]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 | 2 | /* |
d352ac68 | 3 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
4 | */ |
5 | ||
a6b6e75e | 6 | #include <linux/sched.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
bd989ba3 | 8 | #include <linux/rbtree.h> |
adf02123 | 9 | #include <linux/mm.h> |
eb60ceac CM |
10 | #include "ctree.h" |
11 | #include "disk-io.h" | |
7f5c1516 | 12 | #include "transaction.h" |
5f39d397 | 13 | #include "print-tree.h" |
925baedd | 14 | #include "locking.h" |
9a8dd150 | 15 | |
e089f05c CM |
16 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
17 | *root, struct btrfs_path *path, int level); | |
310712b2 OS |
18 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
19 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
20 | int data_size, int extend); | |
5f39d397 | 21 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
22 | struct btrfs_fs_info *fs_info, |
23 | struct extent_buffer *dst, | |
971a1f66 | 24 | struct extent_buffer *src, int empty); |
5f39d397 | 25 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 26 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
27 | struct extent_buffer *dst_buf, |
28 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
29 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
30 | int level, int slot); | |
d97e63b6 | 31 | |
df24a2b9 | 32 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 33 | { |
e2c89907 | 34 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
35 | } |
36 | ||
b4ce94de CM |
37 | /* |
38 | * set all locked nodes in the path to blocking locks. This should | |
39 | * be done before scheduling | |
40 | */ | |
41 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
42 | { | |
43 | int i; | |
44 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
45 | if (!p->nodes[i] || !p->locks[i]) |
46 | continue; | |
47 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
48 | if (p->locks[i] == BTRFS_READ_LOCK) | |
49 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
50 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
51 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
52 | } |
53 | } | |
54 | ||
55 | /* | |
56 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
57 | * |
58 | * held is used to keep lockdep happy, when lockdep is enabled | |
59 | * we set held to a blocking lock before we go around and | |
60 | * retake all the spinlocks in the path. You can safely use NULL | |
61 | * for held | |
b4ce94de | 62 | */ |
4008c04a | 63 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 64 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
65 | { |
66 | int i; | |
4008c04a | 67 | |
bd681513 CM |
68 | if (held) { |
69 | btrfs_set_lock_blocking_rw(held, held_rw); | |
70 | if (held_rw == BTRFS_WRITE_LOCK) | |
71 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
72 | else if (held_rw == BTRFS_READ_LOCK) | |
73 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
74 | } | |
4008c04a | 75 | btrfs_set_path_blocking(p); |
4008c04a CM |
76 | |
77 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
78 | if (p->nodes[i] && p->locks[i]) { |
79 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
80 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
81 | p->locks[i] = BTRFS_WRITE_LOCK; | |
82 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
83 | p->locks[i] = BTRFS_READ_LOCK; | |
84 | } | |
b4ce94de | 85 | } |
4008c04a | 86 | |
4008c04a | 87 | if (held) |
bd681513 | 88 | btrfs_clear_lock_blocking_rw(held, held_rw); |
b4ce94de CM |
89 | } |
90 | ||
d352ac68 | 91 | /* this also releases the path */ |
df24a2b9 | 92 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 93 | { |
ff175d57 JJ |
94 | if (!p) |
95 | return; | |
b3b4aa74 | 96 | btrfs_release_path(p); |
df24a2b9 | 97 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
98 | } |
99 | ||
d352ac68 CM |
100 | /* |
101 | * path release drops references on the extent buffers in the path | |
102 | * and it drops any locks held by this path | |
103 | * | |
104 | * It is safe to call this on paths that no locks or extent buffers held. | |
105 | */ | |
b3b4aa74 | 106 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
107 | { |
108 | int i; | |
a2135011 | 109 | |
234b63a0 | 110 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 111 | p->slots[i] = 0; |
eb60ceac | 112 | if (!p->nodes[i]) |
925baedd CM |
113 | continue; |
114 | if (p->locks[i]) { | |
bd681513 | 115 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
116 | p->locks[i] = 0; |
117 | } | |
5f39d397 | 118 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 119 | p->nodes[i] = NULL; |
eb60ceac CM |
120 | } |
121 | } | |
122 | ||
d352ac68 CM |
123 | /* |
124 | * safely gets a reference on the root node of a tree. A lock | |
125 | * is not taken, so a concurrent writer may put a different node | |
126 | * at the root of the tree. See btrfs_lock_root_node for the | |
127 | * looping required. | |
128 | * | |
129 | * The extent buffer returned by this has a reference taken, so | |
130 | * it won't disappear. It may stop being the root of the tree | |
131 | * at any time because there are no locks held. | |
132 | */ | |
925baedd CM |
133 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
134 | { | |
135 | struct extent_buffer *eb; | |
240f62c8 | 136 | |
3083ee2e JB |
137 | while (1) { |
138 | rcu_read_lock(); | |
139 | eb = rcu_dereference(root->node); | |
140 | ||
141 | /* | |
142 | * RCU really hurts here, we could free up the root node because | |
01327610 | 143 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
144 | * the inc_not_zero dance and if it doesn't work then |
145 | * synchronize_rcu and try again. | |
146 | */ | |
147 | if (atomic_inc_not_zero(&eb->refs)) { | |
148 | rcu_read_unlock(); | |
149 | break; | |
150 | } | |
151 | rcu_read_unlock(); | |
152 | synchronize_rcu(); | |
153 | } | |
925baedd CM |
154 | return eb; |
155 | } | |
156 | ||
d352ac68 CM |
157 | /* loop around taking references on and locking the root node of the |
158 | * tree until you end up with a lock on the root. A locked buffer | |
159 | * is returned, with a reference held. | |
160 | */ | |
925baedd CM |
161 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
162 | { | |
163 | struct extent_buffer *eb; | |
164 | ||
d397712b | 165 | while (1) { |
925baedd CM |
166 | eb = btrfs_root_node(root); |
167 | btrfs_tree_lock(eb); | |
240f62c8 | 168 | if (eb == root->node) |
925baedd | 169 | break; |
925baedd CM |
170 | btrfs_tree_unlock(eb); |
171 | free_extent_buffer(eb); | |
172 | } | |
173 | return eb; | |
174 | } | |
175 | ||
bd681513 CM |
176 | /* loop around taking references on and locking the root node of the |
177 | * tree until you end up with a lock on the root. A locked buffer | |
178 | * is returned, with a reference held. | |
179 | */ | |
84f7d8e6 | 180 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
181 | { |
182 | struct extent_buffer *eb; | |
183 | ||
184 | while (1) { | |
185 | eb = btrfs_root_node(root); | |
186 | btrfs_tree_read_lock(eb); | |
187 | if (eb == root->node) | |
188 | break; | |
189 | btrfs_tree_read_unlock(eb); | |
190 | free_extent_buffer(eb); | |
191 | } | |
192 | return eb; | |
193 | } | |
194 | ||
d352ac68 CM |
195 | /* cowonly root (everything not a reference counted cow subvolume), just get |
196 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
197 | * get properly updated on disk. | |
198 | */ | |
0b86a832 CM |
199 | static void add_root_to_dirty_list(struct btrfs_root *root) |
200 | { | |
0b246afa JM |
201 | struct btrfs_fs_info *fs_info = root->fs_info; |
202 | ||
e7070be1 JB |
203 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
204 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
205 | return; | |
206 | ||
0b246afa | 207 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
208 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
209 | /* Want the extent tree to be the last on the list */ | |
210 | if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
211 | list_move_tail(&root->dirty_list, | |
0b246afa | 212 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
213 | else |
214 | list_move(&root->dirty_list, | |
0b246afa | 215 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 216 | } |
0b246afa | 217 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
218 | } |
219 | ||
d352ac68 CM |
220 | /* |
221 | * used by snapshot creation to make a copy of a root for a tree with | |
222 | * a given objectid. The buffer with the new root node is returned in | |
223 | * cow_ret, and this func returns zero on success or a negative error code. | |
224 | */ | |
be20aa9d CM |
225 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
226 | struct btrfs_root *root, | |
227 | struct extent_buffer *buf, | |
228 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
229 | { | |
0b246afa | 230 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 231 | struct extent_buffer *cow; |
be20aa9d CM |
232 | int ret = 0; |
233 | int level; | |
5d4f98a2 | 234 | struct btrfs_disk_key disk_key; |
be20aa9d | 235 | |
27cdeb70 | 236 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 237 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
238 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
239 | trans->transid != root->last_trans); | |
be20aa9d CM |
240 | |
241 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
242 | if (level == 0) |
243 | btrfs_item_key(buf, &disk_key, 0); | |
244 | else | |
245 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 246 | |
4d75f8a9 DS |
247 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
248 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 249 | if (IS_ERR(cow)) |
be20aa9d CM |
250 | return PTR_ERR(cow); |
251 | ||
58e8012c | 252 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
253 | btrfs_set_header_bytenr(cow, cow->start); |
254 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
255 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
256 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
257 | BTRFS_HEADER_FLAG_RELOC); | |
258 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
259 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
260 | else | |
261 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 262 | |
0b246afa | 263 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 264 | |
be20aa9d | 265 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 266 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 267 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 268 | else |
e339a6b0 | 269 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 270 | |
be20aa9d CM |
271 | if (ret) |
272 | return ret; | |
273 | ||
274 | btrfs_mark_buffer_dirty(cow); | |
275 | *cow_ret = cow; | |
276 | return 0; | |
277 | } | |
278 | ||
bd989ba3 JS |
279 | enum mod_log_op { |
280 | MOD_LOG_KEY_REPLACE, | |
281 | MOD_LOG_KEY_ADD, | |
282 | MOD_LOG_KEY_REMOVE, | |
283 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
284 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
285 | MOD_LOG_MOVE_KEYS, | |
286 | MOD_LOG_ROOT_REPLACE, | |
287 | }; | |
288 | ||
bd989ba3 JS |
289 | struct tree_mod_root { |
290 | u64 logical; | |
291 | u8 level; | |
292 | }; | |
293 | ||
294 | struct tree_mod_elem { | |
295 | struct rb_node node; | |
298cfd36 | 296 | u64 logical; |
097b8a7c | 297 | u64 seq; |
bd989ba3 JS |
298 | enum mod_log_op op; |
299 | ||
300 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
301 | int slot; | |
302 | ||
303 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
304 | u64 generation; | |
305 | ||
306 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
307 | struct btrfs_disk_key key; | |
308 | u64 blockptr; | |
309 | ||
310 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
311 | struct { |
312 | int dst_slot; | |
313 | int nr_items; | |
314 | } move; | |
bd989ba3 JS |
315 | |
316 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
317 | struct tree_mod_root old_root; | |
318 | }; | |
319 | ||
fc36ed7e | 320 | /* |
fcebe456 | 321 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 322 | */ |
fcebe456 | 323 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
324 | { |
325 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
326 | } | |
327 | ||
097b8a7c JS |
328 | /* |
329 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
330 | * passed does not already have a sequence number set. So when a caller expects | |
331 | * to record tree modifications, it should ensure to set elem->seq to zero | |
332 | * before calling btrfs_get_tree_mod_seq. | |
333 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
334 | * blocker was added. | |
335 | */ | |
336 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
337 | struct seq_list *elem) | |
bd989ba3 | 338 | { |
b1a09f1e | 339 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 340 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 341 | if (!elem->seq) { |
fcebe456 | 342 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
343 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
344 | } | |
bd989ba3 | 345 | spin_unlock(&fs_info->tree_mod_seq_lock); |
b1a09f1e | 346 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 347 | |
fcebe456 | 348 | return elem->seq; |
bd989ba3 JS |
349 | } |
350 | ||
351 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
352 | struct seq_list *elem) | |
353 | { | |
354 | struct rb_root *tm_root; | |
355 | struct rb_node *node; | |
356 | struct rb_node *next; | |
357 | struct seq_list *cur_elem; | |
358 | struct tree_mod_elem *tm; | |
359 | u64 min_seq = (u64)-1; | |
360 | u64 seq_putting = elem->seq; | |
361 | ||
362 | if (!seq_putting) | |
363 | return; | |
364 | ||
bd989ba3 JS |
365 | spin_lock(&fs_info->tree_mod_seq_lock); |
366 | list_del(&elem->list); | |
097b8a7c | 367 | elem->seq = 0; |
bd989ba3 JS |
368 | |
369 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 370 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
371 | if (seq_putting > cur_elem->seq) { |
372 | /* | |
373 | * blocker with lower sequence number exists, we | |
374 | * cannot remove anything from the log | |
375 | */ | |
097b8a7c JS |
376 | spin_unlock(&fs_info->tree_mod_seq_lock); |
377 | return; | |
bd989ba3 JS |
378 | } |
379 | min_seq = cur_elem->seq; | |
380 | } | |
381 | } | |
097b8a7c JS |
382 | spin_unlock(&fs_info->tree_mod_seq_lock); |
383 | ||
bd989ba3 JS |
384 | /* |
385 | * anything that's lower than the lowest existing (read: blocked) | |
386 | * sequence number can be removed from the tree. | |
387 | */ | |
b1a09f1e | 388 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
389 | tm_root = &fs_info->tree_mod_log; |
390 | for (node = rb_first(tm_root); node; node = next) { | |
391 | next = rb_next(node); | |
6b4df8b6 | 392 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 393 | if (tm->seq > min_seq) |
bd989ba3 JS |
394 | continue; |
395 | rb_erase(node, tm_root); | |
bd989ba3 JS |
396 | kfree(tm); |
397 | } | |
b1a09f1e | 398 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
399 | } |
400 | ||
401 | /* | |
402 | * key order of the log: | |
298cfd36 | 403 | * node/leaf start address -> sequence |
bd989ba3 | 404 | * |
298cfd36 CR |
405 | * The 'start address' is the logical address of the *new* root node |
406 | * for root replace operations, or the logical address of the affected | |
407 | * block for all other operations. | |
5de865ee | 408 | * |
b1a09f1e | 409 | * Note: must be called with write lock for fs_info::tree_mod_log_lock. |
bd989ba3 JS |
410 | */ |
411 | static noinline int | |
412 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
413 | { | |
414 | struct rb_root *tm_root; | |
415 | struct rb_node **new; | |
416 | struct rb_node *parent = NULL; | |
417 | struct tree_mod_elem *cur; | |
c8cc6341 | 418 | |
fcebe456 | 419 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 420 | |
bd989ba3 JS |
421 | tm_root = &fs_info->tree_mod_log; |
422 | new = &tm_root->rb_node; | |
423 | while (*new) { | |
6b4df8b6 | 424 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 425 | parent = *new; |
298cfd36 | 426 | if (cur->logical < tm->logical) |
bd989ba3 | 427 | new = &((*new)->rb_left); |
298cfd36 | 428 | else if (cur->logical > tm->logical) |
bd989ba3 | 429 | new = &((*new)->rb_right); |
097b8a7c | 430 | else if (cur->seq < tm->seq) |
bd989ba3 | 431 | new = &((*new)->rb_left); |
097b8a7c | 432 | else if (cur->seq > tm->seq) |
bd989ba3 | 433 | new = &((*new)->rb_right); |
5de865ee FDBM |
434 | else |
435 | return -EEXIST; | |
bd989ba3 JS |
436 | } |
437 | ||
438 | rb_link_node(&tm->node, parent, new); | |
439 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 440 | return 0; |
bd989ba3 JS |
441 | } |
442 | ||
097b8a7c JS |
443 | /* |
444 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
445 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
446 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 447 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 448 | */ |
e9b7fd4d JS |
449 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
450 | struct extent_buffer *eb) { | |
451 | smp_mb(); | |
452 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
453 | return 1; | |
097b8a7c JS |
454 | if (eb && btrfs_header_level(eb) == 0) |
455 | return 1; | |
5de865ee | 456 | |
b1a09f1e | 457 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 458 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 459 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
460 | return 1; |
461 | } | |
462 | ||
e9b7fd4d JS |
463 | return 0; |
464 | } | |
465 | ||
5de865ee FDBM |
466 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
467 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
468 | struct extent_buffer *eb) | |
469 | { | |
470 | smp_mb(); | |
471 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
472 | return 0; | |
473 | if (eb && btrfs_header_level(eb) == 0) | |
474 | return 0; | |
475 | ||
476 | return 1; | |
477 | } | |
478 | ||
479 | static struct tree_mod_elem * | |
480 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
481 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 482 | { |
097b8a7c | 483 | struct tree_mod_elem *tm; |
bd989ba3 | 484 | |
c8cc6341 JB |
485 | tm = kzalloc(sizeof(*tm), flags); |
486 | if (!tm) | |
5de865ee | 487 | return NULL; |
bd989ba3 | 488 | |
298cfd36 | 489 | tm->logical = eb->start; |
bd989ba3 JS |
490 | if (op != MOD_LOG_KEY_ADD) { |
491 | btrfs_node_key(eb, &tm->key, slot); | |
492 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
493 | } | |
494 | tm->op = op; | |
495 | tm->slot = slot; | |
496 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 497 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 498 | |
5de865ee | 499 | return tm; |
097b8a7c JS |
500 | } |
501 | ||
e09c2efe DS |
502 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
503 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 504 | { |
5de865ee FDBM |
505 | struct tree_mod_elem *tm; |
506 | int ret; | |
507 | ||
e09c2efe | 508 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
509 | return 0; |
510 | ||
511 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
512 | if (!tm) | |
513 | return -ENOMEM; | |
514 | ||
e09c2efe | 515 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 516 | kfree(tm); |
097b8a7c | 517 | return 0; |
5de865ee FDBM |
518 | } |
519 | ||
e09c2efe | 520 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 521 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
522 | if (ret) |
523 | kfree(tm); | |
097b8a7c | 524 | |
5de865ee | 525 | return ret; |
097b8a7c JS |
526 | } |
527 | ||
6074d45f DS |
528 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
529 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 530 | { |
5de865ee FDBM |
531 | struct tree_mod_elem *tm = NULL; |
532 | struct tree_mod_elem **tm_list = NULL; | |
533 | int ret = 0; | |
bd989ba3 | 534 | int i; |
5de865ee | 535 | int locked = 0; |
bd989ba3 | 536 | |
6074d45f | 537 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 538 | return 0; |
bd989ba3 | 539 | |
176ef8f5 | 540 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
541 | if (!tm_list) |
542 | return -ENOMEM; | |
543 | ||
176ef8f5 | 544 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
545 | if (!tm) { |
546 | ret = -ENOMEM; | |
547 | goto free_tms; | |
548 | } | |
549 | ||
298cfd36 | 550 | tm->logical = eb->start; |
5de865ee FDBM |
551 | tm->slot = src_slot; |
552 | tm->move.dst_slot = dst_slot; | |
553 | tm->move.nr_items = nr_items; | |
554 | tm->op = MOD_LOG_MOVE_KEYS; | |
555 | ||
556 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
557 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 558 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
559 | if (!tm_list[i]) { |
560 | ret = -ENOMEM; | |
561 | goto free_tms; | |
562 | } | |
563 | } | |
564 | ||
6074d45f | 565 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
566 | goto free_tms; |
567 | locked = 1; | |
568 | ||
01763a2e JS |
569 | /* |
570 | * When we override something during the move, we log these removals. | |
571 | * This can only happen when we move towards the beginning of the | |
572 | * buffer, i.e. dst_slot < src_slot. | |
573 | */ | |
bd989ba3 | 574 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 575 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
576 | if (ret) |
577 | goto free_tms; | |
bd989ba3 JS |
578 | } |
579 | ||
6074d45f | 580 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
581 | if (ret) |
582 | goto free_tms; | |
b1a09f1e | 583 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 584 | kfree(tm_list); |
f395694c | 585 | |
5de865ee FDBM |
586 | return 0; |
587 | free_tms: | |
588 | for (i = 0; i < nr_items; i++) { | |
589 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 590 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
591 | kfree(tm_list[i]); |
592 | } | |
593 | if (locked) | |
b1a09f1e | 594 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
595 | kfree(tm_list); |
596 | kfree(tm); | |
bd989ba3 | 597 | |
5de865ee | 598 | return ret; |
bd989ba3 JS |
599 | } |
600 | ||
5de865ee FDBM |
601 | static inline int |
602 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
603 | struct tree_mod_elem **tm_list, | |
604 | int nritems) | |
097b8a7c | 605 | { |
5de865ee | 606 | int i, j; |
097b8a7c JS |
607 | int ret; |
608 | ||
097b8a7c | 609 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
610 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
611 | if (ret) { | |
612 | for (j = nritems - 1; j > i; j--) | |
613 | rb_erase(&tm_list[j]->node, | |
614 | &fs_info->tree_mod_log); | |
615 | return ret; | |
616 | } | |
097b8a7c | 617 | } |
5de865ee FDBM |
618 | |
619 | return 0; | |
097b8a7c JS |
620 | } |
621 | ||
95b757c1 DS |
622 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
623 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 624 | { |
95b757c1 | 625 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
626 | struct tree_mod_elem *tm = NULL; |
627 | struct tree_mod_elem **tm_list = NULL; | |
628 | int nritems = 0; | |
629 | int ret = 0; | |
630 | int i; | |
bd989ba3 | 631 | |
5de865ee | 632 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
633 | return 0; |
634 | ||
5de865ee FDBM |
635 | if (log_removal && btrfs_header_level(old_root) > 0) { |
636 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 637 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 638 | GFP_NOFS); |
5de865ee FDBM |
639 | if (!tm_list) { |
640 | ret = -ENOMEM; | |
641 | goto free_tms; | |
642 | } | |
643 | for (i = 0; i < nritems; i++) { | |
644 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 645 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
646 | if (!tm_list[i]) { |
647 | ret = -ENOMEM; | |
648 | goto free_tms; | |
649 | } | |
650 | } | |
651 | } | |
d9abbf1c | 652 | |
bcc8e07f | 653 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
654 | if (!tm) { |
655 | ret = -ENOMEM; | |
656 | goto free_tms; | |
657 | } | |
bd989ba3 | 658 | |
298cfd36 | 659 | tm->logical = new_root->start; |
bd989ba3 JS |
660 | tm->old_root.logical = old_root->start; |
661 | tm->old_root.level = btrfs_header_level(old_root); | |
662 | tm->generation = btrfs_header_generation(old_root); | |
663 | tm->op = MOD_LOG_ROOT_REPLACE; | |
664 | ||
5de865ee FDBM |
665 | if (tree_mod_dont_log(fs_info, NULL)) |
666 | goto free_tms; | |
667 | ||
668 | if (tm_list) | |
669 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
670 | if (!ret) | |
671 | ret = __tree_mod_log_insert(fs_info, tm); | |
672 | ||
b1a09f1e | 673 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
674 | if (ret) |
675 | goto free_tms; | |
676 | kfree(tm_list); | |
677 | ||
678 | return ret; | |
679 | ||
680 | free_tms: | |
681 | if (tm_list) { | |
682 | for (i = 0; i < nritems; i++) | |
683 | kfree(tm_list[i]); | |
684 | kfree(tm_list); | |
685 | } | |
686 | kfree(tm); | |
687 | ||
688 | return ret; | |
bd989ba3 JS |
689 | } |
690 | ||
691 | static struct tree_mod_elem * | |
692 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
693 | int smallest) | |
694 | { | |
695 | struct rb_root *tm_root; | |
696 | struct rb_node *node; | |
697 | struct tree_mod_elem *cur = NULL; | |
698 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 699 | |
b1a09f1e | 700 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
701 | tm_root = &fs_info->tree_mod_log; |
702 | node = tm_root->rb_node; | |
703 | while (node) { | |
6b4df8b6 | 704 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 705 | if (cur->logical < start) { |
bd989ba3 | 706 | node = node->rb_left; |
298cfd36 | 707 | } else if (cur->logical > start) { |
bd989ba3 | 708 | node = node->rb_right; |
097b8a7c | 709 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
710 | node = node->rb_left; |
711 | } else if (!smallest) { | |
712 | /* we want the node with the highest seq */ | |
713 | if (found) | |
097b8a7c | 714 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
715 | found = cur; |
716 | node = node->rb_left; | |
097b8a7c | 717 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
718 | /* we want the node with the smallest seq */ |
719 | if (found) | |
097b8a7c | 720 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
721 | found = cur; |
722 | node = node->rb_right; | |
723 | } else { | |
724 | found = cur; | |
725 | break; | |
726 | } | |
727 | } | |
b1a09f1e | 728 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
729 | |
730 | return found; | |
731 | } | |
732 | ||
733 | /* | |
734 | * this returns the element from the log with the smallest time sequence | |
735 | * value that's in the log (the oldest log item). any element with a time | |
736 | * sequence lower than min_seq will be ignored. | |
737 | */ | |
738 | static struct tree_mod_elem * | |
739 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
740 | u64 min_seq) | |
741 | { | |
742 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
743 | } | |
744 | ||
745 | /* | |
746 | * this returns the element from the log with the largest time sequence | |
747 | * value that's in the log (the most recent log item). any element with | |
748 | * a time sequence lower than min_seq will be ignored. | |
749 | */ | |
750 | static struct tree_mod_elem * | |
751 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
752 | { | |
753 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
754 | } | |
755 | ||
5de865ee | 756 | static noinline int |
bd989ba3 JS |
757 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
758 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 759 | unsigned long src_offset, int nr_items) |
bd989ba3 | 760 | { |
5de865ee FDBM |
761 | int ret = 0; |
762 | struct tree_mod_elem **tm_list = NULL; | |
763 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 764 | int i; |
5de865ee | 765 | int locked = 0; |
bd989ba3 | 766 | |
5de865ee FDBM |
767 | if (!tree_mod_need_log(fs_info, NULL)) |
768 | return 0; | |
bd989ba3 | 769 | |
c8cc6341 | 770 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
771 | return 0; |
772 | ||
31e818fe | 773 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
774 | GFP_NOFS); |
775 | if (!tm_list) | |
776 | return -ENOMEM; | |
