b4aa74 | 5179 | btrfs_release_path(path); |
3f157a2f CM |
5180 | goto again; |
5181 | } else { | |
5182 | goto out; | |
5183 | } | |
5184 | } | |
5185 | /* save our key for returning back */ | |
5186 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5187 | path->slots[level] = slot; | |
5188 | if (level == path->lowest_level) { | |
5189 | ret = 0; | |
3f157a2f CM |
5190 | goto out; |
5191 | } | |
b4ce94de | 5192 | btrfs_set_path_blocking(path); |
2ff7e61e | 5193 | cur = read_node_slot(fs_info, cur, slot); |
fb770ae4 LB |
5194 | if (IS_ERR(cur)) { |
5195 | ret = PTR_ERR(cur); | |
5196 | goto out; | |
5197 | } | |
3f157a2f | 5198 | |
bd681513 | 5199 | btrfs_tree_read_lock(cur); |
b4ce94de | 5200 | |
bd681513 | 5201 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5202 | path->nodes[level - 1] = cur; |
f7c79f30 | 5203 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5204 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5205 | } |
5206 | out: | |
f98de9b9 FM |
5207 | path->keep_locks = keep_locks; |
5208 | if (ret == 0) { | |
5209 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5210 | btrfs_set_path_blocking(path); | |
3f157a2f | 5211 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5212 | } |
3f157a2f CM |
5213 | return ret; |
5214 | } | |
5215 | ||
2ff7e61e | 5216 | static int tree_move_down(struct btrfs_fs_info *fs_info, |
7069830a | 5217 | struct btrfs_path *path, |
ab6a43e1 | 5218 | int *level) |
7069830a | 5219 | { |
fb770ae4 LB |
5220 | struct extent_buffer *eb; |
5221 | ||
74dd17fb | 5222 | BUG_ON(*level == 0); |
2ff7e61e | 5223 | eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]); |
fb770ae4 LB |
5224 | if (IS_ERR(eb)) |
5225 | return PTR_ERR(eb); | |
5226 | ||
5227 | path->nodes[*level - 1] = eb; | |
7069830a AB |
5228 | path->slots[*level - 1] = 0; |
5229 | (*level)--; | |
fb770ae4 | 5230 | return 0; |
7069830a AB |
5231 | } |
5232 | ||
f1e30261 | 5233 | static int tree_move_next_or_upnext(struct btrfs_path *path, |
7069830a AB |
5234 | int *level, int root_level) |
5235 | { | |
5236 | int ret = 0; | |
5237 | int nritems; | |
5238 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5239 | ||
5240 | path->slots[*level]++; | |
5241 | ||
74dd17fb | 5242 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5243 | if (*level == root_level) |
5244 | return -1; | |
5245 | ||
5246 | /* move upnext */ | |
5247 | path->slots[*level] = 0; | |
5248 | free_extent_buffer(path->nodes[*level]); | |
5249 | path->nodes[*level] = NULL; | |
5250 | (*level)++; | |
5251 | path->slots[*level]++; | |
5252 | ||
5253 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5254 | ret = 1; | |
5255 | } | |
5256 | return ret; | |
5257 | } | |
5258 | ||
5259 | /* | |
5260 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5261 | * or down. | |
5262 | */ | |
2ff7e61e | 5263 | static int tree_advance(struct btrfs_fs_info *fs_info, |
7069830a AB |
5264 | struct btrfs_path *path, |
5265 | int *level, int root_level, | |
5266 | int allow_down, | |
5267 | struct btrfs_key *key) | |
5268 | { | |
5269 | int ret; | |
5270 | ||
5271 | if (*level == 0 || !allow_down) { | |
f1e30261 | 5272 | ret = tree_move_next_or_upnext(path, level, root_level); |
7069830a | 5273 | } else { |
ab6a43e1 | 5274 | ret = tree_move_down(fs_info, path, level); |
7069830a AB |
5275 | } |
5276 | if (ret >= 0) { | |
5277 | if (*level == 0) | |
5278 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5279 | path->slots[*level]); | |
5280 | else | |
5281 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5282 | path->slots[*level]); | |
5283 | } | |
5284 | return ret; | |
5285 | } | |
5286 | ||
2ff7e61e | 5287 | static int tree_compare_item(struct btrfs_path *left_path, |
7069830a AB |
5288 | struct btrfs_path *right_path, |
5289 | char *tmp_buf) | |
5290 | { | |
