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 | ||
d352ac68 | 55 | /* this also releases the path */ |
df24a2b9 | 56 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 57 | { |
ff175d57 JJ |
58 | if (!p) |
59 | return; | |
b3b4aa74 | 60 | btrfs_release_path(p); |
df24a2b9 | 61 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
62 | } |
63 | ||
d352ac68 CM |
64 | /* |
65 | * path release drops references on the extent buffers in the path | |
66 | * and it drops any locks held by this path | |
67 | * | |
68 | * It is safe to call this on paths that no locks or extent buffers held. | |
69 | */ | |
b3b4aa74 | 70 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
71 | { |
72 | int i; | |
a2135011 | 73 | |
234b63a0 | 74 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 75 | p->slots[i] = 0; |
eb60ceac | 76 | if (!p->nodes[i]) |
925baedd CM |
77 | continue; |
78 | if (p->locks[i]) { | |
bd681513 | 79 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
80 | p->locks[i] = 0; |
81 | } | |
5f39d397 | 82 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 83 | p->nodes[i] = NULL; |
eb60ceac CM |
84 | } |
85 | } | |
86 | ||
d352ac68 CM |
87 | /* |
88 | * safely gets a reference on the root node of a tree. A lock | |
89 | * is not taken, so a concurrent writer may put a different node | |
90 | * at the root of the tree. See btrfs_lock_root_node for the | |
91 | * looping required. | |
92 | * | |
93 | * The extent buffer returned by this has a reference taken, so | |
94 | * it won't disappear. It may stop being the root of the tree | |
95 | * at any time because there are no locks held. | |
96 | */ | |
925baedd CM |
97 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
98 | { | |
99 | struct extent_buffer *eb; | |
240f62c8 | 100 | |
3083ee2e JB |
101 | while (1) { |
102 | rcu_read_lock(); | |
103 | eb = rcu_dereference(root->node); | |
104 | ||
105 | /* | |
106 | * RCU really hurts here, we could free up the root node because | |
01327610 | 107 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
108 | * the inc_not_zero dance and if it doesn't work then |
109 | * synchronize_rcu and try again. | |
110 | */ | |
111 | if (atomic_inc_not_zero(&eb->refs)) { | |
112 | rcu_read_unlock(); | |
113 | break; | |
114 | } | |
115 | rcu_read_unlock(); | |
116 | synchronize_rcu(); | |
117 | } | |
925baedd CM |
118 | return eb; |
119 | } | |
120 | ||
d352ac68 CM |
121 | /* loop around taking references on and locking the root node of the |
122 | * tree until you end up with a lock on the root. A locked buffer | |
123 | * is returned, with a reference held. | |
124 | */ | |
925baedd CM |
125 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
126 | { | |
127 | struct extent_buffer *eb; | |
128 | ||
d397712b | 129 | while (1) { |
925baedd CM |
130 | eb = btrfs_root_node(root); |
131 | btrfs_tree_lock(eb); | |
240f62c8 | 132 | if (eb == root->node) |
925baedd | 133 | break; |
925baedd CM |
134 | btrfs_tree_unlock(eb); |
135 | free_extent_buffer(eb); | |
136 | } | |
137 | return eb; | |
138 | } | |
139 | ||
bd681513 CM |
140 | /* loop around taking references on and locking the root node of the |
141 | * tree until you end up with a lock on the root. A locked buffer | |
142 | * is returned, with a reference held. | |
143 | */ | |
84f7d8e6 | 144 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
145 | { |
146 | struct extent_buffer *eb; | |
147 | ||
148 | while (1) { | |
149 | eb = btrfs_root_node(root); | |
150 | btrfs_tree_read_lock(eb); | |
151 | if (eb == root->node) | |
152 | break; | |
153 | btrfs_tree_read_unlock(eb); | |
154 | free_extent_buffer(eb); | |
155 | } | |
156 | return eb; | |
157 | } | |
158 | ||
d352ac68 CM |
159 | /* cowonly root (everything not a reference counted cow subvolume), just get |
160 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
161 | * get properly updated on disk. | |
162 | */ | |
0b86a832 CM |
163 | static void add_root_to_dirty_list(struct btrfs_root *root) |
164 | { | |
0b246afa JM |
165 | struct btrfs_fs_info *fs_info = root->fs_info; |
166 | ||
e7070be1 JB |
167 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
168 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
169 | return; | |
170 | ||
0b246afa | 171 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
172 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
173 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 174 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 175 | list_move_tail(&root->dirty_list, |
0b246afa | 176 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
177 | else |
178 | list_move(&root->dirty_list, | |
0b246afa | 179 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 180 | } |
0b246afa | 181 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
182 | } |
183 | ||
d352ac68 CM |
184 | /* |
185 | * used by snapshot creation to make a copy of a root for a tree with | |
186 | * a given objectid. The buffer with the new root node is returned in | |
187 | * cow_ret, and this func returns zero on success or a negative error code. | |
188 | */ | |
be20aa9d CM |
189 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
190 | struct btrfs_root *root, | |
191 | struct extent_buffer *buf, | |
192 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
193 | { | |
0b246afa | 194 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 195 | struct extent_buffer *cow; |
be20aa9d CM |
196 | int ret = 0; |
197 | int level; | |
5d4f98a2 | 198 | struct btrfs_disk_key disk_key; |
be20aa9d | 199 | |
27cdeb70 | 200 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 201 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
202 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
203 | trans->transid != root->last_trans); | |
be20aa9d CM |
204 | |
205 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
206 | if (level == 0) |
207 | btrfs_item_key(buf, &disk_key, 0); | |
208 | else | |
209 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 210 | |
4d75f8a9 DS |
211 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
212 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 213 | if (IS_ERR(cow)) |
be20aa9d CM |
214 | return PTR_ERR(cow); |
215 | ||
58e8012c | 216 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
217 | btrfs_set_header_bytenr(cow, cow->start); |
218 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
219 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
220 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
221 | BTRFS_HEADER_FLAG_RELOC); | |
222 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
223 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
224 | else | |
225 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 226 | |
0b246afa | 227 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 228 | |
be20aa9d | 229 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 230 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 231 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 232 | else |
e339a6b0 | 233 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 234 | |
be20aa9d CM |
235 | if (ret) |
236 | return ret; | |
237 | ||
238 | btrfs_mark_buffer_dirty(cow); | |
239 | *cow_ret = cow; | |
240 | return 0; | |
241 | } | |
242 | ||
bd989ba3 JS |
243 | enum mod_log_op { |
244 | MOD_LOG_KEY_REPLACE, | |
245 | MOD_LOG_KEY_ADD, | |
246 | MOD_LOG_KEY_REMOVE, | |
247 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
248 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
249 | MOD_LOG_MOVE_KEYS, | |
250 | MOD_LOG_ROOT_REPLACE, | |
251 | }; | |
252 | ||
bd989ba3 JS |
253 | struct tree_mod_root { |
254 | u64 logical; | |
255 | u8 level; | |
256 | }; | |
257 | ||
258 | struct tree_mod_elem { | |
259 | struct rb_node node; | |
298cfd36 | 260 | u64 logical; |
097b8a7c | 261 | u64 seq; |
bd989ba3 JS |
262 | enum mod_log_op op; |
263 | ||
264 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
265 | int slot; | |
266 | ||
267 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
268 | u64 generation; | |
269 | ||
270 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
271 | struct btrfs_disk_key key; | |
272 | u64 blockptr; | |
273 | ||
274 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
275 | struct { |
276 | int dst_slot; | |
277 | int nr_items; | |
278 | } move; | |
bd989ba3 JS |
279 | |
280 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
281 | struct tree_mod_root old_root; | |
282 | }; | |
283 | ||
fc36ed7e | 284 | /* |
fcebe456 | 285 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 286 | */ |
fcebe456 | 287 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
288 | { |
289 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
290 | } | |
291 | ||
097b8a7c JS |
292 | /* |
293 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
294 | * passed does not already have a sequence number set. So when a caller expects | |
295 | * to record tree modifications, it should ensure to set elem->seq to zero | |
296 | * before calling btrfs_get_tree_mod_seq. | |
297 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
298 | * blocker was added. | |
299 | */ | |
300 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
301 | struct seq_list *elem) | |
bd989ba3 | 302 | { |
b1a09f1e | 303 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 304 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 305 | if (!elem->seq) { |
fcebe456 | 306 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
307 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
308 | } | |
bd989ba3 | 309 | spin_unlock(&fs_info->tree_mod_seq_lock); |
b1a09f1e | 310 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 311 | |
fcebe456 | 312 | return elem->seq; |
bd989ba3 JS |
313 | } |
314 | ||
315 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
316 | struct seq_list *elem) | |
317 | { | |
318 | struct rb_root *tm_root; | |
319 | struct rb_node *node; | |
320 | struct rb_node *next; | |
321 | struct seq_list *cur_elem; | |
322 | struct tree_mod_elem *tm; | |
323 | u64 min_seq = (u64)-1; | |
324 | u64 seq_putting = elem->seq; | |
325 | ||
326 | if (!seq_putting) | |
327 | return; | |
328 | ||
bd989ba3 JS |
329 | spin_lock(&fs_info->tree_mod_seq_lock); |
330 | list_del(&elem->list); | |
097b8a7c | 331 | elem->seq = 0; |
bd989ba3 JS |
332 | |
333 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 334 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
335 | if (seq_putting > cur_elem->seq) { |
336 | /* | |
337 | * blocker with lower sequence number exists, we | |
338 | * cannot remove anything from the log | |
339 | */ | |
097b8a7c JS |
340 | spin_unlock(&fs_info->tree_mod_seq_lock); |
341 | return; | |
bd989ba3 JS |
342 | } |
343 | min_seq = cur_elem->seq; | |
344 | } | |
345 | } | |
097b8a7c JS |
346 | spin_unlock(&fs_info->tree_mod_seq_lock); |
347 | ||
bd989ba3 JS |
348 | /* |
349 | * anything that's lower than the lowest existing (read: blocked) | |
350 | * sequence number can be removed from the tree. | |
351 | */ | |
b1a09f1e | 352 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
353 | tm_root = &fs_info->tree_mod_log; |
354 | for (node = rb_first(tm_root); node; node = next) { | |
355 | next = rb_next(node); | |
6b4df8b6 | 356 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 357 | if (tm->seq > min_seq) |
bd989ba3 JS |
358 | continue; |
359 | rb_erase(node, tm_root); | |
bd989ba3 JS |
360 | kfree(tm); |
361 | } | |
b1a09f1e | 362 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
363 | } |
364 | ||
365 | /* | |
366 | * key order of the log: | |
298cfd36 | 367 | * node/leaf start address -> sequence |
bd989ba3 | 368 | * |
298cfd36 CR |
369 | * The 'start address' is the logical address of the *new* root node |
370 | * for root replace operations, or the logical address of the affected | |
371 | * block for all other operations. | |
5de865ee | 372 | * |
b1a09f1e | 373 | * Note: must be called with write lock for fs_info::tree_mod_log_lock. |
bd989ba3 JS |
374 | */ |
375 | static noinline int | |
376 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
377 | { | |
378 | struct rb_root *tm_root; | |
379 | struct rb_node **new; | |
380 | struct rb_node *parent = NULL; | |
381 | struct tree_mod_elem *cur; | |
c8cc6341 | 382 | |
fcebe456 | 383 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 384 | |
bd989ba3 JS |
385 | tm_root = &fs_info->tree_mod_log; |
386 | new = &tm_root->rb_node; | |
387 | while (*new) { | |
6b4df8b6 | 388 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 389 | parent = *new; |
298cfd36 | 390 | if (cur->logical < tm->logical) |
bd989ba3 | 391 | new = &((*new)->rb_left); |
298cfd36 | 392 | else if (cur->logical > tm->logical) |
bd989ba3 | 393 | new = &((*new)->rb_right); |
097b8a7c | 394 | else if (cur->seq < tm->seq) |
bd989ba3 | 395 | new = &((*new)->rb_left); |
097b8a7c | 396 | else if (cur->seq > tm->seq) |
bd989ba3 | 397 | new = &((*new)->rb_right); |
5de865ee FDBM |
398 | else |
399 | return -EEXIST; | |
bd989ba3 JS |
400 | } |
401 | ||
402 | rb_link_node(&tm->node, parent, new); | |
403 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 404 | return 0; |
bd989ba3 JS |
405 | } |
406 | ||
097b8a7c JS |
407 | /* |
408 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
409 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
410 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 411 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 412 | */ |
e9b7fd4d JS |
413 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
414 | struct extent_buffer *eb) { | |
415 | smp_mb(); | |
416 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
417 | return 1; | |
097b8a7c JS |
418 | if (eb && btrfs_header_level(eb) == 0) |
419 | return 1; | |
5de865ee | 420 | |
b1a09f1e | 421 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 422 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 423 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
424 | return 1; |
425 | } | |
426 | ||