bd989ba3 | 777 | |
5de865ee FDBM |
778 | tm_list_add = tm_list; |
779 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 780 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
781 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
782 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
783 | if (!tm_list_rem[i]) { | |
784 | ret = -ENOMEM; | |
785 | goto free_tms; | |
786 | } | |
787 | ||
788 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
789 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
790 | if (!tm_list_add[i]) { | |
791 | ret = -ENOMEM; | |
792 | goto free_tms; | |
793 | } | |
794 | } | |
795 | ||
796 | if (tree_mod_dont_log(fs_info, NULL)) | |
797 | goto free_tms; | |
798 | locked = 1; | |
799 | ||
800 | for (i = 0; i < nr_items; i++) { | |
801 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
802 | if (ret) | |
803 | goto free_tms; | |
804 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
805 | if (ret) | |
806 | goto free_tms; | |
bd989ba3 | 807 | } |
5de865ee | 808 | |
b1a09f1e | 809 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
810 | kfree(tm_list); |
811 | ||
812 | return 0; | |
813 | ||
814 | free_tms: | |
815 | for (i = 0; i < nr_items * 2; i++) { | |
816 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
817 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
818 | kfree(tm_list[i]); | |
819 | } | |
820 | if (locked) | |
b1a09f1e | 821 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
822 | kfree(tm_list); |
823 | ||
824 | return ret; | |
bd989ba3 JS |
825 | } |
826 | ||
db7279a2 | 827 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 828 | { |
5de865ee FDBM |
829 | struct tree_mod_elem **tm_list = NULL; |
830 | int nritems = 0; | |
831 | int i; | |
832 | int ret = 0; | |
833 | ||
834 | if (btrfs_header_level(eb) == 0) | |
835 | return 0; | |
836 | ||
db7279a2 | 837 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
838 | return 0; |
839 | ||
840 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 841 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
842 | if (!tm_list) |
843 | return -ENOMEM; | |
844 | ||
845 | for (i = 0; i < nritems; i++) { | |
846 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
847 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
848 | if (!tm_list[i]) { | |
849 | ret = -ENOMEM; | |
850 | goto free_tms; | |
851 | } | |
852 | } | |
853 | ||
db7279a2 | 854 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
855 | goto free_tms; |
856 | ||
db7279a2 | 857 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 858 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
859 | if (ret) |
860 | goto free_tms; | |
861 | kfree(tm_list); | |
862 | ||
863 | return 0; | |
864 | ||
865 | free_tms: | |
866 | for (i = 0; i < nritems; i++) | |
867 | kfree(tm_list[i]); | |
868 | kfree(tm_list); | |
869 | ||
870 | return ret; | |
bd989ba3 JS |
871 | } |
872 | ||
5d4f98a2 YZ |
873 | /* |
874 | * check if the tree block can be shared by multiple trees | |
875 | */ | |
876 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
877 | struct extent_buffer *buf) | |
878 | { | |
879 | /* | |
01327610 | 880 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
881 | * are never shared. If a block was allocated after the last |
882 | * snapshot and the block was not allocated by tree relocation, | |
883 | * we know the block is not shared. | |
884 | */ | |
27cdeb70 | 885 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
886 | buf != root->node && buf != root->commit_root && |
887 | (btrfs_header_generation(buf) <= | |
888 | btrfs_root_last_snapshot(&root->root_item) || | |
889 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
890 | return 1; | |
891 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 892 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
893 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
894 | return 1; | |
895 | #endif | |
896 | return 0; | |
897 | } | |
898 | ||
899 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
900 | struct btrfs_root *root, | |
901 | struct extent_buffer *buf, | |
f0486c68 YZ |
902 | struct extent_buffer *cow, |
903 | int *last_ref) | |
5d4f98a2 | 904 | { |
0b246afa | 905 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
906 | u64 refs; |
907 | u64 owner; | |
908 | u64 flags; | |
909 | u64 new_flags = 0; | |
910 | int ret; | |
911 | ||
912 | /* | |
913 | * Backrefs update rules: | |
914 | * | |
915 | * Always use full backrefs for extent pointers in tree block | |
916 | * allocated by tree relocation. | |
917 | * | |
918 | * If a shared tree block is no longer referenced by its owner | |
919 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
920 | * use full backrefs for extent pointers in tree block. | |
921 | * | |
922 | * If a tree block is been relocating | |
923 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
924 | * use full backrefs for extent pointers in tree block. | |
925 | * The reason for this is some operations (such as drop tree) | |
926 | * are only allowed for blocks use full backrefs. | |
927 | */ | |
928 | ||
929 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 930 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
931 | btrfs_header_level(buf), 1, |
932 | &refs, &flags); | |
be1a5564 MF |
933 | if (ret) |
934 | return ret; | |
e5df9573 MF |
935 | if (refs == 0) { |
936 | ret = -EROFS; | |
0b246afa | 937 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
938 | return ret; |
939 | } | |
5d4f98a2 YZ |
940 | } else { |
941 | refs = 1; | |
942 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
943 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
944 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
945 | else | |
946 | flags = 0; | |
947 | } | |
948 | ||
949 | owner = btrfs_header_owner(buf); | |
950 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
951 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
952 | ||
953 | if (refs > 1) { | |
954 | if ((owner == root->root_key.objectid || | |
955 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
956 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 957 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
958 | if (ret) |
959 | return ret; | |
5d4f98a2 YZ |
960 | |
961 | if (root->root_key.objectid == | |
962 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 963 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
964 | if (ret) |
965 | return ret; | |
e339a6b0 | 966 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
967 | if (ret) |
968 | return ret; | |
5d4f98a2 YZ |
969 | } |
970 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
971 | } else { | |
972 | ||
973 | if (root->root_key.objectid == | |
974 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 975 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 976 | else |
e339a6b0 | 977 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
978 | if (ret) |
979 | return ret; | |
5d4f98a2 YZ |
980 | } |
981 | if (new_flags != 0) { | |
b1c79e09 JB |
982 | int level = btrfs_header_level(buf); |
983 | ||
2ff7e61e | 984 | ret = btrfs_set_disk_extent_flags(trans, fs_info, |
5d4f98a2 YZ |
985 | buf->start, |
986 | buf->len, | |
b1c79e09 | 987 | new_flags, level, 0); |
be1a5564 MF |
988 | if (ret) |
989 | return ret; | |
5d4f98a2 YZ |
990 | } |
991 | } else { | |
992 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
993 | if (root->root_key.objectid == | |
994 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 995 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 996 | else |
e339a6b0 | 997 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
998 | if (ret) |
999 | return ret; | |
e339a6b0 | 1000 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
1001 | if (ret) |
1002 | return ret; | |
5d4f98a2 | 1003 | } |
7c302b49 | 1004 | clean_tree_block(fs_info, buf); |
f0486c68 | 1005 | *last_ref = 1; |
5d4f98a2 YZ |
1006 | } |
1007 | return 0; | |
1008 | } | |
1009 | ||
d352ac68 | 1010 | /* |
d397712b CM |
1011 | * does the dirty work in cow of a single block. The parent block (if |
1012 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1013 | * dirty and returned locked. If you modify the block it needs to be marked | |
1014 | * dirty again. | |
d352ac68 CM |
1015 | * |
1016 | * search_start -- an allocation hint for the new block | |
1017 | * | |
d397712b CM |
1018 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1019 | * bytes the allocator should try to find free next to the block it returns. | |
1020 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1021 | */ |
d397712b | 1022 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1023 | struct btrfs_root *root, |
1024 | struct extent_buffer *buf, | |
1025 | struct extent_buffer *parent, int parent_slot, | |
1026 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1027 | u64 search_start, u64 empty_size) |
02217ed2 | 1028 | { |
0b246afa | 1029 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1030 | struct btrfs_disk_key disk_key; |
5f39d397 | 1031 | struct extent_buffer *cow; |
be1a5564 | 1032 | int level, ret; |
f0486c68 | 1033 | int last_ref = 0; |
925baedd | 1034 | int unlock_orig = 0; |
0f5053eb | 1035 | u64 parent_start = 0; |
7bb86316 | 1036 | |
925baedd CM |
1037 | if (*cow_ret == buf) |
1038 | unlock_orig = 1; | |
1039 | ||
b9447ef8 | 1040 | btrfs_assert_tree_locked(buf); |
925baedd | 1041 | |
27cdeb70 | 1042 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1043 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1044 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1045 | trans->transid != root->last_trans); | |
5f39d397 | 1046 | |
7bb86316 | 1047 | level = btrfs_header_level(buf); |
31840ae1 | 1048 | |
5d4f98a2 YZ |
1049 | if (level == 0) |
1050 | btrfs_item_key(buf, &disk_key, 0); | |
1051 | else | |
1052 | btrfs_node_key(buf, &disk_key, 0); | |
1053 | ||
0f5053eb GR |
1054 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1055 | parent_start = parent->start; | |
5d4f98a2 | 1056 | |
4d75f8a9 DS |
1057 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1058 | root->root_key.objectid, &disk_key, level, | |
1059 | search_start, empty_size); | |
54aa1f4d CM |
1060 | if (IS_ERR(cow)) |
1061 | return PTR_ERR(cow); | |
6702ed49 | 1062 | |
b4ce94de CM |
1063 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1064 | ||
58e8012c | 1065 | copy_extent_buffer_full(cow, buf); |
db94535d | 1066 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1067 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1068 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1069 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1070 | BTRFS_HEADER_FLAG_RELOC); | |
1071 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1072 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1073 | else | |
1074 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1075 | |
0b246afa | 1076 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 1077 | |
be1a5564 | 1078 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1079 | if (ret) { |
66642832 | 1080 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1081 | return ret; |
1082 | } | |
1a40e23b | 1083 | |
27cdeb70 | 1084 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1085 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1086 | if (ret) { |
66642832 | 1087 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1088 | return ret; |
93314e3b | 1089 | } |
83d4cfd4 | 1090 | } |
3fd0a558 | 1091 | |
02217ed2 | 1092 | if (buf == root->node) { |
925baedd | 1093 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1094 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1095 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1096 | parent_start = buf->start; | |
925baedd | 1097 | |
5f39d397 | 1098 | extent_buffer_get(cow); |
d9d19a01 DS |
1099 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1100 | BUG_ON(ret < 0); | |
240f62c8 | 1101 | rcu_assign_pointer(root->node, cow); |
925baedd | 1102 | |
f0486c68 | 1103 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1104 | last_ref); |
5f39d397 | 1105 | free_extent_buffer(buf); |
0b86a832 | 1106 | add_root_to_dirty_list(root); |
02217ed2 | 1107 | } else { |
5d4f98a2 | 1108 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1109 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1110 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1111 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1112 | cow->start); |
74493f7a CM |
1113 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1114 | trans->transid); | |
d6025579 | 1115 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1116 | if (last_ref) { |
db7279a2 | 1117 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1118 | if (ret) { |
66642832 | 1119 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1120 | return ret; |
1121 | } | |
1122 | } | |
f0486c68 | 1123 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1124 | last_ref); |
02217ed2 | 1125 | } |
925baedd CM |
1126 | if (unlock_orig) |
1127 | btrfs_tree_unlock(buf); | |
3083ee2e | 1128 | free_extent_buffer_stale(buf); |
ccd467d6 | 1129 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1130 | *cow_ret = cow; |
02217ed2 CM |
1131 | return 0; |
1132 | } | |
1133 | ||
5d9e75c4 JS |
1134 | /* |
1135 | * returns the logical address of the oldest predecessor of the given root. | |
1136 | * entries older than time_seq are ignored. | |
1137 | */ | |
bcd24dab DS |
1138 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1139 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1140 | { |
1141 | struct tree_mod_elem *tm; | |
1142 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1143 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1144 | int looped = 0; |
1145 | ||
1146 | if (!time_seq) | |
35a3621b | 1147 | return NULL; |
5d9e75c4 JS |
1148 | |
1149 | /* | |
298cfd36 CR |
1150 | * the very last operation that's logged for a root is the |
1151 | * replacement operation (if it is replaced at all). this has | |
1152 | * the logical address of the *new* root, making it the very | |
1153 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1154 | */ |
1155 | while (1) { | |
bcd24dab | 1156 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1157 | time_seq); |
1158 | if (!looped && !tm) | |
35a3621b | 1159 | return NULL; |
5d9e75c4 | 1160 | /* |
28da9fb4 JS |
1161 | * if there are no tree operation for the oldest root, we simply |
1162 | * return it. this should only happen if that (old) root is at | |
1163 | * level 0. | |
5d9e75c4 | 1164 | */ |
28da9fb4 JS |
1165 | if (!tm) |
1166 | break; | |
5d9e75c4 | 1167 | |
28da9fb4 JS |
1168 | /* |
1169 | * if there's an operation that's not a root replacement, we | |
1170 | * found the oldest version of our root. normally, we'll find a | |
1171 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1172 | */ | |
5d9e75c4 JS |
1173 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1174 | break; | |
1175 | ||
1176 | found = tm; | |
1177 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1178 | looped = 1; |
1179 | } | |
1180 | ||
a95236d9 JS |
1181 | /* if there's no old root to return, return what we found instead */ |
1182 | if (!found) | |
1183 | found = tm; | |
1184 | ||
5d9e75c4 JS |
1185 | return found; |
1186 | } | |
1187 | ||
1188 | /* | |
1189 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1190 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1191 | * time_seq). |
1192 | */ | |
1193 | static void | |
f1ca7e98 JB |
1194 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1195 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1196 | { |
1197 | u32 n; | |
1198 | struct rb_node *next; | |
1199 | struct tree_mod_elem *tm = first_tm; | |
1200 | unsigned long o_dst; | |
1201 | unsigned long o_src; | |
1202 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1203 | ||
1204 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1205 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1206 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1207 | /* |
1208 | * all the operations are recorded with the operator used for | |
1209 | * the modification. as we're going backwards, we do the | |
1210 | * opposite of each operation here. | |
1211 | */ | |
1212 | switch (tm->op) { | |
1213 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1214 | BUG_ON(tm->slot < n); | |
1c697d4a | 1215 | /* Fallthrough */ |
95c80bb1 | 1216 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1217 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1218 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1219 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1220 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1221 | tm->generation); | |
4c3e6969 | 1222 | n++; |
5d9e75c4 JS |
1223 | break; |
1224 | case MOD_LOG_KEY_REPLACE: | |
1225 | BUG_ON(tm->slot >= n); | |
1226 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1227 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1228 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1229 | tm->generation); | |
1230 | break; | |
1231 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1232 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1233 | n--; |
1234 | break; | |
1235 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1236 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1237 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1238 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1239 | tm->move.nr_items * p_size); |
1240 | break; | |
1241 | case MOD_LOG_ROOT_REPLACE: | |
1242 | /* | |
1243 | * this operation is special. for roots, this must be | |
1244 | * handled explicitly before rewinding. | |
1245 | * for non-roots, this operation may exist if the node | |
1246 | * was a root: root A -> child B; then A gets empty and | |
1247 | * B is promoted to the new root. in the mod log, we'll | |
1248 | * have a root-replace operation for B, a tree block | |
1249 | * that is no root. we simply ignore that operation. | |
1250 | */ | |
1251 | break; | |
1252 | } | |
1253 | next = rb_next(&tm->node); | |
1254 | if (!next) | |
1255 | break; | |
6b4df8b6 | 1256 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1257 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1258 | break; |
1259 | } | |
b1a09f1e | 1260 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1261 | btrfs_set_header_nritems(eb, n); |
1262 | } | |
1263 | ||
47fb091f | 1264 | /* |
01327610 | 1265 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1266 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1267 | * returned buffer is always read-locked. If the returned buffer is not the | |
1268 | * input buffer, the lock on the input buffer is released and the input buffer | |
1269 | * is freed (its refcount is decremented). | |
1270 | */ | |
5d9e75c4 | 1271 | static struct extent_buffer * |
9ec72677 JB |
1272 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1273 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1274 | { |
1275 | struct extent_buffer *eb_rewin; | |
1276 | struct tree_mod_elem *tm; | |
1277 | ||
1278 | if (!time_seq) | |
1279 | return eb; | |
1280 | ||
1281 | if (btrfs_header_level(eb) == 0) | |
1282 | return eb; | |
1283 | ||
1284 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1285 | if (!tm) | |
1286 | return eb; | |
1287 | ||
9ec72677 JB |
1288 | btrfs_set_path_blocking(path); |
1289 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1290 | ||
5d9e75c4 JS |
1291 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1292 | BUG_ON(tm->slot != 0); | |
da17066c | 1293 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1294 | if (!eb_rewin) { |
9ec72677 | 1295 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1296 | free_extent_buffer(eb); |
1297 | return NULL; | |
1298 | } | |
5d9e75c4 JS |
1299 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1300 | btrfs_set_header_backref_rev(eb_rewin, | |
1301 | btrfs_header_backref_rev(eb)); | |
1302 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1303 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1304 | } else { |
1305 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1306 | if (!eb_rewin) { |
9ec72677 | 1307 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1308 | free_extent_buffer(eb); |
1309 | return NULL; | |
1310 | } | |
5d9e75c4 JS |
1311 | } |
1312 | ||
9ec72677 JB |
1313 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1314 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1315 | free_extent_buffer(eb); |
1316 | ||
47fb091f JS |
1317 | extent_buffer_get(eb_rewin); |
1318 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1319 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1320 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1321 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1322 | |
1323 | return eb_rewin; | |
1324 | } | |
1325 | ||
8ba97a15 JS |
1326 | /* |
1327 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1328 | * value. If there are no changes, the current root->root_node is returned. If | |
1329 | * anything changed in between, there's a fresh buffer allocated on which the | |
1330 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1331 | * Returns NULL on error (with no locks held). | |
1332 | */ | |
5d9e75c4 JS |
1333 | static inline struct extent_buffer * |
1334 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1335 | { | |
0b246afa | 1336 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1337 | struct tree_mod_elem *tm; |
30b0463a JS |
1338 | struct extent_buffer *eb = NULL; |
1339 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1340 | struct extent_buffer *old; |
a95236d9 | 1341 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1342 | u64 old_generation = 0; |
a95236d9 | 1343 | u64 logical; |
581c1760 | 1344 | int level; |
5d9e75c4 | 1345 | |
30b0463a | 1346 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1347 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1348 | if (!tm) |
30b0463a | 1349 | return eb_root; |
5d9e75c4 | 1350 | |
a95236d9 JS |
1351 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1352 | old_root = &tm->old_root; | |
1353 | old_generation = tm->generation; | |
1354 | logical = old_root->logical; | |
581c1760 | 1355 | level = old_root->level; |
a95236d9 | 1356 | } else { |
30b0463a | 1357 | logical = eb_root->start; |
581c1760 | 1358 | level = btrfs_header_level(eb_root); |
a95236d9 | 1359 | } |
5d9e75c4 | 1360 | |
0b246afa | 1361 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1362 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1363 | btrfs_tree_read_unlock(eb_root); |
1364 | free_extent_buffer(eb_root); | |
581c1760 | 1365 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1366 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1367 | if (!IS_ERR(old)) | |
1368 | free_extent_buffer(old); | |
0b246afa JM |
1369 | btrfs_warn(fs_info, |
1370 | "failed to read tree block %llu from get_old_root", | |
1371 | logical); | |
834328a8 | 1372 | } else { |
7bfdcf7f LB |
1373 | eb = btrfs_clone_extent_buffer(old); |
1374 | free_extent_buffer(old); | |
834328a8 JS |
1375 | } |
1376 | } else if (old_root) { | |
30b0463a JS |
1377 | btrfs_tree_read_unlock(eb_root); |
1378 | free_extent_buffer(eb_root); | |
0b246afa | 1379 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1380 | } else { |
9ec72677 | 1381 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1382 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1383 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1384 | free_extent_buffer(eb_root); |
834328a8 JS |
1385 | } |
1386 | ||
8ba97a15 JS |
1387 | if (!eb) |
1388 | return NULL; | |
d6381084 | 1389 | extent_buffer_get(eb); |
8ba97a15 | 1390 | btrfs_tree_read_lock(eb); |
a95236d9 | 1391 | if (old_root) { |
5d9e75c4 JS |
1392 | btrfs_set_header_bytenr(eb, eb->start); |
1393 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1394 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1395 | btrfs_set_header_level(eb, old_root->level); |
1396 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1397 | } |
28da9fb4 | 1398 | if (tm) |
0b246afa | 1399 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1400 | else |
1401 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1402 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1403 | |
1404 | return eb; | |
1405 | } | |
1406 | ||
5b6602e7 JS |
1407 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1408 | { | |
1409 | struct tree_mod_elem *tm; | |
1410 | int level; | |
30b0463a | 1411 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1412 | |
bcd24dab | 1413 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1414 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1415 | level = tm->old_root.level; | |
1416 | } else { | |
30b0463a | 1417 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1418 | } |
30b0463a | 1419 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1420 | |
1421 | return level; | |
1422 | } | |
1423 | ||
5d4f98a2 YZ |
1424 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1425 | struct btrfs_root *root, | |
1426 | struct extent_buffer *buf) | |
1427 | { | |
f5ee5c9a | 1428 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1429 | return 0; |
fccb84c9 | 1430 | |
d1980131 DS |
1431 | /* Ensure we can see the FORCE_COW bit */ |
1432 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1433 | |
1434 | /* | |
1435 | * We do not need to cow a block if | |
1436 | * 1) this block is not created or changed in this transaction; | |
1437 | * 2) this block does not belong to TREE_RELOC tree; | |
1438 | * 3) the root is not forced COW. | |
1439 | * | |
1440 | * What is forced COW: | |
01327610 | 1441 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1442 | * after we've finished coping src root, we must COW the shared |
1443 | * block to ensure the metadata consistency. | |
1444 | */ | |
5d4f98a2 YZ |
1445 | if (btrfs_header_generation(buf) == trans->transid && |
1446 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1447 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1448 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1449 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1450 | return 0; |
1451 | return 1; | |
1452 | } | |
1453 | ||
d352ac68 CM |
1454 | /* |
1455 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1456 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1457 | * once per transaction, as long as it hasn't been written yet |
1458 | */ | |
d397712b | 1459 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1460 | struct btrfs_root *root, struct extent_buffer *buf, |
1461 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1462 | struct extent_buffer **cow_ret) |
6702ed49 | 1463 | { |
0b246afa | 1464 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1465 | u64 search_start; |
f510cfec | 1466 | int ret; |
dc17ff8f | 1467 | |
0b246afa | 1468 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1469 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1470 | trans->transid, |
0b246afa | 1471 | fs_info->running_transaction->transid); |
31b1a2bd | 1472 | |
0b246afa | 1473 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1474 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1475 | trans->transid, fs_info->generation); |
dc17ff8f | 1476 | |
5d4f98a2 | 1477 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1478 | trans->dirty = true; |
6702ed49 CM |
1479 | *cow_ret = buf; |
1480 | return 0; | |
1481 | } | |
c487685d | 1482 | |
ee22184b | 1483 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1484 | |
1485 | if (parent) | |
1486 | btrfs_set_lock_blocking(parent); | |
1487 | btrfs_set_lock_blocking(buf); | |
1488 | ||
f510cfec | 1489 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1490 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1491 | |
1492 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1493 | ||
f510cfec | 1494 | return ret; |
6702ed49 CM |
1495 | } |
1496 | ||
d352ac68 CM |
1497 | /* |
1498 | * helper function for defrag to decide if two blocks pointed to by a | |
1499 | * node are actually close by | |
1500 | */ | |
6b80053d | 1501 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1502 | { |