5291 | int cmp; | |
5292 | int len1, len2; | |
5293 | unsigned long off1, off2; | |
5294 | ||
5295 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5296 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5297 | if (len1 != len2) | |
5298 | return 1; | |
5299 | ||
5300 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5301 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5302 | right_path->slots[0]); | |
5303 | ||
5304 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5305 | ||
5306 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5307 | if (cmp) | |
5308 | return 1; | |
5309 | return 0; | |
5310 | } | |
5311 | ||
5312 | #define ADVANCE 1 | |
5313 | #define ADVANCE_ONLY_NEXT -1 | |
5314 | ||
5315 | /* | |
5316 | * This function compares two trees and calls the provided callback for | |
5317 | * every changed/new/deleted item it finds. | |
5318 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5319 | * the compare pretty fast on snapshotted subvolumes. | |
5320 | * | |
5321 | * This currently works on commit roots only. As commit roots are read only, | |
5322 | * we don't do any locking. The commit roots are protected with transactions. | |
5323 | * Transactions are ended and rejoined when a commit is tried in between. | |
5324 | * | |
5325 | * This function checks for modifications done to the trees while comparing. | |
5326 | * If it detects a change, it aborts immediately. | |
5327 | */ | |
5328 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5329 | struct btrfs_root *right_root, | |
5330 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5331 | { | |
0b246afa | 5332 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5333 | int ret; |
5334 | int cmp; | |
7069830a AB |
5335 | struct btrfs_path *left_path = NULL; |
5336 | struct btrfs_path *right_path = NULL; | |
5337 | struct btrfs_key left_key; | |
5338 | struct btrfs_key right_key; | |
5339 | char *tmp_buf = NULL; | |
5340 | int left_root_level; | |
5341 | int right_root_level; | |
5342 | int left_level; | |
5343 | int right_level; | |
5344 | int left_end_reached; | |
5345 | int right_end_reached; | |
5346 | int advance_left; | |
5347 | int advance_right; | |
5348 | u64 left_blockptr; | |
5349 | u64 right_blockptr; | |
6baa4293 FM |
5350 | u64 left_gen; |
5351 | u64 right_gen; | |
7069830a AB |
5352 | |
5353 | left_path = btrfs_alloc_path(); | |
5354 | if (!left_path) { | |
5355 | ret = -ENOMEM; | |
5356 | goto out; | |
5357 | } | |
5358 | right_path = btrfs_alloc_path(); | |
5359 | if (!right_path) { | |
5360 | ret = -ENOMEM; | |
5361 | goto out; | |
5362 | } | |
5363 | ||
752ade68 | 5364 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); |
7069830a | 5365 | if (!tmp_buf) { |
752ade68 MH |
5366 | ret = -ENOMEM; |
5367 | goto out; | |
7069830a AB |
5368 | } |
5369 | ||
5370 | left_path->search_commit_root = 1; | |
5371 | left_path->skip_locking = 1; | |
5372 | right_path->search_commit_root = 1; | |
5373 | right_path->skip_locking = 1; | |
5374 | ||
7069830a AB |
5375 | /* |
5376 | * Strategy: Go to the first items of both trees. Then do | |
5377 | * | |
5378 | * If both trees are at level 0 | |
5379 | * Compare keys of current items | |
5380 | * If left < right treat left item as new, advance left tree | |
5381 | * and repeat | |
5382 | * If left > right treat right item as deleted, advance right tree | |
5383 | * and repeat | |
5384 | * If left == right do deep compare of items, treat as changed if | |
5385 | * needed, advance both trees and repeat | |
5386 | * If both trees are at the same level but not at level 0 | |
5387 | * Compare keys of current nodes/leafs | |
5388 | * If left < right advance left tree and repeat | |
5389 | * If left > right advance right tree and repeat | |
5390 | * If left == right compare blockptrs of the next nodes/leafs | |