e9b7fd4d JS |
427 | return 0; |
428 | } | |
429 | ||
5de865ee FDBM |
430 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
431 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
432 | struct extent_buffer *eb) | |
433 | { | |
434 | smp_mb(); | |
435 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
436 | return 0; | |
437 | if (eb && btrfs_header_level(eb) == 0) | |
438 | return 0; | |
439 | ||
440 | return 1; | |
441 | } | |
442 | ||
443 | static struct tree_mod_elem * | |
444 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
445 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 446 | { |
097b8a7c | 447 | struct tree_mod_elem *tm; |
bd989ba3 | 448 | |
c8cc6341 JB |
449 | tm = kzalloc(sizeof(*tm), flags); |
450 | if (!tm) | |
5de865ee | 451 | return NULL; |
bd989ba3 | 452 | |
298cfd36 | 453 | tm->logical = eb->start; |
bd989ba3 JS |
454 | if (op != MOD_LOG_KEY_ADD) { |
455 | btrfs_node_key(eb, &tm->key, slot); | |
456 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
457 | } | |
458 | tm->op = op; | |
459 | tm->slot = slot; | |
460 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 461 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 462 | |
5de865ee | 463 | return tm; |
097b8a7c JS |
464 | } |
465 | ||
e09c2efe DS |
466 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
467 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 468 | { |
5de865ee FDBM |
469 | struct tree_mod_elem *tm; |
470 | int ret; | |
471 | ||
e09c2efe | 472 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
473 | return 0; |
474 | ||
475 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
476 | if (!tm) | |
477 | return -ENOMEM; | |
478 | ||
e09c2efe | 479 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 480 | kfree(tm); |
097b8a7c | 481 | return 0; |
5de865ee FDBM |
482 | } |
483 | ||
e09c2efe | 484 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 485 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
486 | if (ret) |
487 | kfree(tm); | |
097b8a7c | 488 | |
5de865ee | 489 | return ret; |
097b8a7c JS |
490 | } |
491 | ||
6074d45f DS |
492 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
493 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 494 | { |
5de865ee FDBM |
495 | struct tree_mod_elem *tm = NULL; |
496 | struct tree_mod_elem **tm_list = NULL; | |
497 | int ret = 0; | |
bd989ba3 | 498 | int i; |
5de865ee | 499 | int locked = 0; |
bd989ba3 | 500 | |
6074d45f | 501 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 502 | return 0; |
bd989ba3 | 503 | |
176ef8f5 | 504 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
505 | if (!tm_list) |
506 | return -ENOMEM; | |
507 | ||
176ef8f5 | 508 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
509 | if (!tm) { |
510 | ret = -ENOMEM; | |
511 | goto free_tms; | |
512 | } | |
513 | ||
298cfd36 | 514 | tm->logical = eb->start; |
5de865ee FDBM |
515 | tm->slot = src_slot; |
516 | tm->move.dst_slot = dst_slot; | |
517 | tm->move.nr_items = nr_items; | |
518 | tm->op = MOD_LOG_MOVE_KEYS; | |
519 | ||
520 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
521 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 522 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
523 | if (!tm_list[i]) { |
524 | ret = -ENOMEM; | |
525 | goto free_tms; | |
526 | } | |
527 | } | |
528 | ||
6074d45f | 529 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
530 | goto free_tms; |
531 | locked = 1; | |
532 | ||
01763a2e JS |
533 | /* |
534 | * When we override something during the move, we log these removals. | |
535 | * This can only happen when we move towards the beginning of the | |
536 | * buffer, i.e. dst_slot < src_slot. | |
537 | */ | |
bd989ba3 | 538 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 539 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
540 | if (ret) |
541 | goto free_tms; | |
bd989ba3 JS |
542 | } |
543 | ||
6074d45f | 544 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
545 | if (ret) |
546 | goto free_tms; | |
b1a09f1e | 547 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 548 | kfree(tm_list); |
f395694c | 549 | |
5de865ee FDBM |
550 | return 0; |
551 | free_tms: | |
552 | for (i = 0; i < nr_items; i++) { | |
553 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 554 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
555 | kfree(tm_list[i]); |
556 | } | |
557 | if (locked) | |
b1a09f1e | 558 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
559 | kfree(tm_list); |
560 | kfree(tm); | |
bd989ba3 | 561 | |
5de865ee | 562 | return ret; |
bd989ba3 JS |
563 | } |
564 | ||
5de865ee FDBM |
565 | static inline int |
566 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
567 | struct tree_mod_elem **tm_list, | |
568 | int nritems) | |
097b8a7c | 569 | { |
5de865ee | 570 | int i, j; |
097b8a7c JS |
571 | int ret; |
572 | ||
097b8a7c | 573 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
574 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
575 | if (ret) { | |
576 | for (j = nritems - 1; j > i; j--) | |
577 | rb_erase(&tm_list[j]->node, | |
578 | &fs_info->tree_mod_log); | |
579 | return ret; | |
580 | } | |
097b8a7c | 581 | } |
5de865ee FDBM |
582 | |
583 | return 0; | |
097b8a7c JS |
584 | } |
585 | ||
95b757c1 DS |
586 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
587 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 588 | { |
95b757c1 | 589 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
590 | struct tree_mod_elem *tm = NULL; |
591 | struct tree_mod_elem **tm_list = NULL; | |
592 | int nritems = 0; | |
593 | int ret = 0; | |
594 | int i; | |
bd989ba3 | 595 | |
5de865ee | 596 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
597 | return 0; |
598 | ||
5de865ee FDBM |
599 | if (log_removal && btrfs_header_level(old_root) > 0) { |
600 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 601 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 602 | GFP_NOFS); |
5de865ee FDBM |
603 | if (!tm_list) { |
604 | ret = -ENOMEM; | |
605 | goto free_tms; | |
606 | } | |
607 | for (i = 0; i < nritems; i++) { | |
608 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 609 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
610 | if (!tm_list[i]) { |
611 | ret = -ENOMEM; | |
612 | goto free_tms; | |
613 | } | |
614 | } | |
615 | } | |
d9abbf1c | 616 | |
bcc8e07f | 617 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
618 | if (!tm) { |
619 | ret = -ENOMEM; | |
620 | goto free_tms; | |
621 | } | |
bd989ba3 | 622 | |
298cfd36 | 623 | tm->logical = new_root->start; |
bd989ba3 JS |
624 | tm->old_root.logical = old_root->start; |
625 | tm->old_root.level = btrfs_header_level(old_root); | |
626 | tm->generation = btrfs_header_generation(old_root); | |
627 | tm->op = MOD_LOG_ROOT_REPLACE; | |
628 | ||
5de865ee FDBM |
629 | if (tree_mod_dont_log(fs_info, NULL)) |
630 | goto free_tms; | |
631 | ||
632 | if (tm_list) | |
633 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
634 | if (!ret) | |
635 | ret = __tree_mod_log_insert(fs_info, tm); | |
636 | ||
b1a09f1e | 637 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
638 | if (ret) |
639 | goto free_tms; | |
640 | kfree(tm_list); | |
641 | ||
642 | return ret; | |
643 | ||
644 | free_tms: | |
645 | if (tm_list) { | |
646 | for (i = 0; i < nritems; i++) | |
647 | kfree(tm_list[i]); | |
648 | kfree(tm_list); | |
649 | } | |
650 | kfree(tm); | |
651 | ||
652 | return ret; | |
bd989ba3 JS |
653 | } |
654 | ||
655 | static struct tree_mod_elem * | |
656 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
657 | int smallest) | |
658 | { | |
659 | struct rb_root *tm_root; | |
660 | struct rb_node *node; | |
661 | struct tree_mod_elem *cur = NULL; | |
662 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 663 | |
b1a09f1e | 664 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
665 | tm_root = &fs_info->tree_mod_log; |
666 | node = tm_root->rb_node; | |
667 | while (node) { | |
6b4df8b6 | 668 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 669 | if (cur->logical < start) { |
bd989ba3 | 670 | node = node->rb_left; |
298cfd36 | 671 | } else if (cur->logical > start) { |
bd989ba3 | 672 | node = node->rb_right; |
097b8a7c | 673 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
674 | node = node->rb_left; |
675 | } else if (!smallest) { | |
676 | /* we want the node with the highest seq */ | |
677 | if (found) | |
097b8a7c | 678 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
679 | found = cur; |
680 | node = node->rb_left; | |
097b8a7c | 681 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
682 | /* we want the node with the smallest seq */ |
683 | if (found) | |
097b8a7c | 684 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
685 | found = cur; |
686 | node = node->rb_right; | |
687 | } else { | |
688 | found = cur; | |
689 | break; | |
690 | } | |
691 | } | |
b1a09f1e | 692 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
693 | |
694 | return found; | |
695 | } | |
696 | ||
697 | /* | |
698 | * this returns the element from the log with the smallest time sequence | |
699 | * value that's in the log (the oldest log item). any element with a time | |
700 | * sequence lower than min_seq will be ignored. | |
701 | */ | |
702 | static struct tree_mod_elem * | |
703 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
704 | u64 min_seq) | |
705 | { | |
706 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
707 | } | |
708 | ||
709 | /* | |
710 | * this returns the element from the log with the largest time sequence | |
711 | * value that's in the log (the most recent log item). any element with | |
712 | * a time sequence lower than min_seq will be ignored. | |
713 | */ | |
714 | static struct tree_mod_elem * | |
715 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
716 | { | |
717 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
718 | } | |
719 | ||
5de865ee | 720 | static noinline int |
bd989ba3 JS |
721 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
722 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 723 | unsigned long src_offset, int nr_items) |
bd989ba3 | 724 | { |
5de865ee FDBM |
725 | int ret = 0; |
726 | struct tree_mod_elem **tm_list = NULL; | |
727 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 728 | int i; |
5de865ee | 729 | int locked = 0; |
bd989ba3 | 730 | |
5de865ee FDBM |
731 | if (!tree_mod_need_log(fs_info, NULL)) |
732 | return 0; | |
bd989ba3 | 733 | |
c8cc6341 | 734 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
735 | return 0; |
736 | ||
31e818fe | 737 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
738 | GFP_NOFS); |
739 | if (!tm_list) | |
740 | return -ENOMEM; | |
bd989ba3 | 741 | |
5de865ee FDBM |
742 | tm_list_add = tm_list; |
743 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 744 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
745 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
746 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
747 | if (!tm_list_rem[i]) { | |
748 | ret = -ENOMEM; | |
749 | goto free_tms; | |
750 | } | |
751 | ||
752 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
753 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
754 | if (!tm_list_add[i]) { | |
755 | ret = -ENOMEM; | |
756 | goto free_tms; | |
757 | } | |
758 | } | |
759 | ||
760 | if (tree_mod_dont_log(fs_info, NULL)) | |
761 | goto free_tms; | |
762 | locked = 1; | |
763 | ||
764 | for (i = 0; i < nr_items; i++) { | |
765 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
766 | if (ret) | |
767 | goto free_tms; | |
768 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
769 | if (ret) | |
770 | goto free_tms; | |
bd989ba3 | 771 | } |
5de865ee | 772 | |
b1a09f1e | 773 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
774 | kfree(tm_list); |
775 | ||
776 | return 0; | |
777 | ||
778 | free_tms: | |
779 | for (i = 0; i < nr_items * 2; i++) { | |
780 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
781 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
782 | kfree(tm_list[i]); | |
783 | } | |
784 | if (locked) | |
b1a09f1e | 785 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
786 | kfree(tm_list); |
787 | ||
788 | return ret; | |
bd989ba3 JS |
789 | } |
790 | ||
db7279a2 | 791 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 792 | { |
5de865ee FDBM |
793 | struct tree_mod_elem **tm_list = NULL; |
794 | int nritems = 0; | |
795 | int i; | |
796 | int ret = 0; | |
797 | ||
798 | if (btrfs_header_level(eb) == 0) | |
799 | return 0; | |
800 | ||
db7279a2 | 801 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
802 | return 0; |
803 | ||
804 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 805 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
806 | if (!tm_list) |
807 | return -ENOMEM; | |
808 | ||
809 | for (i = 0; i < nritems; i++) { | |
810 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
811 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
812 | if (!tm_list[i]) { | |
813 | ret = -ENOMEM; | |
814 | goto free_tms; | |
815 | } | |
816 | } | |
817 | ||
db7279a2 | 818 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
819 | goto free_tms; |
820 | ||
db7279a2 | 821 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 822 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
823 | if (ret) |
824 | goto free_tms; | |
825 | kfree(tm_list); | |
826 | ||
827 | return 0; | |
828 | ||
829 | free_tms: | |
830 | for (i = 0; i < nritems; i++) | |
831 | kfree(tm_list[i]); | |
832 | kfree(tm_list); | |
833 | ||
834 | return ret; | |
bd989ba3 JS |
835 | } |
836 | ||
5d4f98a2 YZ |
837 | /* |
838 | * check if the tree block can be shared by multiple trees | |
839 | */ | |
840 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
841 | struct extent_buffer *buf) | |
842 | { | |
843 | /* | |
01327610 | 844 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
845 | * are never shared. If a block was allocated after the last |