6b80053d | 1503 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1504 | return 1; |
6b80053d | 1505 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1506 | return 1; |
1507 | return 0; | |
1508 | } | |
1509 | ||
081e9573 CM |
1510 | /* |
1511 | * compare two keys in a memcmp fashion | |
1512 | */ | |
310712b2 OS |
1513 | static int comp_keys(const struct btrfs_disk_key *disk, |
1514 | const struct btrfs_key *k2) | |
081e9573 CM |
1515 | { |
1516 | struct btrfs_key k1; | |
1517 | ||
1518 | btrfs_disk_key_to_cpu(&k1, disk); | |
1519 | ||
20736aba | 1520 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1521 | } |
1522 | ||
f3465ca4 JB |
1523 | /* |
1524 | * same as comp_keys only with two btrfs_key's | |
1525 | */ | |
310712b2 | 1526 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1527 | { |
1528 | if (k1->objectid > k2->objectid) | |
1529 | return 1; | |
1530 | if (k1->objectid < k2->objectid) | |
1531 | return -1; | |
1532 | if (k1->type > k2->type) | |
1533 | return 1; | |
1534 | if (k1->type < k2->type) | |
1535 | return -1; | |
1536 | if (k1->offset > k2->offset) | |
1537 | return 1; | |
1538 | if (k1->offset < k2->offset) | |
1539 | return -1; | |
1540 | return 0; | |
1541 | } | |
081e9573 | 1542 | |
d352ac68 CM |
1543 | /* |
1544 | * this is used by the defrag code to go through all the | |
1545 | * leaves pointed to by a node and reallocate them so that | |
1546 | * disk order is close to key order | |
1547 | */ | |
6702ed49 | 1548 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1549 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1550 | int start_slot, u64 *last_ret, |
a6b6e75e | 1551 | struct btrfs_key *progress) |
6702ed49 | 1552 | { |
0b246afa | 1553 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1554 | struct extent_buffer *cur; |
6702ed49 | 1555 | u64 blocknr; |
ca7a79ad | 1556 | u64 gen; |
e9d0b13b CM |
1557 | u64 search_start = *last_ret; |
1558 | u64 last_block = 0; | |
6702ed49 CM |
1559 | u64 other; |
1560 | u32 parent_nritems; | |
6702ed49 CM |
1561 | int end_slot; |
1562 | int i; | |
1563 | int err = 0; | |
f2183bde | 1564 | int parent_level; |
6b80053d CM |
1565 | int uptodate; |
1566 | u32 blocksize; | |
081e9573 CM |
1567 | int progress_passed = 0; |
1568 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1569 | |
5708b959 | 1570 | parent_level = btrfs_header_level(parent); |
5708b959 | 1571 | |
0b246afa JM |
1572 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1573 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1574 | |
6b80053d | 1575 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1576 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1577 | end_slot = parent_nritems - 1; |
6702ed49 | 1578 | |
5dfe2be7 | 1579 | if (parent_nritems <= 1) |
6702ed49 CM |
1580 | return 0; |
1581 | ||
b4ce94de CM |
1582 | btrfs_set_lock_blocking(parent); |
1583 | ||
5dfe2be7 | 1584 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1585 | struct btrfs_key first_key; |
6702ed49 | 1586 | int close = 1; |
a6b6e75e | 1587 | |
081e9573 CM |
1588 | btrfs_node_key(parent, &disk_key, i); |
1589 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1590 | continue; | |
1591 | ||
1592 | progress_passed = 1; | |
6b80053d | 1593 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1594 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1595 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1596 | if (last_block == 0) |
1597 | last_block = blocknr; | |
5708b959 | 1598 | |
6702ed49 | 1599 | if (i > 0) { |
6b80053d CM |
1600 | other = btrfs_node_blockptr(parent, i - 1); |
1601 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1602 | } |
5dfe2be7 | 1603 | if (!close && i < end_slot) { |
6b80053d CM |
1604 | other = btrfs_node_blockptr(parent, i + 1); |
1605 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1606 | } |
e9d0b13b CM |
1607 | if (close) { |
1608 | last_block = blocknr; | |
6702ed49 | 1609 | continue; |
e9d0b13b | 1610 | } |
6702ed49 | 1611 | |
0b246afa | 1612 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1613 | if (cur) |
b9fab919 | 1614 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1615 | else |
1616 | uptodate = 0; | |
5708b959 | 1617 | if (!cur || !uptodate) { |
6b80053d | 1618 | if (!cur) { |
581c1760 QW |
1619 | cur = read_tree_block(fs_info, blocknr, gen, |
1620 | parent_level - 1, | |
1621 | &first_key); | |
64c043de LB |
1622 | if (IS_ERR(cur)) { |
1623 | return PTR_ERR(cur); | |
1624 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1625 | free_extent_buffer(cur); |
97d9a8a4 | 1626 | return -EIO; |
416bc658 | 1627 | } |
6b80053d | 1628 | } else if (!uptodate) { |
581c1760 QW |
1629 | err = btrfs_read_buffer(cur, gen, |
1630 | parent_level - 1,&first_key); | |
018642a1 TI |
1631 | if (err) { |
1632 | free_extent_buffer(cur); | |
1633 | return err; | |
1634 | } | |
f2183bde | 1635 | } |
6702ed49 | 1636 | } |
e9d0b13b | 1637 | if (search_start == 0) |
6b80053d | 1638 | search_start = last_block; |
e9d0b13b | 1639 | |
e7a84565 | 1640 | btrfs_tree_lock(cur); |
b4ce94de | 1641 | btrfs_set_lock_blocking(cur); |
6b80053d | 1642 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1643 | &cur, search_start, |
6b80053d | 1644 | min(16 * blocksize, |
9fa8cfe7 | 1645 | (end_slot - i) * blocksize)); |
252c38f0 | 1646 | if (err) { |
e7a84565 | 1647 | btrfs_tree_unlock(cur); |
6b80053d | 1648 | free_extent_buffer(cur); |
6702ed49 | 1649 | break; |
252c38f0 | 1650 | } |
e7a84565 CM |
1651 | search_start = cur->start; |
1652 | last_block = cur->start; | |
f2183bde | 1653 | *last_ret = search_start; |
e7a84565 CM |
1654 | btrfs_tree_unlock(cur); |
1655 | free_extent_buffer(cur); | |
6702ed49 CM |
1656 | } |
1657 | return err; | |
1658 | } | |
1659 | ||
74123bd7 | 1660 | /* |
5f39d397 CM |
1661 | * search for key in the extent_buffer. The items start at offset p, |
1662 | * and they are item_size apart. There are 'max' items in p. | |
1663 | * | |
74123bd7 CM |
1664 | * the slot in the array is returned via slot, and it points to |
1665 | * the place where you would insert key if it is not found in | |
1666 | * the array. | |
1667 | * | |
1668 | * slot may point to max if the key is bigger than all of the keys | |
1669 | */ | |
e02119d5 | 1670 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1671 | unsigned long p, int item_size, |
1672 | const struct btrfs_key *key, | |
e02119d5 | 1673 | int max, int *slot) |
be0e5c09 CM |
1674 | { |
1675 | int low = 0; | |
1676 | int high = max; | |
1677 | int mid; | |
1678 | int ret; | |
479965d6 | 1679 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1680 | struct btrfs_disk_key unaligned; |
1681 | unsigned long offset; | |
5f39d397 CM |
1682 | char *kaddr = NULL; |
1683 | unsigned long map_start = 0; | |
1684 | unsigned long map_len = 0; | |
479965d6 | 1685 | int err; |
be0e5c09 | 1686 | |
5e24e9af LB |
1687 | if (low > high) { |
1688 | btrfs_err(eb->fs_info, | |
1689 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1690 | __func__, low, high, eb->start, | |
1691 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1692 | return -EINVAL; | |
1693 | } | |
1694 | ||
d397712b | 1695 | while (low < high) { |
be0e5c09 | 1696 | mid = (low + high) / 2; |
5f39d397 CM |
1697 | offset = p + mid * item_size; |
1698 | ||
a6591715 | 1699 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1700 | (offset + sizeof(struct btrfs_disk_key)) > |
1701 | map_start + map_len) { | |
934d375b CM |
1702 | |
1703 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1704 | sizeof(struct btrfs_disk_key), |
a6591715 | 1705 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1706 | |
1707 | if (!err) { | |
1708 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1709 | map_start); | |
415b35a5 | 1710 | } else if (err == 1) { |
479965d6 CM |
1711 | read_extent_buffer(eb, &unaligned, |
1712 | offset, sizeof(unaligned)); | |
1713 | tmp = &unaligned; | |
415b35a5 LB |
1714 | } else { |
1715 | return err; | |
479965d6 | 1716 | } |
5f39d397 | 1717 | |
5f39d397 CM |
1718 | } else { |
1719 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1720 | map_start); | |
1721 | } | |
be0e5c09 CM |
1722 | ret = comp_keys(tmp, key); |
1723 | ||
1724 | if (ret < 0) | |
1725 | low = mid + 1; | |
1726 | else if (ret > 0) | |
1727 | high = mid; | |
1728 | else { | |
1729 | *slot = mid; | |
1730 | return 0; | |
1731 | } | |
1732 | } | |
1733 | *slot = low; | |
1734 | return 1; | |
1735 | } | |
1736 | ||
97571fd0 CM |
1737 | /* |
1738 | * simple bin_search frontend that does the right thing for | |
1739 | * leaves vs nodes | |
1740 | */ | |
a74b35ec NB |
1741 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1742 | int level, int *slot) | |
be0e5c09 | 1743 | { |
f775738f | 1744 | if (level == 0) |
5f39d397 CM |
1745 | return generic_bin_search(eb, |
1746 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1747 | sizeof(struct btrfs_item), |
5f39d397 | 1748 | key, btrfs_header_nritems(eb), |
7518a238 | 1749 | slot); |
f775738f | 1750 | else |
5f39d397 CM |
1751 | return generic_bin_search(eb, |
1752 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1753 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1754 | key, btrfs_header_nritems(eb), |
7518a238 | 1755 | slot); |
be0e5c09 CM |
1756 | } |
1757 | ||
f0486c68 YZ |
1758 | static void root_add_used(struct btrfs_root *root, u32 size) |
1759 | { | |
1760 | spin_lock(&root->accounting_lock); | |
1761 | btrfs_set_root_used(&root->root_item, | |
1762 | btrfs_root_used(&root->root_item) + size); | |
1763 | spin_unlock(&root->accounting_lock); | |
1764 | } | |
1765 | ||
1766 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1767 | { | |
1768 | spin_lock(&root->accounting_lock); | |
1769 | btrfs_set_root_used(&root->root_item, | |
1770 | btrfs_root_used(&root->root_item) - size); | |
1771 | spin_unlock(&root->accounting_lock); | |
1772 | } | |
1773 | ||
d352ac68 CM |
1774 | /* given a node and slot number, this reads the blocks it points to. The |
1775 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1776 | */ |
2ff7e61e JM |
1777 | static noinline struct extent_buffer * |
1778 | read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent, | |
1779 | int slot) | |
bb803951 | 1780 | { |
ca7a79ad | 1781 | int level = btrfs_header_level(parent); |
416bc658 | 1782 | struct extent_buffer *eb; |
581c1760 | 1783 | struct btrfs_key first_key; |
416bc658 | 1784 | |
fb770ae4 LB |
1785 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1786 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1787 | |
1788 | BUG_ON(level == 0); | |
1789 | ||
581c1760 | 1790 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
2ff7e61e | 1791 | eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1792 | btrfs_node_ptr_generation(parent, slot), |
1793 | level - 1, &first_key); | |
fb770ae4 LB |
1794 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1795 | free_extent_buffer(eb); | |
1796 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1797 | } |
1798 | ||
1799 | return eb; | |
bb803951 CM |
1800 | } |
1801 | ||
d352ac68 CM |
1802 | /* |
1803 | * node level balancing, used to make sure nodes are in proper order for | |
1804 | * item deletion. We balance from the top down, so we have to make sure | |
1805 | * that a deletion won't leave an node completely empty later on. | |
1806 | */ | |
e02119d5 | 1807 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1808 | struct btrfs_root *root, |
1809 | struct btrfs_path *path, int level) | |
bb803951 | 1810 | { |
0b246afa | 1811 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1812 | struct extent_buffer *right = NULL; |
1813 | struct extent_buffer *mid; | |
1814 | struct extent_buffer *left = NULL; | |
1815 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1816 | int ret = 0; |
1817 | int wret; | |
1818 | int pslot; | |
bb803951 | 1819 | int orig_slot = path->slots[level]; |
79f95c82 | 1820 | u64 orig_ptr; |
bb803951 CM |
1821 | |
1822 | if (level == 0) | |
1823 | return 0; | |
1824 | ||
5f39d397 | 1825 | mid = path->nodes[level]; |
b4ce94de | 1826 | |
bd681513 CM |
1827 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1828 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1829 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1830 | ||
1d4f8a0c | 1831 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1832 | |
a05a9bb1 | 1833 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1834 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1835 | pslot = path->slots[level + 1]; |
1836 | } | |
bb803951 | 1837 | |
40689478 CM |
1838 | /* |
1839 | * deal with the case where there is only one pointer in the root | |
1840 | * by promoting the node below to a root | |
1841 | */ | |
5f39d397 CM |
1842 | if (!parent) { |
1843 | struct extent_buffer *child; | |
bb803951 | 1844 | |
5f39d397 | 1845 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1846 | return 0; |
1847 | ||
1848 | /* promote the child to a root */ | |
2ff7e61e | 1849 | child = read_node_slot(fs_info, mid, 0); |
fb770ae4 LB |
1850 | if (IS_ERR(child)) { |
1851 | ret = PTR_ERR(child); | |
0b246afa | 1852 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1853 | goto enospc; |
1854 | } | |
1855 | ||
925baedd | 1856 | btrfs_tree_lock(child); |
b4ce94de | 1857 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1858 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1859 | if (ret) { |
1860 | btrfs_tree_unlock(child); | |
1861 | free_extent_buffer(child); | |
1862 | goto enospc; | |
1863 | } | |
2f375ab9 | 1864 | |
d9d19a01 DS |
1865 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1866 | BUG_ON(ret < 0); | |
240f62c8 | 1867 | rcu_assign_pointer(root->node, child); |
925baedd | 1868 | |
0b86a832 | 1869 | add_root_to_dirty_list(root); |
925baedd | 1870 | btrfs_tree_unlock(child); |
b4ce94de | 1871 | |
925baedd | 1872 | path->locks[level] = 0; |
bb803951 | 1873 | path->nodes[level] = NULL; |
7c302b49 | 1874 | clean_tree_block(fs_info, mid); |
925baedd | 1875 | btrfs_tree_unlock(mid); |
bb803951 | 1876 | /* once for the path */ |
5f39d397 | 1877 | free_extent_buffer(mid); |
f0486c68 YZ |
1878 | |
1879 | root_sub_used(root, mid->len); | |
5581a51a | 1880 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1881 | /* once for the root ptr */ |
3083ee2e | 1882 | free_extent_buffer_stale(mid); |
f0486c68 | 1883 | return 0; |
bb803951 | 1884 | } |
5f39d397 | 1885 | if (btrfs_header_nritems(mid) > |
0b246afa | 1886 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1887 | return 0; |
1888 | ||
2ff7e61e | 1889 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
1890 | if (IS_ERR(left)) |
1891 | left = NULL; | |
1892 | ||
5f39d397 | 1893 | if (left) { |
925baedd | 1894 | btrfs_tree_lock(left); |
b4ce94de | 1895 | btrfs_set_lock_blocking(left); |
5f39d397 | 1896 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1897 | parent, pslot - 1, &left); |
54aa1f4d CM |
1898 | if (wret) { |
1899 | ret = wret; | |
1900 | goto enospc; | |
1901 | } | |
2cc58cf2 | 1902 | } |
fb770ae4 | 1903 | |
2ff7e61e | 1904 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
1905 | if (IS_ERR(right)) |
1906 | right = NULL; | |
1907 | ||
5f39d397 | 1908 | if (right) { |
925baedd | 1909 | btrfs_tree_lock(right); |
b4ce94de | 1910 | btrfs_set_lock_blocking(right); |
5f39d397 | 1911 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1912 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1913 | if (wret) { |
1914 | ret = wret; | |
1915 | goto enospc; | |
1916 | } | |
1917 | } | |
1918 | ||
1919 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1920 | if (left) { |
1921 | orig_slot += btrfs_header_nritems(left); | |
2ff7e61e | 1922 | wret = push_node_left(trans, fs_info, left, mid, 1); |
79f95c82 CM |
1923 | if (wret < 0) |
1924 | ret = wret; | |
bb803951 | 1925 | } |
79f95c82 CM |
1926 | |
1927 | /* | |
1928 | * then try to empty the right most buffer into the middle | |
1929 | */ | |
5f39d397 | 1930 | if (right) { |
2ff7e61e | 1931 | wret = push_node_left(trans, fs_info, mid, right, 1); |
54aa1f4d | 1932 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1933 | ret = wret; |
5f39d397 | 1934 | if (btrfs_header_nritems(right) == 0) { |
7c302b49 | 1935 | clean_tree_block(fs_info, right); |
925baedd | 1936 | btrfs_tree_unlock(right); |
afe5fea7 | 1937 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1938 | root_sub_used(root, right->len); |
5581a51a | 1939 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1940 | free_extent_buffer_stale(right); |
f0486c68 | 1941 | right = NULL; |
bb803951 | 1942 | } else { |
5f39d397 CM |
1943 | struct btrfs_disk_key right_key; |
1944 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1945 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1946 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1947 | BUG_ON(ret < 0); | |
5f39d397 CM |
1948 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1949 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1950 | } |
1951 | } | |
5f39d397 | 1952 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1953 | /* |
1954 | * we're not allowed to leave a node with one item in the | |
1955 | * tree during a delete. A deletion from lower in the tree | |
1956 | * could try to delete the only pointer in this node. | |
1957 | * So, pull some keys from the left. | |
1958 | * There has to be a left pointer at this point because | |
1959 | * otherwise we would have pulled some pointers from the | |
1960 | * right | |
1961 | */ | |
305a26af MF |
1962 | if (!left) { |
1963 | ret = -EROFS; | |
0b246afa | 1964 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1965 | goto enospc; |
1966 | } | |
2ff7e61e | 1967 | wret = balance_node_right(trans, fs_info, mid, left); |
54aa1f4d | 1968 | if (wret < 0) { |
79f95c82 | 1969 | ret = wret; |
54aa1f4d CM |
1970 | goto enospc; |
1971 | } | |
bce4eae9 | 1972 | if (wret == 1) { |
2ff7e61e | 1973 | wret = push_node_left(trans, fs_info, left, mid, 1); |
bce4eae9 CM |
1974 | if (wret < 0) |
1975 | ret = wret; | |
1976 | } | |
79f95c82 CM |
1977 | BUG_ON(wret == 1); |
1978 | } | |
5f39d397 | 1979 | if (btrfs_header_nritems(mid) == 0) { |
7c302b49 | 1980 | clean_tree_block(fs_info, mid); |
925baedd | 1981 | btrfs_tree_unlock(mid); |
afe5fea7 | 1982 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1983 | root_sub_used(root, mid->len); |
5581a51a | 1984 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1985 | free_extent_buffer_stale(mid); |
f0486c68 | 1986 | mid = NULL; |
79f95c82 CM |
1987 | } else { |
1988 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1989 | struct btrfs_disk_key mid_key; |
1990 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
1991 | ret = tree_mod_log_insert_key(parent, pslot, |
1992 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1993 | BUG_ON(ret < 0); | |
5f39d397 CM |
1994 | btrfs_set_node_key(parent, &mid_key, pslot); |
1995 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1996 | } |
bb803951 | 1997 | |
79f95c82 | 1998 | /* update the path */ |
5f39d397 CM |
1999 | if (left) { |
2000 | if (btrfs_header_nritems(left) > orig_slot) { | |
2001 | extent_buffer_get(left); | |
925baedd | 2002 | /* left was locked after cow */ |
5f39d397 | 2003 | path->nodes[level] = left; |
bb803951 CM |
2004 | path->slots[level + 1] -= 1; |
2005 | path->slots[level] = orig_slot; | |
925baedd CM |
2006 | if (mid) { |
2007 | btrfs_tree_unlock(mid); | |
5f39d397 | 2008 | free_extent_buffer(mid); |
925baedd | 2009 | } |
bb803951 | 2010 | } else { |
5f39d397 | 2011 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2012 | path->slots[level] = orig_slot; |
2013 | } | |
2014 | } | |
79f95c82 | 2015 | /* double check we haven't messed things up */ |
e20d96d6 | 2016 | if (orig_ptr != |
5f39d397 | 2017 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2018 | BUG(); |
54aa1f4d | 2019 | enospc: |
925baedd CM |
2020 | if (right) { |
2021 | btrfs_tree_unlock(right); | |
5f39d397 | 2022 | free_extent_buffer(right); |
925baedd CM |
2023 | } |
2024 | if (left) { | |
2025 | if (path->nodes[level] != left) | |
2026 | btrfs_tree_unlock(left); | |
5f39d397 | 2027 | free_extent_buffer(left); |
925baedd | 2028 | } |
bb803951 CM |
2029 | return ret; |
2030 | } | |
2031 | ||
d352ac68 CM |
2032 | /* Node balancing for insertion. Here we only split or push nodes around |
2033 | * when they are completely full. This is also done top down, so we | |
2034 | * have to be pessimistic. | |
2035 | */ | |
d397712b | 2036 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2037 | struct btrfs_root *root, |
2038 | struct btrfs_path *path, int level) | |
e66f709b | 2039 | { |
0b246afa | 2040 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2041 | struct extent_buffer *right = NULL; |
2042 | struct extent_buffer *mid; | |
2043 | struct extent_buffer *left = NULL; | |
2044 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2045 | int ret = 0; |
2046 | int wret; | |
2047 | int pslot; | |
2048 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2049 | |
2050 | if (level == 0) | |
2051 | return 1; | |
2052 | ||
5f39d397 | 2053 | mid = path->nodes[level]; |
7bb86316 | 2054 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2055 | |
a05a9bb1 | 2056 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2057 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2058 | pslot = path->slots[level + 1]; |
2059 | } | |
e66f709b | 2060 | |
5f39d397 | 2061 | if (!parent) |
e66f709b | 2062 | return 1; |
e66f709b | 2063 | |
2ff7e61e | 2064 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
2065 | if (IS_ERR(left)) |
2066 | left = NULL; | |
e66f709b CM |
2067 | |
2068 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2069 | if (left) { |
e66f709b | 2070 | u32 left_nr; |
925baedd CM |
2071 | |
2072 | btrfs_tree_lock(left); | |
b4ce94de CM |
2073 | btrfs_set_lock_blocking(left); |
2074 | ||
5f39d397 | 2075 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2076 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2077 | wret = 1; |
2078 | } else { | |
5f39d397 | 2079 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2080 | pslot - 1, &left); |
54aa1f4d CM |
2081 | if (ret) |
2082 | wret = 1; | |
2083 | else { | |
2ff7e61e | 2084 | wret = push_node_left(trans, fs_info, |
971a1f66 | 2085 | left, mid, 0); |
54aa1f4d | 2086 | } |
33ade1f8 | 2087 | } |
e66f709b CM |
2088 | if (wret < 0) |
2089 | ret = wret; | |
2090 | if (wret == 0) { | |
5f39d397 | 2091 | struct btrfs_disk_key disk_key; |
e66f709b | 2092 | orig_slot += left_nr; |
5f39d397 | 2093 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2094 | ret = tree_mod_log_insert_key(parent, pslot, |
2095 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2096 | BUG_ON(ret < 0); | |
5f39d397 CM |
2097 | btrfs_set_node_key(parent, &disk_key, pslot); |
2098 | btrfs_mark_buffer_dirty(parent); | |
2099 | if (btrfs_header_nritems(left) > orig_slot) { | |
2100 | path->nodes[level] = left; | |
e66f709b CM |
2101 | path->slots[level + 1] -= 1; |
2102 | path->slots[level] = orig_slot; | |
925baedd | 2103 | btrfs_tree_unlock(mid); |
5f39d397 | 2104 | free_extent_buffer(mid); |
e66f709b CM |
2105 | } else { |
2106 | orig_slot -= | |
5f39d397 | 2107 | btrfs_header_nritems(left); |
e66f709b | 2108 | path->slots[level] = orig_slot; |
925baedd | 2109 | btrfs_tree_unlock(left); |
5f39d397 | 2110 | free_extent_buffer(left); |
e66f709b | 2111 | } |
e66f709b CM |
2112 | return 0; |
2113 | } | |
925baedd | 2114 | btrfs_tree_unlock(left); |
5f39d397 | 2115 | free_extent_buffer(left); |
e66f709b | 2116 | } |
2ff7e61e | 2117 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
2118 | if (IS_ERR(right)) |
2119 | right = NULL; | |
e66f709b CM |
2120 | |
2121 | /* | |
2122 | * then try to empty the right most buffer into the middle | |
2123 | */ | |
5f39d397 | 2124 | if (right) { |
33ade1f8 | 2125 | u32 right_nr; |
b4ce94de | 2126 | |
925baedd | 2127 | btrfs_tree_lock(right); |
b4ce94de CM |
2128 | btrfs_set_lock_blocking(right); |
2129 | ||
5f39d397 | 2130 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2131 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2132 | wret = 1; |
2133 | } else { | |
5f39d397 CM |
2134 | ret = btrfs_cow_block(trans, root, right, |
2135 | parent, pslot + 1, | |
9fa8cfe7 | 2136 | &right); |
54aa1f4d CM |
2137 | if (ret) |
2138 | wret = 1; | |
2139 | else { | |
2ff7e61e | 2140 | wret = balance_node_right(trans, fs_info, |
5f39d397 | 2141 | right, mid); |
54aa1f4d | 2142 | } |
33ade1f8 | 2143 | } |
e66f709b CM |
2144 | if (wret < 0) |
2145 | ret = wret; | |
2146 | if (wret == 0) { | |
5f39d397 CM |
2147 | struct btrfs_disk_key disk_key; |
2148 | ||
2149 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2150 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2151 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2152 | BUG_ON(ret < 0); | |
5f39d397 CM |
2153 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2154 | btrfs_mark_buffer_dirty(parent); | |
2155 | ||
2156 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2157 | path->nodes[level] = right; | |
e66f709b CM |
2158 | path->slots[level + 1] += 1; |
2159 | path->slots[level] = orig_slot - | |
5f39d397 | 2160 | btrfs_header_nritems(mid); |
925baedd | 2161 | btrfs_tree_unlock(mid); |
5f39d397 | 2162 | free_extent_buffer(mid); |
e66f709b | 2163 | } else { |
925baedd | 2164 | btrfs_tree_unlock(right); |
5f39d397 | 2165 | free_extent_buffer(right); |
e66f709b | 2166 | } |
e66f709b CM |
2167 | return 0; |
2168 | } | |
925baedd | 2169 | btrfs_tree_unlock(right); |
5f39d397 | 2170 | free_extent_buffer(right); |
e66f709b | 2171 | } |
e66f709b CM |
2172 | return 1; |
2173 | } | |
2174 | ||
3c69faec | 2175 | /* |
d352ac68 CM |
2176 | * readahead one full node of leaves, finding things that are close |
2177 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2178 | */ |
2ff7e61e | 2179 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2180 | struct btrfs_path *path, |
2181 | int level, int slot, u64 objectid) | |
3c69faec | 2182 | { |
5f39d397 | 2183 | struct extent_buffer *node; |
01f46658 | 2184 | struct btrfs_disk_key disk_key; |
3c69faec | 2185 | u32 nritems; |
3c69faec | 2186 | u64 search; |
a7175319 | 2187 | u64 target; |
6b80053d | 2188 | u64 nread = 0; |
5f39d397 | 2189 | struct extent_buffer *eb; |
6b80053d CM |
2190 | u32 nr; |
2191 | u32 blocksize; | |
2192 | u32 nscan = 0; | |
db94535d | 2193 | |
a6b6e75e | 2194 | if (level != 1) |
6702ed49 CM |
2195 | return; |
2196 | ||
2197 | if (!path->nodes[level]) | |
3c69faec CM |
2198 | return; |
2199 | ||
5f39d397 | 2200 | node = path->nodes[level]; |
925baedd | 2201 | |
3c69faec | 2202 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2203 | blocksize = fs_info->nodesize; |
2204 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2205 | if (eb) { |
2206 | free_extent_buffer(eb); | |
3c69faec CM |
2207 | return; |
2208 | } | |
2209 | ||
a7175319 | 2210 | target = search; |
6b80053d | 2211 | |
5f39d397 | 2212 | nritems = btrfs_header_nritems(node); |
6b80053d | 2213 | nr = slot; |
25b8b936 | 2214 | |
d397712b | 2215 | while (1) { |
e4058b54 | 2216 | if (path->reada == READA_BACK) { |
6b80053d CM |
2217 | if (nr == 0) |
2218 | break; | |
2219 | nr--; | |
e4058b54 | 2220 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2221 | nr++; |
2222 | if (nr >= nritems) | |
2223 | break; | |
3c69faec | 2224 | } |
e4058b54 | 2225 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2226 | btrfs_node_key(node, &disk_key, nr); |
2227 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2228 | break; | |
2229 | } | |
6b80053d | 2230 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2231 | if ((search <= target && target - search <= 65536) || |
2232 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2233 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2234 | nread += blocksize; |
2235 | } | |
2236 | nscan++; | |
a7175319 | 2237 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2238 | break; |
3c69faec CM |
2239 | } |
2240 | } | |
925baedd | 2241 | |
2ff7e61e | 2242 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2243 | struct btrfs_path *path, int level) |
b4ce94de CM |