5391 | * If they match advance both trees but stay at the same level | |
5392 | * and repeat | |
5393 | * If they don't match advance both trees while allowing to go | |
5394 | * deeper and repeat | |
5395 | * If tree levels are different | |
5396 | * Advance the tree that needs it and repeat | |
5397 | * | |
5398 | * Advancing a tree means: | |
5399 | * If we are at level 0, try to go to the next slot. If that's not | |
5400 | * possible, go one level up and repeat. Stop when we found a level | |
5401 | * where we could go to the next slot. We may at this point be on a | |
5402 | * node or a leaf. | |
5403 | * | |
5404 | * If we are not at level 0 and not on shared tree blocks, go one | |
5405 | * level deeper. | |
5406 | * | |
5407 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5408 | * the right if possible or go up and right. | |
5409 | */ | |
5410 | ||
0b246afa | 5411 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5412 | left_level = btrfs_header_level(left_root->commit_root); |
5413 | left_root_level = left_level; | |
5414 | left_path->nodes[left_level] = left_root->commit_root; | |
5415 | extent_buffer_get(left_path->nodes[left_level]); | |
5416 | ||
5417 | right_level = btrfs_header_level(right_root->commit_root); | |
5418 | right_root_level = right_level; | |
5419 | right_path->nodes[right_level] = right_root->commit_root; | |
5420 | extent_buffer_get(right_path->nodes[right_level]); | |
0b246afa | 5421 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5422 | |
5423 | if (left_level == 0) | |
5424 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5425 | &left_key, left_path->slots[left_level]); | |
5426 | else | |
5427 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5428 | &left_key, left_path->slots[left_level]); | |
5429 | if (right_level == 0) | |
5430 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5431 | &right_key, right_path->slots[right_level]); | |
5432 | else | |
5433 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5434 | &right_key, right_path->slots[right_level]); | |
5435 | ||
5436 | left_end_reached = right_end_reached = 0; | |
5437 | advance_left = advance_right = 0; | |
5438 | ||
5439 | while (1) { | |
7069830a | 5440 | if (advance_left && !left_end_reached) { |
2ff7e61e | 5441 | ret = tree_advance(fs_info, left_path, &left_level, |
7069830a AB |
5442 | left_root_level, |
5443 | advance_left != ADVANCE_ONLY_NEXT, | |
5444 | &left_key); | |
fb770ae4 | 5445 | if (ret == -1) |
7069830a | 5446 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5447 | else if (ret < 0) |
5448 | goto out; | |
7069830a AB |
5449 | advance_left = 0; |
5450 | } | |
5451 | if (advance_right && !right_end_reached) { | |
2ff7e61e | 5452 | ret = tree_advance(fs_info, right_path, &right_level, |
7069830a AB |
5453 | right_root_level, |
5454 | advance_right != ADVANCE_ONLY_NEXT, | |
5455 | &right_key); | |
fb770ae4 | 5456 | if (ret == -1) |
7069830a | 5457 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5458 | else if (ret < 0) |
5459 | goto out; | |
7069830a AB |
5460 | advance_right = 0; |
5461 | } | |
5462 | ||
5463 | if (left_end_reached && right_end_reached) { | |
5464 | ret = 0; | |
5465 | goto out; | |
5466 | } else if (left_end_reached) { | |
5467 | if (right_level == 0) { | |
ee8c494f | 5468 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5469 | &right_key, |
5470 | BTRFS_COMPARE_TREE_DELETED, | |
5471 | ctx); | |
5472 | if (ret < 0) | |
5473 | goto out; | |
5474 | } | |
5475 | advance_right = ADVANCE; | |
5476 | continue; | |
5477 | } else if (right_end_reached) { | |
5478 | if (left_level == 0) { | |
ee8c494f | 5479 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5480 | &left_key, |
5481 | BTRFS_COMPARE_TREE_NEW, | |
5482 | ctx); | |
5483 | if (ret < 0) | |