846 | * snapshot and the block was not allocated by tree relocation, | |
847 | * we know the block is not shared. | |
848 | */ | |
27cdeb70 | 849 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
850 | buf != root->node && buf != root->commit_root && |
851 | (btrfs_header_generation(buf) <= | |
852 | btrfs_root_last_snapshot(&root->root_item) || | |
853 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
854 | return 1; | |
a79865c6 | 855 | |
5d4f98a2 YZ |
856 | return 0; |
857 | } | |
858 | ||
859 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
860 | struct btrfs_root *root, | |
861 | struct extent_buffer *buf, | |
f0486c68 YZ |
862 | struct extent_buffer *cow, |
863 | int *last_ref) | |
5d4f98a2 | 864 | { |
0b246afa | 865 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
866 | u64 refs; |
867 | u64 owner; | |
868 | u64 flags; | |
869 | u64 new_flags = 0; | |
870 | int ret; | |
871 | ||
872 | /* | |
873 | * Backrefs update rules: | |
874 | * | |
875 | * Always use full backrefs for extent pointers in tree block | |
876 | * allocated by tree relocation. | |
877 | * | |
878 | * If a shared tree block is no longer referenced by its owner | |
879 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
880 | * use full backrefs for extent pointers in tree block. | |
881 | * | |
882 | * If a tree block is been relocating | |
883 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
884 | * use full backrefs for extent pointers in tree block. | |
885 | * The reason for this is some operations (such as drop tree) | |
886 | * are only allowed for blocks use full backrefs. | |
887 | */ | |
888 | ||
889 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 890 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
891 | btrfs_header_level(buf), 1, |
892 | &refs, &flags); | |
be1a5564 MF |
893 | if (ret) |
894 | return ret; | |
e5df9573 MF |
895 | if (refs == 0) { |
896 | ret = -EROFS; | |
0b246afa | 897 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
898 | return ret; |
899 | } | |
5d4f98a2 YZ |
900 | } else { |
901 | refs = 1; | |
902 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
903 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
904 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
905 | else | |
906 | flags = 0; | |
907 | } | |
908 | ||
909 | owner = btrfs_header_owner(buf); | |
910 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
911 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
912 | ||
913 | if (refs > 1) { | |
914 | if ((owner == root->root_key.objectid || | |
915 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
916 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 917 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
918 | if (ret) |
919 | return ret; | |
5d4f98a2 YZ |
920 | |
921 | if (root->root_key.objectid == | |
922 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 923 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
924 | if (ret) |
925 | return ret; | |
e339a6b0 | 926 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
927 | if (ret) |
928 | return ret; | |
5d4f98a2 YZ |
929 | } |
930 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
931 | } else { | |
932 | ||
933 | if (root->root_key.objectid == | |
934 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 935 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 936 | else |
e339a6b0 | 937 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
938 | if (ret) |
939 | return ret; | |
5d4f98a2 YZ |
940 | } |
941 | if (new_flags != 0) { | |
b1c79e09 JB |
942 | int level = btrfs_header_level(buf); |
943 | ||
2ff7e61e | 944 | ret = btrfs_set_disk_extent_flags(trans, fs_info, |
5d4f98a2 YZ |
945 | buf->start, |
946 | buf->len, | |
b1c79e09 | 947 | new_flags, level, 0); |
be1a5564 MF |
948 | if (ret) |
949 | return ret; | |
5d4f98a2 YZ |
950 | } |
951 | } else { | |
952 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
953 | if (root->root_key.objectid == | |
954 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 955 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 956 | else |
e339a6b0 | 957 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
958 | if (ret) |
959 | return ret; | |
e339a6b0 | 960 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
961 | if (ret) |
962 | return ret; | |
5d4f98a2 | 963 | } |
7c302b49 | 964 | clean_tree_block(fs_info, buf); |
f0486c68 | 965 | *last_ref = 1; |
5d4f98a2 YZ |
966 | } |
967 | return 0; | |
968 | } | |
969 | ||
d352ac68 | 970 | /* |
d397712b CM |
971 | * does the dirty work in cow of a single block. The parent block (if |
972 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
973 | * dirty and returned locked. If you modify the block it needs to be marked | |
974 | * dirty again. | |
d352ac68 CM |
975 | * |
976 | * search_start -- an allocation hint for the new block | |
977 | * | |
d397712b CM |
978 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
979 | * bytes the allocator should try to find free next to the block it returns. | |
980 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 981 | */ |
d397712b | 982 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
983 | struct btrfs_root *root, |
984 | struct extent_buffer *buf, | |
985 | struct extent_buffer *parent, int parent_slot, | |
986 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 987 | u64 search_start, u64 empty_size) |
02217ed2 | 988 | { |
0b246afa | 989 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 990 | struct btrfs_disk_key disk_key; |
5f39d397 | 991 | struct extent_buffer *cow; |
be1a5564 | 992 | int level, ret; |
f0486c68 | 993 | int last_ref = 0; |
925baedd | 994 | int unlock_orig = 0; |
0f5053eb | 995 | u64 parent_start = 0; |
7bb86316 | 996 | |
925baedd CM |
997 | if (*cow_ret == buf) |
998 | unlock_orig = 1; | |
999 | ||
b9447ef8 | 1000 | btrfs_assert_tree_locked(buf); |
925baedd | 1001 | |
27cdeb70 | 1002 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1003 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1004 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1005 | trans->transid != root->last_trans); | |
5f39d397 | 1006 | |
7bb86316 | 1007 | level = btrfs_header_level(buf); |
31840ae1 | 1008 | |
5d4f98a2 YZ |
1009 | if (level == 0) |
1010 | btrfs_item_key(buf, &disk_key, 0); | |
1011 | else | |
1012 | btrfs_node_key(buf, &disk_key, 0); | |
1013 | ||
0f5053eb GR |
1014 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1015 | parent_start = parent->start; | |
5d4f98a2 | 1016 | |
4d75f8a9 DS |
1017 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1018 | root->root_key.objectid, &disk_key, level, | |
1019 | search_start, empty_size); | |
54aa1f4d CM |
1020 | if (IS_ERR(cow)) |
1021 | return PTR_ERR(cow); | |
6702ed49 | 1022 | |
b4ce94de CM |
1023 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1024 | ||
58e8012c | 1025 | copy_extent_buffer_full(cow, buf); |
db94535d | 1026 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1027 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1028 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1029 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1030 | BTRFS_HEADER_FLAG_RELOC); | |
1031 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1032 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1033 | else | |
1034 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1035 | |
0b246afa | 1036 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 1037 | |
be1a5564 | 1038 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1039 | if (ret) { |
66642832 | 1040 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1041 | return ret; |
1042 | } | |
1a40e23b | 1043 | |
27cdeb70 | 1044 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1045 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1046 | if (ret) { |
66642832 | 1047 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1048 | return ret; |
93314e3b | 1049 | } |
83d4cfd4 | 1050 | } |
3fd0a558 | 1051 | |
02217ed2 | 1052 | if (buf == root->node) { |
925baedd | 1053 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1054 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1055 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1056 | parent_start = buf->start; | |
925baedd | 1057 | |
5f39d397 | 1058 | extent_buffer_get(cow); |
d9d19a01 DS |
1059 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1060 | BUG_ON(ret < 0); | |
240f62c8 | 1061 | rcu_assign_pointer(root->node, cow); |
925baedd | 1062 | |
f0486c68 | 1063 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1064 | last_ref); |
5f39d397 | 1065 | free_extent_buffer(buf); |
0b86a832 | 1066 | add_root_to_dirty_list(root); |
02217ed2 | 1067 | } else { |
5d4f98a2 | 1068 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1069 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1070 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1071 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1072 | cow->start); |
74493f7a CM |
1073 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1074 | trans->transid); | |
d6025579 | 1075 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1076 | if (last_ref) { |
db7279a2 | 1077 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1078 | if (ret) { |
66642832 | 1079 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1080 | return ret; |
1081 | } | |
1082 | } | |
f0486c68 | 1083 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1084 | last_ref); |
02217ed2 | 1085 | } |
925baedd CM |
1086 | if (unlock_orig) |
1087 | btrfs_tree_unlock(buf); | |
3083ee2e | 1088 | free_extent_buffer_stale(buf); |
ccd467d6 | 1089 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1090 | *cow_ret = cow; |
02217ed2 CM |
1091 | return 0; |
1092 | } | |
1093 | ||
5d9e75c4 JS |
1094 | /* |
1095 | * returns the logical address of the oldest predecessor of the given root. | |
1096 | * entries older than time_seq are ignored. | |
1097 | */ | |
bcd24dab DS |
1098 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1099 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1100 | { |
1101 | struct tree_mod_elem *tm; | |
1102 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1103 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1104 | int looped = 0; |
1105 | ||
1106 | if (!time_seq) | |
35a3621b | 1107 | return NULL; |
5d9e75c4 JS |
1108 | |
1109 | /* | |
298cfd36 CR |
1110 | * the very last operation that's logged for a root is the |
1111 | * replacement operation (if it is replaced at all). this has | |
1112 | * the logical address of the *new* root, making it the very | |
1113 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1114 | */ |
1115 | while (1) { | |
bcd24dab | 1116 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1117 | time_seq); |
1118 | if (!looped && !tm) | |
35a3621b | 1119 | return NULL; |
5d9e75c4 | 1120 | /* |
28da9fb4 JS |
1121 | * if there are no tree operation for the oldest root, we simply |
1122 | * return it. this should only happen if that (old) root is at | |
1123 | * level 0. | |
5d9e75c4 | 1124 | */ |
28da9fb4 JS |
1125 | if (!tm) |
1126 | break; | |
5d9e75c4 | 1127 | |
28da9fb4 JS |
1128 | /* |
1129 | * if there's an operation that's not a root replacement, we | |
1130 | * found the oldest version of our root. normally, we'll find a | |
1131 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1132 | */ | |
5d9e75c4 JS |
1133 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1134 | break; | |
1135 | ||
1136 | found = tm; | |
1137 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1138 | looped = 1; |
1139 | } | |
1140 | ||
a95236d9 JS |
1141 | /* if there's no old root to return, return what we found instead */ |
1142 | if (!found) | |
1143 | found = tm; | |
1144 | ||
5d9e75c4 JS |
1145 | return found; |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1150 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1151 | * time_seq). |
1152 | */ | |
1153 | static void | |
f1ca7e98 JB |
1154 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1155 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1156 | { |
1157 | u32 n; | |
1158 | struct rb_node *next; | |
1159 | struct tree_mod_elem *tm = first_tm; | |
1160 | unsigned long o_dst; | |
1161 | unsigned long o_src; | |
1162 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1163 | ||
1164 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1165 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1166 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1167 | /* |
1168 | * all the operations are recorded with the operator used for | |
1169 | * the modification. as we're going backwards, we do the | |
1170 | * opposite of each operation here. | |
1171 | */ | |
1172 | switch (tm->op) { | |
1173 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1174 | BUG_ON(tm->slot < n); | |
1c697d4a | 1175 | /* Fallthrough */ |
95c80bb1 | 1176 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1177 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1178 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1179 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1180 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1181 | tm->generation); | |
4c3e6969 | 1182 | n++; |
5d9e75c4 JS |
1183 | break; |
1184 | case MOD_LOG_KEY_REPLACE: | |
1185 | BUG_ON(tm->slot >= n); | |
1186 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1187 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1188 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1189 | tm->generation); | |
1190 | break; | |
1191 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1192 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1193 | n--; |
1194 | break; | |
1195 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1196 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1197 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1198 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1199 | tm->move.nr_items * p_size); |
1200 | break; | |
1201 | case MOD_LOG_ROOT_REPLACE: | |