2244 | { |
2245 | int slot; | |
2246 | int nritems; | |
2247 | struct extent_buffer *parent; | |
2248 | struct extent_buffer *eb; | |
2249 | u64 gen; | |
2250 | u64 block1 = 0; | |
2251 | u64 block2 = 0; | |
b4ce94de | 2252 | |
8c594ea8 | 2253 | parent = path->nodes[level + 1]; |
b4ce94de | 2254 | if (!parent) |
0b08851f | 2255 | return; |
b4ce94de CM |
2256 | |
2257 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2258 | slot = path->slots[level + 1]; |
b4ce94de CM |
2259 | |
2260 | if (slot > 0) { | |
2261 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2262 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2263 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2264 | /* |
2265 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2266 | * don't want to return eagain here. That will loop | |
2267 | * forever | |
2268 | */ | |
2269 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2270 | block1 = 0; |
2271 | free_extent_buffer(eb); | |
2272 | } | |
8c594ea8 | 2273 | if (slot + 1 < nritems) { |
b4ce94de CM |
2274 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2275 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2276 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2277 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2278 | block2 = 0; |
2279 | free_extent_buffer(eb); | |
2280 | } | |
8c594ea8 | 2281 | |
0b08851f | 2282 | if (block1) |
2ff7e61e | 2283 | readahead_tree_block(fs_info, block1); |
0b08851f | 2284 | if (block2) |
2ff7e61e | 2285 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2286 | } |
2287 | ||
2288 | ||
d352ac68 | 2289 | /* |
d397712b CM |
2290 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2291 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2292 | * operations on the tree might require changing key pointers higher up in the | |
2293 | * tree. | |
d352ac68 | 2294 | * |
d397712b CM |
2295 | * callers might also have set path->keep_locks, which tells this code to keep |
2296 | * the lock if the path points to the last slot in the block. This is part of | |
2297 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2298 | * |
d397712b CM |
2299 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2300 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2301 | */ |
e02119d5 | 2302 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2303 | int lowest_unlock, int min_write_lock_level, |
2304 | int *write_lock_level) | |
925baedd CM |
2305 | { |
2306 | int i; | |
2307 | int skip_level = level; | |
051e1b9f | 2308 | int no_skips = 0; |
925baedd CM |
2309 | struct extent_buffer *t; |
2310 | ||
2311 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2312 | if (!path->nodes[i]) | |
2313 | break; | |
2314 | if (!path->locks[i]) | |
2315 | break; | |
051e1b9f | 2316 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2317 | skip_level = i + 1; |
2318 | continue; | |
2319 | } | |
051e1b9f | 2320 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2321 | u32 nritems; |
2322 | t = path->nodes[i]; | |
2323 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2324 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2325 | skip_level = i + 1; |
2326 | continue; | |
2327 | } | |
2328 | } | |
051e1b9f CM |
2329 | if (skip_level < i && i >= lowest_unlock) |
2330 | no_skips = 1; | |
2331 | ||
925baedd | 2332 | t = path->nodes[i]; |
d80bb3f9 | 2333 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2334 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2335 | path->locks[i] = 0; |
f7c79f30 CM |
2336 | if (write_lock_level && |
2337 | i > min_write_lock_level && | |
2338 | i <= *write_lock_level) { | |
2339 | *write_lock_level = i - 1; | |
2340 | } | |
925baedd CM |
2341 | } |
2342 | } | |
2343 | } | |
2344 | ||
b4ce94de CM |
2345 | /* |
2346 | * This releases any locks held in the path starting at level and | |
2347 | * going all the way up to the root. | |
2348 | * | |
2349 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2350 | * corner cases, such as COW of the block at slot zero in the node. This | |
2351 | * ignores those rules, and it should only be called when there are no | |
2352 | * more updates to be done higher up in the tree. | |
2353 | */ | |
2354 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2355 | { | |
2356 | int i; | |
2357 | ||
09a2a8f9 | 2358 | if (path->keep_locks) |
b4ce94de CM |
2359 | return; |
2360 | ||
2361 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2362 | if (!path->nodes[i]) | |
12f4dacc | 2363 | continue; |
b4ce94de | 2364 | if (!path->locks[i]) |
12f4dacc | 2365 | continue; |
bd681513 | 2366 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2367 | path->locks[i] = 0; |
2368 | } | |
2369 | } | |
2370 | ||
c8c42864 CM |
2371 | /* |
2372 | * helper function for btrfs_search_slot. The goal is to find a block | |
2373 | * in cache without setting the path to blocking. If we find the block | |
2374 | * we return zero and the path is unchanged. | |
2375 | * | |
2376 | * If we can't find the block, we set the path blocking and do some | |
2377 | * reada. -EAGAIN is returned and the search must be repeated. | |
2378 | */ | |
2379 | static int | |
d07b8528 LB |
2380 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2381 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2382 | const struct btrfs_key *key) |
c8c42864 | 2383 | { |
0b246afa | 2384 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2385 | u64 blocknr; |
2386 | u64 gen; | |
c8c42864 CM |
2387 | struct extent_buffer *b = *eb_ret; |
2388 | struct extent_buffer *tmp; | |
581c1760 | 2389 | struct btrfs_key first_key; |
76a05b35 | 2390 | int ret; |
581c1760 | 2391 | int parent_level; |
c8c42864 CM |
2392 | |
2393 | blocknr = btrfs_node_blockptr(b, slot); | |
2394 | gen = btrfs_node_ptr_generation(b, slot); | |
581c1760 QW |
2395 | parent_level = btrfs_header_level(b); |
2396 | btrfs_node_key_to_cpu(b, &first_key, slot); | |
c8c42864 | 2397 | |
0b246afa | 2398 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2399 | if (tmp) { |
b9fab919 | 2400 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2401 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2402 | *eb_ret = tmp; | |
2403 | return 0; | |
2404 | } | |
2405 | ||
2406 | /* the pages were up to date, but we failed | |
2407 | * the generation number check. Do a full | |
2408 | * read for the generation number that is correct. | |
2409 | * We must do this without dropping locks so | |
2410 | * we can trust our generation number | |
2411 | */ | |
2412 | btrfs_set_path_blocking(p); | |
2413 | ||
2414 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2415 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2416 | if (!ret) { |
2417 | *eb_ret = tmp; | |
2418 | return 0; | |
cb44921a | 2419 | } |
bdf7c00e JB |
2420 | free_extent_buffer(tmp); |
2421 | btrfs_release_path(p); | |
2422 | return -EIO; | |
c8c42864 CM |
2423 | } |
2424 | ||
2425 | /* | |
2426 | * reduce lock contention at high levels | |
2427 | * of the btree by dropping locks before | |
76a05b35 CM |
2428 | * we read. Don't release the lock on the current |
2429 | * level because we need to walk this node to figure | |
2430 | * out which blocks to read. | |
c8c42864 | 2431 | */ |
8c594ea8 CM |
2432 | btrfs_unlock_up_safe(p, level + 1); |
2433 | btrfs_set_path_blocking(p); | |
2434 | ||
e4058b54 | 2435 | if (p->reada != READA_NONE) |
2ff7e61e | 2436 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2437 | |
76a05b35 | 2438 | ret = -EAGAIN; |
02a3307a | 2439 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2440 | &first_key); |
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 | */ | |
e6a1d6fd | 2448 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2449 | ret = -EIO; |
c8c42864 | 2450 | free_extent_buffer(tmp); |
c871b0f2 LB |
2451 | } else { |
2452 | ret = PTR_ERR(tmp); | |
76a05b35 | 2453 | } |
02a3307a LB |
2454 | |
2455 | btrfs_release_path(p); | |
76a05b35 | 2456 | return ret; |
c8c42864 CM |
2457 | } |
2458 | ||
2459 | /* | |
2460 | * helper function for btrfs_search_slot. This does all of the checks | |
2461 | * for node-level blocks and does any balancing required based on | |
2462 | * the ins_len. | |
2463 | * | |
2464 | * If no extra work was required, zero is returned. If we had to | |
2465 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2466 | * start over | |
2467 | */ | |
2468 | static int | |
2469 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2470 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2471 | struct extent_buffer *b, int level, int ins_len, |
2472 | int *write_lock_level) | |
c8c42864 | 2473 | { |
0b246afa | 2474 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2475 | int ret; |
0b246afa | 2476 | |
c8c42864 | 2477 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2478 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2479 | int sret; |
2480 | ||
bd681513 CM |
2481 | if (*write_lock_level < level + 1) { |
2482 | *write_lock_level = level + 1; | |
2483 | btrfs_release_path(p); | |
2484 | goto again; | |
2485 | } | |
2486 | ||
c8c42864 | 2487 | btrfs_set_path_blocking(p); |
2ff7e61e | 2488 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2489 | sret = split_node(trans, root, p, level); |
bd681513 | 2490 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2491 | |
2492 | BUG_ON(sret > 0); | |
2493 | if (sret) { | |
2494 | ret = sret; | |
2495 | goto done; | |
2496 | } | |
2497 | b = p->nodes[level]; | |
2498 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2499 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2500 | int sret; |
2501 | ||
bd681513 CM |
2502 | if (*write_lock_level < level + 1) { |
2503 | *write_lock_level = level + 1; | |
2504 | btrfs_release_path(p); | |
2505 | goto again; | |
2506 | } | |
2507 | ||
c8c42864 | 2508 | btrfs_set_path_blocking(p); |
2ff7e61e | 2509 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2510 | sret = balance_level(trans, root, p, level); |
bd681513 | 2511 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2512 | |
2513 | if (sret) { | |
2514 | ret = sret; | |
2515 | goto done; | |
2516 | } | |
2517 | b = p->nodes[level]; | |
2518 | if (!b) { | |
b3b4aa74 | 2519 | btrfs_release_path(p); |
c8c42864 CM |
2520 | goto again; |
2521 | } | |
2522 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2523 | } | |
2524 | return 0; | |
2525 | ||
2526 | again: | |
2527 | ret = -EAGAIN; | |
2528 | done: | |
2529 | return ret; | |
2530 | } | |
2531 | ||
d7396f07 | 2532 | static void key_search_validate(struct extent_buffer *b, |
310712b2 | 2533 | const struct btrfs_key *key, |
d7396f07 FDBM |
2534 | int level) |
2535 | { | |
2536 | #ifdef CONFIG_BTRFS_ASSERT | |
2537 | struct btrfs_disk_key disk_key; | |
2538 | ||
2539 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2540 | ||
2541 | if (level == 0) | |
2542 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2543 | offsetof(struct btrfs_leaf, items[0].key), | |
2544 | sizeof(disk_key))); | |
2545 | else | |
2546 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2547 | offsetof(struct btrfs_node, ptrs[0].key), | |
2548 | sizeof(disk_key))); | |
2549 | #endif | |
2550 | } | |
2551 | ||
310712b2 | 2552 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2553 | int level, int *prev_cmp, int *slot) |
2554 | { | |
2555 | if (*prev_cmp != 0) { | |
a74b35ec | 2556 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2557 | return *prev_cmp; |
2558 | } | |
2559 | ||
2560 | key_search_validate(b, key, level); | |
2561 | *slot = 0; | |
2562 | ||
2563 | return 0; | |
2564 | } | |
2565 | ||
381cf658 | 2566 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2567 | u64 iobjectid, u64 ioff, u8 key_type, |
2568 | struct btrfs_key *found_key) | |
2569 | { | |
2570 | int ret; | |
2571 | struct btrfs_key key; | |
2572 | struct extent_buffer *eb; | |
381cf658 DS |
2573 | |
2574 | ASSERT(path); | |
1d4c08e0 | 2575 | ASSERT(found_key); |
e33d5c3d KN |
2576 | |
2577 | key.type = key_type; | |
2578 | key.objectid = iobjectid; | |
2579 | key.offset = ioff; | |
2580 | ||
2581 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2582 | if (ret < 0) |
e33d5c3d KN |
2583 | return ret; |
2584 | ||
2585 | eb = path->nodes[0]; | |
2586 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2587 | ret = btrfs_next_leaf(fs_root, path); | |
2588 | if (ret) | |
2589 | return ret; | |
2590 | eb = path->nodes[0]; | |
2591 | } | |
2592 | ||
2593 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2594 | if (found_key->type != key.type || | |
2595 | found_key->objectid != key.objectid) | |
2596 | return 1; | |
2597 | ||
2598 | return 0; | |
2599 | } | |
2600 | ||
1fc28d8e LB |
2601 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2602 | struct btrfs_path *p, | |
2603 | int write_lock_level) | |
2604 | { | |
2605 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2606 | struct extent_buffer *b; | |
2607 | int root_lock; | |
2608 | int level = 0; | |
2609 | ||
2610 | /* We try very hard to do read locks on the root */ | |
2611 | root_lock = BTRFS_READ_LOCK; | |
2612 | ||
2613 | if (p->search_commit_root) { | |
2614 | /* The commit roots are read only so we always do read locks */ | |
2615 | if (p->need_commit_sem) | |
2616 | down_read(&fs_info->commit_root_sem); | |
2617 | b = root->commit_root; | |
2618 | extent_buffer_get(b); | |
2619 | level = btrfs_header_level(b); | |
2620 | if (p->need_commit_sem) | |
2621 | up_read(&fs_info->commit_root_sem); | |
2622 | if (!p->skip_locking) | |
2623 | btrfs_tree_read_lock(b); | |
2624 | ||
2625 | goto out; | |
2626 | } | |
2627 | ||
2628 | if (p->skip_locking) { | |
2629 | b = btrfs_root_node(root); | |
2630 | level = btrfs_header_level(b); | |
2631 | goto out; | |
2632 | } | |
2633 | ||
2634 | /* | |
662c653b LB |
2635 | * If the level is set to maximum, we can skip trying to get the read |
2636 | * lock. | |
1fc28d8e | 2637 | */ |
662c653b LB |
2638 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2639 | /* | |
2640 | * We don't know the level of the root node until we actually | |
2641 | * have it read locked | |
2642 | */ | |
2643 | b = btrfs_read_lock_root_node(root); | |
2644 | level = btrfs_header_level(b); | |
2645 | if (level > write_lock_level) | |
2646 | goto out; | |
2647 | ||
2648 | /* Whoops, must trade for write lock */ | |
2649 | btrfs_tree_read_unlock(b); | |
2650 | free_extent_buffer(b); | |
2651 | } | |
1fc28d8e | 2652 | |
1fc28d8e LB |
2653 | b = btrfs_lock_root_node(root); |
2654 | root_lock = BTRFS_WRITE_LOCK; | |
2655 | ||
2656 | /* The level might have changed, check again */ | |
2657 | level = btrfs_header_level(b); | |
2658 | ||
2659 | out: | |
2660 | p->nodes[level] = b; | |
2661 | if (!p->skip_locking) | |
2662 | p->locks[level] = root_lock; | |
2663 | /* | |
2664 | * Callers are responsible for dropping b's references. | |
2665 | */ | |
2666 | return b; | |
2667 | } | |
2668 | ||
2669 | ||
74123bd7 | 2670 | /* |
4271ecea NB |
2671 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2672 | * modifications to preserve tree invariants. | |
74123bd7 | 2673 | * |
4271ecea NB |
2674 | * @trans: Handle of transaction, used when modifying the tree |
2675 | * @p: Holds all btree nodes along the search path | |
2676 | * @root: The root node of the tree | |
2677 | * @key: The key we are looking for | |
2678 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2679 | * deletions it's -1. 0 for plain searches | |
2680 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2681 | * when modifying the tree. | |
97571fd0 | 2682 | * |
4271ecea NB |
2683 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2684 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2685 | * | |
2686 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2687 | * of the path (level 0) | |
2688 | * | |
2689 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2690 | * points to the slot where it should be inserted | |
2691 | * | |
2692 | * If an error is encountered while searching the tree a negative error number | |
2693 | * is returned | |
74123bd7 | 2694 | */ |
310712b2 OS |
2695 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2696 | const struct btrfs_key *key, struct btrfs_path *p, | |
2697 | int ins_len, int cow) | |
be0e5c09 | 2698 | { |
0b246afa | 2699 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2700 | struct extent_buffer *b; |
be0e5c09 CM |
2701 | int slot; |
2702 | int ret; | |
33c66f43 | 2703 | int err; |
be0e5c09 | 2704 | int level; |
925baedd | 2705 | int lowest_unlock = 1; |
bd681513 CM |
2706 | /* everything at write_lock_level or lower must be write locked */ |
2707 | int write_lock_level = 0; | |
9f3a7427 | 2708 | u8 lowest_level = 0; |
f7c79f30 | 2709 | int min_write_lock_level; |
d7396f07 | 2710 | int prev_cmp; |
9f3a7427 | 2711 | |
6702ed49 | 2712 | lowest_level = p->lowest_level; |
323ac95b | 2713 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2714 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2715 | BUG_ON(!cow && ins_len); |
25179201 | 2716 | |
bd681513 | 2717 | if (ins_len < 0) { |
925baedd | 2718 | lowest_unlock = 2; |
65b51a00 | 2719 | |
bd681513 CM |
2720 | /* when we are removing items, we might have to go up to level |
2721 | * two as we update tree pointers Make sure we keep write | |
2722 | * for those levels as well | |
2723 | */ | |
2724 | write_lock_level = 2; | |
2725 | } else if (ins_len > 0) { | |
2726 | /* | |
2727 | * for inserting items, make sure we have a write lock on | |
2728 | * level 1 so we can update keys | |
2729 | */ | |
2730 | write_lock_level = 1; | |
2731 | } | |
2732 | ||
2733 | if (!cow) | |
2734 | write_lock_level = -1; | |
2735 | ||
09a2a8f9 | 2736 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2737 | write_lock_level = BTRFS_MAX_LEVEL; |
2738 | ||
f7c79f30 CM |
2739 | min_write_lock_level = write_lock_level; |
2740 | ||
bb803951 | 2741 | again: |
d7396f07 | 2742 | prev_cmp = -1; |
1fc28d8e | 2743 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
925baedd | 2744 | |
eb60ceac | 2745 | while (b) { |
5f39d397 | 2746 | level = btrfs_header_level(b); |
65b51a00 CM |
2747 | |
2748 | /* | |
2749 | * setup the path here so we can release it under lock | |
2750 | * contention with the cow code | |
2751 | */ | |
02217ed2 | 2752 | if (cow) { |
9ea2c7c9 NB |
2753 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2754 | ||
c8c42864 CM |
2755 | /* |
2756 | * if we don't really need to cow this block | |
2757 | * then we don't want to set the path blocking, | |
2758 | * so we test it here | |
2759 | */ | |
64c12921 JM |
2760 | if (!should_cow_block(trans, root, b)) { |
2761 | trans->dirty = true; | |
65b51a00 | 2762 | goto cow_done; |
64c12921 | 2763 | } |
5d4f98a2 | 2764 | |
bd681513 CM |
2765 | /* |
2766 | * must have write locks on this node and the | |
2767 | * parent | |
2768 | */ | |
5124e00e JB |
2769 | if (level > write_lock_level || |
2770 | (level + 1 > write_lock_level && | |
2771 | level + 1 < BTRFS_MAX_LEVEL && | |
2772 | p->nodes[level + 1])) { | |
bd681513 CM |
2773 | write_lock_level = level + 1; |
2774 | btrfs_release_path(p); | |
2775 | goto again; | |
2776 | } | |
2777 | ||
160f4089 | 2778 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2779 | if (last_level) |
2780 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2781 | &b); | |
2782 | else | |
2783 | err = btrfs_cow_block(trans, root, b, | |
2784 | p->nodes[level + 1], | |
2785 | p->slots[level + 1], &b); | |
33c66f43 | 2786 | if (err) { |
33c66f43 | 2787 | ret = err; |
65b51a00 | 2788 | goto done; |
54aa1f4d | 2789 | } |
02217ed2 | 2790 | } |
65b51a00 | 2791 | cow_done: |
eb60ceac | 2792 | p->nodes[level] = b; |
bd681513 | 2793 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2794 | |
2795 | /* | |
2796 | * we have a lock on b and as long as we aren't changing | |
2797 | * the tree, there is no way to for the items in b to change. | |
2798 | * It is safe to drop the lock on our parent before we | |
2799 | * go through the expensive btree search on b. | |
2800 | * | |
eb653de1 FDBM |
2801 | * If we're inserting or deleting (ins_len != 0), then we might |
2802 | * be changing slot zero, which may require changing the parent. | |
2803 | * So, we can't drop the lock until after we know which slot | |
2804 | * we're operating on. | |
b4ce94de | 2805 | */ |
eb653de1 FDBM |
2806 | if (!ins_len && !p->keep_locks) { |
2807 | int u = level + 1; | |
2808 | ||
2809 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2810 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2811 | p->locks[u] = 0; | |
2812 | } | |
2813 | } | |
b4ce94de | 2814 | |
d7396f07 | 2815 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2816 | if (ret < 0) |
2817 | goto done; | |
b4ce94de | 2818 | |
5f39d397 | 2819 | if (level != 0) { |
33c66f43 YZ |
2820 | int dec = 0; |
2821 | if (ret && slot > 0) { | |
2822 | dec = 1; | |
be0e5c09 | 2823 | slot -= 1; |
33c66f43 | 2824 | } |
be0e5c09 | 2825 | p->slots[level] = slot; |
33c66f43 | 2826 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2827 | ins_len, &write_lock_level); |
33c66f43 | 2828 | if (err == -EAGAIN) |
c8c42864 | 2829 | goto again; |
33c66f43 YZ |
2830 | if (err) { |
2831 | ret = err; | |
c8c42864 | 2832 | goto done; |
33c66f43 | 2833 | } |
c8c42864 CM |
2834 | b = p->nodes[level]; |
2835 | slot = p->slots[level]; | |
b4ce94de | 2836 | |
bd681513 CM |
2837 | /* |
2838 | * slot 0 is special, if we change the key | |
2839 | * we have to update the parent pointer | |
2840 | * which means we must have a write lock | |
2841 | * on the parent | |
2842 | */ | |
eb653de1 | 2843 | if (slot == 0 && ins_len && |
bd681513 CM |
2844 | write_lock_level < level + 1) { |
2845 | write_lock_level = level + 1; | |
2846 | btrfs_release_path(p); | |
2847 | goto again; | |
2848 | } | |
2849 | ||
f7c79f30 CM |
2850 | unlock_up(p, level, lowest_unlock, |
2851 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2852 | |
925baedd | 2853 | if (level == lowest_level) { |
33c66f43 YZ |
2854 | if (dec) |
2855 | p->slots[level]++; | |
5b21f2ed | 2856 | goto done; |
925baedd | 2857 | } |
ca7a79ad | 2858 | |
d07b8528 | 2859 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2860 | slot, key); |
33c66f43 | 2861 | if (err == -EAGAIN) |
c8c42864 | 2862 | goto again; |
33c66f43 YZ |
2863 | if (err) { |
2864 | ret = err; | |
76a05b35 | 2865 | goto done; |
33c66f43 | 2866 | } |
76a05b35 | 2867 | |
b4ce94de | 2868 | if (!p->skip_locking) { |
bd681513 CM |
2869 | level = btrfs_header_level(b); |
2870 | if (level <= write_lock_level) { | |
2871 | err = btrfs_try_tree_write_lock(b); | |
2872 | if (!err) { | |
2873 | btrfs_set_path_blocking(p); | |
2874 | btrfs_tree_lock(b); | |
2875 | btrfs_clear_path_blocking(p, b, | |
2876 | BTRFS_WRITE_LOCK); | |
2877 | } | |
2878 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2879 | } else { | |
f82c458a | 2880 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2881 | if (!err) { |
2882 | btrfs_set_path_blocking(p); | |
2883 | btrfs_tree_read_lock(b); | |
2884 | btrfs_clear_path_blocking(p, b, | |
2885 | BTRFS_READ_LOCK); | |
2886 | } | |
2887 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2888 | } |
bd681513 | 2889 | p->nodes[level] = b; |
b4ce94de | 2890 | } |
be0e5c09 CM |
2891 | } else { |
2892 | p->slots[level] = slot; | |
87b29b20 | 2893 | if (ins_len > 0 && |
2ff7e61e | 2894 | btrfs_leaf_free_space(fs_info, b) < ins_len) { |
bd681513 CM |
2895 | if (write_lock_level < 1) { |
2896 | write_lock_level = 1; | |
2897 | btrfs_release_path(p); | |
2898 | goto again; | |
2899 | } | |
2900 | ||
b4ce94de | 2901 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2902 | err = split_leaf(trans, root, key, |
2903 | p, ins_len, ret == 0); | |
bd681513 | 2904 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2905 | |
33c66f43 YZ |
2906 | BUG_ON(err > 0); |
2907 | if (err) { | |
2908 | ret = err; | |
65b51a00 CM |
2909 | goto done; |
2910 | } | |
5c680ed6 | 2911 | } |
459931ec | 2912 | if (!p->search_for_split) |
f7c79f30 CM |
2913 | unlock_up(p, level, lowest_unlock, |
2914 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2915 | goto done; |
be0e5c09 CM |
2916 | } |
2917 | } | |
65b51a00 CM |
2918 | ret = 1; |
2919 | done: | |
b4ce94de CM |
2920 | /* |
2921 | * we don't really know what they plan on doing with the path | |
2922 | * from here on, so for now just mark it as blocking | |
2923 | */ | |
b9473439 CM |
2924 | if (!p->leave_spinning) |
2925 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2926 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2927 | btrfs_release_path(p); |
65b51a00 | 2928 | return ret; |
be0e5c09 CM |
2929 | } |
2930 | ||
5d9e75c4 JS |
2931 | /* |
2932 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2933 | * current state of the tree together with the operations recorded in the tree | |
2934 | * modification log to search for the key in a previous version of this tree, as | |
2935 | * denoted by the time_seq parameter. | |
2936 | * | |
2937 | * Naturally, there is no support for insert, delete or cow operations. | |
2938 | * | |
2939 | * The resulting path and return value will be set up as if we called | |
2940 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2941 | */ | |
310712b2 | 2942 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2943 | struct btrfs_path *p, u64 time_seq) |
2944 | { | |
0b246afa | 2945 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2946 | struct extent_buffer *b; |
2947 | int slot; | |
2948 | int ret; | |
2949 | int err; | |
2950 | int level; | |
2951 | int lowest_unlock = 1; | |
2952 | u8 lowest_level = 0; | |
d4b4087c | 2953 | int prev_cmp = -1; |
5d9e75c4 JS |
2954 | |
2955 | lowest_level = p->lowest_level; | |
2956 | WARN_ON(p->nodes[0] != NULL); | |
2957 | ||
2958 | if (p->search_commit_root) { | |
2959 | BUG_ON(time_seq); | |
2960 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2961 | } | |
2962 | ||
2963 | again: | |
5d9e75c4 | 2964 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2965 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2966 | p->locks[level] = BTRFS_READ_LOCK; |
2967 | ||
2968 | while (b) { | |
2969 | level = btrfs_header_level(b); | |
2970 | p->nodes[level] = b; | |
2971 | btrfs_clear_path_blocking(p, NULL, 0); | |
2972 | ||
2973 | /* | |
2974 | * we have a lock on b and as long as we aren't changing | |
2975 | * the tree, there is no way to for the items in b to change. | |
2976 | * It is safe to drop the lock on our parent before we | |
2977 | * go through the expensive btree search on b. | |
2978 | */ | |
2979 | btrfs_unlock_up_safe(p, level + 1); | |
2980 | ||
d4b4087c | 2981 | /* |
01327610 | 2982 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2983 | * time. |
2984 | */ | |
2985 | prev_cmp = -1; | |
d7396f07 | 2986 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
2987 | |
2988 | if (level != 0) { | |
2989 | int dec = 0; | |
2990 | if (ret && slot > 0) { | |
2991 | dec = 1; | |
2992 | slot -= 1; | |
2993 | } | |
2994 | p->slots[level] = slot; | |
2995 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2996 | ||
2997 | if (level == lowest_level) { | |
2998 | if (dec) | |
2999 | p->slots[level]++; | |
3000 | goto done; | |
3001 | } | |
3002 | ||
d07b8528 | 3003 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 3004 | slot, key); |
5d9e75c4 JS |
3005 | if (err == -EAGAIN) |
3006 | goto again; | |
3007 | if (err) { | |
3008 | ret = err; | |
3009 | goto done; | |
3010 | } | |
3011 | ||
3012 | level = btrfs_header_level(b); | |
f82c458a | 3013 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3014 | if (!err) { |
3015 | btrfs_set_path_blocking(p); | |
3016 | btrfs_tree_read_lock(b); | |
3017 | btrfs_clear_path_blocking(p, b, | |
3018 | BTRFS_READ_LOCK); | |
3019 | } | |
0b246afa | 3020 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3021 | if (!b) { |
3022 | ret = -ENOMEM; | |
3023 | goto done; | |
3024 | } | |
5d9e75c4 JS |
3025 | p->locks[level] = BTRFS_READ_LOCK; |
3026 | p->nodes[level] = b; | |
5d9e75c4 JS |
3027 | } else { |
3028 | p->slots[level] = slot; | |
3029 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3030 | goto done; | |
3031 | } | |
3032 | } | |
3033 | ret = 1; | |