5484 | goto out; | |
5485 | } | |
5486 | advance_left = ADVANCE; | |
5487 | continue; | |
5488 | } | |
5489 | ||
5490 | if (left_level == 0 && right_level == 0) { | |
5491 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5492 | if (cmp < 0) { | |
ee8c494f | 5493 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5494 | &left_key, |
5495 | BTRFS_COMPARE_TREE_NEW, | |
5496 | ctx); | |
5497 | if (ret < 0) | |
5498 | goto out; | |
5499 | advance_left = ADVANCE; | |
5500 | } else if (cmp > 0) { | |
ee8c494f | 5501 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5502 | &right_key, |
5503 | BTRFS_COMPARE_TREE_DELETED, | |
5504 | ctx); | |
5505 | if (ret < 0) | |
5506 | goto out; | |
5507 | advance_right = ADVANCE; | |
5508 | } else { | |
b99d9a6a | 5509 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5510 | |
74dd17fb | 5511 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
2ff7e61e JM |
5512 | ret = tree_compare_item(left_path, right_path, |
5513 | tmp_buf); | |
ba5e8f2e | 5514 | if (ret) |
b99d9a6a | 5515 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5516 | else |
b99d9a6a | 5517 | result = BTRFS_COMPARE_TREE_SAME; |
ee8c494f | 5518 | ret = changed_cb(left_path, right_path, |
b99d9a6a | 5519 | &left_key, result, ctx); |
ba5e8f2e JB |
5520 | if (ret < 0) |
5521 | goto out; | |
7069830a AB |
5522 | advance_left = ADVANCE; |
5523 | advance_right = ADVANCE; | |
5524 | } | |
5525 | } else if (left_level == right_level) { | |
5526 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5527 | if (cmp < 0) { | |
5528 | advance_left = ADVANCE; | |
5529 | } else if (cmp > 0) { | |
5530 | advance_right = ADVANCE; | |
5531 | } else { | |
5532 | left_blockptr = btrfs_node_blockptr( | |
5533 | left_path->nodes[left_level], | |
5534 | left_path->slots[left_level]); | |
5535 | right_blockptr = btrfs_node_blockptr( | |
5536 | right_path->nodes[right_level], | |
5537 | right_path->slots[right_level]); | |
6baa4293 FM |
5538 | left_gen = btrfs_node_ptr_generation( |
5539 | left_path->nodes[left_level], | |
5540 | left_path->slots[left_level]); | |
5541 | right_gen = btrfs_node_ptr_generation( | |
5542 | right_path->nodes[right_level], | |
5543 | right_path->slots[right_level]); | |
5544 | if (left_blockptr == right_blockptr && | |
5545 | left_gen == right_gen) { | |
7069830a AB |
5546 | /* |
5547 | * As we're on a shared block, don't | |
5548 | * allow to go deeper. | |
5549 | */ | |
5550 | advance_left = ADVANCE_ONLY_NEXT; | |
5551 | advance_right = ADVANCE_ONLY_NEXT; | |
5552 | } else { | |
5553 | advance_left = ADVANCE; | |
5554 | advance_right = ADVANCE; | |
5555 | } | |
5556 | } | |
5557 | } else if (left_level < right_level) { | |
5558 | advance_right = ADVANCE; | |
5559 | } else { | |
5560 | advance_left = ADVANCE; | |
5561 | } | |
5562 | } | |
5563 | ||
5564 | out: | |
5565 | btrfs_free_path(left_path); | |
5566 | btrfs_free_path(right_path); | |
8f282f71 | 5567 | kvfree(tmp_buf); |
7069830a AB |
5568 | return ret; |
5569 | } | |
5570 | ||
3f157a2f CM |
5571 | /* |
5572 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5573 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5574 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5575 | * |
5576 | * 0 is returned if another key is found, < 0 if there are any errors | |
5577 | * and 1 is returned if there are no higher keys in the tree | |
5578 | * | |
5579 | * path->keep_locks should be set to 1 on the search made before | |
5580 | * calling this function. | |
5581 | */ | |
e7a84565 | 5582 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5583 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5584 | { |
e7a84565 CM |
5585 | int slot; |
5586 | struct extent_buffer *c; | |
5587 | ||
934d375b | 5588 | WARN_ON(!path->keep_locks); |