1202 | /* | |
1203 | * this operation is special. for roots, this must be | |
1204 | * handled explicitly before rewinding. | |
1205 | * for non-roots, this operation may exist if the node | |
1206 | * was a root: root A -> child B; then A gets empty and | |
1207 | * B is promoted to the new root. in the mod log, we'll | |
1208 | * have a root-replace operation for B, a tree block | |
1209 | * that is no root. we simply ignore that operation. | |
1210 | */ | |
1211 | break; | |
1212 | } | |
1213 | next = rb_next(&tm->node); | |
1214 | if (!next) | |
1215 | break; | |
6b4df8b6 | 1216 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1217 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1218 | break; |
1219 | } | |
b1a09f1e | 1220 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1221 | btrfs_set_header_nritems(eb, n); |
1222 | } | |
1223 | ||
47fb091f | 1224 | /* |
01327610 | 1225 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1226 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1227 | * returned buffer is always read-locked. If the returned buffer is not the | |
1228 | * input buffer, the lock on the input buffer is released and the input buffer | |
1229 | * is freed (its refcount is decremented). | |
1230 | */ | |
5d9e75c4 | 1231 | static struct extent_buffer * |
9ec72677 JB |
1232 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1233 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1234 | { |
1235 | struct extent_buffer *eb_rewin; | |
1236 | struct tree_mod_elem *tm; | |
1237 | ||
1238 | if (!time_seq) | |
1239 | return eb; | |
1240 | ||
1241 | if (btrfs_header_level(eb) == 0) | |
1242 | return eb; | |
1243 | ||
1244 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1245 | if (!tm) | |
1246 | return eb; | |
1247 | ||
9ec72677 JB |
1248 | btrfs_set_path_blocking(path); |
1249 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1250 | ||
5d9e75c4 JS |
1251 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1252 | BUG_ON(tm->slot != 0); | |
da17066c | 1253 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1254 | if (!eb_rewin) { |
9ec72677 | 1255 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1256 | free_extent_buffer(eb); |
1257 | return NULL; | |
1258 | } | |
5d9e75c4 JS |
1259 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1260 | btrfs_set_header_backref_rev(eb_rewin, | |
1261 | btrfs_header_backref_rev(eb)); | |
1262 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1263 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1264 | } else { |
1265 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1266 | if (!eb_rewin) { |
9ec72677 | 1267 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1268 | free_extent_buffer(eb); |
1269 | return NULL; | |
1270 | } | |
5d9e75c4 JS |
1271 | } |
1272 | ||
9ec72677 | 1273 | btrfs_tree_read_unlock_blocking(eb); |
5d9e75c4 JS |
1274 | free_extent_buffer(eb); |
1275 | ||
47fb091f JS |
1276 | extent_buffer_get(eb_rewin); |
1277 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1278 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1279 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1280 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1281 | |
1282 | return eb_rewin; | |
1283 | } | |
1284 | ||
8ba97a15 JS |
1285 | /* |
1286 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1287 | * value. If there are no changes, the current root->root_node is returned. If | |
1288 | * anything changed in between, there's a fresh buffer allocated on which the | |
1289 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1290 | * Returns NULL on error (with no locks held). | |
1291 | */ | |
5d9e75c4 JS |
1292 | static inline struct extent_buffer * |
1293 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1294 | { | |
0b246afa | 1295 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1296 | struct tree_mod_elem *tm; |
30b0463a JS |
1297 | struct extent_buffer *eb = NULL; |
1298 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1299 | struct extent_buffer *old; |
a95236d9 | 1300 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1301 | u64 old_generation = 0; |
a95236d9 | 1302 | u64 logical; |
581c1760 | 1303 | int level; |
5d9e75c4 | 1304 | |
30b0463a | 1305 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1306 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1307 | if (!tm) |
30b0463a | 1308 | return eb_root; |
5d9e75c4 | 1309 | |
a95236d9 JS |
1310 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1311 | old_root = &tm->old_root; | |
1312 | old_generation = tm->generation; | |
1313 | logical = old_root->logical; | |
581c1760 | 1314 | level = old_root->level; |
a95236d9 | 1315 | } else { |
30b0463a | 1316 | logical = eb_root->start; |
581c1760 | 1317 | level = btrfs_header_level(eb_root); |
a95236d9 | 1318 | } |
5d9e75c4 | 1319 | |
0b246afa | 1320 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1321 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1322 | btrfs_tree_read_unlock(eb_root); |
1323 | free_extent_buffer(eb_root); | |
581c1760 | 1324 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1325 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1326 | if (!IS_ERR(old)) | |
1327 | free_extent_buffer(old); | |
0b246afa JM |
1328 | btrfs_warn(fs_info, |
1329 | "failed to read tree block %llu from get_old_root", | |
1330 | logical); | |
834328a8 | 1331 | } else { |
7bfdcf7f LB |
1332 | eb = btrfs_clone_extent_buffer(old); |
1333 | free_extent_buffer(old); | |
834328a8 JS |
1334 | } |
1335 | } else if (old_root) { | |
30b0463a JS |
1336 | btrfs_tree_read_unlock(eb_root); |
1337 | free_extent_buffer(eb_root); | |
0b246afa | 1338 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1339 | } else { |
9ec72677 | 1340 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1341 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1342 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1343 | free_extent_buffer(eb_root); |
834328a8 JS |
1344 | } |
1345 | ||
8ba97a15 JS |
1346 | if (!eb) |
1347 | return NULL; | |
d6381084 | 1348 | extent_buffer_get(eb); |
8ba97a15 | 1349 | btrfs_tree_read_lock(eb); |
a95236d9 | 1350 | if (old_root) { |
5d9e75c4 JS |
1351 | btrfs_set_header_bytenr(eb, eb->start); |
1352 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1353 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1354 | btrfs_set_header_level(eb, old_root->level); |
1355 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1356 | } |
28da9fb4 | 1357 | if (tm) |
0b246afa | 1358 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1359 | else |
1360 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1361 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1362 | |
1363 | return eb; | |
1364 | } | |
1365 | ||
5b6602e7 JS |
1366 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1367 | { | |
1368 | struct tree_mod_elem *tm; | |
1369 | int level; | |
30b0463a | 1370 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1371 | |
bcd24dab | 1372 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1373 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1374 | level = tm->old_root.level; | |
1375 | } else { | |
30b0463a | 1376 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1377 | } |
30b0463a | 1378 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1379 | |
1380 | return level; | |
1381 | } | |
1382 | ||
5d4f98a2 YZ |
1383 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1384 | struct btrfs_root *root, | |
1385 | struct extent_buffer *buf) | |
1386 | { | |
f5ee5c9a | 1387 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1388 | return 0; |
fccb84c9 | 1389 | |
d1980131 DS |
1390 | /* Ensure we can see the FORCE_COW bit */ |
1391 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1392 | |
1393 | /* | |
1394 | * We do not need to cow a block if | |
1395 | * 1) this block is not created or changed in this transaction; | |
1396 | * 2) this block does not belong to TREE_RELOC tree; | |
1397 | * 3) the root is not forced COW. | |
1398 | * | |
1399 | * What is forced COW: | |
01327610 | 1400 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1401 | * after we've finished coping src root, we must COW the shared |
1402 | * block to ensure the metadata consistency. | |
1403 | */ | |
5d4f98a2 YZ |
1404 | if (btrfs_header_generation(buf) == trans->transid && |
1405 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1406 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1407 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1408 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1409 | return 0; |
1410 | return 1; | |
1411 | } | |
1412 | ||
d352ac68 CM |
1413 | /* |
1414 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1415 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1416 | * once per transaction, as long as it hasn't been written yet |
1417 | */ | |
d397712b | 1418 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1419 | struct btrfs_root *root, struct extent_buffer *buf, |
1420 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1421 | struct extent_buffer **cow_ret) |
6702ed49 | 1422 | { |
0b246afa | 1423 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1424 | u64 search_start; |
f510cfec | 1425 | int ret; |
dc17ff8f | 1426 | |
0b246afa | 1427 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1428 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1429 | trans->transid, |
0b246afa | 1430 | fs_info->running_transaction->transid); |
31b1a2bd | 1431 | |
0b246afa | 1432 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1433 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1434 | trans->transid, fs_info->generation); |
dc17ff8f | 1435 | |
5d4f98a2 | 1436 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1437 | trans->dirty = true; |
6702ed49 CM |
1438 | *cow_ret = buf; |
1439 | return 0; | |
1440 | } | |
c487685d | 1441 | |
ee22184b | 1442 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1443 | |
1444 | if (parent) | |
1445 | btrfs_set_lock_blocking(parent); | |
1446 | btrfs_set_lock_blocking(buf); | |
1447 | ||
f510cfec | 1448 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1449 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1450 | |
1451 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1452 | ||
f510cfec | 1453 | return ret; |
6702ed49 CM |
1454 | } |
1455 | ||
d352ac68 CM |
1456 | /* |
1457 | * helper function for defrag to decide if two blocks pointed to by a | |
1458 | * node are actually close by | |
1459 | */ | |
6b80053d | 1460 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1461 | { |
6b80053d | 1462 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1463 | return 1; |
6b80053d | 1464 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1465 | return 1; |
1466 | return 0; | |
1467 | } | |
1468 | ||
081e9573 CM |
1469 | /* |
1470 | * compare two keys in a memcmp fashion | |
1471 | */ | |
310712b2 OS |
1472 | static int comp_keys(const struct btrfs_disk_key *disk, |
1473 | const struct btrfs_key *k2) | |
081e9573 CM |
1474 | { |
1475 | struct btrfs_key k1; | |
1476 | ||
1477 | btrfs_disk_key_to_cpu(&k1, disk); | |
1478 | ||
20736aba | 1479 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1480 | } |
1481 | ||
f3465ca4 JB |
1482 | /* |
1483 | * same as comp_keys only with two btrfs_key's | |
1484 | */ | |
310712b2 | 1485 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1486 | { |
1487 | if (k1->objectid > k2->objectid) | |
1488 | return 1; | |
1489 | if (k1->objectid < k2->objectid) | |
1490 | return -1; | |
1491 | if (k1->type > k2->type) | |
1492 | return 1; | |
1493 | if (k1->type < k2->type) | |
1494 | return -1; | |
1495 | if (k1->offset > k2->offset) | |
1496 | return 1; | |
1497 | if (k1->offset < k2->offset) | |
1498 | return -1; | |
1499 | return 0; | |
1500 | } | |
081e9573 | 1501 | |
d352ac68 CM |
1502 | /* |
1503 | * this is used by the defrag code to go through all the | |
1504 | * leaves pointed to by a node and reallocate them so that | |
1505 | * disk order is close to key order | |
1506 | */ | |
6702ed49 | 1507 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1508 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1509 | int start_slot, u64 *last_ret, |
a6b6e75e | 1510 | struct btrfs_key *progress) |
6702ed49 | 1511 | { |
0b246afa | 1512 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1513 | struct extent_buffer *cur; |
6702ed49 | 1514 | u64 blocknr; |
ca7a79ad | 1515 | u64 gen; |
e9d0b13b CM |
1516 | u64 search_start = *last_ret; |
1517 | u64 last_block = 0; | |
6702ed49 CM |
1518 | u64 other; |
1519 | u32 parent_nritems; | |
6702ed49 CM |
1520 | int end_slot; |
1521 | int i; | |
1522 | int err = 0; | |
f2183bde | 1523 | int parent_level; |
6b80053d CM |
1524 | int uptodate; |
1525 | u32 blocksize; | |
081e9573 CM |
1526 | int progress_passed = 0; |
1527 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1528 | |
5708b959 | 1529 | parent_level = btrfs_header_level(parent); |
5708b959 | 1530 | |
0b246afa JM |
1531 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1532 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1533 | |
6b80053d | 1534 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1535 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1536 | end_slot = parent_nritems - 1; |
6702ed49 | 1537 | |
5dfe2be7 | 1538 | if (parent_nritems <= 1) |
6702ed49 CM |
1539 | return 0; |
1540 | ||
b4ce94de CM |
1541 | btrfs_set_lock_blocking(parent); |
1542 | ||
5dfe2be7 | 1543 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1544 | struct btrfs_key first_key; |
6702ed49 | 1545 | int close = 1; |
a6b6e75e | 1546 | |
081e9573 CM |
1547 | btrfs_node_key(parent, &disk_key, i); |
1548 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1549 | continue; | |
1550 | ||
1551 | progress_passed = 1; | |
6b80053d | 1552 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1553 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1554 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1555 | if (last_block == 0) |