3034 | done: | |
3035 | if (!p->leave_spinning) | |
3036 | btrfs_set_path_blocking(p); | |
3037 | if (ret < 0) | |
3038 | btrfs_release_path(p); | |
3039 | ||
3040 | return ret; | |
3041 | } | |
3042 | ||
2f38b3e1 AJ |
3043 | /* |
3044 | * helper to use instead of search slot if no exact match is needed but | |
3045 | * instead the next or previous item should be returned. | |
3046 | * When find_higher is true, the next higher item is returned, the next lower | |
3047 | * otherwise. | |
3048 | * When return_any and find_higher are both true, and no higher item is found, | |
3049 | * return the next lower instead. | |
3050 | * When return_any is true and find_higher is false, and no lower item is found, | |
3051 | * return the next higher instead. | |
3052 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3053 | * < 0 on error | |
3054 | */ | |
3055 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3056 | const struct btrfs_key *key, |
3057 | struct btrfs_path *p, int find_higher, | |
3058 | int return_any) | |
2f38b3e1 AJ |
3059 | { |
3060 | int ret; | |
3061 | struct extent_buffer *leaf; | |
3062 | ||
3063 | again: | |
3064 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3065 | if (ret <= 0) | |
3066 | return ret; | |
3067 | /* | |
3068 | * a return value of 1 means the path is at the position where the | |
3069 | * item should be inserted. Normally this is the next bigger item, | |
3070 | * but in case the previous item is the last in a leaf, path points | |
3071 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3072 | * item. | |
3073 | */ | |
3074 | leaf = p->nodes[0]; | |
3075 | ||
3076 | if (find_higher) { | |
3077 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3078 | ret = btrfs_next_leaf(root, p); | |
3079 | if (ret <= 0) | |
3080 | return ret; | |
3081 | if (!return_any) | |
3082 | return 1; | |
3083 | /* | |
3084 | * no higher item found, return the next | |
3085 | * lower instead | |
3086 | */ | |
3087 | return_any = 0; | |
3088 | find_higher = 0; | |
3089 | btrfs_release_path(p); | |
3090 | goto again; | |
3091 | } | |
3092 | } else { | |
e6793769 AJ |
3093 | if (p->slots[0] == 0) { |
3094 | ret = btrfs_prev_leaf(root, p); | |
3095 | if (ret < 0) | |
3096 | return ret; | |
3097 | if (!ret) { | |
23c6bf6a FDBM |
3098 | leaf = p->nodes[0]; |
3099 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3100 | p->slots[0]--; | |
e6793769 | 3101 | return 0; |
2f38b3e1 | 3102 | } |
e6793769 AJ |
3103 | if (!return_any) |
3104 | return 1; | |
3105 | /* | |
3106 | * no lower item found, return the next | |
3107 | * higher instead | |
3108 | */ | |
3109 | return_any = 0; | |
3110 | find_higher = 1; | |
3111 | btrfs_release_path(p); | |
3112 | goto again; | |
3113 | } else { | |
2f38b3e1 AJ |
3114 | --p->slots[0]; |
3115 | } | |
3116 | } | |
3117 | return 0; | |
3118 | } | |
3119 | ||
74123bd7 CM |
3120 | /* |
3121 | * adjust the pointers going up the tree, starting at level | |
3122 | * making sure the right key of each node is points to 'key'. | |
3123 | * This is used after shifting pointers to the left, so it stops | |
3124 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3125 | * higher levels | |
aa5d6bed | 3126 | * |
74123bd7 | 3127 | */ |
b7a0365e DD |
3128 | static void fixup_low_keys(struct btrfs_fs_info *fs_info, |
3129 | struct btrfs_path *path, | |
143bede5 | 3130 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3131 | { |
3132 | int i; | |
5f39d397 | 3133 | struct extent_buffer *t; |
0e82bcfe | 3134 | int ret; |
5f39d397 | 3135 | |
234b63a0 | 3136 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3137 | int tslot = path->slots[i]; |
0e82bcfe | 3138 | |
eb60ceac | 3139 | if (!path->nodes[i]) |
be0e5c09 | 3140 | break; |
5f39d397 | 3141 | t = path->nodes[i]; |
0e82bcfe DS |
3142 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3143 | GFP_ATOMIC); | |
3144 | BUG_ON(ret < 0); | |
5f39d397 | 3145 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3146 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3147 | if (tslot != 0) |
3148 | break; | |
3149 | } | |
3150 | } | |
3151 | ||
31840ae1 ZY |
3152 | /* |
3153 | * update item key. | |
3154 | * | |
3155 | * This function isn't completely safe. It's the caller's responsibility | |
3156 | * that the new key won't break the order | |
3157 | */ | |
b7a0365e DD |
3158 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3159 | struct btrfs_path *path, | |
310712b2 | 3160 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3161 | { |
3162 | struct btrfs_disk_key disk_key; | |
3163 | struct extent_buffer *eb; | |
3164 | int slot; | |
3165 | ||
3166 | eb = path->nodes[0]; | |
3167 | slot = path->slots[0]; | |
3168 | if (slot > 0) { | |
3169 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3170 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3171 | } |
3172 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3173 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3174 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3175 | } |
3176 | ||
3177 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3178 | btrfs_set_item_key(eb, &disk_key, slot); | |
3179 | btrfs_mark_buffer_dirty(eb); | |
3180 | if (slot == 0) | |
b7a0365e | 3181 | fixup_low_keys(fs_info, path, &disk_key, 1); |
31840ae1 ZY |
3182 | } |
3183 | ||
74123bd7 CM |
3184 | /* |
3185 | * try to push data from one node into the next node left in the | |
79f95c82 | 3186 | * tree. |
aa5d6bed CM |
3187 | * |
3188 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3189 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3190 | */ |
98ed5174 | 3191 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
3192 | struct btrfs_fs_info *fs_info, |
3193 | struct extent_buffer *dst, | |
971a1f66 | 3194 | struct extent_buffer *src, int empty) |
be0e5c09 | 3195 | { |
be0e5c09 | 3196 | int push_items = 0; |
bb803951 CM |
3197 | int src_nritems; |
3198 | int dst_nritems; | |
aa5d6bed | 3199 | int ret = 0; |
be0e5c09 | 3200 | |
5f39d397 CM |
3201 | src_nritems = btrfs_header_nritems(src); |
3202 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3203 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3204 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3205 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3206 | |
bce4eae9 | 3207 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3208 | return 1; |
3209 | ||
d397712b | 3210 | if (push_items <= 0) |
be0e5c09 CM |
3211 | return 1; |
3212 | ||
bce4eae9 | 3213 | if (empty) { |
971a1f66 | 3214 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3215 | if (push_items < src_nritems) { |
3216 | /* leave at least 8 pointers in the node if | |
3217 | * we aren't going to empty it | |
3218 | */ | |
3219 | if (src_nritems - push_items < 8) { | |
3220 | if (push_items <= 8) | |
3221 | return 1; | |
3222 | push_items -= 8; | |
3223 | } | |
3224 | } | |
3225 | } else | |
3226 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3227 | |
0b246afa | 3228 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3229 | push_items); |
3230 | if (ret) { | |
66642832 | 3231 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3232 | return ret; |
3233 | } | |
5f39d397 CM |
3234 | copy_extent_buffer(dst, src, |
3235 | btrfs_node_key_ptr_offset(dst_nritems), | |
3236 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3237 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3238 | |
bb803951 | 3239 | if (push_items < src_nritems) { |
57911b8b | 3240 | /* |
bf1d3425 DS |
3241 | * Don't call tree_mod_log_insert_move here, key removal was |
3242 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3243 | */ |
5f39d397 CM |
3244 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3245 | btrfs_node_key_ptr_offset(push_items), | |
3246 | (src_nritems - push_items) * | |
3247 | sizeof(struct btrfs_key_ptr)); | |
3248 | } | |
3249 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3250 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3251 | btrfs_mark_buffer_dirty(src); | |
3252 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3253 | |
79f95c82 CM |
3254 | return ret; |
3255 | } | |
3256 | ||
3257 | /* | |
3258 | * try to push data from one node into the next node right in the | |
3259 | * tree. | |
3260 | * | |
3261 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3262 | * error, and > 0 if there was no room in the right hand block. | |
3263 | * | |
3264 | * this will only push up to 1/2 the contents of the left node over | |
3265 | */ | |
5f39d397 | 3266 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 3267 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
3268 | struct extent_buffer *dst, |
3269 | struct extent_buffer *src) | |
79f95c82 | 3270 | { |
79f95c82 CM |
3271 | int push_items = 0; |
3272 | int max_push; | |
3273 | int src_nritems; | |
3274 | int dst_nritems; | |
3275 | int ret = 0; | |
79f95c82 | 3276 | |
7bb86316 CM |
3277 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3278 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3279 | ||
5f39d397 CM |
3280 | src_nritems = btrfs_header_nritems(src); |
3281 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3282 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3283 | if (push_items <= 0) |
79f95c82 | 3284 | return 1; |
bce4eae9 | 3285 | |
d397712b | 3286 | if (src_nritems < 4) |
bce4eae9 | 3287 | return 1; |
79f95c82 CM |
3288 | |
3289 | max_push = src_nritems / 2 + 1; | |
3290 | /* don't try to empty the node */ | |
d397712b | 3291 | if (max_push >= src_nritems) |
79f95c82 | 3292 | return 1; |
252c38f0 | 3293 | |
79f95c82 CM |
3294 | if (max_push < push_items) |
3295 | push_items = max_push; | |
3296 | ||
bf1d3425 DS |
3297 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3298 | BUG_ON(ret < 0); | |
5f39d397 CM |
3299 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3300 | btrfs_node_key_ptr_offset(0), | |
3301 | (dst_nritems) * | |
3302 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3303 | |
0b246afa | 3304 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3305 | src_nritems - push_items, push_items); |
3306 | if (ret) { | |
66642832 | 3307 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3308 | return ret; |
3309 | } | |
5f39d397 CM |
3310 | copy_extent_buffer(dst, src, |
3311 | btrfs_node_key_ptr_offset(0), | |
3312 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3313 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3314 | |
5f39d397 CM |
3315 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3316 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3317 | |
5f39d397 CM |
3318 | btrfs_mark_buffer_dirty(src); |
3319 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3320 | |
aa5d6bed | 3321 | return ret; |
be0e5c09 CM |
3322 | } |
3323 | ||
97571fd0 CM |
3324 | /* |
3325 | * helper function to insert a new root level in the tree. | |
3326 | * A new node is allocated, and a single item is inserted to | |
3327 | * point to the existing root | |
aa5d6bed CM |
3328 | * |
3329 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3330 | */ |
d397712b | 3331 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3332 | struct btrfs_root *root, |
fdd99c72 | 3333 | struct btrfs_path *path, int level) |
5c680ed6 | 3334 | { |
0b246afa | 3335 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3336 | u64 lower_gen; |
5f39d397 CM |
3337 | struct extent_buffer *lower; |
3338 | struct extent_buffer *c; | |
925baedd | 3339 | struct extent_buffer *old; |
5f39d397 | 3340 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3341 | int ret; |
5c680ed6 CM |
3342 | |
3343 | BUG_ON(path->nodes[level]); | |
3344 | BUG_ON(path->nodes[level-1] != root->node); | |
3345 | ||
7bb86316 CM |
3346 | lower = path->nodes[level-1]; |
3347 | if (level == 1) | |
3348 | btrfs_item_key(lower, &lower_key, 0); | |
3349 | else | |
3350 | btrfs_node_key(lower, &lower_key, 0); | |
3351 | ||
4d75f8a9 DS |
3352 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3353 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3354 | if (IS_ERR(c)) |
3355 | return PTR_ERR(c); | |
925baedd | 3356 | |
0b246afa | 3357 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3358 | |
b159fa28 | 3359 | memzero_extent_buffer(c, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3360 | btrfs_set_header_nritems(c, 1); |
3361 | btrfs_set_header_level(c, level); | |
db94535d | 3362 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3363 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3364 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3365 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3366 | |
0b246afa JM |
3367 | write_extent_buffer_fsid(c, fs_info->fsid); |
3368 | write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid); | |
e17cade2 | 3369 | |
5f39d397 | 3370 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3371 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3372 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3373 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3374 | |
3375 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3376 | |
5f39d397 | 3377 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3378 | |
925baedd | 3379 | old = root->node; |
d9d19a01 DS |
3380 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3381 | BUG_ON(ret < 0); | |
240f62c8 | 3382 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3383 | |
3384 | /* the super has an extra ref to root->node */ | |
3385 | free_extent_buffer(old); | |
3386 | ||
0b86a832 | 3387 | add_root_to_dirty_list(root); |
5f39d397 CM |
3388 | extent_buffer_get(c); |
3389 | path->nodes[level] = c; | |
95449a16 | 3390 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3391 | path->slots[level] = 0; |
3392 | return 0; | |
3393 | } | |
3394 | ||
74123bd7 CM |
3395 | /* |
3396 | * worker function to insert a single pointer in a node. | |
3397 | * the node should have enough room for the pointer already | |
97571fd0 | 3398 | * |
74123bd7 CM |
3399 | * slot and level indicate where you want the key to go, and |
3400 | * blocknr is the block the key points to. | |
3401 | */ | |
143bede5 | 3402 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2ff7e61e | 3403 | struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 3404 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3405 | int slot, int level) |
74123bd7 | 3406 | { |
5f39d397 | 3407 | struct extent_buffer *lower; |
74123bd7 | 3408 | int nritems; |
f3ea38da | 3409 | int ret; |
5c680ed6 CM |
3410 | |
3411 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3412 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3413 | lower = path->nodes[level]; |
3414 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3415 | BUG_ON(slot > nritems); |
0b246afa | 3416 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3417 | if (slot != nritems) { |
bf1d3425 DS |
3418 | if (level) { |
3419 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3420 | nritems - slot); |
bf1d3425 DS |
3421 | BUG_ON(ret < 0); |
3422 | } | |
5f39d397 CM |
3423 | memmove_extent_buffer(lower, |
3424 | btrfs_node_key_ptr_offset(slot + 1), | |
3425 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3426 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3427 | } |
c3e06965 | 3428 | if (level) { |
e09c2efe DS |
3429 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3430 | GFP_NOFS); | |
f3ea38da JS |
3431 | BUG_ON(ret < 0); |
3432 | } | |
5f39d397 | 3433 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3434 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3435 | WARN_ON(trans->transid == 0); |
3436 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3437 | btrfs_set_header_nritems(lower, nritems + 1); |
3438 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3439 | } |
3440 | ||
97571fd0 CM |
3441 | /* |
3442 | * split the node at the specified level in path in two. | |
3443 | * The path is corrected to point to the appropriate node after the split | |
3444 | * | |
3445 | * Before splitting this tries to make some room in the node by pushing | |
3446 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3447 | * |
3448 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3449 | */ |
e02119d5 CM |
3450 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3451 | struct btrfs_root *root, | |
3452 | struct btrfs_path *path, int level) | |
be0e5c09 | 3453 | { |
0b246afa | 3454 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3455 | struct extent_buffer *c; |
3456 | struct extent_buffer *split; | |
3457 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3458 | int mid; |
5c680ed6 | 3459 | int ret; |
7518a238 | 3460 | u32 c_nritems; |
eb60ceac | 3461 | |
5f39d397 | 3462 | c = path->nodes[level]; |
7bb86316 | 3463 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3464 | if (c == root->node) { |
d9abbf1c | 3465 | /* |
90f8d62e JS |
3466 | * trying to split the root, lets make a new one |
3467 | * | |
fdd99c72 | 3468 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3469 | * insert_new_root, because that root buffer will be kept as a |
3470 | * normal node. We are going to log removal of half of the | |
3471 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3472 | * tree lock on the buffer, which is why we cannot race with | |
3473 | * other tree_mod_log users. | |
d9abbf1c | 3474 | */ |
fdd99c72 | 3475 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3476 | if (ret) |
3477 | return ret; | |
b3612421 | 3478 | } else { |
e66f709b | 3479 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3480 | c = path->nodes[level]; |
3481 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3482 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3483 | return 0; |
54aa1f4d CM |
3484 | if (ret < 0) |
3485 | return ret; | |
be0e5c09 | 3486 | } |
e66f709b | 3487 | |
5f39d397 | 3488 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3489 | mid = (c_nritems + 1) / 2; |
3490 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3491 | |
4d75f8a9 DS |
3492 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3493 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3494 | if (IS_ERR(split)) |
3495 | return PTR_ERR(split); | |
3496 | ||
0b246afa | 3497 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3498 | |
b159fa28 | 3499 | memzero_extent_buffer(split, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3500 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3501 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3502 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3503 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3504 | btrfs_set_header_owner(split, root->root_key.objectid); |
0b246afa JM |
3505 | write_extent_buffer_fsid(split, fs_info->fsid); |
3506 | write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid); | |
54aa1f4d | 3507 | |
0b246afa | 3508 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3509 | if (ret) { |
66642832 | 3510 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3511 | return ret; |
3512 | } | |
5f39d397 CM |
3513 | copy_extent_buffer(split, c, |
3514 | btrfs_node_key_ptr_offset(0), | |
3515 | btrfs_node_key_ptr_offset(mid), | |
3516 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3517 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3518 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3519 | ret = 0; |
3520 | ||
5f39d397 CM |
3521 | btrfs_mark_buffer_dirty(c); |
3522 | btrfs_mark_buffer_dirty(split); | |
3523 | ||
2ff7e61e | 3524 | insert_ptr(trans, fs_info, path, &disk_key, split->start, |
c3e06965 | 3525 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3526 | |
5de08d7d | 3527 | if (path->slots[level] >= mid) { |
5c680ed6 | 3528 | path->slots[level] -= mid; |
925baedd | 3529 | btrfs_tree_unlock(c); |
5f39d397 CM |
3530 | free_extent_buffer(c); |
3531 | path->nodes[level] = split; | |
5c680ed6 CM |
3532 | path->slots[level + 1] += 1; |
3533 | } else { | |
925baedd | 3534 | btrfs_tree_unlock(split); |
5f39d397 | 3535 | free_extent_buffer(split); |
be0e5c09 | 3536 | } |
aa5d6bed | 3537 | return ret; |
be0e5c09 CM |
3538 | } |
3539 | ||
74123bd7 CM |
3540 | /* |
3541 | * how many bytes are required to store the items in a leaf. start | |
3542 | * and nr indicate which items in the leaf to check. This totals up the | |
3543 | * space used both by the item structs and the item data | |
3544 | */ | |
5f39d397 | 3545 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3546 | { |
41be1f3b JB |
3547 | struct btrfs_item *start_item; |
3548 | struct btrfs_item *end_item; | |
3549 | struct btrfs_map_token token; | |
be0e5c09 | 3550 | int data_len; |
5f39d397 | 3551 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3552 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3553 | |
3554 | if (!nr) | |
3555 | return 0; | |
41be1f3b | 3556 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3557 | start_item = btrfs_item_nr(start); |
3558 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3559 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3560 | btrfs_token_item_size(l, start_item, &token); | |
3561 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3562 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3563 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3564 | return data_len; |
3565 | } | |
3566 | ||
d4dbff95 CM |
3567 | /* |
3568 | * The space between the end of the leaf items and | |
3569 | * the start of the leaf data. IOW, how much room | |
3570 | * the leaf has left for both items and data | |
3571 | */ | |
2ff7e61e | 3572 | noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info, |
e02119d5 | 3573 | struct extent_buffer *leaf) |
d4dbff95 | 3574 | { |
5f39d397 CM |
3575 | int nritems = btrfs_header_nritems(leaf); |
3576 | int ret; | |
0b246afa JM |
3577 | |
3578 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3579 | if (ret < 0) { |
0b246afa JM |
3580 | btrfs_crit(fs_info, |
3581 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3582 | ret, | |
3583 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3584 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3585 | } |
3586 | return ret; | |
d4dbff95 CM |
3587 | } |
3588 | ||
99d8f83c CM |
3589 | /* |
3590 | * min slot controls the lowest index we're willing to push to the | |
3591 | * right. We'll push up to and including min_slot, but no lower | |
3592 | */ | |
1e47eef2 | 3593 | static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3594 | struct btrfs_path *path, |
3595 | int data_size, int empty, | |
3596 | struct extent_buffer *right, | |
99d8f83c CM |
3597 | int free_space, u32 left_nritems, |
3598 | u32 min_slot) | |
00ec4c51 | 3599 | { |
5f39d397 | 3600 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3601 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3602 | struct btrfs_map_token token; |
5f39d397 | 3603 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3604 | int slot; |
34a38218 | 3605 | u32 i; |
00ec4c51 CM |
3606 | int push_space = 0; |
3607 | int push_items = 0; | |
0783fcfc | 3608 | struct btrfs_item *item; |
34a38218 | 3609 | u32 nr; |
7518a238 | 3610 | u32 right_nritems; |
5f39d397 | 3611 | u32 data_end; |
db94535d | 3612 | u32 this_item_size; |
00ec4c51 | 3613 | |
cfed81a0 CM |
3614 | btrfs_init_map_token(&token); |
3615 | ||
34a38218 CM |
3616 | if (empty) |
3617 | nr = 0; | |
3618 | else | |
99d8f83c | 3619 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3620 | |
31840ae1 | 3621 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3622 | push_space += data_size; |
31840ae1 | 3623 | |
44871b1b | 3624 | slot = path->slots[1]; |
34a38218 CM |
3625 | i = left_nritems - 1; |
3626 | while (i >= nr) { | |
dd3cc16b | 3627 | item = btrfs_item_nr(i); |
db94535d | 3628 | |
31840ae1 ZY |
3629 | if (!empty && push_items > 0) { |
3630 | if (path->slots[0] > i) | |
3631 | break; | |
3632 | if (path->slots[0] == i) { | |
2ff7e61e | 3633 | int space = btrfs_leaf_free_space(fs_info, left); |
31840ae1 ZY |
3634 | if (space + push_space * 2 > free_space) |
3635 | break; | |
3636 | } | |
3637 | } | |
3638 | ||
00ec4c51 | 3639 | if (path->slots[0] == i) |
87b29b20 | 3640 | push_space += data_size; |
db94535d | 3641 | |
db94535d CM |
3642 | this_item_size = btrfs_item_size(left, item); |
3643 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3644 | break; |
31840ae1 | 3645 | |
00ec4c51 | 3646 | push_items++; |
db94535d | 3647 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3648 | if (i == 0) |
3649 | break; | |
3650 | i--; | |
db94535d | 3651 | } |
5f39d397 | 3652 | |
925baedd CM |
3653 | if (push_items == 0) |
3654 | goto out_unlock; | |
5f39d397 | 3655 | |
6c1500f2 | 3656 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3657 | |
00ec4c51 | 3658 | /* push left to right */ |
5f39d397 | 3659 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3660 | |
5f39d397 | 3661 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
2ff7e61e | 3662 | push_space -= leaf_data_end(fs_info, left); |
5f39d397 | 3663 | |
00ec4c51 | 3664 | /* make room in the right data area */ |
2ff7e61e | 3665 | data_end = leaf_data_end(fs_info, right); |
5f39d397 | 3666 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3667 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3668 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3669 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3670 | |
00ec4c51 | 3671 | /* copy from the left data area */ |
3d9ec8c4 | 3672 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3673 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3674 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left), |
d6025579 | 3675 | push_space); |
5f39d397 CM |
3676 | |
3677 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3678 | btrfs_item_nr_offset(0), | |
3679 | right_nritems * sizeof(struct btrfs_item)); | |
3680 | ||
00ec4c51 | 3681 | /* copy the items from left to right */ |
5f39d397 CM |
3682 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3683 | btrfs_item_nr_offset(left_nritems - push_items), | |
3684 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3685 | |
3686 | /* update the item pointers */ | |
7518a238 | 3687 | right_nritems += push_items; |
5f39d397 | 3688 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3689 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3690 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3691 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3692 | push_space -= btrfs_token_item_size(right, item, &token); |
3693 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3694 | } |
3695 | ||
7518a238 | 3696 | left_nritems -= push_items; |
5f39d397 | 3697 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3698 | |
34a38218 CM |
3699 | if (left_nritems) |
3700 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3701 | else |
7c302b49 | 3702 | clean_tree_block(fs_info, left); |
f0486c68 | 3703 | |
5f39d397 | 3704 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3705 | |
5f39d397 CM |
3706 | btrfs_item_key(right, &disk_key, 0); |
3707 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3708 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3709 | |
00ec4c51 | 3710 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3711 | if (path->slots[0] >= left_nritems) { |
3712 | path->slots[0] -= left_nritems; | |
925baedd | 3713 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
7c302b49 | 3714 | clean_tree_block(fs_info, path->nodes[0]); |
925baedd | 3715 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3716 | free_extent_buffer(path->nodes[0]); |
3717 | path->nodes[0] = right; | |
00ec4c51 CM |
3718 | path->slots[1] += 1; |
3719 | } else { | |
925baedd | 3720 | btrfs_tree_unlock(right); |
5f39d397 | 3721 | free_extent_buffer(right); |
00ec4c51 CM |
3722 | } |
3723 | return 0; | |
925baedd CM |
3724 | |
3725 | out_unlock: | |
3726 | btrfs_tree_unlock(right); | |
3727 | free_extent_buffer(right); | |
3728 | return 1; | |
00ec4c51 | 3729 | } |
925baedd | 3730 | |
44871b1b CM |
3731 | /* |
3732 | * push some data in the path leaf to the right, trying to free up at | |
3733 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3734 | * | |
3735 | * returns 1 if the push failed because the other node didn't have enough | |