d397712b | 5589 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5590 | if (!path->nodes[level]) |
5591 | return 1; | |
5592 | ||
5593 | slot = path->slots[level] + 1; | |
5594 | c = path->nodes[level]; | |
3f157a2f | 5595 | next: |
e7a84565 | 5596 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5597 | int ret; |
5598 | int orig_lowest; | |
5599 | struct btrfs_key cur_key; | |
5600 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5601 | !path->nodes[level + 1]) | |
e7a84565 | 5602 | return 1; |
33c66f43 YZ |
5603 | |
5604 | if (path->locks[level + 1]) { | |
5605 | level++; | |
5606 | continue; | |
5607 | } | |
5608 | ||
5609 | slot = btrfs_header_nritems(c) - 1; | |
5610 | if (level == 0) | |
5611 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5612 | else | |
5613 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5614 | ||
5615 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5616 | btrfs_release_path(path); |
33c66f43 YZ |
5617 | path->lowest_level = level; |
5618 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5619 | 0, 0); | |
5620 | path->lowest_level = orig_lowest; | |
5621 | if (ret < 0) | |
5622 | return ret; | |
5623 | ||
5624 | c = path->nodes[level]; | |
5625 | slot = path->slots[level]; | |
5626 | if (ret == 0) | |
5627 | slot++; | |
5628 | goto next; | |
e7a84565 | 5629 | } |
33c66f43 | 5630 | |
e7a84565 CM |
5631 | if (level == 0) |
5632 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5633 | else { |
3f157a2f CM |
5634 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5635 | ||
3f157a2f CM |
5636 | if (gen < min_trans) { |
5637 | slot++; | |
5638 | goto next; | |
5639 | } | |
e7a84565 | 5640 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5641 | } |
e7a84565 CM |
5642 | return 0; |
5643 | } | |
5644 | return 1; | |
5645 | } | |
5646 | ||
97571fd0 | 5647 | /* |
925baedd | 5648 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5649 | * returns 0 if it found something or 1 if there are no greater leaves. |
5650 | * returns < 0 on io errors. | |
97571fd0 | 5651 | */ |
234b63a0 | 5652 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5653 | { |
5654 | return btrfs_next_old_leaf(root, path, 0); | |
5655 | } | |
5656 | ||
5657 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5658 | u64 time_seq) | |
d97e63b6 CM |
5659 | { |
5660 | int slot; | |
8e73f275 | 5661 | int level; |
5f39d397 | 5662 | struct extent_buffer *c; |
8e73f275 | 5663 | struct extent_buffer *next; |
925baedd CM |
5664 | struct btrfs_key key; |
5665 | u32 nritems; | |
5666 | int ret; | |
8e73f275 | 5667 | int old_spinning = path->leave_spinning; |
bd681513 | 5668 | int next_rw_lock = 0; |
925baedd CM |
5669 | |
5670 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5671 | if (nritems == 0) |
925baedd | 5672 | return 1; |
925baedd | 5673 | |
8e73f275 CM |
5674 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5675 | again: | |
5676 | level = 1; | |
5677 | next = NULL; | |
bd681513 | 5678 | next_rw_lock = 0; |
b3b4aa74 | 5679 | btrfs_release_path(path); |
8e73f275 | 5680 | |
a2135011 | 5681 | path->keep_locks = 1; |
31533fb2 | 5682 | path->leave_spinning = 1; |
8e73f275 | 5683 | |
3d7806ec JS |
5684 | if (time_seq) |
5685 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5686 | else | |
5687 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5688 | path->keep_locks = 0; |
5689 | ||
5690 | if (ret < 0) | |
5691 | return ret; | |
5692 | ||
a2135011 | 5693 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5694 | /* |
5695 | * by releasing the path above we dropped all our locks. A balance | |
5696 | * could have added more items next to the key that used to be | |
5697 | * at the very end of the block. So, check again here and | |
5698 | * advance the path if there are now more items available. | |
5699 | */ | |