1556 | last_block = blocknr; | |
5708b959 | 1557 | |
6702ed49 | 1558 | if (i > 0) { |
6b80053d CM |
1559 | other = btrfs_node_blockptr(parent, i - 1); |
1560 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1561 | } |
5dfe2be7 | 1562 | if (!close && i < end_slot) { |
6b80053d CM |
1563 | other = btrfs_node_blockptr(parent, i + 1); |
1564 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1565 | } |
e9d0b13b CM |
1566 | if (close) { |
1567 | last_block = blocknr; | |
6702ed49 | 1568 | continue; |
e9d0b13b | 1569 | } |
6702ed49 | 1570 | |
0b246afa | 1571 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1572 | if (cur) |
b9fab919 | 1573 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1574 | else |
1575 | uptodate = 0; | |
5708b959 | 1576 | if (!cur || !uptodate) { |
6b80053d | 1577 | if (!cur) { |
581c1760 QW |
1578 | cur = read_tree_block(fs_info, blocknr, gen, |
1579 | parent_level - 1, | |
1580 | &first_key); | |
64c043de LB |
1581 | if (IS_ERR(cur)) { |
1582 | return PTR_ERR(cur); | |
1583 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1584 | free_extent_buffer(cur); |
97d9a8a4 | 1585 | return -EIO; |
416bc658 | 1586 | } |
6b80053d | 1587 | } else if (!uptodate) { |
581c1760 QW |
1588 | err = btrfs_read_buffer(cur, gen, |
1589 | parent_level - 1,&first_key); | |
018642a1 TI |
1590 | if (err) { |
1591 | free_extent_buffer(cur); | |
1592 | return err; | |
1593 | } | |
f2183bde | 1594 | } |
6702ed49 | 1595 | } |
e9d0b13b | 1596 | if (search_start == 0) |
6b80053d | 1597 | search_start = last_block; |
e9d0b13b | 1598 | |
e7a84565 | 1599 | btrfs_tree_lock(cur); |
b4ce94de | 1600 | btrfs_set_lock_blocking(cur); |
6b80053d | 1601 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1602 | &cur, search_start, |
6b80053d | 1603 | min(16 * blocksize, |
9fa8cfe7 | 1604 | (end_slot - i) * blocksize)); |
252c38f0 | 1605 | if (err) { |
e7a84565 | 1606 | btrfs_tree_unlock(cur); |
6b80053d | 1607 | free_extent_buffer(cur); |
6702ed49 | 1608 | break; |
252c38f0 | 1609 | } |
e7a84565 CM |
1610 | search_start = cur->start; |
1611 | last_block = cur->start; | |
f2183bde | 1612 | *last_ret = search_start; |
e7a84565 CM |
1613 | btrfs_tree_unlock(cur); |
1614 | free_extent_buffer(cur); | |
6702ed49 CM |
1615 | } |
1616 | return err; | |
1617 | } | |
1618 | ||
74123bd7 | 1619 | /* |
5f39d397 CM |
1620 | * search for key in the extent_buffer. The items start at offset p, |
1621 | * and they are item_size apart. There are 'max' items in p. | |
1622 | * | |
74123bd7 CM |
1623 | * the slot in the array is returned via slot, and it points to |
1624 | * the place where you would insert key if it is not found in | |
1625 | * the array. | |
1626 | * | |
1627 | * slot may point to max if the key is bigger than all of the keys | |
1628 | */ | |
e02119d5 | 1629 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1630 | unsigned long p, int item_size, |
1631 | const struct btrfs_key *key, | |
e02119d5 | 1632 | int max, int *slot) |
be0e5c09 CM |
1633 | { |
1634 | int low = 0; | |
1635 | int high = max; | |
1636 | int mid; | |
1637 | int ret; | |
479965d6 | 1638 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1639 | struct btrfs_disk_key unaligned; |
1640 | unsigned long offset; | |
5f39d397 CM |
1641 | char *kaddr = NULL; |
1642 | unsigned long map_start = 0; | |
1643 | unsigned long map_len = 0; | |
479965d6 | 1644 | int err; |
be0e5c09 | 1645 | |
5e24e9af LB |
1646 | if (low > high) { |
1647 | btrfs_err(eb->fs_info, | |
1648 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1649 | __func__, low, high, eb->start, | |
1650 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1651 | return -EINVAL; | |
1652 | } | |
1653 | ||
d397712b | 1654 | while (low < high) { |
be0e5c09 | 1655 | mid = (low + high) / 2; |
5f39d397 CM |
1656 | offset = p + mid * item_size; |
1657 | ||
a6591715 | 1658 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1659 | (offset + sizeof(struct btrfs_disk_key)) > |
1660 | map_start + map_len) { | |
934d375b CM |
1661 | |
1662 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1663 | sizeof(struct btrfs_disk_key), |
a6591715 | 1664 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1665 | |
1666 | if (!err) { | |
1667 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1668 | map_start); | |
415b35a5 | 1669 | } else if (err == 1) { |
479965d6 CM |
1670 | read_extent_buffer(eb, &unaligned, |
1671 | offset, sizeof(unaligned)); | |
1672 | tmp = &unaligned; | |
415b35a5 LB |
1673 | } else { |
1674 | return err; | |
479965d6 | 1675 | } |
5f39d397 | 1676 | |
5f39d397 CM |
1677 | } else { |
1678 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1679 | map_start); | |
1680 | } | |
be0e5c09 CM |
1681 | ret = comp_keys(tmp, key); |
1682 | ||
1683 | if (ret < 0) | |
1684 | low = mid + 1; | |
1685 | else if (ret > 0) | |
1686 | high = mid; | |
1687 | else { | |
1688 | *slot = mid; | |
1689 | return 0; | |
1690 | } | |
1691 | } | |
1692 | *slot = low; | |
1693 | return 1; | |
1694 | } | |
1695 | ||
97571fd0 CM |
1696 | /* |
1697 | * simple bin_search frontend that does the right thing for | |
1698 | * leaves vs nodes | |
1699 | */ | |
a74b35ec NB |
1700 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1701 | int level, int *slot) | |
be0e5c09 | 1702 | { |
f775738f | 1703 | if (level == 0) |
5f39d397 CM |
1704 | return generic_bin_search(eb, |
1705 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1706 | sizeof(struct btrfs_item), |
5f39d397 | 1707 | key, btrfs_header_nritems(eb), |
7518a238 | 1708 | slot); |
f775738f | 1709 | else |
5f39d397 CM |
1710 | return generic_bin_search(eb, |
1711 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1712 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1713 | key, btrfs_header_nritems(eb), |
7518a238 | 1714 | slot); |
be0e5c09 CM |
1715 | } |
1716 | ||
f0486c68 YZ |
1717 | static void root_add_used(struct btrfs_root *root, u32 size) |
1718 | { | |
1719 | spin_lock(&root->accounting_lock); | |
1720 | btrfs_set_root_used(&root->root_item, | |
1721 | btrfs_root_used(&root->root_item) + size); | |
1722 | spin_unlock(&root->accounting_lock); | |
1723 | } | |
1724 | ||
1725 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1726 | { | |
1727 | spin_lock(&root->accounting_lock); | |
1728 | btrfs_set_root_used(&root->root_item, | |
1729 | btrfs_root_used(&root->root_item) - size); | |
1730 | spin_unlock(&root->accounting_lock); | |
1731 | } | |
1732 | ||
d352ac68 CM |
1733 | /* given a node and slot number, this reads the blocks it points to. The |
1734 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1735 | */ |
2ff7e61e JM |
1736 | static noinline struct extent_buffer * |
1737 | read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent, | |
1738 | int slot) | |
bb803951 | 1739 | { |
ca7a79ad | 1740 | int level = btrfs_header_level(parent); |
416bc658 | 1741 | struct extent_buffer *eb; |
581c1760 | 1742 | struct btrfs_key first_key; |
416bc658 | 1743 | |
fb770ae4 LB |
1744 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1745 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1746 | |
1747 | BUG_ON(level == 0); | |
1748 | ||
581c1760 | 1749 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
2ff7e61e | 1750 | eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1751 | btrfs_node_ptr_generation(parent, slot), |
1752 | level - 1, &first_key); | |
fb770ae4 LB |
1753 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1754 | free_extent_buffer(eb); | |
1755 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1756 | } |
1757 | ||
1758 | return eb; | |
bb803951 CM |
1759 | } |
1760 | ||
d352ac68 CM |
1761 | /* |
1762 | * node level balancing, used to make sure nodes are in proper order for | |
1763 | * item deletion. We balance from the top down, so we have to make sure | |
1764 | * that a deletion won't leave an node completely empty later on. | |
1765 | */ | |
e02119d5 | 1766 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1767 | struct btrfs_root *root, |
1768 | struct btrfs_path *path, int level) | |
bb803951 | 1769 | { |
0b246afa | 1770 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1771 | struct extent_buffer *right = NULL; |
1772 | struct extent_buffer *mid; | |
1773 | struct extent_buffer *left = NULL; | |
1774 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1775 | int ret = 0; |
1776 | int wret; | |
1777 | int pslot; | |
bb803951 | 1778 | int orig_slot = path->slots[level]; |
79f95c82 | 1779 | u64 orig_ptr; |
bb803951 | 1780 | |
98e6b1eb | 1781 | ASSERT(level > 0); |
bb803951 | 1782 | |
5f39d397 | 1783 | mid = path->nodes[level]; |
b4ce94de | 1784 | |
bd681513 CM |
1785 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1786 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1787 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1788 | ||
1d4f8a0c | 1789 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1790 | |
a05a9bb1 | 1791 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1792 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1793 | pslot = path->slots[level + 1]; |
1794 | } | |
bb803951 | 1795 | |
40689478 CM |
1796 | /* |
1797 | * deal with the case where there is only one pointer in the root | |
1798 | * by promoting the node below to a root | |
1799 | */ | |
5f39d397 CM |
1800 | if (!parent) { |
1801 | struct extent_buffer *child; | |
bb803951 | 1802 | |
5f39d397 | 1803 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1804 | return 0; |
1805 | ||
1806 | /* promote the child to a root */ | |
2ff7e61e | 1807 | child = read_node_slot(fs_info, mid, 0); |
fb770ae4 LB |
1808 | if (IS_ERR(child)) { |
1809 | ret = PTR_ERR(child); | |
0b246afa | 1810 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1811 | goto enospc; |
1812 | } | |
1813 | ||
925baedd | 1814 | btrfs_tree_lock(child); |
b4ce94de | 1815 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1816 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1817 | if (ret) { |
1818 | btrfs_tree_unlock(child); | |
1819 | free_extent_buffer(child); | |
1820 | goto enospc; | |
1821 | } | |
2f375ab9 | 1822 | |
d9d19a01 DS |
1823 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1824 | BUG_ON(ret < 0); | |
240f62c8 | 1825 | rcu_assign_pointer(root->node, child); |
925baedd | 1826 | |
0b86a832 | 1827 | add_root_to_dirty_list(root); |
925baedd | 1828 | btrfs_tree_unlock(child); |
b4ce94de | 1829 | |
925baedd | 1830 | path->locks[level] = 0; |
bb803951 | 1831 | path->nodes[level] = NULL; |
7c302b49 | 1832 | clean_tree_block(fs_info, mid); |
925baedd | 1833 | btrfs_tree_unlock(mid); |
bb803951 | 1834 | /* once for the path */ |
5f39d397 | 1835 | free_extent_buffer(mid); |
f0486c68 YZ |
1836 | |
1837 | root_sub_used(root, mid->len); | |
5581a51a | 1838 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1839 | /* once for the root ptr */ |
3083ee2e | 1840 | free_extent_buffer_stale(mid); |
f0486c68 | 1841 | return 0; |
bb803951 | 1842 | } |
5f39d397 | 1843 | if (btrfs_header_nritems(mid) > |
0b246afa | 1844 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1845 | return 0; |
1846 | ||
2ff7e61e | 1847 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
1848 | if (IS_ERR(left)) |
1849 | left = NULL; | |
1850 | ||
5f39d397 | 1851 | if (left) { |
925baedd | 1852 | btrfs_tree_lock(left); |
b4ce94de | 1853 | btrfs_set_lock_blocking(left); |
5f39d397 | 1854 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1855 | parent, pslot - 1, &left); |
54aa1f4d CM |
1856 | if (wret) { |
1857 | ret = wret; | |
1858 | goto enospc; | |
1859 | } | |
2cc58cf2 | 1860 | } |
fb770ae4 | 1861 | |
2ff7e61e | 1862 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
1863 | if (IS_ERR(right)) |
1864 | right = NULL; | |
1865 | ||
5f39d397 | 1866 | if (right) { |
925baedd | 1867 | btrfs_tree_lock(right); |
b4ce94de | 1868 | btrfs_set_lock_blocking(right); |
5f39d397 | 1869 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1870 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1871 | if (wret) { |
1872 | ret = wret; | |
1873 | goto enospc; | |
1874 | } | |
1875 | } | |
1876 | ||
1877 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1878 | if (left) { |
1879 | orig_slot += btrfs_header_nritems(left); | |
2ff7e61e | 1880 | wret = push_node_left(trans, fs_info, left, mid, 1); |
79f95c82 CM |
1881 | if (wret < 0) |
1882 | ret = wret; | |
bb803951 | 1883 | } |
79f95c82 CM |
1884 | |
1885 | /* | |
1886 | * then try to empty the right most buffer into the middle | |
1887 | */ | |
5f39d397 | 1888 | if (right) { |
2ff7e61e | 1889 | wret = push_node_left(trans, fs_info, mid, right, 1); |
54aa1f4d | 1890 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1891 | ret = wret; |
5f39d397 | 1892 | if (btrfs_header_nritems(right) == 0) { |
7c302b49 | 1893 | clean_tree_block(fs_info, right); |
925baedd | 1894 | btrfs_tree_unlock(right); |
afe5fea7 | 1895 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1896 | root_sub_used(root, right->len); |
5581a51a | 1897 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1898 | free_extent_buffer_stale(right); |
f0486c68 | 1899 | right = NULL; |
bb803951 | 1900 | } else { |
5f39d397 CM |
1901 | struct btrfs_disk_key right_key; |
1902 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1903 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1904 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1905 | BUG_ON(ret < 0); | |
5f39d397 CM |
1906 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1907 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1908 | } |
1909 | } | |
5f39d397 | 1910 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1911 | /* |
1912 | * we're not allowed to leave a node with one item in the | |
1913 | * tree during a delete. A deletion from lower in the tree | |
1914 | * could try to delete the only pointer in this node. | |
1915 | * So, pull some keys from the left. | |