3736 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3737 | * |
3738 | * this will push starting from min_slot to the end of the leaf. It won't | |
3739 | * push any slot lower than min_slot | |
44871b1b CM |
3740 | */ |
3741 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3742 | *root, struct btrfs_path *path, |
3743 | int min_data_size, int data_size, | |
3744 | int empty, u32 min_slot) | |
44871b1b | 3745 | { |
2ff7e61e | 3746 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3747 | struct extent_buffer *left = path->nodes[0]; |
3748 | struct extent_buffer *right; | |
3749 | struct extent_buffer *upper; | |
3750 | int slot; | |
3751 | int free_space; | |
3752 | u32 left_nritems; | |
3753 | int ret; | |
3754 | ||
3755 | if (!path->nodes[1]) | |
3756 | return 1; | |
3757 | ||
3758 | slot = path->slots[1]; | |
3759 | upper = path->nodes[1]; | |
3760 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3761 | return 1; | |
3762 | ||
3763 | btrfs_assert_tree_locked(path->nodes[1]); | |
3764 | ||
2ff7e61e | 3765 | right = read_node_slot(fs_info, upper, slot + 1); |
fb770ae4 LB |
3766 | /* |
3767 | * slot + 1 is not valid or we fail to read the right node, | |
3768 | * no big deal, just return. | |
3769 | */ | |
3770 | if (IS_ERR(right)) | |
91ca338d TI |
3771 | return 1; |
3772 | ||
44871b1b CM |
3773 | btrfs_tree_lock(right); |
3774 | btrfs_set_lock_blocking(right); | |
3775 | ||
2ff7e61e | 3776 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3777 | if (free_space < data_size) |
3778 | goto out_unlock; | |
3779 | ||
3780 | /* cow and double check */ | |
3781 | ret = btrfs_cow_block(trans, root, right, upper, | |
3782 | slot + 1, &right); | |
3783 | if (ret) | |
3784 | goto out_unlock; | |
3785 | ||
2ff7e61e | 3786 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3787 | if (free_space < data_size) |
3788 | goto out_unlock; | |
3789 | ||
3790 | left_nritems = btrfs_header_nritems(left); | |
3791 | if (left_nritems == 0) | |
3792 | goto out_unlock; | |
3793 | ||
2ef1fed2 FDBM |
3794 | if (path->slots[0] == left_nritems && !empty) { |
3795 | /* Key greater than all keys in the leaf, right neighbor has | |
3796 | * enough room for it and we're not emptying our leaf to delete | |
3797 | * it, therefore use right neighbor to insert the new item and | |
3798 | * no need to touch/dirty our left leaft. */ | |
3799 | btrfs_tree_unlock(left); | |
3800 | free_extent_buffer(left); | |
3801 | path->nodes[0] = right; | |
3802 | path->slots[0] = 0; | |
3803 | path->slots[1]++; | |
3804 | return 0; | |
3805 | } | |
3806 | ||
1e47eef2 | 3807 | return __push_leaf_right(fs_info, path, min_data_size, empty, |
99d8f83c | 3808 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3809 | out_unlock: |
3810 | btrfs_tree_unlock(right); | |
3811 | free_extent_buffer(right); | |
3812 | return 1; | |
3813 | } | |
3814 | ||
74123bd7 CM |
3815 | /* |
3816 | * push some data in the path leaf to the left, trying to free up at | |
3817 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3818 | * |
3819 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3820 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3821 | * items | |
74123bd7 | 3822 | */ |
66cb7ddb | 3823 | static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3824 | struct btrfs_path *path, int data_size, |
3825 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3826 | int free_space, u32 right_nritems, |
3827 | u32 max_slot) | |
be0e5c09 | 3828 | { |
5f39d397 CM |
3829 | struct btrfs_disk_key disk_key; |
3830 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3831 | int i; |
be0e5c09 CM |
3832 | int push_space = 0; |
3833 | int push_items = 0; | |
0783fcfc | 3834 | struct btrfs_item *item; |
7518a238 | 3835 | u32 old_left_nritems; |
34a38218 | 3836 | u32 nr; |
aa5d6bed | 3837 | int ret = 0; |
db94535d CM |
3838 | u32 this_item_size; |
3839 | u32 old_left_item_size; | |
cfed81a0 CM |
3840 | struct btrfs_map_token token; |
3841 | ||
3842 | btrfs_init_map_token(&token); | |
be0e5c09 | 3843 | |
34a38218 | 3844 | if (empty) |
99d8f83c | 3845 | nr = min(right_nritems, max_slot); |
34a38218 | 3846 | else |
99d8f83c | 3847 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3848 | |
3849 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3850 | item = btrfs_item_nr(i); |
db94535d | 3851 | |
31840ae1 ZY |
3852 | if (!empty && push_items > 0) { |
3853 | if (path->slots[0] < i) | |
3854 | break; | |
3855 | if (path->slots[0] == i) { | |
2ff7e61e | 3856 | int space = btrfs_leaf_free_space(fs_info, right); |
31840ae1 ZY |
3857 | if (space + push_space * 2 > free_space) |
3858 | break; | |
3859 | } | |
3860 | } | |
3861 | ||
be0e5c09 | 3862 | if (path->slots[0] == i) |
87b29b20 | 3863 | push_space += data_size; |
db94535d CM |
3864 | |
3865 | this_item_size = btrfs_item_size(right, item); | |
3866 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3867 | break; |
db94535d | 3868 | |
be0e5c09 | 3869 | push_items++; |
db94535d CM |
3870 | push_space += this_item_size + sizeof(*item); |
3871 | } | |
3872 | ||
be0e5c09 | 3873 | if (push_items == 0) { |
925baedd CM |
3874 | ret = 1; |
3875 | goto out; | |
be0e5c09 | 3876 | } |
fae7f21c | 3877 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3878 | |
be0e5c09 | 3879 | /* push data from right to left */ |
5f39d397 CM |
3880 | copy_extent_buffer(left, right, |
3881 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3882 | btrfs_item_nr_offset(0), | |
3883 | push_items * sizeof(struct btrfs_item)); | |
3884 | ||
0b246afa | 3885 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3886 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3887 | |
3d9ec8c4 | 3888 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3889 | leaf_data_end(fs_info, left) - push_space, |
3d9ec8c4 | 3890 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3891 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3892 | push_space); |
5f39d397 | 3893 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3894 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3895 | |
db94535d | 3896 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3897 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3898 | u32 ioff; |
db94535d | 3899 | |
dd3cc16b | 3900 | item = btrfs_item_nr(i); |
db94535d | 3901 | |
cfed81a0 CM |
3902 | ioff = btrfs_token_item_offset(left, item, &token); |
3903 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3904 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3905 | &token); |
be0e5c09 | 3906 | } |
5f39d397 | 3907 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3908 | |
3909 | /* fixup right node */ | |
31b1a2bd JL |
3910 | if (push_items > right_nritems) |
3911 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3912 | right_nritems); |
34a38218 CM |
3913 | |
3914 | if (push_items < right_nritems) { | |
3915 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2ff7e61e | 3916 | leaf_data_end(fs_info, right); |
3d9ec8c4 | 3917 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3918 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3919 | BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3920 | leaf_data_end(fs_info, right), push_space); |
34a38218 CM |
3921 | |
3922 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3923 | btrfs_item_nr_offset(push_items), |
3924 | (btrfs_header_nritems(right) - push_items) * | |
3925 | sizeof(struct btrfs_item)); | |
34a38218 | 3926 | } |
eef1c494 Y |
3927 | right_nritems -= push_items; |
3928 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3929 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3930 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3931 | item = btrfs_item_nr(i); |
db94535d | 3932 | |
cfed81a0 CM |
3933 | push_space = push_space - btrfs_token_item_size(right, |
3934 | item, &token); | |
3935 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3936 | } |
eb60ceac | 3937 | |
5f39d397 | 3938 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3939 | if (right_nritems) |
3940 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3941 | else |
7c302b49 | 3942 | clean_tree_block(fs_info, right); |
098f59c2 | 3943 | |
5f39d397 | 3944 | btrfs_item_key(right, &disk_key, 0); |
0b246afa | 3945 | fixup_low_keys(fs_info, path, &disk_key, 1); |
be0e5c09 CM |
3946 | |
3947 | /* then fixup the leaf pointer in the path */ | |
3948 | if (path->slots[0] < push_items) { | |
3949 | path->slots[0] += old_left_nritems; | |
925baedd | 3950 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3951 | free_extent_buffer(path->nodes[0]); |
3952 | path->nodes[0] = left; | |
be0e5c09 CM |
3953 | path->slots[1] -= 1; |
3954 | } else { | |
925baedd | 3955 | btrfs_tree_unlock(left); |
5f39d397 | 3956 | free_extent_buffer(left); |
be0e5c09 CM |
3957 | path->slots[0] -= push_items; |
3958 | } | |
eb60ceac | 3959 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3960 | return ret; |
925baedd CM |
3961 | out: |
3962 | btrfs_tree_unlock(left); | |
3963 | free_extent_buffer(left); | |
3964 | return ret; | |
be0e5c09 CM |
3965 | } |
3966 | ||
44871b1b CM |
3967 | /* |
3968 | * push some data in the path leaf to the left, trying to free up at | |
3969 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3970 | * |
3971 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3972 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3973 | * items | |
44871b1b CM |
3974 | */ |
3975 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3976 | *root, struct btrfs_path *path, int min_data_size, |
3977 | int data_size, int empty, u32 max_slot) | |
44871b1b | 3978 | { |
2ff7e61e | 3979 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3980 | struct extent_buffer *right = path->nodes[0]; |
3981 | struct extent_buffer *left; | |
3982 | int slot; | |
3983 | int free_space; | |
3984 | u32 right_nritems; | |
3985 | int ret = 0; | |
3986 | ||
3987 | slot = path->slots[1]; | |
3988 | if (slot == 0) | |
3989 | return 1; | |
3990 | if (!path->nodes[1]) | |
3991 | return 1; | |
3992 | ||
3993 | right_nritems = btrfs_header_nritems(right); | |
3994 | if (right_nritems == 0) | |
3995 | return 1; | |
3996 | ||
3997 | btrfs_assert_tree_locked(path->nodes[1]); | |
3998 | ||
2ff7e61e | 3999 | left = read_node_slot(fs_info, path->nodes[1], slot - 1); |
fb770ae4 LB |
4000 | /* |
4001 | * slot - 1 is not valid or we fail to read the left node, | |
4002 | * no big deal, just return. | |
4003 | */ | |
4004 | if (IS_ERR(left)) | |
91ca338d TI |
4005 | return 1; |
4006 | ||
44871b1b CM |
4007 | btrfs_tree_lock(left); |
4008 | btrfs_set_lock_blocking(left); | |
4009 | ||
2ff7e61e | 4010 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4011 | if (free_space < data_size) { |
4012 | ret = 1; | |
4013 | goto out; | |
4014 | } | |
4015 | ||
4016 | /* cow and double check */ | |
4017 | ret = btrfs_cow_block(trans, root, left, | |
4018 | path->nodes[1], slot - 1, &left); | |
4019 | if (ret) { | |
4020 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4021 | if (ret == -ENOSPC) |
4022 | ret = 1; | |
44871b1b CM |
4023 | goto out; |
4024 | } | |
4025 | ||
2ff7e61e | 4026 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4027 | if (free_space < data_size) { |
4028 | ret = 1; | |
4029 | goto out; | |
4030 | } | |
4031 | ||
66cb7ddb | 4032 | return __push_leaf_left(fs_info, path, min_data_size, |
99d8f83c CM |
4033 | empty, left, free_space, right_nritems, |
4034 | max_slot); | |
44871b1b CM |
4035 | out: |
4036 | btrfs_tree_unlock(left); | |
4037 | free_extent_buffer(left); | |
4038 | return ret; | |
4039 | } | |
4040 | ||
4041 | /* | |
4042 | * split the path's leaf in two, making sure there is at least data_size | |
4043 | * available for the resulting leaf level of the path. | |
44871b1b | 4044 | */ |
143bede5 | 4045 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2ff7e61e | 4046 | struct btrfs_fs_info *fs_info, |
143bede5 JM |
4047 | struct btrfs_path *path, |
4048 | struct extent_buffer *l, | |
4049 | struct extent_buffer *right, | |
4050 | int slot, int mid, int nritems) | |
44871b1b CM |
4051 | { |
4052 | int data_copy_size; | |
4053 | int rt_data_off; | |
4054 | int i; | |
44871b1b | 4055 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4056 | struct btrfs_map_token token; |
4057 | ||
4058 | btrfs_init_map_token(&token); | |
44871b1b CM |
4059 | |
4060 | nritems = nritems - mid; | |
4061 | btrfs_set_header_nritems(right, nritems); | |
2ff7e61e | 4062 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l); |
44871b1b CM |
4063 | |
4064 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4065 | btrfs_item_nr_offset(mid), | |
4066 | nritems * sizeof(struct btrfs_item)); | |
4067 | ||
4068 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4069 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4070 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
2ff7e61e | 4071 | leaf_data_end(fs_info, l), data_copy_size); |
44871b1b | 4072 | |
0b246afa | 4073 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4074 | |
4075 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4076 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4077 | u32 ioff; |
4078 | ||
cfed81a0 CM |
4079 | ioff = btrfs_token_item_offset(right, item, &token); |
4080 | btrfs_set_token_item_offset(right, item, | |
4081 | ioff + rt_data_off, &token); | |
44871b1b CM |
4082 | } |
4083 | ||
44871b1b | 4084 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4085 | btrfs_item_key(right, &disk_key, 0); |
2ff7e61e | 4086 | insert_ptr(trans, fs_info, path, &disk_key, right->start, |
c3e06965 | 4087 | path->slots[1] + 1, 1); |
44871b1b CM |
4088 | |
4089 | btrfs_mark_buffer_dirty(right); | |
4090 | btrfs_mark_buffer_dirty(l); | |
4091 | BUG_ON(path->slots[0] != slot); | |
4092 | ||
44871b1b CM |
4093 | if (mid <= slot) { |
4094 | btrfs_tree_unlock(path->nodes[0]); | |
4095 | free_extent_buffer(path->nodes[0]); | |
4096 | path->nodes[0] = right; | |
4097 | path->slots[0] -= mid; | |
4098 | path->slots[1] += 1; | |
4099 | } else { | |
4100 | btrfs_tree_unlock(right); | |
4101 | free_extent_buffer(right); | |
4102 | } | |
4103 | ||
4104 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4105 | } |
4106 | ||
99d8f83c CM |
4107 | /* |
4108 | * double splits happen when we need to insert a big item in the middle | |
4109 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4110 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4111 | * A B C | |
4112 | * | |
4113 | * We avoid this by trying to push the items on either side of our target | |
4114 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4115 | * completely. | |
4116 | */ | |
4117 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4118 | struct btrfs_root *root, | |
4119 | struct btrfs_path *path, | |
4120 | int data_size) | |
4121 | { | |
2ff7e61e | 4122 | struct btrfs_fs_info *fs_info = root->fs_info; |
99d8f83c CM |
4123 | int ret; |
4124 | int progress = 0; | |
4125 | int slot; | |
4126 | u32 nritems; | |
5a4267ca | 4127 | int space_needed = data_size; |
99d8f83c CM |
4128 | |
4129 | slot = path->slots[0]; | |
5a4267ca | 4130 | if (slot < btrfs_header_nritems(path->nodes[0])) |
2ff7e61e | 4131 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); |
99d8f83c CM |
4132 | |
4133 | /* | |
4134 | * try to push all the items after our slot into the | |
4135 | * right leaf | |
4136 | */ | |
5a4267ca | 4137 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4138 | if (ret < 0) |
4139 | return ret; | |
4140 | ||
4141 | if (ret == 0) | |
4142 | progress++; | |
4143 | ||
4144 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4145 | /* | |
4146 | * our goal is to get our slot at the start or end of a leaf. If | |
4147 | * we've done so we're done | |
4148 | */ | |
4149 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4150 | return 0; | |
4151 | ||
2ff7e61e | 4152 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4153 | return 0; |
4154 | ||
4155 | /* try to push all the items before our slot into the next leaf */ | |
4156 | slot = path->slots[0]; | |
263d3995 FM |
4157 | space_needed = data_size; |
4158 | if (slot > 0) | |
4159 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); | |
5a4267ca | 4160 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4161 | if (ret < 0) |
4162 | return ret; | |
4163 | ||
4164 | if (ret == 0) | |
4165 | progress++; | |
4166 | ||
4167 | if (progress) | |
4168 | return 0; | |
4169 | return 1; | |
4170 | } | |
4171 | ||
74123bd7 CM |
4172 | /* |
4173 | * split the path's leaf in two, making sure there is at least data_size | |
4174 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4175 | * |
4176 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4177 | */ |
e02119d5 CM |
4178 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4179 | struct btrfs_root *root, | |
310712b2 | 4180 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4181 | struct btrfs_path *path, int data_size, |
4182 | int extend) | |
be0e5c09 | 4183 | { |
5d4f98a2 | 4184 | struct btrfs_disk_key disk_key; |
5f39d397 | 4185 | struct extent_buffer *l; |
7518a238 | 4186 | u32 nritems; |
eb60ceac CM |
4187 | int mid; |
4188 | int slot; | |
5f39d397 | 4189 | struct extent_buffer *right; |
b7a0365e | 4190 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4191 | int ret = 0; |
aa5d6bed | 4192 | int wret; |
5d4f98a2 | 4193 | int split; |
cc0c5538 | 4194 | int num_doubles = 0; |
99d8f83c | 4195 | int tried_avoid_double = 0; |
aa5d6bed | 4196 | |
a5719521 YZ |
4197 | l = path->nodes[0]; |
4198 | slot = path->slots[0]; | |
4199 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4200 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4201 | return -EOVERFLOW; |
4202 | ||
40689478 | 4203 | /* first try to make some room by pushing left and right */ |
33157e05 | 4204 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4205 | int space_needed = data_size; |
4206 | ||
4207 | if (slot < btrfs_header_nritems(l)) | |
2ff7e61e | 4208 | space_needed -= btrfs_leaf_free_space(fs_info, l); |
5a4267ca FDBM |
4209 | |
4210 | wret = push_leaf_right(trans, root, path, space_needed, | |
4211 | space_needed, 0, 0); | |
d397712b | 4212 | if (wret < 0) |
eaee50e8 | 4213 | return wret; |
3685f791 | 4214 | if (wret) { |
263d3995 FM |
4215 | space_needed = data_size; |
4216 | if (slot > 0) | |
4217 | space_needed -= btrfs_leaf_free_space(fs_info, | |
4218 | l); | |
5a4267ca FDBM |
4219 | wret = push_leaf_left(trans, root, path, space_needed, |
4220 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4221 | if (wret < 0) |
4222 | return wret; | |
4223 | } | |
4224 | l = path->nodes[0]; | |
aa5d6bed | 4225 | |
3685f791 | 4226 | /* did the pushes work? */ |
2ff7e61e | 4227 | if (btrfs_leaf_free_space(fs_info, l) >= data_size) |
3685f791 | 4228 | return 0; |
3326d1b0 | 4229 | } |
aa5d6bed | 4230 | |
5c680ed6 | 4231 | if (!path->nodes[1]) { |
fdd99c72 | 4232 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4233 | if (ret) |
4234 | return ret; | |
4235 | } | |
cc0c5538 | 4236 | again: |
5d4f98a2 | 4237 | split = 1; |
cc0c5538 | 4238 | l = path->nodes[0]; |
eb60ceac | 4239 | slot = path->slots[0]; |
5f39d397 | 4240 | nritems = btrfs_header_nritems(l); |
d397712b | 4241 | mid = (nritems + 1) / 2; |
54aa1f4d | 4242 | |
5d4f98a2 YZ |
4243 | if (mid <= slot) { |
4244 | if (nritems == 1 || | |
4245 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4246 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4247 | if (slot >= nritems) { |
4248 | split = 0; | |
4249 | } else { | |
4250 | mid = slot; | |
4251 | if (mid != nritems && | |
4252 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4253 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4254 | if (data_size && !tried_avoid_double) |
4255 | goto push_for_double; | |
5d4f98a2 YZ |
4256 | split = 2; |
4257 | } | |
4258 | } | |
4259 | } | |
4260 | } else { | |
4261 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4262 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4263 | if (!extend && data_size && slot == 0) { |
4264 | split = 0; | |
4265 | } else if ((extend || !data_size) && slot == 0) { | |
4266 | mid = 1; | |
4267 | } else { | |
4268 | mid = slot; | |
4269 | if (mid != nritems && | |
4270 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4271 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4272 | if (data_size && !tried_avoid_double) |
4273 | goto push_for_double; | |
67871254 | 4274 | split = 2; |
5d4f98a2 YZ |
4275 | } |
4276 | } | |
4277 | } | |
4278 | } | |
4279 | ||
4280 | if (split == 0) | |
4281 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4282 | else | |
4283 | btrfs_item_key(l, &disk_key, mid); | |
4284 | ||
4d75f8a9 DS |
4285 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4286 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4287 | if (IS_ERR(right)) |
5f39d397 | 4288 | return PTR_ERR(right); |
f0486c68 | 4289 | |
0b246afa | 4290 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4291 | |
b159fa28 | 4292 | memzero_extent_buffer(right, 0, sizeof(struct btrfs_header)); |
db94535d | 4293 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4294 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4295 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4296 | btrfs_set_header_owner(right, root->root_key.objectid); |
4297 | btrfs_set_header_level(right, 0); | |
d24ee97b DS |
4298 | write_extent_buffer_fsid(right, fs_info->fsid); |
4299 | write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid); | |
44871b1b | 4300 | |
5d4f98a2 YZ |
4301 | if (split == 0) { |
4302 | if (mid <= slot) { | |
4303 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4304 | insert_ptr(trans, fs_info, path, &disk_key, |
4305 | right->start, path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
4306 | btrfs_tree_unlock(path->nodes[0]); |
4307 | free_extent_buffer(path->nodes[0]); | |
4308 | path->nodes[0] = right; | |
4309 | path->slots[0] = 0; | |
4310 | path->slots[1] += 1; | |
4311 | } else { | |
4312 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4313 | insert_ptr(trans, fs_info, path, &disk_key, |
4314 | right->start, path->slots[1], 1); | |
5d4f98a2 YZ |
4315 | btrfs_tree_unlock(path->nodes[0]); |
4316 | free_extent_buffer(path->nodes[0]); | |
4317 | path->nodes[0] = right; | |
4318 | path->slots[0] = 0; | |
143bede5 | 4319 | if (path->slots[1] == 0) |
b7a0365e | 4320 | fixup_low_keys(fs_info, path, &disk_key, 1); |
d4dbff95 | 4321 | } |
196e0249 LB |
4322 | /* |
4323 | * We create a new leaf 'right' for the required ins_len and | |
4324 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4325 | * the content of ins_len to 'right'. | |
4326 | */ | |
5d4f98a2 | 4327 | return ret; |
d4dbff95 | 4328 | } |
74123bd7 | 4329 | |
2ff7e61e | 4330 | copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems); |
31840ae1 | 4331 | |
5d4f98a2 | 4332 | if (split == 2) { |
cc0c5538 CM |
4333 | BUG_ON(num_doubles != 0); |
4334 | num_doubles++; | |
4335 | goto again; | |
a429e513 | 4336 | } |
44871b1b | 4337 | |
143bede5 | 4338 | return 0; |
99d8f83c CM |
4339 | |
4340 | push_for_double: | |
4341 | push_for_double_split(trans, root, path, data_size); | |
4342 | tried_avoid_double = 1; | |
2ff7e61e | 4343 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4344 | return 0; |
4345 | goto again; | |
be0e5c09 CM |
4346 | } |
4347 | ||
ad48fd75 YZ |
4348 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4349 | struct btrfs_root *root, | |
4350 | struct btrfs_path *path, int ins_len) | |
459931ec | 4351 | { |
2ff7e61e | 4352 | struct btrfs_fs_info *fs_info = root->fs_info; |
ad48fd75 | 4353 | struct btrfs_key key; |
459931ec | 4354 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4355 | struct btrfs_file_extent_item *fi; |
4356 | u64 extent_len = 0; | |
4357 | u32 item_size; | |
4358 | int ret; | |
459931ec CM |
4359 | |
4360 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4361 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4362 | ||
4363 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4364 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4365 | ||
2ff7e61e | 4366 | if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len) |
ad48fd75 | 4367 | return 0; |
459931ec CM |
4368 | |
4369 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4370 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4371 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4372 | struct btrfs_file_extent_item); | |
4373 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4374 | } | |
b3b4aa74 | 4375 | btrfs_release_path(path); |
459931ec | 4376 | |
459931ec | 4377 | path->keep_locks = 1; |
ad48fd75 YZ |
4378 | path->search_for_split = 1; |
4379 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4380 | path->search_for_split = 0; |
a8df6fe6 FM |
4381 | if (ret > 0) |
4382 | ret = -EAGAIN; | |
ad48fd75 YZ |
4383 | if (ret < 0) |
4384 | goto err; | |
459931ec | 4385 | |
ad48fd75 YZ |
4386 | ret = -EAGAIN; |
4387 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4388 | /* if our item isn't there, return now */ |
4389 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4390 | goto err; |
4391 | ||
109f6aef | 4392 | /* the leaf has changed, it now has room. return now */ |
2ff7e61e | 4393 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len) |
109f6aef CM |
4394 | goto err; |
4395 | ||
ad48fd75 YZ |
4396 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4397 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4398 | struct btrfs_file_extent_item); | |
4399 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4400 | goto err; | |
459931ec CM |
4401 | } |
4402 | ||
b9473439 | 4403 | btrfs_set_path_blocking(path); |
ad48fd75 | 4404 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4405 | if (ret) |
4406 | goto err; | |
459931ec | 4407 | |
ad48fd75 | 4408 | path->keep_locks = 0; |
b9473439 | 4409 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4410 | return 0; |
4411 | err: | |
4412 | path->keep_locks = 0; | |
4413 | return ret; | |
4414 | } | |
4415 | ||
4961e293 | 4416 | static noinline int split_item(struct btrfs_fs_info *fs_info, |
ad48fd75 | 4417 | struct btrfs_path *path, |
310712b2 | 4418 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4419 | unsigned long split_offset) |
4420 | { | |
4421 | struct extent_buffer *leaf; | |
4422 | struct btrfs_item *item; | |
4423 | struct btrfs_item *new_item; | |
4424 | int slot; | |
4425 | char *buf; | |
4426 | u32 nritems; | |
4427 | u32 item_size; | |
4428 | u32 orig_offset; | |
4429 | struct btrfs_disk_key disk_key; | |
4430 | ||
b9473439 | 4431 | leaf = path->nodes[0]; |
2ff7e61e | 4432 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4433 | |
b4ce94de CM |
4434 | btrfs_set_path_blocking(path); |
4435 | ||
dd3cc16b | 4436 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4437 | orig_offset = btrfs_item_offset(leaf, item); |
4438 | item_size = btrfs_item_size(leaf, item); | |
4439 | ||
459931ec | 4440 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4441 | if (!buf) |
4442 | return -ENOMEM; | |
4443 | ||
459931ec CM |
4444 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4445 | path->slots[0]), item_size); | |
459931ec | 4446 | |
ad48fd75 | 4447 | slot = path->slots[0] + 1; |
459931ec | 4448 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4449 | if (slot != nritems) { |
4450 | /* shift the items */ | |
4451 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4452 | btrfs_item_nr_offset(slot), |
4453 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4454 | } |
4455 | ||
4456 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4457 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4458 | ||
dd3cc16b | 4459 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4460 | |
4461 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4462 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4463 | ||
4464 | btrfs_set_item_offset(leaf, item, | |
4465 | orig_offset + item_size - split_offset); | |
4466 | btrfs_set_item_size(leaf, item, split_offset); | |
4467 | ||
4468 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4469 | ||