a2135011 | 5700 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5701 | if (ret == 0) |
5702 | path->slots[0]++; | |
8e73f275 | 5703 | ret = 0; |
925baedd CM |
5704 | goto done; |
5705 | } | |
0b43e04f LB |
5706 | /* |
5707 | * So the above check misses one case: | |
5708 | * - after releasing the path above, someone has removed the item that | |
5709 | * used to be at the very end of the block, and balance between leafs | |
5710 | * gets another one with bigger key.offset to replace it. | |
5711 | * | |
5712 | * This one should be returned as well, or we can get leaf corruption | |
5713 | * later(esp. in __btrfs_drop_extents()). | |
5714 | * | |
5715 | * And a bit more explanation about this check, | |
5716 | * with ret > 0, the key isn't found, the path points to the slot | |
5717 | * where it should be inserted, so the path->slots[0] item must be the | |
5718 | * bigger one. | |
5719 | */ | |
5720 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5721 | ret = 0; | |
5722 | goto done; | |
5723 | } | |
d97e63b6 | 5724 | |
d397712b | 5725 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5726 | if (!path->nodes[level]) { |
5727 | ret = 1; | |
5728 | goto done; | |
5729 | } | |
5f39d397 | 5730 | |
d97e63b6 CM |
5731 | slot = path->slots[level] + 1; |
5732 | c = path->nodes[level]; | |
5f39d397 | 5733 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5734 | level++; |
8e73f275 CM |
5735 | if (level == BTRFS_MAX_LEVEL) { |
5736 | ret = 1; | |
5737 | goto done; | |
5738 | } | |
d97e63b6 CM |
5739 | continue; |
5740 | } | |
5f39d397 | 5741 | |
925baedd | 5742 | if (next) { |
bd681513 | 5743 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5744 | free_extent_buffer(next); |
925baedd | 5745 | } |
5f39d397 | 5746 | |
8e73f275 | 5747 | next = c; |
bd681513 | 5748 | next_rw_lock = path->locks[level]; |
d07b8528 | 5749 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5750 | slot, &key); |
8e73f275 CM |
5751 | if (ret == -EAGAIN) |
5752 | goto again; | |
5f39d397 | 5753 | |
76a05b35 | 5754 | if (ret < 0) { |
b3b4aa74 | 5755 | btrfs_release_path(path); |
76a05b35 CM |
5756 | goto done; |
5757 | } | |
5758 | ||
5cd57b2c | 5759 | if (!path->skip_locking) { |
bd681513 | 5760 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5761 | if (!ret && time_seq) { |
5762 | /* | |
5763 | * If we don't get the lock, we may be racing | |
5764 | * with push_leaf_left, holding that lock while | |
5765 | * itself waiting for the leaf we've currently | |
5766 | * locked. To solve this situation, we give up | |
5767 | * on our lock and cycle. | |
5768 | */ | |
cf538830 | 5769 | free_extent_buffer(next); |
d42244a0 JS |
5770 | btrfs_release_path(path); |
5771 | cond_resched(); | |
5772 | goto again; | |
5773 | } | |
8e73f275 CM |
5774 | if (!ret) { |
5775 | btrfs_set_path_blocking(path); | |
bd681513 | 5776 | btrfs_tree_read_lock(next); |
31533fb2 | 5777 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5778 | BTRFS_READ_LOCK); |
8e73f275 | 5779 | } |
31533fb2 | 5780 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5781 | } |
d97e63b6 CM |
5782 | break; |
5783 | } | |
5784 | path->slots[level] = slot; | |
d397712b | 5785 | while (1) { |
d97e63b6 CM |
5786 | level--; |
5787 | c = path->nodes[level]; | |
925baedd | 5788 | if (path->locks[level]) |
bd681513 | 5789 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5790 | |
5f39d397 | 5791 | free_extent_buffer(c); |
d97e63b6 CM |
5792 | path->nodes[level] = next; |
5793 | path->slots[level] = 0; | |
a74a4b97 | 5794 | if (!path->skip_locking) |
bd681513 | 5795 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5796 | if (!level) |
5797 | break; | |
b4ce94de | 5798 | |
d07b8528 | 5799 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5800 | 0, &key); |
8e73f275 CM |
5801 | if (ret == -EAGAIN) |