1916 | * There has to be a left pointer at this point because | |
1917 | * otherwise we would have pulled some pointers from the | |
1918 | * right | |
1919 | */ | |
305a26af MF |
1920 | if (!left) { |
1921 | ret = -EROFS; | |
0b246afa | 1922 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1923 | goto enospc; |
1924 | } | |
2ff7e61e | 1925 | wret = balance_node_right(trans, fs_info, mid, left); |
54aa1f4d | 1926 | if (wret < 0) { |
79f95c82 | 1927 | ret = wret; |
54aa1f4d CM |
1928 | goto enospc; |
1929 | } | |
bce4eae9 | 1930 | if (wret == 1) { |
2ff7e61e | 1931 | wret = push_node_left(trans, fs_info, left, mid, 1); |
bce4eae9 CM |
1932 | if (wret < 0) |
1933 | ret = wret; | |
1934 | } | |
79f95c82 CM |
1935 | BUG_ON(wret == 1); |
1936 | } | |
5f39d397 | 1937 | if (btrfs_header_nritems(mid) == 0) { |
7c302b49 | 1938 | clean_tree_block(fs_info, mid); |
925baedd | 1939 | btrfs_tree_unlock(mid); |
afe5fea7 | 1940 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1941 | root_sub_used(root, mid->len); |
5581a51a | 1942 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1943 | free_extent_buffer_stale(mid); |
f0486c68 | 1944 | mid = NULL; |
79f95c82 CM |
1945 | } else { |
1946 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1947 | struct btrfs_disk_key mid_key; |
1948 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
1949 | ret = tree_mod_log_insert_key(parent, pslot, |
1950 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1951 | BUG_ON(ret < 0); | |
5f39d397 CM |
1952 | btrfs_set_node_key(parent, &mid_key, pslot); |
1953 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1954 | } |
bb803951 | 1955 | |
79f95c82 | 1956 | /* update the path */ |
5f39d397 CM |
1957 | if (left) { |
1958 | if (btrfs_header_nritems(left) > orig_slot) { | |
1959 | extent_buffer_get(left); | |
925baedd | 1960 | /* left was locked after cow */ |
5f39d397 | 1961 | path->nodes[level] = left; |
bb803951 CM |
1962 | path->slots[level + 1] -= 1; |
1963 | path->slots[level] = orig_slot; | |
925baedd CM |
1964 | if (mid) { |
1965 | btrfs_tree_unlock(mid); | |
5f39d397 | 1966 | free_extent_buffer(mid); |
925baedd | 1967 | } |
bb803951 | 1968 | } else { |
5f39d397 | 1969 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1970 | path->slots[level] = orig_slot; |
1971 | } | |
1972 | } | |
79f95c82 | 1973 | /* double check we haven't messed things up */ |
e20d96d6 | 1974 | if (orig_ptr != |
5f39d397 | 1975 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1976 | BUG(); |
54aa1f4d | 1977 | enospc: |
925baedd CM |
1978 | if (right) { |
1979 | btrfs_tree_unlock(right); | |
5f39d397 | 1980 | free_extent_buffer(right); |
925baedd CM |
1981 | } |
1982 | if (left) { | |
1983 | if (path->nodes[level] != left) | |
1984 | btrfs_tree_unlock(left); | |
5f39d397 | 1985 | free_extent_buffer(left); |
925baedd | 1986 | } |
bb803951 CM |
1987 | return ret; |
1988 | } | |
1989 | ||
d352ac68 CM |
1990 | /* Node balancing for insertion. Here we only split or push nodes around |
1991 | * when they are completely full. This is also done top down, so we | |
1992 | * have to be pessimistic. | |
1993 | */ | |
d397712b | 1994 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1995 | struct btrfs_root *root, |
1996 | struct btrfs_path *path, int level) | |
e66f709b | 1997 | { |
0b246afa | 1998 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1999 | struct extent_buffer *right = NULL; |
2000 | struct extent_buffer *mid; | |
2001 | struct extent_buffer *left = NULL; | |
2002 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2003 | int ret = 0; |
2004 | int wret; | |
2005 | int pslot; | |
2006 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2007 | |
2008 | if (level == 0) | |
2009 | return 1; | |
2010 | ||
5f39d397 | 2011 | mid = path->nodes[level]; |
7bb86316 | 2012 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2013 | |
a05a9bb1 | 2014 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2015 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2016 | pslot = path->slots[level + 1]; |
2017 | } | |
e66f709b | 2018 | |
5f39d397 | 2019 | if (!parent) |
e66f709b | 2020 | return 1; |
e66f709b | 2021 | |
2ff7e61e | 2022 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
2023 | if (IS_ERR(left)) |
2024 | left = NULL; | |
e66f709b CM |
2025 | |
2026 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2027 | if (left) { |
e66f709b | 2028 | u32 left_nr; |
925baedd CM |
2029 | |
2030 | btrfs_tree_lock(left); | |
b4ce94de CM |
2031 | btrfs_set_lock_blocking(left); |
2032 | ||
5f39d397 | 2033 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2034 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2035 | wret = 1; |
2036 | } else { | |
5f39d397 | 2037 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2038 | pslot - 1, &left); |
54aa1f4d CM |
2039 | if (ret) |
2040 | wret = 1; | |
2041 | else { | |
2ff7e61e | 2042 | wret = push_node_left(trans, fs_info, |
971a1f66 | 2043 | left, mid, 0); |
54aa1f4d | 2044 | } |
33ade1f8 | 2045 | } |
e66f709b CM |
2046 | if (wret < 0) |
2047 | ret = wret; | |
2048 | if (wret == 0) { | |
5f39d397 | 2049 | struct btrfs_disk_key disk_key; |
e66f709b | 2050 | orig_slot += left_nr; |
5f39d397 | 2051 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2052 | ret = tree_mod_log_insert_key(parent, pslot, |
2053 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2054 | BUG_ON(ret < 0); | |
5f39d397 CM |
2055 | btrfs_set_node_key(parent, &disk_key, pslot); |
2056 | btrfs_mark_buffer_dirty(parent); | |
2057 | if (btrfs_header_nritems(left) > orig_slot) { | |
2058 | path->nodes[level] = left; | |
e66f709b CM |
2059 | path->slots[level + 1] -= 1; |
2060 | path->slots[level] = orig_slot; | |
925baedd | 2061 | btrfs_tree_unlock(mid); |
5f39d397 | 2062 | free_extent_buffer(mid); |
e66f709b CM |
2063 | } else { |
2064 | orig_slot -= | |
5f39d397 | 2065 | btrfs_header_nritems(left); |
e66f709b | 2066 | path->slots[level] = orig_slot; |
925baedd | 2067 | btrfs_tree_unlock(left); |
5f39d397 | 2068 | free_extent_buffer(left); |
e66f709b | 2069 | } |
e66f709b CM |
2070 | return 0; |
2071 | } | |
925baedd | 2072 | btrfs_tree_unlock(left); |
5f39d397 | 2073 | free_extent_buffer(left); |
e66f709b | 2074 | } |
2ff7e61e | 2075 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
2076 | if (IS_ERR(right)) |
2077 | right = NULL; | |
e66f709b CM |
2078 | |
2079 | /* | |
2080 | * then try to empty the right most buffer into the middle | |
2081 | */ | |
5f39d397 | 2082 | if (right) { |
33ade1f8 | 2083 | u32 right_nr; |
b4ce94de | 2084 | |
925baedd | 2085 | btrfs_tree_lock(right); |
b4ce94de CM |
2086 | btrfs_set_lock_blocking(right); |
2087 | ||
5f39d397 | 2088 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2089 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2090 | wret = 1; |
2091 | } else { | |
5f39d397 CM |
2092 | ret = btrfs_cow_block(trans, root, right, |
2093 | parent, pslot + 1, | |
9fa8cfe7 | 2094 | &right); |
54aa1f4d CM |
2095 | if (ret) |
2096 | wret = 1; | |
2097 | else { | |
2ff7e61e | 2098 | wret = balance_node_right(trans, fs_info, |
5f39d397 | 2099 | right, mid); |
54aa1f4d | 2100 | } |
33ade1f8 | 2101 | } |
e66f709b CM |
2102 | if (wret < 0) |
2103 | ret = wret; | |
2104 | if (wret == 0) { | |
5f39d397 CM |
2105 | struct btrfs_disk_key disk_key; |
2106 | ||
2107 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2108 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2109 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2110 | BUG_ON(ret < 0); | |
5f39d397 CM |
2111 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2112 | btrfs_mark_buffer_dirty(parent); | |
2113 | ||
2114 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2115 | path->nodes[level] = right; | |
e66f709b CM |
2116 | path->slots[level + 1] += 1; |
2117 | path->slots[level] = orig_slot - | |
5f39d397 | 2118 | btrfs_header_nritems(mid); |
925baedd | 2119 | btrfs_tree_unlock(mid); |
5f39d397 | 2120 | free_extent_buffer(mid); |
e66f709b | 2121 | } else { |
925baedd | 2122 | btrfs_tree_unlock(right); |
5f39d397 | 2123 | free_extent_buffer(right); |
e66f709b | 2124 | } |
e66f709b CM |
2125 | return 0; |
2126 | } | |
925baedd | 2127 | btrfs_tree_unlock(right); |
5f39d397 | 2128 | free_extent_buffer(right); |
e66f709b | 2129 | } |
e66f709b CM |
2130 | return 1; |
2131 | } | |
2132 | ||
3c69faec | 2133 | /* |
d352ac68 CM |
2134 | * readahead one full node of leaves, finding things that are close |
2135 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2136 | */ |
2ff7e61e | 2137 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2138 | struct btrfs_path *path, |
2139 | int level, int slot, u64 objectid) | |
3c69faec | 2140 | { |
5f39d397 | 2141 | struct extent_buffer *node; |
01f46658 | 2142 | struct btrfs_disk_key disk_key; |
3c69faec | 2143 | u32 nritems; |
3c69faec | 2144 | u64 search; |
a7175319 | 2145 | u64 target; |
6b80053d | 2146 | u64 nread = 0; |
5f39d397 | 2147 | struct extent_buffer *eb; |
6b80053d CM |
2148 | u32 nr; |
2149 | u32 blocksize; | |
2150 | u32 nscan = 0; | |
db94535d | 2151 | |
a6b6e75e | 2152 | if (level != 1) |
6702ed49 CM |
2153 | return; |
2154 | ||
2155 | if (!path->nodes[level]) | |
3c69faec CM |
2156 | return; |
2157 | ||
5f39d397 | 2158 | node = path->nodes[level]; |
925baedd | 2159 | |
3c69faec | 2160 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2161 | blocksize = fs_info->nodesize; |
2162 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2163 | if (eb) { |
2164 | free_extent_buffer(eb); | |
3c69faec CM |
2165 | return; |
2166 | } | |
2167 | ||
a7175319 | 2168 | target = search; |
6b80053d | 2169 | |
5f39d397 | 2170 | nritems = btrfs_header_nritems(node); |
6b80053d | 2171 | nr = slot; |
25b8b936 | 2172 | |
d397712b | 2173 | while (1) { |
e4058b54 | 2174 | if (path->reada == READA_BACK) { |
6b80053d CM |
2175 | if (nr == 0) |
2176 | break; | |
2177 | nr--; | |
e4058b54 | 2178 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2179 | nr++; |
2180 | if (nr >= nritems) | |
2181 | break; | |
3c69faec | 2182 | } |
e4058b54 | 2183 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2184 | btrfs_node_key(node, &disk_key, nr); |
2185 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2186 | break; | |
2187 | } | |
6b80053d | 2188 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2189 | if ((search <= target && target - search <= 65536) || |
2190 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2191 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2192 | nread += blocksize; |
2193 | } | |
2194 | nscan++; | |
a7175319 | 2195 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2196 | break; |
3c69faec CM |
2197 | } |
2198 | } | |
925baedd | 2199 | |
2ff7e61e | 2200 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2201 | struct btrfs_path *path, int level) |
b4ce94de CM |
2202 | { |
2203 | int slot; | |
2204 | int nritems; | |
2205 | struct extent_buffer *parent; | |
2206 | struct extent_buffer *eb; | |
2207 | u64 gen; | |
2208 | u64 block1 = 0; | |
2209 | u64 block2 = 0; | |
b4ce94de | 2210 | |
8c594ea8 | 2211 | parent = path->nodes[level + 1]; |
b4ce94de | 2212 | if (!parent) |
0b08851f | 2213 | return; |
b4ce94de CM |
2214 | |
2215 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2216 | slot = path->slots[level + 1]; |
b4ce94de CM |
2217 | |
2218 | if (slot > 0) { | |
2219 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2220 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2221 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2222 | /* |
2223 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2224 | * don't want to return eagain here. That will loop | |
2225 | * forever | |
2226 | */ | |
2227 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2228 | block1 = 0; |
2229 | free_extent_buffer(eb); | |
2230 | } | |
8c594ea8 | 2231 | if (slot + 1 < nritems) { |
b4ce94de CM |
2232 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2233 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2234 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2235 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2236 | block2 = 0; |
2237 | free_extent_buffer(eb); | |
2238 | } | |
8c594ea8 | 2239 | |
0b08851f | 2240 | if (block1) |
2ff7e61e | 2241 | readahead_tree_block(fs_info, block1); |
0b08851f | 2242 | if (block2) |
2ff7e61e | 2243 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2244 | } |
2245 | ||
2246 | ||
d352ac68 | 2247 | /* |
d397712b CM |
2248 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2249 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2250 | * operations on the tree might require changing key pointers higher up in the | |
2251 | * tree. | |
d352ac68 | 2252 | * |
d397712b CM |
2253 | * callers might also have set path->keep_locks, which tells this code to keep |
2254 | * the lock if the path points to the last slot in the block. This is part of | |
2255 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2256 | * |
d397712b CM |
2257 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2258 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2259 | */ |
e02119d5 | 2260 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2261 | int lowest_unlock, int min_write_lock_level, |
2262 | int *write_lock_level) | |
925baedd CM |
2263 | { |
2264 | int i; | |
2265 | int skip_level = level; | |
051e1b9f | 2266 | int no_skips = 0; |
925baedd CM |
2267 | struct extent_buffer *t; |
2268 | ||
2269 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2270 | if (!path->nodes[i]) | |
2271 | break; | |
2272 | if (!path->locks[i]) | |
2273 | break; | |
051e1b9f | 2274 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2275 | skip_level = i + 1; |
2276 | continue; | |
2277 | } | |
051e1b9f | 2278 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2279 | u32 nritems; |
2280 | t = path->nodes[i]; | |
2281 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2282 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2283 | skip_level = i + 1; |