4470 | /* write the data for the start of the original item */ | |
4471 | write_extent_buffer(leaf, buf, | |
4472 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4473 | split_offset); | |
4474 | ||
4475 | /* write the data for the new item */ | |
4476 | write_extent_buffer(leaf, buf + split_offset, | |
4477 | btrfs_item_ptr_offset(leaf, slot), | |
4478 | item_size - split_offset); | |
4479 | btrfs_mark_buffer_dirty(leaf); | |
4480 | ||
2ff7e61e | 4481 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0); |
459931ec | 4482 | kfree(buf); |
ad48fd75 YZ |
4483 | return 0; |
4484 | } | |
4485 | ||
4486 | /* | |
4487 | * This function splits a single item into two items, | |
4488 | * giving 'new_key' to the new item and splitting the | |
4489 | * old one at split_offset (from the start of the item). | |
4490 | * | |
4491 | * The path may be released by this operation. After | |
4492 | * the split, the path is pointing to the old item. The | |
4493 | * new item is going to be in the same node as the old one. | |
4494 | * | |
4495 | * Note, the item being split must be smaller enough to live alone on | |
4496 | * a tree block with room for one extra struct btrfs_item | |
4497 | * | |
4498 | * This allows us to split the item in place, keeping a lock on the | |
4499 | * leaf the entire time. | |
4500 | */ | |
4501 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4502 | struct btrfs_root *root, | |
4503 | struct btrfs_path *path, | |
310712b2 | 4504 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4505 | unsigned long split_offset) |
4506 | { | |
4507 | int ret; | |
4508 | ret = setup_leaf_for_split(trans, root, path, | |
4509 | sizeof(struct btrfs_item)); | |
4510 | if (ret) | |
4511 | return ret; | |
4512 | ||
4961e293 | 4513 | ret = split_item(root->fs_info, path, new_key, split_offset); |
459931ec CM |
4514 | return ret; |
4515 | } | |
4516 | ||
ad48fd75 YZ |
4517 | /* |
4518 | * This function duplicate a item, giving 'new_key' to the new item. | |
4519 | * It guarantees both items live in the same tree leaf and the new item | |
4520 | * is contiguous with the original item. | |
4521 | * | |
4522 | * This allows us to split file extent in place, keeping a lock on the | |
4523 | * leaf the entire time. | |
4524 | */ | |
4525 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4526 | struct btrfs_root *root, | |
4527 | struct btrfs_path *path, | |
310712b2 | 4528 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4529 | { |
4530 | struct extent_buffer *leaf; | |
4531 | int ret; | |
4532 | u32 item_size; | |
4533 | ||
4534 | leaf = path->nodes[0]; | |
4535 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4536 | ret = setup_leaf_for_split(trans, root, path, | |
4537 | item_size + sizeof(struct btrfs_item)); | |
4538 | if (ret) | |
4539 | return ret; | |
4540 | ||
4541 | path->slots[0]++; | |
afe5fea7 | 4542 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4543 | item_size, item_size + |
4544 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4545 | leaf = path->nodes[0]; |
4546 | memcpy_extent_buffer(leaf, | |
4547 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4548 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4549 | item_size); | |
4550 | return 0; | |
4551 | } | |
4552 | ||
d352ac68 CM |
4553 | /* |
4554 | * make the item pointed to by the path smaller. new_size indicates | |
4555 | * how small to make it, and from_end tells us if we just chop bytes | |
4556 | * off the end of the item or if we shift the item to chop bytes off | |
4557 | * the front. | |
4558 | */ | |
2ff7e61e JM |
4559 | void btrfs_truncate_item(struct btrfs_fs_info *fs_info, |
4560 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 4561 | { |
b18c6685 | 4562 | int slot; |
5f39d397 CM |
4563 | struct extent_buffer *leaf; |
4564 | struct btrfs_item *item; | |
b18c6685 CM |
4565 | u32 nritems; |
4566 | unsigned int data_end; | |
4567 | unsigned int old_data_start; | |
4568 | unsigned int old_size; | |
4569 | unsigned int size_diff; | |
4570 | int i; | |
cfed81a0 CM |
4571 | struct btrfs_map_token token; |
4572 | ||
4573 | btrfs_init_map_token(&token); | |
b18c6685 | 4574 | |
5f39d397 | 4575 | leaf = path->nodes[0]; |
179e29e4 CM |
4576 | slot = path->slots[0]; |
4577 | ||
4578 | old_size = btrfs_item_size_nr(leaf, slot); | |
4579 | if (old_size == new_size) | |
143bede5 | 4580 | return; |
b18c6685 | 4581 | |
5f39d397 | 4582 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4583 | data_end = leaf_data_end(fs_info, leaf); |
b18c6685 | 4584 | |
5f39d397 | 4585 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4586 | |
b18c6685 CM |
4587 | size_diff = old_size - new_size; |
4588 | ||
4589 | BUG_ON(slot < 0); | |
4590 | BUG_ON(slot >= nritems); | |
4591 | ||
4592 | /* | |
4593 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4594 | */ | |
4595 | /* first correct the data pointers */ | |
4596 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4597 | u32 ioff; |
dd3cc16b | 4598 | item = btrfs_item_nr(i); |
db94535d | 4599 | |
cfed81a0 CM |
4600 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4601 | btrfs_set_token_item_offset(leaf, item, | |
4602 | ioff + size_diff, &token); | |
b18c6685 | 4603 | } |
db94535d | 4604 | |
b18c6685 | 4605 | /* shift the data */ |
179e29e4 | 4606 | if (from_end) { |
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 + new_size - data_end); |
4610 | } else { | |
4611 | struct btrfs_disk_key disk_key; | |
4612 | u64 offset; | |
4613 | ||
4614 | btrfs_item_key(leaf, &disk_key, slot); | |
4615 | ||
4616 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4617 | unsigned long ptr; | |
4618 | struct btrfs_file_extent_item *fi; | |
4619 | ||
4620 | fi = btrfs_item_ptr(leaf, slot, | |
4621 | struct btrfs_file_extent_item); | |
4622 | fi = (struct btrfs_file_extent_item *)( | |
4623 | (unsigned long)fi - size_diff); | |
4624 | ||
4625 | if (btrfs_file_extent_type(leaf, fi) == | |
4626 | BTRFS_FILE_EXTENT_INLINE) { | |
4627 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4628 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4629 | (unsigned long)fi, |
7ec20afb | 4630 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4631 | } |
4632 | } | |
4633 | ||
3d9ec8c4 NB |
4634 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4635 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4636 | data_end, old_data_start - data_end); |
4637 | ||
4638 | offset = btrfs_disk_key_offset(&disk_key); | |
4639 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4640 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4641 | if (slot == 0) | |
0b246afa | 4642 | fixup_low_keys(fs_info, path, &disk_key, 1); |
179e29e4 | 4643 | } |
5f39d397 | 4644 | |
dd3cc16b | 4645 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4646 | btrfs_set_item_size(leaf, item, new_size); |
4647 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4648 | |
2ff7e61e | 4649 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4650 | btrfs_print_leaf(leaf); |
b18c6685 | 4651 | BUG(); |
5f39d397 | 4652 | } |
b18c6685 CM |
4653 | } |
4654 | ||
d352ac68 | 4655 | /* |
8f69dbd2 | 4656 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4657 | */ |
2ff7e61e | 4658 | void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 4659 | u32 data_size) |
6567e837 | 4660 | { |
6567e837 | 4661 | int slot; |
5f39d397 CM |
4662 | struct extent_buffer *leaf; |
4663 | struct btrfs_item *item; | |
6567e837 CM |
4664 | u32 nritems; |
4665 | unsigned int data_end; | |
4666 | unsigned int old_data; | |
4667 | unsigned int old_size; | |
4668 | int i; | |
cfed81a0 CM |
4669 | struct btrfs_map_token token; |
4670 | ||
4671 | btrfs_init_map_token(&token); | |
6567e837 | 4672 | |
5f39d397 | 4673 | leaf = path->nodes[0]; |
6567e837 | 4674 | |
5f39d397 | 4675 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4676 | data_end = leaf_data_end(fs_info, leaf); |
6567e837 | 4677 | |
2ff7e61e | 4678 | if (btrfs_leaf_free_space(fs_info, leaf) < data_size) { |
a4f78750 | 4679 | btrfs_print_leaf(leaf); |
6567e837 | 4680 | BUG(); |
5f39d397 | 4681 | } |
6567e837 | 4682 | slot = path->slots[0]; |
5f39d397 | 4683 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4684 | |
4685 | BUG_ON(slot < 0); | |
3326d1b0 | 4686 | if (slot >= nritems) { |
a4f78750 | 4687 | btrfs_print_leaf(leaf); |
0b246afa JM |
4688 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4689 | slot, nritems); | |
3326d1b0 CM |
4690 | BUG_ON(1); |
4691 | } | |
6567e837 CM |
4692 | |
4693 | /* | |
4694 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4695 | */ | |
4696 | /* first correct the data pointers */ | |
4697 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4698 | u32 ioff; |
dd3cc16b | 4699 | item = btrfs_item_nr(i); |
db94535d | 4700 | |
cfed81a0 CM |
4701 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4702 | btrfs_set_token_item_offset(leaf, item, | |
4703 | ioff - data_size, &token); | |
6567e837 | 4704 | } |
5f39d397 | 4705 | |
6567e837 | 4706 | /* shift the data */ |
3d9ec8c4 NB |
4707 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4708 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4709 | data_end, old_data - data_end); |
5f39d397 | 4710 | |
6567e837 | 4711 | data_end = old_data; |
5f39d397 | 4712 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4713 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4714 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4715 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4716 | |
2ff7e61e | 4717 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4718 | btrfs_print_leaf(leaf); |
6567e837 | 4719 | BUG(); |
5f39d397 | 4720 | } |
6567e837 CM |
4721 | } |
4722 | ||
74123bd7 | 4723 | /* |
44871b1b CM |
4724 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4725 | * to save stack depth by doing the bulk of the work in a function | |
4726 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4727 | */ |
afe5fea7 | 4728 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4729 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4730 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4731 | { |
0b246afa | 4732 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4733 | struct btrfs_item *item; |
9c58309d | 4734 | int i; |
7518a238 | 4735 | u32 nritems; |
be0e5c09 | 4736 | unsigned int data_end; |
e2fa7227 | 4737 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4738 | struct extent_buffer *leaf; |
4739 | int slot; | |
cfed81a0 CM |
4740 | struct btrfs_map_token token; |
4741 | ||
24cdc847 FM |
4742 | if (path->slots[0] == 0) { |
4743 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
0b246afa | 4744 | fixup_low_keys(fs_info, path, &disk_key, 1); |
24cdc847 FM |
4745 | } |
4746 | btrfs_unlock_up_safe(path, 1); | |
4747 | ||
cfed81a0 | 4748 | btrfs_init_map_token(&token); |
e2fa7227 | 4749 | |
5f39d397 | 4750 | leaf = path->nodes[0]; |
44871b1b | 4751 | slot = path->slots[0]; |
74123bd7 | 4752 | |
5f39d397 | 4753 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4754 | data_end = leaf_data_end(fs_info, leaf); |
eb60ceac | 4755 | |
2ff7e61e | 4756 | if (btrfs_leaf_free_space(fs_info, leaf) < total_size) { |
a4f78750 | 4757 | btrfs_print_leaf(leaf); |
0b246afa | 4758 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
2ff7e61e | 4759 | total_size, btrfs_leaf_free_space(fs_info, leaf)); |
be0e5c09 | 4760 | BUG(); |
d4dbff95 | 4761 | } |
5f39d397 | 4762 | |
be0e5c09 | 4763 | if (slot != nritems) { |
5f39d397 | 4764 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4765 | |
5f39d397 | 4766 | if (old_data < data_end) { |
a4f78750 | 4767 | btrfs_print_leaf(leaf); |
0b246afa | 4768 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4769 | slot, old_data, data_end); |
5f39d397 CM |
4770 | BUG_ON(1); |
4771 | } | |
be0e5c09 CM |
4772 | /* |
4773 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4774 | */ | |
4775 | /* first correct the data pointers */ | |
0783fcfc | 4776 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4777 | u32 ioff; |
db94535d | 4778 | |
62e85577 | 4779 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4780 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4781 | btrfs_set_token_item_offset(leaf, item, | |
4782 | ioff - total_data, &token); | |
0783fcfc | 4783 | } |
be0e5c09 | 4784 | /* shift the items */ |
9c58309d | 4785 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4786 | btrfs_item_nr_offset(slot), |
d6025579 | 4787 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4788 | |
4789 | /* shift the data */ | |
3d9ec8c4 NB |
4790 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4791 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4792 | data_end, old_data - data_end); |
be0e5c09 CM |
4793 | data_end = old_data; |
4794 | } | |
5f39d397 | 4795 | |
62e2749e | 4796 | /* setup the item for the new data */ |
9c58309d CM |
4797 | for (i = 0; i < nr; i++) { |
4798 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4799 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4800 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4801 | btrfs_set_token_item_offset(leaf, item, |
4802 | data_end - data_size[i], &token); | |
9c58309d | 4803 | data_end -= data_size[i]; |
cfed81a0 | 4804 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4805 | } |
44871b1b | 4806 | |
9c58309d | 4807 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4808 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4809 | |
2ff7e61e | 4810 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4811 | btrfs_print_leaf(leaf); |
be0e5c09 | 4812 | BUG(); |
5f39d397 | 4813 | } |
44871b1b CM |
4814 | } |
4815 | ||
4816 | /* | |
4817 | * Given a key and some data, insert items into the tree. | |
4818 | * This does all the path init required, making room in the tree if needed. | |
4819 | */ | |
4820 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4821 | struct btrfs_root *root, | |
4822 | struct btrfs_path *path, | |
310712b2 | 4823 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4824 | int nr) |
4825 | { | |
44871b1b CM |
4826 | int ret = 0; |
4827 | int slot; | |
4828 | int i; | |
4829 | u32 total_size = 0; | |
4830 | u32 total_data = 0; | |
4831 | ||
4832 | for (i = 0; i < nr; i++) | |
4833 | total_data += data_size[i]; | |
4834 | ||
4835 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4836 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4837 | if (ret == 0) | |
4838 | return -EEXIST; | |
4839 | if (ret < 0) | |
143bede5 | 4840 | return ret; |
44871b1b | 4841 | |
44871b1b CM |
4842 | slot = path->slots[0]; |
4843 | BUG_ON(slot < 0); | |
4844 | ||
afe5fea7 | 4845 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4846 | total_data, total_size, nr); |
143bede5 | 4847 | return 0; |
62e2749e CM |
4848 | } |
4849 | ||
4850 | /* | |
4851 | * Given a key and some data, insert an item into the tree. | |
4852 | * This does all the path init required, making room in the tree if needed. | |
4853 | */ | |
310712b2 OS |
4854 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4855 | const struct btrfs_key *cpu_key, void *data, | |
4856 | u32 data_size) | |
62e2749e CM |
4857 | { |
4858 | int ret = 0; | |
2c90e5d6 | 4859 | struct btrfs_path *path; |
5f39d397 CM |
4860 | struct extent_buffer *leaf; |
4861 | unsigned long ptr; | |
62e2749e | 4862 | |
2c90e5d6 | 4863 | path = btrfs_alloc_path(); |
db5b493a TI |
4864 | if (!path) |
4865 | return -ENOMEM; | |
2c90e5d6 | 4866 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4867 | if (!ret) { |
5f39d397 CM |
4868 | leaf = path->nodes[0]; |
4869 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4870 | write_extent_buffer(leaf, data, ptr, data_size); | |
4871 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4872 | } |
2c90e5d6 | 4873 | btrfs_free_path(path); |
aa5d6bed | 4874 | return ret; |
be0e5c09 CM |
4875 | } |
4876 | ||
74123bd7 | 4877 | /* |
5de08d7d | 4878 | * delete the pointer from a given node. |
74123bd7 | 4879 | * |
d352ac68 CM |
4880 | * the tree should have been previously balanced so the deletion does not |
4881 | * empty a node. | |
74123bd7 | 4882 | */ |
afe5fea7 TI |
4883 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4884 | int level, int slot) | |
be0e5c09 | 4885 | { |
0b246afa | 4886 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4887 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4888 | u32 nritems; |
f3ea38da | 4889 | int ret; |
be0e5c09 | 4890 | |
5f39d397 | 4891 | nritems = btrfs_header_nritems(parent); |
d397712b | 4892 | if (slot != nritems - 1) { |
bf1d3425 DS |
4893 | if (level) { |
4894 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4895 | nritems - slot - 1); |
bf1d3425 DS |
4896 | BUG_ON(ret < 0); |
4897 | } | |
5f39d397 CM |
4898 | memmove_extent_buffer(parent, |
4899 | btrfs_node_key_ptr_offset(slot), | |
4900 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4901 | sizeof(struct btrfs_key_ptr) * |
4902 | (nritems - slot - 1)); | |
57ba86c0 | 4903 | } else if (level) { |
e09c2efe DS |
4904 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4905 | GFP_NOFS); | |
57ba86c0 | 4906 | BUG_ON(ret < 0); |
bb803951 | 4907 | } |
f3ea38da | 4908 | |
7518a238 | 4909 | nritems--; |
5f39d397 | 4910 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4911 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4912 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4913 | /* just turn the root into a leaf and break */ |
5f39d397 | 4914 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4915 | } else if (slot == 0) { |
5f39d397 CM |
4916 | struct btrfs_disk_key disk_key; |
4917 | ||
4918 | btrfs_node_key(parent, &disk_key, 0); | |
0b246afa | 4919 | fixup_low_keys(fs_info, path, &disk_key, level + 1); |
be0e5c09 | 4920 | } |
d6025579 | 4921 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4922 | } |
4923 | ||
323ac95b CM |
4924 | /* |
4925 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4926 | * path->nodes[1]. |
323ac95b CM |
4927 | * |
4928 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4929 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4930 | * | |
4931 | * The path must have already been setup for deleting the leaf, including | |
4932 | * all the proper balancing. path->nodes[1] must be locked. | |
4933 | */ | |
143bede5 JM |
4934 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4935 | struct btrfs_root *root, | |
4936 | struct btrfs_path *path, | |
4937 | struct extent_buffer *leaf) | |
323ac95b | 4938 | { |
5d4f98a2 | 4939 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4940 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4941 | |
4d081c41 CM |
4942 | /* |
4943 | * btrfs_free_extent is expensive, we want to make sure we | |
4944 | * aren't holding any locks when we call it | |
4945 | */ | |
4946 | btrfs_unlock_up_safe(path, 0); | |
4947 | ||
f0486c68 YZ |
4948 | root_sub_used(root, leaf->len); |
4949 | ||
3083ee2e | 4950 | extent_buffer_get(leaf); |
5581a51a | 4951 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4952 | free_extent_buffer_stale(leaf); |
323ac95b | 4953 | } |
74123bd7 CM |
4954 | /* |
4955 | * delete the item at the leaf level in path. If that empties | |
4956 | * the leaf, remove it from the tree | |
4957 | */ | |
85e21bac CM |
4958 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4959 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4960 | { |
0b246afa | 4961 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4962 | struct extent_buffer *leaf; |
4963 | struct btrfs_item *item; | |
ce0eac2a AM |
4964 | u32 last_off; |
4965 | u32 dsize = 0; | |
aa5d6bed CM |
4966 | int ret = 0; |
4967 | int wret; | |
85e21bac | 4968 | int i; |
7518a238 | 4969 | u32 nritems; |
cfed81a0 CM |
4970 | struct btrfs_map_token token; |
4971 | ||
4972 | btrfs_init_map_token(&token); | |
be0e5c09 | 4973 | |
5f39d397 | 4974 | leaf = path->nodes[0]; |
85e21bac CM |
4975 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4976 | ||
4977 | for (i = 0; i < nr; i++) | |
4978 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4979 | ||
5f39d397 | 4980 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4981 | |
85e21bac | 4982 | if (slot + nr != nritems) { |
2ff7e61e | 4983 | int data_end = leaf_data_end(fs_info, leaf); |
5f39d397 | 4984 | |
3d9ec8c4 | 4985 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4986 | data_end + dsize, |
3d9ec8c4 | 4987 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4988 | last_off - data_end); |
5f39d397 | 4989 | |
85e21bac | 4990 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4991 | u32 ioff; |
db94535d | 4992 | |
dd3cc16b | 4993 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4994 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4995 | btrfs_set_token_item_offset(leaf, item, | |
4996 | ioff + dsize, &token); | |
0783fcfc | 4997 | } |
db94535d | 4998 | |
5f39d397 | 4999 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5000 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5001 | sizeof(struct btrfs_item) * |
85e21bac | 5002 | (nritems - slot - nr)); |
be0e5c09 | 5003 | } |
85e21bac CM |
5004 | btrfs_set_header_nritems(leaf, nritems - nr); |
5005 | nritems -= nr; | |
5f39d397 | 5006 | |
74123bd7 | 5007 | /* delete the leaf if we've emptied it */ |
7518a238 | 5008 | if (nritems == 0) { |
5f39d397 CM |
5009 | if (leaf == root->node) { |
5010 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5011 | } else { |
f0486c68 | 5012 | btrfs_set_path_blocking(path); |
7c302b49 | 5013 | clean_tree_block(fs_info, leaf); |
143bede5 | 5014 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5015 | } |
be0e5c09 | 5016 | } else { |
7518a238 | 5017 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5018 | if (slot == 0) { |
5f39d397 CM |
5019 | struct btrfs_disk_key disk_key; |
5020 | ||
5021 | btrfs_item_key(leaf, &disk_key, 0); | |
0b246afa | 5022 | fixup_low_keys(fs_info, path, &disk_key, 1); |
aa5d6bed | 5023 | } |
aa5d6bed | 5024 | |
74123bd7 | 5025 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5026 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5027 | /* push_leaf_left fixes the path. |
5028 | * make sure the path still points to our leaf | |
5029 | * for possible call to del_ptr below | |
5030 | */ | |
4920c9ac | 5031 | slot = path->slots[1]; |
5f39d397 CM |
5032 | extent_buffer_get(leaf); |
5033 | ||
b9473439 | 5034 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5035 | wret = push_leaf_left(trans, root, path, 1, 1, |
5036 | 1, (u32)-1); | |
54aa1f4d | 5037 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5038 | ret = wret; |
5f39d397 CM |
5039 | |
5040 | if (path->nodes[0] == leaf && | |
5041 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5042 | wret = push_leaf_right(trans, root, path, 1, |
5043 | 1, 1, 0); | |
54aa1f4d | 5044 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5045 | ret = wret; |
5046 | } | |
5f39d397 CM |
5047 | |
5048 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5049 | path->slots[1] = slot; |
143bede5 | 5050 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5051 | free_extent_buffer(leaf); |
143bede5 | 5052 | ret = 0; |
5de08d7d | 5053 | } else { |
925baedd CM |
5054 | /* if we're still in the path, make sure |
5055 | * we're dirty. Otherwise, one of the | |
5056 | * push_leaf functions must have already | |
5057 | * dirtied this buffer | |
5058 | */ | |
5059 | if (path->nodes[0] == leaf) | |
5060 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5061 | free_extent_buffer(leaf); |
be0e5c09 | 5062 | } |
d5719762 | 5063 | } else { |
5f39d397 | 5064 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5065 | } |
5066 | } | |
aa5d6bed | 5067 | return ret; |
be0e5c09 CM |
5068 | } |
5069 | ||
7bb86316 | 5070 | /* |
925baedd | 5071 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5072 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5073 | * returns < 0 on io errors. | |
d352ac68 CM |
5074 | * |
5075 | * This may release the path, and so you may lose any locks held at the | |
5076 | * time you call it. | |
7bb86316 | 5077 | */ |
16e7549f | 5078 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5079 | { |
925baedd CM |
5080 | struct btrfs_key key; |
5081 | struct btrfs_disk_key found_key; | |
5082 | int ret; | |
7bb86316 | 5083 | |
925baedd | 5084 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5085 | |
e8b0d724 | 5086 | if (key.offset > 0) { |
925baedd | 5087 | key.offset--; |
e8b0d724 | 5088 | } else if (key.type > 0) { |
925baedd | 5089 | key.type--; |
e8b0d724 FDBM |
5090 | key.offset = (u64)-1; |
5091 | } else if (key.objectid > 0) { | |
925baedd | 5092 | key.objectid--; |
e8b0d724 FDBM |
5093 | key.type = (u8)-1; |
5094 | key.offset = (u64)-1; | |
5095 | } else { | |
925baedd | 5096 | return 1; |
e8b0d724 | 5097 | } |
7bb86316 | 5098 | |
b3b4aa74 | 5099 | btrfs_release_path(path); |
925baedd CM |
5100 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5101 | if (ret < 0) | |
5102 | return ret; | |
5103 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5104 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5105 | /* |
5106 | * We might have had an item with the previous key in the tree right | |
5107 | * before we released our path. And after we released our path, that | |
5108 | * item might have been pushed to the first slot (0) of the leaf we | |
5109 | * were holding due to a tree balance. Alternatively, an item with the | |
5110 | * previous key can exist as the only element of a leaf (big fat item). | |
5111 | * Therefore account for these 2 cases, so that our callers (like | |
5112 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5113 | * the previous key we computed above. | |
5114 | */ | |
5115 | if (ret <= 0) | |
925baedd CM |
5116 | return 0; |
5117 | return 1; | |
7bb86316 CM |
5118 | } |
5119 | ||
3f157a2f CM |
5120 | /* |
5121 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5122 | * for nodes or leaves that are have a minimum transaction id. |
5123 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5124 | * |
5125 | * This does not cow, but it does stuff the starting key it finds back | |
5126 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5127 | * key and get a writable path. | |
5128 | * | |
3f157a2f CM |
5129 | * This honors path->lowest_level to prevent descent past a given level |
5130 | * of the tree. | |
5131 | * | |
d352ac68 CM |
5132 | * min_trans indicates the oldest transaction that you are interested |
5133 | * in walking through. Any nodes or leaves older than min_trans are | |
5134 | * skipped over (without reading them). | |
5135 | * | |
3f157a2f CM |
5136 | * returns zero if something useful was found, < 0 on error and 1 if there |
5137 | * was nothing in the tree that matched the search criteria. | |
5138 | */ | |
5139 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5140 | struct btrfs_path *path, |
3f157a2f CM |
5141 | u64 min_trans) |
5142 | { | |
2ff7e61e | 5143 | struct btrfs_fs_info *fs_info = root->fs_info; |
3f157a2f CM |
5144 | struct extent_buffer *cur; |
5145 | struct btrfs_key found_key; | |
5146 | int slot; | |
9652480b | 5147 | int sret; |
3f157a2f CM |
5148 | u32 nritems; |
5149 | int level; | |
5150 | int ret = 1; | |
f98de9b9 | 5151 | int keep_locks = path->keep_locks; |
3f157a2f | 5152 | |
f98de9b9 | 5153 | path->keep_locks = 1; |
3f157a2f | 5154 | again: |
bd681513 | 5155 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5156 | level = btrfs_header_level(cur); |
e02119d5 | 5157 | WARN_ON(path->nodes[level]); |
3f157a2f | 5158 | path->nodes[level] = cur; |
bd681513 | 5159 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5160 | |
5161 | if (btrfs_header_generation(cur) < min_trans) { | |
5162 | ret = 1; | |
5163 | goto out; | |
5164 | } | |
d397712b | 5165 | while (1) { |
3f157a2f CM |
5166 | nritems = btrfs_header_nritems(cur); |
5167 | level = btrfs_header_level(cur); | |
a74b35ec | 5168 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
3f157a2f | 5169 | |
323ac95b CM |
5170 | /* at the lowest level, we're done, setup the path and exit */ |
5171 | if (level == path->lowest_level) { | |
e02119d5 CM |
5172 | if (slot >= nritems) |
5173 | goto find_next_key; | |
3f157a2f CM |
5174 | ret = 0; |