5802 | goto again; | |
5803 | ||
76a05b35 | 5804 | if (ret < 0) { |
b3b4aa74 | 5805 | btrfs_release_path(path); |
76a05b35 CM |
5806 | goto done; |
5807 | } | |
5808 | ||
5cd57b2c | 5809 | if (!path->skip_locking) { |
bd681513 | 5810 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5811 | if (!ret) { |
5812 | btrfs_set_path_blocking(path); | |
bd681513 | 5813 | btrfs_tree_read_lock(next); |
31533fb2 | 5814 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5815 | BTRFS_READ_LOCK); |
5816 | } | |
31533fb2 | 5817 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5818 | } |
d97e63b6 | 5819 | } |
8e73f275 | 5820 | ret = 0; |
925baedd | 5821 | done: |
f7c79f30 | 5822 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5823 | path->leave_spinning = old_spinning; |
5824 | if (!old_spinning) | |
5825 | btrfs_set_path_blocking(path); | |
5826 | ||
5827 | return ret; | |
d97e63b6 | 5828 | } |
0b86a832 | 5829 | |
3f157a2f CM |
5830 | /* |
5831 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5832 | * searching until it gets past min_objectid or finds an item of 'type' | |
5833 | * | |
5834 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5835 | */ | |
0b86a832 CM |
5836 | int btrfs_previous_item(struct btrfs_root *root, |
5837 | struct btrfs_path *path, u64 min_objectid, | |
5838 | int type) | |
5839 | { | |
5840 | struct btrfs_key found_key; | |
5841 | struct extent_buffer *leaf; | |
e02119d5 | 5842 | u32 nritems; |
0b86a832 CM |
5843 | int ret; |
5844 | ||
d397712b | 5845 | while (1) { |
0b86a832 | 5846 | if (path->slots[0] == 0) { |
b4ce94de | 5847 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5848 | ret = btrfs_prev_leaf(root, path); |
5849 | if (ret != 0) | |
5850 | return ret; | |
5851 | } else { | |
5852 | path->slots[0]--; | |
5853 | } | |
5854 | leaf = path->nodes[0]; | |
e02119d5 CM |
5855 | nritems = btrfs_header_nritems(leaf); |
5856 | if (nritems == 0) | |
5857 | return 1; | |
5858 | if (path->slots[0] == nritems) | |
5859 | path->slots[0]--; | |
5860 | ||
0b86a832 | 5861 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5862 | if (found_key.objectid < min_objectid) |
5863 | break; | |
0a4eefbb YZ |
5864 | if (found_key.type == type) |
5865 | return 0; | |
e02119d5 CM |
5866 | if (found_key.objectid == min_objectid && |
5867 | found_key.type < type) | |
5868 | break; | |
0b86a832 CM |
5869 | } |
5870 | return 1; | |
5871 | } | |
ade2e0b3 WS |
5872 | |
5873 | /* | |
5874 | * search in extent tree to find a previous Metadata/Data extent item with | |
5875 | * min objecitd. | |
5876 | * | |
5877 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5878 | */ | |
5879 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5880 | struct btrfs_path *path, u64 min_objectid) | |
5881 | { | |
5882 | struct btrfs_key found_key; | |
5883 | struct extent_buffer *leaf; | |
5884 | u32 nritems; | |
5885 | int ret; | |
5886 | ||
5887 | while (1) { | |
5888 | if (path->slots[0] == 0) { | |
5889 | btrfs_set_path_blocking(path); | |
5890 | ret = btrfs_prev_leaf(root, path); | |
5891 | if (ret != 0) | |
5892 | return ret; | |
5893 | } else { | |
5894 | path->slots[0]--; | |
5895 | } | |
5896 | leaf = path->nodes[0]; | |
5897 | nritems = btrfs_header_nritems(leaf); | |
5898 | if (nritems == 0) | |
5899 | return 1; | |
5900 | if (path->slots[0] == nritems) | |
5901 | path->slots[0]--; | |
5902 | ||
5903 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5904 | if (found_key.objectid < min_objectid) | |
5905 | break; | |
5906 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5907 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5908 | return 0; | |
5909 | if (found_key.objectid == min_objectid && | |
5910 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5911 | break; | |
5912 | } | |
5913 | return 1; | |
5914 | } |