2284 | continue; | |
2285 | } | |
2286 | } | |
051e1b9f CM |
2287 | if (skip_level < i && i >= lowest_unlock) |
2288 | no_skips = 1; | |
2289 | ||
925baedd | 2290 | t = path->nodes[i]; |
d80bb3f9 | 2291 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2292 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2293 | path->locks[i] = 0; |
f7c79f30 CM |
2294 | if (write_lock_level && |
2295 | i > min_write_lock_level && | |
2296 | i <= *write_lock_level) { | |
2297 | *write_lock_level = i - 1; | |
2298 | } | |
925baedd CM |
2299 | } |
2300 | } | |
2301 | } | |
2302 | ||
b4ce94de CM |
2303 | /* |
2304 | * This releases any locks held in the path starting at level and | |
2305 | * going all the way up to the root. | |
2306 | * | |
2307 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2308 | * corner cases, such as COW of the block at slot zero in the node. This | |
2309 | * ignores those rules, and it should only be called when there are no | |
2310 | * more updates to be done higher up in the tree. | |
2311 | */ | |
2312 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2313 | { | |
2314 | int i; | |
2315 | ||
09a2a8f9 | 2316 | if (path->keep_locks) |
b4ce94de CM |
2317 | return; |
2318 | ||
2319 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2320 | if (!path->nodes[i]) | |
12f4dacc | 2321 | continue; |
b4ce94de | 2322 | if (!path->locks[i]) |
12f4dacc | 2323 | continue; |
bd681513 | 2324 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2325 | path->locks[i] = 0; |
2326 | } | |
2327 | } | |
2328 | ||
c8c42864 CM |
2329 | /* |
2330 | * helper function for btrfs_search_slot. The goal is to find a block | |
2331 | * in cache without setting the path to blocking. If we find the block | |
2332 | * we return zero and the path is unchanged. | |
2333 | * | |
2334 | * If we can't find the block, we set the path blocking and do some | |
2335 | * reada. -EAGAIN is returned and the search must be repeated. | |
2336 | */ | |
2337 | static int | |
d07b8528 LB |
2338 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2339 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2340 | const struct btrfs_key *key) |
c8c42864 | 2341 | { |
0b246afa | 2342 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2343 | u64 blocknr; |
2344 | u64 gen; | |
c8c42864 CM |
2345 | struct extent_buffer *b = *eb_ret; |
2346 | struct extent_buffer *tmp; | |
581c1760 | 2347 | struct btrfs_key first_key; |
76a05b35 | 2348 | int ret; |
581c1760 | 2349 | int parent_level; |
c8c42864 CM |
2350 | |
2351 | blocknr = btrfs_node_blockptr(b, slot); | |
2352 | gen = btrfs_node_ptr_generation(b, slot); | |
581c1760 QW |
2353 | parent_level = btrfs_header_level(b); |
2354 | btrfs_node_key_to_cpu(b, &first_key, slot); | |
c8c42864 | 2355 | |
0b246afa | 2356 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2357 | if (tmp) { |
b9fab919 | 2358 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2359 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2360 | *eb_ret = tmp; | |
2361 | return 0; | |
2362 | } | |
2363 | ||
2364 | /* the pages were up to date, but we failed | |
2365 | * the generation number check. Do a full | |
2366 | * read for the generation number that is correct. | |
2367 | * We must do this without dropping locks so | |
2368 | * we can trust our generation number | |
2369 | */ | |
2370 | btrfs_set_path_blocking(p); | |
2371 | ||
2372 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2373 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2374 | if (!ret) { |
2375 | *eb_ret = tmp; | |
2376 | return 0; | |
cb44921a | 2377 | } |
bdf7c00e JB |
2378 | free_extent_buffer(tmp); |
2379 | btrfs_release_path(p); | |
2380 | return -EIO; | |
c8c42864 CM |
2381 | } |
2382 | ||
2383 | /* | |
2384 | * reduce lock contention at high levels | |
2385 | * of the btree by dropping locks before | |
76a05b35 CM |
2386 | * we read. Don't release the lock on the current |
2387 | * level because we need to walk this node to figure | |
2388 | * out which blocks to read. | |
c8c42864 | 2389 | */ |
8c594ea8 CM |
2390 | btrfs_unlock_up_safe(p, level + 1); |
2391 | btrfs_set_path_blocking(p); | |
2392 | ||
e4058b54 | 2393 | if (p->reada != READA_NONE) |
2ff7e61e | 2394 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2395 | |
76a05b35 | 2396 | ret = -EAGAIN; |
02a3307a | 2397 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2398 | &first_key); |
64c043de | 2399 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2400 | /* |
2401 | * If the read above didn't mark this buffer up to date, | |
2402 | * it will never end up being up to date. Set ret to EIO now | |
2403 | * and give up so that our caller doesn't loop forever | |
2404 | * on our EAGAINs. | |
2405 | */ | |
e6a1d6fd | 2406 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2407 | ret = -EIO; |
c8c42864 | 2408 | free_extent_buffer(tmp); |
c871b0f2 LB |
2409 | } else { |
2410 | ret = PTR_ERR(tmp); | |
76a05b35 | 2411 | } |
02a3307a LB |
2412 | |
2413 | btrfs_release_path(p); | |
76a05b35 | 2414 | return ret; |
c8c42864 CM |
2415 | } |
2416 | ||
2417 | /* | |
2418 | * helper function for btrfs_search_slot. This does all of the checks | |
2419 | * for node-level blocks and does any balancing required based on | |
2420 | * the ins_len. | |
2421 | * | |
2422 | * If no extra work was required, zero is returned. If we had to | |
2423 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2424 | * start over | |
2425 | */ | |
2426 | static int | |
2427 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2428 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2429 | struct extent_buffer *b, int level, int ins_len, |
2430 | int *write_lock_level) | |
c8c42864 | 2431 | { |
0b246afa | 2432 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2433 | int ret; |
0b246afa | 2434 | |
c8c42864 | 2435 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2436 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2437 | int sret; |
2438 | ||
bd681513 CM |
2439 | if (*write_lock_level < level + 1) { |
2440 | *write_lock_level = level + 1; | |
2441 | btrfs_release_path(p); | |
2442 | goto again; | |
2443 | } | |
2444 | ||
c8c42864 | 2445 | btrfs_set_path_blocking(p); |
2ff7e61e | 2446 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2447 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2448 | |
2449 | BUG_ON(sret > 0); | |
2450 | if (sret) { | |
2451 | ret = sret; | |
2452 | goto done; | |
2453 | } | |
2454 | b = p->nodes[level]; | |
2455 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2456 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2457 | int sret; |
2458 | ||
bd681513 CM |
2459 | if (*write_lock_level < level + 1) { |
2460 | *write_lock_level = level + 1; | |
2461 | btrfs_release_path(p); | |
2462 | goto again; | |
2463 | } | |
2464 | ||
c8c42864 | 2465 | btrfs_set_path_blocking(p); |
2ff7e61e | 2466 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2467 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2468 | |
2469 | if (sret) { | |
2470 | ret = sret; | |
2471 | goto done; | |
2472 | } | |
2473 | b = p->nodes[level]; | |
2474 | if (!b) { | |
b3b4aa74 | 2475 | btrfs_release_path(p); |
c8c42864 CM |
2476 | goto again; |
2477 | } | |
2478 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2479 | } | |
2480 | return 0; | |
2481 | ||
2482 | again: | |
2483 | ret = -EAGAIN; | |
2484 | done: | |
2485 | return ret; | |
2486 | } | |
2487 | ||
d7396f07 | 2488 | static void key_search_validate(struct extent_buffer *b, |
310712b2 | 2489 | const struct btrfs_key *key, |
d7396f07 FDBM |
2490 | int level) |
2491 | { | |
2492 | #ifdef CONFIG_BTRFS_ASSERT | |
2493 | struct btrfs_disk_key disk_key; | |
2494 | ||
2495 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2496 | ||
2497 | if (level == 0) | |
2498 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2499 | offsetof(struct btrfs_leaf, items[0].key), | |
2500 | sizeof(disk_key))); | |
2501 | else | |
2502 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2503 | offsetof(struct btrfs_node, ptrs[0].key), | |
2504 | sizeof(disk_key))); | |
2505 | #endif | |
2506 | } | |
2507 | ||
310712b2 | 2508 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2509 | int level, int *prev_cmp, int *slot) |
2510 | { | |
2511 | if (*prev_cmp != 0) { | |
a74b35ec | 2512 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2513 | return *prev_cmp; |
2514 | } | |
2515 | ||
2516 | key_search_validate(b, key, level); | |
2517 | *slot = 0; | |
2518 | ||
2519 | return 0; | |
2520 | } | |
2521 | ||
381cf658 | 2522 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2523 | u64 iobjectid, u64 ioff, u8 key_type, |
2524 | struct btrfs_key *found_key) | |
2525 | { | |
2526 | int ret; | |
2527 | struct btrfs_key key; | |
2528 | struct extent_buffer *eb; | |
381cf658 DS |
2529 | |
2530 | ASSERT(path); | |
1d4c08e0 | 2531 | ASSERT(found_key); |
e33d5c3d KN |
2532 | |
2533 | key.type = key_type; | |
2534 | key.objectid = iobjectid; | |
2535 | key.offset = ioff; | |
2536 | ||
2537 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2538 | if (ret < 0) |
e33d5c3d KN |
2539 | return ret; |
2540 | ||
2541 | eb = path->nodes[0]; | |
2542 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2543 | ret = btrfs_next_leaf(fs_root, path); | |
2544 | if (ret) | |
2545 | return ret; | |
2546 | eb = path->nodes[0]; | |
2547 | } | |
2548 | ||
2549 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2550 | if (found_key->type != key.type || | |
2551 | found_key->objectid != key.objectid) | |
2552 | return 1; | |
2553 | ||
2554 | return 0; | |
2555 | } | |
2556 | ||
1fc28d8e LB |
2557 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2558 | struct btrfs_path *p, | |
2559 | int write_lock_level) | |
2560 | { | |
2561 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2562 | struct extent_buffer *b; | |
2563 | int root_lock; | |
2564 | int level = 0; | |
2565 | ||
2566 | /* We try very hard to do read locks on the root */ | |
2567 | root_lock = BTRFS_READ_LOCK; | |
2568 | ||
2569 | if (p->search_commit_root) { | |
2570 | /* The commit roots are read only so we always do read locks */ | |
2571 | if (p->need_commit_sem) | |
2572 | down_read(&fs_info->commit_root_sem); | |
2573 | b = root->commit_root; | |
2574 | extent_buffer_get(b); | |
2575 | level = btrfs_header_level(b); | |
2576 | if (p->need_commit_sem) | |
2577 | up_read(&fs_info->commit_root_sem); | |
f9ddfd05 LB |
2578 | /* |
2579 | * Ensure that all callers have set skip_locking when | |
2580 | * p->search_commit_root = 1. | |
2581 | */ | |
2582 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2583 | |
2584 | goto out; | |
2585 | } | |
2586 | ||
2587 | if (p->skip_locking) { | |
2588 | b = btrfs_root_node(root); | |
2589 | level = btrfs_header_level(b); | |
2590 | goto out; | |
2591 | } | |
2592 | ||
2593 | /* | |
662c653b LB |
2594 | * If the level is set to maximum, we can skip trying to get the read |
2595 | * lock. | |
1fc28d8e | 2596 | */ |
662c653b LB |
2597 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2598 | /* | |
2599 | * We don't know the level of the root node until we actually | |
2600 | * have it read locked | |
2601 | */ | |
2602 | b = btrfs_read_lock_root_node(root); | |
2603 | level = btrfs_header_level(b); | |
2604 | if (level > write_lock_level) | |
2605 | goto out; | |
2606 | ||
2607 | /* Whoops, must trade for write lock */ | |
2608 | btrfs_tree_read_unlock(b); | |
2609 | free_extent_buffer(b); | |
2610 | } | |
1fc28d8e | 2611 | |
1fc28d8e LB |
2612 | b = btrfs_lock_root_node(root); |
2613 | root_lock = BTRFS_WRITE_LOCK; | |
2614 | ||
2615 | /* The level might have changed, check again */ | |
2616 | level = btrfs_header_level(b); | |
2617 | ||
2618 | out: | |
2619 | p->nodes[level] = b; | |
2620 | if (!p->skip_locking) | |
2621 | p->locks[level] = root_lock; | |
2622 | /* | |
2623 | * Callers are responsible for dropping b's references. | |
2624 | */ | |
2625 | return b; | |
2626 | } | |
2627 | ||
2628 | ||
74123bd7 | 2629 | /* |
4271ecea NB |
2630 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2631 | * modifications to preserve tree invariants. | |
74123bd7 | 2632 | * |
4271ecea NB |
2633 | * @trans: Handle of transaction, used when modifying the tree |
2634 | * @p: Holds all btree nodes along the search path | |
2635 | * @root: The root node of the tree | |
2636 | * @key: The key we are looking for | |
2637 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2638 | * deletions it's -1. 0 for plain searches | |
2639 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2640 | * when modifying the tree. | |
97571fd0 | 2641 | * |
4271ecea NB |
2642 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2643 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2644 | * | |
2645 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2646 | * of the path (level 0) | |
2647 | * | |
2648 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2649 | * points to the slot where it should be inserted | |
2650 | * | |
2651 | * If an error is encountered while searching the tree a negative error number | |
2652 | * is returned | |
74123bd7 | 2653 | */ |
310712b2 OS |
2654 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2655 | const struct btrfs_key *key, struct btrfs_path *p, | |
2656 | int ins_len, int cow) | |
be0e5c09 | 2657 | { |
0b246afa | 2658 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2659 | struct extent_buffer *b; |
be0e5c09 CM |
2660 | int slot; |
2661 | int ret; | |
33c66f43 | 2662 | int err; |
be0e5c09 | 2663 | int level; |
925baedd | 2664 | int lowest_unlock = 1; |
bd681513 CM |
2665 | /* everything at write_lock_level or lower must be write locked */ |
2666 | int write_lock_level = 0; | |
9f3a7427 | 2667 | u8 lowest_level = 0; |
f7c79f30 | 2668 | int min_write_lock_level; |
d7396f07 | 2669 | int prev_cmp; |
9f3a7427 | 2670 | |
6702ed49 | 2671 | lowest_level = p->lowest_level; |