5175 | path->slots[level] = slot; | |
5176 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5177 | goto out; | |
5178 | } | |
9652480b Y |
5179 | if (sret && slot > 0) |
5180 | slot--; | |
3f157a2f | 5181 | /* |
de78b51a ES |
5182 | * check this node pointer against the min_trans parameters. |
5183 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5184 | */ |
d397712b | 5185 | while (slot < nritems) { |
3f157a2f | 5186 | u64 gen; |
e02119d5 | 5187 | |
3f157a2f CM |
5188 | gen = btrfs_node_ptr_generation(cur, slot); |
5189 | if (gen < min_trans) { | |
5190 | slot++; | |
5191 | continue; | |
5192 | } | |
de78b51a | 5193 | break; |
3f157a2f | 5194 | } |
e02119d5 | 5195 | find_next_key: |
3f157a2f CM |
5196 | /* |
5197 | * we didn't find a candidate key in this node, walk forward | |
5198 | * and find another one | |
5199 | */ | |
5200 | if (slot >= nritems) { | |
e02119d5 | 5201 | path->slots[level] = slot; |
b4ce94de | 5202 | btrfs_set_path_blocking(path); |
e02119d5 | 5203 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5204 | min_trans); |
e02119d5 | 5205 | if (sret == 0) { |
b3b4aa74 | 5206 | btrfs_release_path(path); |
3f157a2f CM |
5207 | goto again; |
5208 | } else { | |
5209 | goto out; | |
5210 | } | |
5211 | } | |
5212 | /* save our key for returning back */ | |
5213 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5214 | path->slots[level] = slot; | |
5215 | if (level == path->lowest_level) { | |
5216 | ret = 0; | |
3f157a2f CM |
5217 | goto out; |
5218 | } | |
b4ce94de | 5219 | btrfs_set_path_blocking(path); |
2ff7e61e | 5220 | cur = read_node_slot(fs_info, cur, slot); |
fb770ae4 LB |
5221 | if (IS_ERR(cur)) { |
5222 | ret = PTR_ERR(cur); | |
5223 | goto out; | |
5224 | } | |
3f157a2f | 5225 | |
bd681513 | 5226 | btrfs_tree_read_lock(cur); |
b4ce94de | 5227 | |
bd681513 | 5228 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5229 | path->nodes[level - 1] = cur; |
f7c79f30 | 5230 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5231 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5232 | } |
5233 | out: | |
f98de9b9 FM |
5234 | path->keep_locks = keep_locks; |
5235 | if (ret == 0) { | |
5236 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5237 | btrfs_set_path_blocking(path); | |
3f157a2f | 5238 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5239 | } |
3f157a2f CM |
5240 | return ret; |
5241 | } | |
5242 | ||
2ff7e61e | 5243 | static int tree_move_down(struct btrfs_fs_info *fs_info, |
7069830a | 5244 | struct btrfs_path *path, |
ab6a43e1 | 5245 | int *level) |
7069830a | 5246 | { |
fb770ae4 LB |
5247 | struct extent_buffer *eb; |
5248 | ||
74dd17fb | 5249 | BUG_ON(*level == 0); |
2ff7e61e | 5250 | eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]); |
fb770ae4 LB |
5251 | if (IS_ERR(eb)) |
5252 | return PTR_ERR(eb); | |
5253 | ||
5254 | path->nodes[*level - 1] = eb; | |
7069830a AB |
5255 | path->slots[*level - 1] = 0; |
5256 | (*level)--; | |
fb770ae4 | 5257 | return 0; |
7069830a AB |
5258 | } |
5259 | ||
f1e30261 | 5260 | static int tree_move_next_or_upnext(struct btrfs_path *path, |
7069830a AB |
5261 | int *level, int root_level) |
5262 | { | |
5263 | int ret = 0; | |
5264 | int nritems; | |
5265 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5266 | ||
5267 | path->slots[*level]++; | |
5268 | ||
74dd17fb | 5269 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5270 | if (*level == root_level) |
5271 | return -1; | |
5272 | ||
5273 | /* move upnext */ | |
5274 | path->slots[*level] = 0; | |
5275 | free_extent_buffer(path->nodes[*level]); | |
5276 | path->nodes[*level] = NULL; | |
5277 | (*level)++; | |
5278 | path->slots[*level]++; | |
5279 | ||
5280 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5281 | ret = 1; | |
5282 | } | |
5283 | return ret; | |
5284 | } | |
5285 | ||
5286 | /* | |
5287 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5288 | * or down. | |
5289 | */ | |
2ff7e61e | 5290 | static int tree_advance(struct btrfs_fs_info *fs_info, |
7069830a AB |
5291 | struct btrfs_path *path, |
5292 | int *level, int root_level, | |
5293 | int allow_down, | |
5294 | struct btrfs_key *key) | |
5295 | { | |
5296 | int ret; | |
5297 | ||
5298 | if (*level == 0 || !allow_down) { | |
f1e30261 | 5299 | ret = tree_move_next_or_upnext(path, level, root_level); |
7069830a | 5300 | } else { |
ab6a43e1 | 5301 | ret = tree_move_down(fs_info, path, level); |
7069830a AB |
5302 | } |
5303 | if (ret >= 0) { | |
5304 | if (*level == 0) | |
5305 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5306 | path->slots[*level]); | |
5307 | else | |
5308 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5309 | path->slots[*level]); | |
5310 | } | |
5311 | return ret; | |
5312 | } | |
5313 | ||
2ff7e61e | 5314 | static int tree_compare_item(struct btrfs_path *left_path, |
7069830a AB |
5315 | struct btrfs_path *right_path, |
5316 | char *tmp_buf) | |
5317 | { | |
5318 | int cmp; | |
5319 | int len1, len2; | |
5320 | unsigned long off1, off2; | |
5321 | ||
5322 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5323 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5324 | if (len1 != len2) | |
5325 | return 1; | |
5326 | ||
5327 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5328 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5329 | right_path->slots[0]); | |
5330 | ||
5331 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5332 | ||
5333 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5334 | if (cmp) | |
5335 | return 1; | |
5336 | return 0; | |
5337 | } | |
5338 | ||
5339 | #define ADVANCE 1 | |
5340 | #define ADVANCE_ONLY_NEXT -1 | |
5341 | ||
5342 | /* | |
5343 | * This function compares two trees and calls the provided callback for | |
5344 | * every changed/new/deleted item it finds. | |
5345 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5346 | * the compare pretty fast on snapshotted subvolumes. | |
5347 | * | |
5348 | * This currently works on commit roots only. As commit roots are read only, | |
5349 | * we don't do any locking. The commit roots are protected with transactions. | |
5350 | * Transactions are ended and rejoined when a commit is tried in between. | |
5351 | * | |
5352 | * This function checks for modifications done to the trees while comparing. | |
5353 | * If it detects a change, it aborts immediately. | |
5354 | */ | |
5355 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5356 | struct btrfs_root *right_root, | |
5357 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5358 | { | |
0b246afa | 5359 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5360 | int ret; |
5361 | int cmp; | |
7069830a AB |
5362 | struct btrfs_path *left_path = NULL; |
5363 | struct btrfs_path *right_path = NULL; | |
5364 | struct btrfs_key left_key; | |
5365 | struct btrfs_key right_key; | |
5366 | char *tmp_buf = NULL; | |
5367 | int left_root_level; | |
5368 | int right_root_level; | |
5369 | int left_level; | |
5370 | int right_level; | |
5371 | int left_end_reached; | |
5372 | int right_end_reached; | |
5373 | int advance_left; | |
5374 | int advance_right; | |
5375 | u64 left_blockptr; | |
5376 | u64 right_blockptr; | |
6baa4293 FM |
5377 | u64 left_gen; |
5378 | u64 right_gen; | |
7069830a AB |
5379 | |
5380 | left_path = btrfs_alloc_path(); | |
5381 | if (!left_path) { | |
5382 | ret = -ENOMEM; | |
5383 | goto out; | |
5384 | } | |
5385 | right_path = btrfs_alloc_path(); | |
5386 | if (!right_path) { | |
5387 | ret = -ENOMEM; | |
5388 | goto out; | |
5389 | } | |
5390 | ||
752ade68 | 5391 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); |
7069830a | 5392 | if (!tmp_buf) { |
752ade68 MH |
5393 | ret = -ENOMEM; |
5394 | goto out; | |
7069830a AB |
5395 | } |
5396 | ||
5397 | left_path->search_commit_root = 1; | |
5398 | left_path->skip_locking = 1; | |
5399 | right_path->search_commit_root = 1; | |
5400 | right_path->skip_locking = 1; | |
5401 | ||
7069830a AB |
5402 | /* |
5403 | * Strategy: Go to the first items of both trees. Then do | |
5404 | * | |
5405 | * If both trees are at level 0 | |
5406 | * Compare keys of current items | |
5407 | * If left < right treat left item as new, advance left tree | |
5408 | * and repeat | |
5409 | * If left > right treat right item as deleted, advance right tree | |
5410 | * and repeat | |
5411 | * If left == right do deep compare of items, treat as changed if | |
5412 | * needed, advance both trees and repeat | |
5413 | * If both trees are at the same level but not at level 0 | |
5414 | * Compare keys of current nodes/leafs | |
5415 | * If left < right advance left tree and repeat | |
5416 | * If left > right advance right tree and repeat | |
5417 | * If left == right compare blockptrs of the next nodes/leafs | |
5418 | * If they match advance both trees but stay at the same level | |
5419 | * and repeat | |
5420 | * If they don't match advance both trees while allowing to go | |
5421 | * deeper and repeat | |
5422 | * If tree levels are different | |
5423 | * Advance the tree that needs it and repeat | |
5424 | * | |
5425 | * Advancing a tree means: | |
5426 | * If we are at level 0, try to go to the next slot. If that's not | |
5427 | * possible, go one level up and repeat. Stop when we found a level | |
5428 | * where we could go to the next slot. We may at this point be on a | |
5429 | * node or a leaf. | |
5430 | * | |
5431 | * If we are not at level 0 and not on shared tree blocks, go one | |
5432 | * level deeper. | |
5433 | * | |
5434 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5435 | * the right if possible or go up and right. | |
5436 | */ | |
5437 | ||
0b246afa | 5438 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5439 | left_level = btrfs_header_level(left_root->commit_root); |
5440 | left_root_level = left_level; | |
6f2f0b39 RK |
5441 | left_path->nodes[left_level] = |
5442 | btrfs_clone_extent_buffer(left_root->commit_root); | |
5443 | if (!left_path->nodes[left_level]) { | |
5444 | up_read(&fs_info->commit_root_sem); | |
5445 | ret = -ENOMEM; | |
5446 | goto out; | |
5447 | } | |
7069830a AB |
5448 | extent_buffer_get(left_path->nodes[left_level]); |
5449 | ||
5450 | right_level = btrfs_header_level(right_root->commit_root); | |
5451 | right_root_level = right_level; | |
6f2f0b39 RK |
5452 | right_path->nodes[right_level] = |
5453 | btrfs_clone_extent_buffer(right_root->commit_root); | |
5454 | if (!right_path->nodes[right_level]) { | |
5455 | up_read(&fs_info->commit_root_sem); | |
5456 | ret = -ENOMEM; | |
5457 | goto out; | |
5458 | } | |
7069830a | 5459 | extent_buffer_get(right_path->nodes[right_level]); |
0b246afa | 5460 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5461 | |
5462 | if (left_level == 0) | |
5463 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5464 | &left_key, left_path->slots[left_level]); | |
5465 | else | |
5466 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5467 | &left_key, left_path->slots[left_level]); | |
5468 | if (right_level == 0) | |
5469 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5470 | &right_key, right_path->slots[right_level]); | |
5471 | else | |
5472 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5473 | &right_key, right_path->slots[right_level]); | |
5474 | ||
5475 | left_end_reached = right_end_reached = 0; | |
5476 | advance_left = advance_right = 0; | |
5477 | ||
5478 | while (1) { | |
7069830a | 5479 | if (advance_left && !left_end_reached) { |
2ff7e61e | 5480 | ret = tree_advance(fs_info, left_path, &left_level, |
7069830a AB |
5481 | left_root_level, |
5482 | advance_left != ADVANCE_ONLY_NEXT, | |
5483 | &left_key); | |
fb770ae4 | 5484 | if (ret == -1) |
7069830a | 5485 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5486 | else if (ret < 0) |
5487 | goto out; | |
7069830a AB |
5488 | advance_left = 0; |
5489 | } | |
5490 | if (advance_right && !right_end_reached) { | |
2ff7e61e | 5491 | ret = tree_advance(fs_info, right_path, &right_level, |
7069830a AB |
5492 | right_root_level, |
5493 | advance_right != ADVANCE_ONLY_NEXT, | |
5494 | &right_key); | |
fb770ae4 | 5495 | if (ret == -1) |
7069830a | 5496 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5497 | else if (ret < 0) |
5498 | goto out; | |
7069830a AB |
5499 | advance_right = 0; |
5500 | } | |
5501 | ||
5502 | if (left_end_reached && right_end_reached) { | |
5503 | ret = 0; | |
5504 | goto out; | |
5505 | } else if (left_end_reached) { | |
5506 | if (right_level == 0) { | |
ee8c494f | 5507 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5508 | &right_key, |
5509 | BTRFS_COMPARE_TREE_DELETED, | |
5510 | ctx); | |
5511 | if (ret < 0) | |
5512 | goto out; | |
5513 | } | |
5514 | advance_right = ADVANCE; | |
5515 | continue; | |
5516 | } else if (right_end_reached) { | |
5517 | if (left_level == 0) { | |
ee8c494f | 5518 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5519 | &left_key, |
5520 | BTRFS_COMPARE_TREE_NEW, | |
5521 | ctx); | |
5522 | if (ret < 0) | |
5523 | goto out; | |
5524 | } | |
5525 | advance_left = ADVANCE; | |
5526 | continue; | |
5527 | } | |
5528 | ||
5529 | if (left_level == 0 && right_level == 0) { | |
5530 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5531 | if (cmp < 0) { | |
ee8c494f | 5532 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5533 | &left_key, |
5534 | BTRFS_COMPARE_TREE_NEW, | |
5535 | ctx); | |
5536 | if (ret < 0) | |
5537 | goto out; | |
5538 | advance_left = ADVANCE; | |
5539 | } else if (cmp > 0) { | |
ee8c494f | 5540 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5541 | &right_key, |
5542 | BTRFS_COMPARE_TREE_DELETED, | |
5543 | ctx); | |
5544 | if (ret < 0) | |
5545 | goto out; | |
5546 | advance_right = ADVANCE; | |
5547 | } else { | |
b99d9a6a | 5548 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5549 | |
74dd17fb | 5550 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
2ff7e61e JM |
5551 | ret = tree_compare_item(left_path, right_path, |
5552 | tmp_buf); | |
ba5e8f2e | 5553 | if (ret) |
b99d9a6a | 5554 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5555 | else |
b99d9a6a | 5556 | result = BTRFS_COMPARE_TREE_SAME; |
ee8c494f | 5557 | ret = changed_cb(left_path, right_path, |
b99d9a6a | 5558 | &left_key, result, ctx); |
ba5e8f2e JB |
5559 | if (ret < 0) |
5560 | goto out; | |
7069830a AB |
5561 | advance_left = ADVANCE; |
5562 | advance_right = ADVANCE; | |
5563 | } | |
5564 | } else if (left_level == right_level) { | |
5565 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5566 | if (cmp < 0) { | |
5567 | advance_left = ADVANCE; | |
5568 | } else if (cmp > 0) { | |
5569 | advance_right = ADVANCE; | |
5570 | } else { | |
5571 | left_blockptr = btrfs_node_blockptr( | |
5572 | left_path->nodes[left_level], | |
5573 | left_path->slots[left_level]); | |
5574 | right_blockptr = btrfs_node_blockptr( | |
5575 | right_path->nodes[right_level], | |
5576 | right_path->slots[right_level]); | |
6baa4293 FM |
5577 | left_gen = btrfs_node_ptr_generation( |
5578 | left_path->nodes[left_level], | |
5579 | left_path->slots[left_level]); | |
5580 | right_gen = btrfs_node_ptr_generation( | |
5581 | right_path->nodes[right_level], | |
5582 | right_path->slots[right_level]); | |
5583 | if (left_blockptr == right_blockptr && | |
5584 | left_gen == right_gen) { | |
7069830a AB |
5585 | /* |
5586 | * As we're on a shared block, don't | |
5587 | * allow to go deeper. | |
5588 | */ | |
5589 | advance_left = ADVANCE_ONLY_NEXT; | |
5590 | advance_right = ADVANCE_ONLY_NEXT; | |
5591 | } else { | |
5592 | advance_left = ADVANCE; | |
5593 | advance_right = ADVANCE; | |
5594 | } | |
5595 | } | |
5596 | } else if (left_level < right_level) { | |
5597 | advance_right = ADVANCE; | |
5598 | } else { | |
5599 | advance_left = ADVANCE; | |
5600 | } | |
5601 | } | |
5602 | ||
5603 | out: | |
5604 | btrfs_free_path(left_path); | |
5605 | btrfs_free_path(right_path); | |
8f282f71 | 5606 | kvfree(tmp_buf); |
7069830a AB |
5607 | return ret; |
5608 | } | |
5609 | ||
3f157a2f CM |
5610 | /* |
5611 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5612 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5613 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5614 | * |
5615 | * 0 is returned if another key is found, < 0 if there are any errors | |
5616 | * and 1 is returned if there are no higher keys in the tree | |
5617 | * | |
5618 | * path->keep_locks should be set to 1 on the search made before | |
5619 | * calling this function. | |
5620 | */ | |
e7a84565 | 5621 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5622 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5623 | { |
e7a84565 CM |
5624 | int slot; |
5625 | struct extent_buffer *c; | |
5626 | ||
934d375b | 5627 | WARN_ON(!path->keep_locks); |
d397712b | 5628 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5629 | if (!path->nodes[level]) |
5630 | return 1; | |
5631 | ||
5632 | slot = path->slots[level] + 1; | |
5633 | c = path->nodes[level]; | |
3f157a2f | 5634 | next: |
e7a84565 | 5635 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5636 | int ret; |
5637 | int orig_lowest; | |
5638 | struct btrfs_key cur_key; | |
5639 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5640 | !path->nodes[level + 1]) | |
e7a84565 | 5641 | return 1; |
33c66f43 YZ |
5642 | |
5643 | if (path->locks[level + 1]) { | |
5644 | level++; | |
5645 | continue; | |
5646 | } | |
5647 | ||
5648 | slot = btrfs_header_nritems(c) - 1; | |
5649 | if (level == 0) | |
5650 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5651 | else | |
5652 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5653 | ||
5654 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5655 | btrfs_release_path(path); |
33c66f43 YZ |
5656 | path->lowest_level = level; |
5657 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5658 | 0, 0); | |
5659 | path->lowest_level = orig_lowest; | |
5660 | if (ret < 0) | |
5661 | return ret; | |
5662 | ||
5663 | c = path->nodes[level]; | |
5664 | slot = path->slots[level]; | |
5665 | if (ret == 0) | |
5666 | slot++; | |
5667 | goto next; | |
e7a84565 | 5668 | } |
33c66f43 | 5669 | |
e7a84565 CM |
5670 | if (level == 0) |
5671 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5672 | else { |
3f157a2f CM |
5673 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5674 | ||
3f157a2f CM |
5675 | if (gen < min_trans) { |
5676 | slot++; | |
5677 | goto next; | |
5678 | } | |
e7a84565 | 5679 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5680 | } |
e7a84565 CM |
5681 | return 0; |
5682 | } | |
5683 | return 1; | |
5684 | } | |
5685 | ||
97571fd0 | 5686 | /* |
925baedd | 5687 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5688 | * returns 0 if it found something or 1 if there are no greater leaves. |
5689 | * returns < 0 on io errors. | |
97571fd0 | 5690 | */ |
234b63a0 | 5691 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5692 | { |
5693 | return btrfs_next_old_leaf(root, path, 0); | |
5694 | } | |
5695 | ||
5696 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5697 | u64 time_seq) | |
d97e63b6 CM |
5698 | { |
5699 | int slot; | |
8e73f275 | 5700 | int level; |
5f39d397 | 5701 | struct extent_buffer *c; |
8e73f275 | 5702 | struct extent_buffer *next; |
925baedd CM |
5703 | struct btrfs_key key; |
5704 | u32 nritems; | |
5705 | int ret; | |
8e73f275 | 5706 | int old_spinning = path->leave_spinning; |
bd681513 | 5707 | int next_rw_lock = 0; |
925baedd CM |
5708 | |
5709 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5710 | if (nritems == 0) |
925baedd | 5711 | return 1; |
925baedd | 5712 | |
8e73f275 CM |
5713 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5714 | again: | |
5715 | level = 1; | |
5716 | next = NULL; | |
bd681513 | 5717 | next_rw_lock = 0; |
b3b4aa74 | 5718 | btrfs_release_path(path); |
8e73f275 | 5719 | |
a2135011 | 5720 | path->keep_locks = 1; |
31533fb2 | 5721 | path->leave_spinning = 1; |
8e73f275 | 5722 | |
3d7806ec JS |
5723 | if (time_seq) |
5724 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5725 | else | |
5726 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5727 | path->keep_locks = 0; |
5728 | ||
5729 | if (ret < 0) | |
5730 | return ret; | |
5731 | ||
a2135011 | 5732 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5733 | /* |
5734 | * by releasing the path above we dropped all our locks. A balance | |
5735 | * could have added more items next to the key that used to be | |
5736 | * at the very end of the block. So, check again here and | |
5737 | * advance the path if there are now more items available. | |
5738 | */ | |
a2135011 | 5739 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5740 | if (ret == 0) |
5741 | path->slots[0]++; | |
8e73f275 | 5742 | ret = 0; |
925baedd CM |
5743 | goto done; |
5744 | } | |
0b43e04f LB |
5745 | /* |
5746 | * So the above check misses one case: | |
5747 | * - after releasing the path above, someone has removed the item that | |
5748 | * used to be at the very end of the block, and balance between leafs | |
5749 | * gets another one with bigger key.offset to replace it. | |
5750 | * | |
5751 | * This one should be returned as well, or we can get leaf corruption | |
5752 | * later(esp. in __btrfs_drop_extents()). | |
5753 | * | |
5754 | * And a bit more explanation about this check, | |
5755 | * with ret > 0, the key isn't found, the path points to the slot | |
5756 | * where it should be inserted, so the path->slots[0] item must be the | |
5757 | * bigger one. | |
5758 | */ | |
5759 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5760 | ret = 0; | |
5761 | goto done; | |
5762 | } | |
d97e63b6 | 5763 | |
d397712b | 5764 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5765 | if (!path->nodes[level]) { |
5766 | ret = 1; | |
5767 | goto done; | |
5768 | } | |
5f39d397 | 5769 | |
d97e63b6 CM |
5770 | slot = path->slots[level] + 1; |
5771 | c = path->nodes[level]; | |
5f39d397 | 5772 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5773 | level++; |
8e73f275 CM |
5774 | if (level == BTRFS_MAX_LEVEL) { |
5775 | ret = 1; | |
5776 | goto done; | |
5777 | } | |
d97e63b6 CM |
5778 | continue; |
5779 | } | |
5f39d397 | 5780 | |
925baedd | 5781 | if (next) { |
bd681513 | 5782 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5783 | free_extent_buffer(next); |
925baedd | 5784 | } |
5f39d397 | 5785 | |
8e73f275 | 5786 | next = c; |
bd681513 | 5787 | next_rw_lock = path->locks[level]; |
d07b8528 | 5788 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5789 | slot, &key); |
8e73f275 CM |
5790 | if (ret == -EAGAIN) |
5791 | goto again; | |
5f39d397 | 5792 | |
76a05b35 | 5793 | if (ret < 0) { |
b3b4aa74 | 5794 | btrfs_release_path(path); |
76a05b35 CM |
5795 | goto done; |
5796 | } | |
5797 | ||
5cd57b2c | 5798 | if (!path->skip_locking) { |
bd681513 | 5799 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5800 | if (!ret && time_seq) { |
5801 | /* | |
5802 | * If we don't get the lock, we may be racing | |
5803 | * with push_leaf_left, holding that lock while | |
5804 | * itself waiting for the leaf we've currently | |
5805 | * locked. To solve this situation, we give up | |
5806 | * on our lock and cycle. | |
5807 | */ | |
cf538830 | 5808 | free_extent_buffer(next); |
d42244a0 JS |
5809 | btrfs_release_path(path); |
5810 | cond_resched(); | |
5811 | goto again; | |
5812 | } | |
8e73f275 CM |
5813 | if (!ret) { |
5814 | btrfs_set_path_blocking(path); | |
bd681513 | 5815 | btrfs_tree_read_lock(next); |
31533fb2 | 5816 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5817 | BTRFS_READ_LOCK); |
8e73f275 | 5818 | } |
31533fb2 | 5819 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5820 | } |
d97e63b6 CM |
5821 | break; |
5822 | } | |
5823 | path->slots[level] = slot; | |
d397712b | 5824 | while (1) { |
d97e63b6 CM |
5825 | level--; |
5826 | c = path->nodes[level]; | |
925baedd | 5827 | if (path->locks[level]) |
bd681513 | 5828 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5829 | |
5f39d397 | 5830 | free_extent_buffer(c); |
d97e63b6 CM |
5831 | path->nodes[level] = next; |
5832 | path->slots[level] = 0; | |
a74a4b97 | 5833 | if (!path->skip_locking) |
bd681513 | 5834 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5835 | if (!level) |
5836 | break; | |
b4ce94de | 5837 | |
d07b8528 | 5838 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5839 | 0, &key); |
8e73f275 CM |
5840 | if (ret == -EAGAIN) |
5841 | goto again; | |
5842 | ||
76a05b35 | 5843 | if (ret < 0) { |
b3b4aa74 | 5844 | btrfs_release_path(path); |
76a05b35 CM |
5845 | goto done; |
5846 | } | |
5847 | ||
5cd57b2c | 5848 | if (!path->skip_locking) { |
bd681513 | 5849 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5850 | if (!ret) { |
5851 | btrfs_set_path_blocking(path); | |
bd681513 | 5852 | btrfs_tree_read_lock(next); |
31533fb2 | 5853 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5854 | BTRFS_READ_LOCK); |
5855 | } | |
31533fb2 | 5856 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5857 | } |
d97e63b6 | 5858 | } |
8e73f275 | 5859 | ret = 0; |
925baedd | 5860 | done: |
f7c79f30 | 5861 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5862 | path->leave_spinning = old_spinning; |
5863 | if (!old_spinning) | |
5864 | btrfs_set_path_blocking(path); | |
5865 | ||
5866 | return ret; | |
d97e63b6 | 5867 | } |
0b86a832 | 5868 | |
3f157a2f CM |
5869 | /* |
5870 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5871 | * searching until it gets past min_objectid or finds an item of 'type' | |
5872 | * | |
5873 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5874 | */ | |
0b86a832 CM |
5875 | int btrfs_previous_item(struct btrfs_root *root, |
5876 | struct btrfs_path *path, u64 min_objectid, | |
5877 | int type) | |
5878 | { | |
5879 | struct btrfs_key found_key; | |
5880 | struct extent_buffer *leaf; | |
e02119d5 | 5881 | u32 nritems; |
0b86a832 CM |
5882 | int ret; |
5883 | ||
d397712b | 5884 | while (1) { |
0b86a832 | 5885 | if (path->slots[0] == 0) { |
b4ce94de | 5886 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5887 | ret = btrfs_prev_leaf(root, path); |
5888 | if (ret != 0) | |
5889 | return ret; | |
5890 | } else { | |
5891 | path->slots[0]--; | |
5892 | } | |
5893 | leaf = path->nodes[0]; | |
e02119d5 CM |
5894 | nritems = btrfs_header_nritems(leaf); |
5895 | if (nritems == 0) | |
5896 | return 1; | |
5897 | if (path->slots[0] == nritems) | |
5898 | path->slots[0]--; | |
5899 | ||
0b86a832 | 5900 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5901 | if (found_key.objectid < min_objectid) |
5902 | break; | |
0a4eefbb YZ |
5903 | if (found_key.type == type) |
5904 | return 0; | |
e02119d5 CM |
5905 | if (found_key.objectid == min_objectid && |
5906 | found_key.type < type) | |
5907 | break; | |
0b86a832 CM |
5908 | } |
5909 | return 1; | |
5910 | } | |
ade2e0b3 WS |
5911 | |
5912 | /* | |
5913 | * search in extent tree to find a previous Metadata/Data extent item with | |
5914 | * min objecitd. | |
5915 | * | |
5916 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5917 | */ | |
5918 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5919 | struct btrfs_path *path, u64 min_objectid) | |
5920 | { | |
5921 | struct btrfs_key found_key; | |
5922 | struct extent_buffer *leaf; | |
5923 | u32 nritems; | |
5924 | int ret; | |
5925 | ||
5926 | while (1) { | |
5927 | if (path->slots[0] == 0) { | |
5928 | btrfs_set_path_blocking(path); | |
5929 | ret = btrfs_prev_leaf(root, path); | |
5930 | if (ret != 0) | |
5931 | return ret; | |
5932 | } else { | |
5933 | path->slots[0]--; | |
5934 | } | |
5935 | leaf = path->nodes[0]; | |
5936 | nritems = btrfs_header_nritems(leaf); | |
5937 | if (nritems == 0) | |
5938 | return 1; | |
5939 | if (path->slots[0] == nritems) | |
5940 | path->slots[0]--; | |
5941 | ||
5942 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5943 | if (found_key.objectid < min_objectid) | |
5944 | break; | |
5945 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5946 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5947 | return 0; | |
5948 | if (found_key.objectid == min_objectid && | |
5949 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5950 | break; | |
5951 | } | |
5952 | return 1; | |
5953 | } |