323ac95b | 2672 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2673 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2674 | BUG_ON(!cow && ins_len); |
25179201 | 2675 | |
bd681513 | 2676 | if (ins_len < 0) { |
925baedd | 2677 | lowest_unlock = 2; |
65b51a00 | 2678 | |
bd681513 CM |
2679 | /* when we are removing items, we might have to go up to level |
2680 | * two as we update tree pointers Make sure we keep write | |
2681 | * for those levels as well | |
2682 | */ | |
2683 | write_lock_level = 2; | |
2684 | } else if (ins_len > 0) { | |
2685 | /* | |
2686 | * for inserting items, make sure we have a write lock on | |
2687 | * level 1 so we can update keys | |
2688 | */ | |
2689 | write_lock_level = 1; | |
2690 | } | |
2691 | ||
2692 | if (!cow) | |
2693 | write_lock_level = -1; | |
2694 | ||
09a2a8f9 | 2695 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2696 | write_lock_level = BTRFS_MAX_LEVEL; |
2697 | ||
f7c79f30 CM |
2698 | min_write_lock_level = write_lock_level; |
2699 | ||
bb803951 | 2700 | again: |
d7396f07 | 2701 | prev_cmp = -1; |
1fc28d8e | 2702 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
925baedd | 2703 | |
eb60ceac | 2704 | while (b) { |
5f39d397 | 2705 | level = btrfs_header_level(b); |
65b51a00 CM |
2706 | |
2707 | /* | |
2708 | * setup the path here so we can release it under lock | |
2709 | * contention with the cow code | |
2710 | */ | |
02217ed2 | 2711 | if (cow) { |
9ea2c7c9 NB |
2712 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2713 | ||
c8c42864 CM |
2714 | /* |
2715 | * if we don't really need to cow this block | |
2716 | * then we don't want to set the path blocking, | |
2717 | * so we test it here | |
2718 | */ | |
64c12921 JM |
2719 | if (!should_cow_block(trans, root, b)) { |
2720 | trans->dirty = true; | |
65b51a00 | 2721 | goto cow_done; |
64c12921 | 2722 | } |
5d4f98a2 | 2723 | |
bd681513 CM |
2724 | /* |
2725 | * must have write locks on this node and the | |
2726 | * parent | |
2727 | */ | |
5124e00e JB |
2728 | if (level > write_lock_level || |
2729 | (level + 1 > write_lock_level && | |
2730 | level + 1 < BTRFS_MAX_LEVEL && | |
2731 | p->nodes[level + 1])) { | |
bd681513 CM |
2732 | write_lock_level = level + 1; |
2733 | btrfs_release_path(p); | |
2734 | goto again; | |
2735 | } | |
2736 | ||
160f4089 | 2737 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2738 | if (last_level) |
2739 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2740 | &b); | |
2741 | else | |
2742 | err = btrfs_cow_block(trans, root, b, | |
2743 | p->nodes[level + 1], | |
2744 | p->slots[level + 1], &b); | |
33c66f43 | 2745 | if (err) { |
33c66f43 | 2746 | ret = err; |
65b51a00 | 2747 | goto done; |
54aa1f4d | 2748 | } |
02217ed2 | 2749 | } |
65b51a00 | 2750 | cow_done: |
eb60ceac | 2751 | p->nodes[level] = b; |
52398340 LB |
2752 | /* |
2753 | * Leave path with blocking locks to avoid massive | |
2754 | * lock context switch, this is made on purpose. | |
2755 | */ | |
b4ce94de CM |
2756 | |
2757 | /* | |
2758 | * we have a lock on b and as long as we aren't changing | |
2759 | * the tree, there is no way to for the items in b to change. | |
2760 | * It is safe to drop the lock on our parent before we | |
2761 | * go through the expensive btree search on b. | |
2762 | * | |
eb653de1 FDBM |
2763 | * If we're inserting or deleting (ins_len != 0), then we might |
2764 | * be changing slot zero, which may require changing the parent. | |
2765 | * So, we can't drop the lock until after we know which slot | |
2766 | * we're operating on. | |
b4ce94de | 2767 | */ |
eb653de1 FDBM |
2768 | if (!ins_len && !p->keep_locks) { |
2769 | int u = level + 1; | |
2770 | ||
2771 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2772 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2773 | p->locks[u] = 0; | |
2774 | } | |
2775 | } | |
b4ce94de | 2776 | |
d7396f07 | 2777 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2778 | if (ret < 0) |
2779 | goto done; | |
b4ce94de | 2780 | |
5f39d397 | 2781 | if (level != 0) { |
33c66f43 YZ |
2782 | int dec = 0; |
2783 | if (ret && slot > 0) { | |
2784 | dec = 1; | |
be0e5c09 | 2785 | slot -= 1; |
33c66f43 | 2786 | } |
be0e5c09 | 2787 | p->slots[level] = slot; |
33c66f43 | 2788 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2789 | ins_len, &write_lock_level); |
33c66f43 | 2790 | if (err == -EAGAIN) |
c8c42864 | 2791 | goto again; |
33c66f43 YZ |
2792 | if (err) { |
2793 | ret = err; | |
c8c42864 | 2794 | goto done; |
33c66f43 | 2795 | } |
c8c42864 CM |
2796 | b = p->nodes[level]; |
2797 | slot = p->slots[level]; | |
b4ce94de | 2798 | |
bd681513 CM |
2799 | /* |
2800 | * slot 0 is special, if we change the key | |
2801 | * we have to update the parent pointer | |
2802 | * which means we must have a write lock | |
2803 | * on the parent | |
2804 | */ | |
eb653de1 | 2805 | if (slot == 0 && ins_len && |
bd681513 CM |
2806 | write_lock_level < level + 1) { |
2807 | write_lock_level = level + 1; | |
2808 | btrfs_release_path(p); | |
2809 | goto again; | |
2810 | } | |
2811 | ||
f7c79f30 CM |
2812 | unlock_up(p, level, lowest_unlock, |
2813 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2814 | |
925baedd | 2815 | if (level == lowest_level) { |
33c66f43 YZ |
2816 | if (dec) |
2817 | p->slots[level]++; | |
5b21f2ed | 2818 | goto done; |
925baedd | 2819 | } |
ca7a79ad | 2820 | |
d07b8528 | 2821 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2822 | slot, key); |
33c66f43 | 2823 | if (err == -EAGAIN) |
c8c42864 | 2824 | goto again; |
33c66f43 YZ |
2825 | if (err) { |
2826 | ret = err; | |
76a05b35 | 2827 | goto done; |
33c66f43 | 2828 | } |
76a05b35 | 2829 | |
b4ce94de | 2830 | if (!p->skip_locking) { |
bd681513 CM |
2831 | level = btrfs_header_level(b); |
2832 | if (level <= write_lock_level) { | |
2833 | err = btrfs_try_tree_write_lock(b); | |
2834 | if (!err) { | |
2835 | btrfs_set_path_blocking(p); | |
2836 | btrfs_tree_lock(b); | |
bd681513 CM |
2837 | } |
2838 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2839 | } else { | |
f82c458a | 2840 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2841 | if (!err) { |
2842 | btrfs_set_path_blocking(p); | |
2843 | btrfs_tree_read_lock(b); | |
bd681513 CM |
2844 | } |
2845 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2846 | } |
bd681513 | 2847 | p->nodes[level] = b; |
b4ce94de | 2848 | } |
be0e5c09 CM |
2849 | } else { |
2850 | p->slots[level] = slot; | |
87b29b20 | 2851 | if (ins_len > 0 && |
2ff7e61e | 2852 | btrfs_leaf_free_space(fs_info, b) < ins_len) { |
bd681513 CM |
2853 | if (write_lock_level < 1) { |
2854 | write_lock_level = 1; | |
2855 | btrfs_release_path(p); | |
2856 | goto again; | |
2857 | } | |
2858 | ||
b4ce94de | 2859 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2860 | err = split_leaf(trans, root, key, |
2861 | p, ins_len, ret == 0); | |
b4ce94de | 2862 | |
33c66f43 YZ |
2863 | BUG_ON(err > 0); |
2864 | if (err) { | |
2865 | ret = err; | |
65b51a00 CM |
2866 | goto done; |
2867 | } | |
5c680ed6 | 2868 | } |
459931ec | 2869 | if (!p->search_for_split) |
f7c79f30 | 2870 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2871 | min_write_lock_level, NULL); |
65b51a00 | 2872 | goto done; |
be0e5c09 CM |
2873 | } |
2874 | } | |
65b51a00 CM |
2875 | ret = 1; |
2876 | done: | |
b4ce94de CM |
2877 | /* |
2878 | * we don't really know what they plan on doing with the path | |
2879 | * from here on, so for now just mark it as blocking | |
2880 | */ | |
b9473439 CM |
2881 | if (!p->leave_spinning) |
2882 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2883 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2884 | btrfs_release_path(p); |
65b51a00 | 2885 | return ret; |
be0e5c09 CM |
2886 | } |
2887 | ||
5d9e75c4 JS |
2888 | /* |
2889 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2890 | * current state of the tree together with the operations recorded in the tree | |
2891 | * modification log to search for the key in a previous version of this tree, as | |
2892 | * denoted by the time_seq parameter. | |
2893 | * | |
2894 | * Naturally, there is no support for insert, delete or cow operations. | |
2895 | * | |
2896 | * The resulting path and return value will be set up as if we called | |
2897 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2898 | */ | |
310712b2 | 2899 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2900 | struct btrfs_path *p, u64 time_seq) |
2901 | { | |
0b246afa | 2902 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2903 | struct extent_buffer *b; |
2904 | int slot; | |
2905 | int ret; | |
2906 | int err; | |
2907 | int level; | |
2908 | int lowest_unlock = 1; | |
2909 | u8 lowest_level = 0; | |
d4b4087c | 2910 | int prev_cmp = -1; |
5d9e75c4 JS |
2911 | |
2912 | lowest_level = p->lowest_level; | |
2913 | WARN_ON(p->nodes[0] != NULL); | |
2914 | ||
2915 | if (p->search_commit_root) { | |
2916 | BUG_ON(time_seq); | |
2917 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2918 | } | |
2919 | ||
2920 | again: | |
5d9e75c4 | 2921 | b = get_old_root(root, time_seq); |
315bed43 NB |
2922 | if (!b) { |
2923 | ret = -EIO; | |
2924 | goto done; | |
2925 | } | |
5d9e75c4 | 2926 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2927 | p->locks[level] = BTRFS_READ_LOCK; |
2928 | ||
2929 | while (b) { | |
2930 | level = btrfs_header_level(b); | |
2931 | p->nodes[level] = b; | |
5d9e75c4 JS |
2932 | |
2933 | /* | |
2934 | * we have a lock on b and as long as we aren't changing | |
2935 | * the tree, there is no way to for the items in b to change. | |
2936 | * It is safe to drop the lock on our parent before we | |
2937 | * go through the expensive btree search on b. | |
2938 | */ | |
2939 | btrfs_unlock_up_safe(p, level + 1); | |
2940 | ||
d4b4087c | 2941 | /* |
01327610 | 2942 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2943 | * time. |
2944 | */ | |
2945 | prev_cmp = -1; | |
d7396f07 | 2946 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
2947 | |
2948 | if (level != 0) { | |
2949 | int dec = 0; | |
2950 | if (ret && slot > 0) { | |
2951 | dec = 1; | |
2952 | slot -= 1; | |
2953 | } | |
2954 | p->slots[level] = slot; | |
2955 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2956 | ||
2957 | if (level == lowest_level) { | |
2958 | if (dec) | |
2959 | p->slots[level]++; | |
2960 | goto done; | |
2961 | } | |
2962 | ||
d07b8528 | 2963 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2964 | slot, key); |
5d9e75c4 JS |
2965 | if (err == -EAGAIN) |
2966 | goto again; | |
2967 | if (err) { | |
2968 | ret = err; | |
2969 | goto done; | |
2970 | } | |
2971 | ||
2972 | level = btrfs_header_level(b); | |
f82c458a | 2973 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
2974 | if (!err) { |
2975 | btrfs_set_path_blocking(p); | |
2976 | btrfs_tree_read_lock(b); | |
5d9e75c4 | 2977 | } |
0b246afa | 2978 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
2979 | if (!b) { |
2980 | ret = -ENOMEM; | |
2981 | goto done; | |
2982 | } | |
5d9e75c4 JS |
2983 | p->locks[level] = BTRFS_READ_LOCK; |
2984 | p->nodes[level] = b; | |
5d9e75c4 JS |
2985 | } else { |
2986 | p->slots[level] = slot; | |
2987 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2988 | goto done; | |
2989 | } | |
2990 | } | |
2991 | ret = 1; | |
2992 | done: | |
2993 | if (!p->leave_spinning) | |
2994 | btrfs_set_path_blocking(p); | |
2995 | if (ret < 0) | |
2996 | btrfs_release_path(p); | |
2997 | ||
2998 | return ret; | |
2999 | } | |
3000 | ||
2f38b3e1 AJ |
3001 | /* |
3002 | * helper to use instead of search slot if no exact match is needed but | |
3003 | * instead the next or previous item should be returned. | |
3004 | * When find_higher is true, the next higher item is returned, the next lower | |
3005 | * otherwise. | |
3006 | * When return_any and find_higher are both true, and no higher item is found, | |
3007 | * return the next lower instead. | |
3008 | * When return_any is true and find_higher is false, and no lower item is found, | |
3009 | * return the next higher instead. | |
3010 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3011 | * < 0 on error | |
3012 | */ | |
3013 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3014 | const struct btrfs_key *key, |
3015 | struct btrfs_path *p, int find_higher, | |
3016 | int return_any) | |
2f38b3e1 AJ |
3017 | { |
3018 | int ret; | |
3019 | struct extent_buffer *leaf; | |
3020 | ||
3021 | again: | |
3022 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3023 | if (ret <= 0) | |
3024 | return ret; | |
3025 | /* | |
3026 | * a return value of 1 means the path is at the position where the | |
3027 | * item should be inserted. Normally this is the next bigger item, | |
3028 | * but in case the previous item is the last in a leaf, path points | |
3029 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3030 | * item. | |
3031 | */ | |
3032 | leaf = p->nodes[0]; | |
3033 | ||
3034 | if (find_higher) { | |
3035 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3036 | ret = btrfs_next_leaf(root, p); | |
3037 | if (ret <= 0) | |
3038 | return ret; | |
3039 | if (!return_any) | |
3040 | return 1; | |
3041 | /* | |
3042 | * no higher item found, return the next | |
3043 | * lower instead | |
3044 | */ | |
3045 | return_any = 0; | |
3046 | find_higher = 0; | |
3047 | btrfs_release_path(p); | |
3048 | goto again; | |
3049 | } | |
3050 | } else { | |
e6793769 AJ |
3051 | if (p->slots[0] == 0) { |
3052 | ret = btrfs_prev_leaf(root, p); | |
3053 | if (ret < 0) | |
3054 | return ret; | |
3055 | if (!ret) { | |
23c6bf6a FDBM |
3056 | leaf = p->nodes[0]; |
3057 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3058 | p->slots[0]--; | |
e6793769 | 3059 | return 0; |
2f38b3e1 | 3060 | } |
e6793769 AJ |
3061 | if (!return_any) |
3062 | return 1; | |
3063 | /* | |
3064 | * no lower item found, return the next | |
3065 | * higher instead | |
3066 | */ | |
3067 | return_any = 0; | |
3068 | find_higher = 1; | |
3069 | btrfs_release_path(p); | |
3070 | goto again; | |
3071 | } else { | |
2f38b3e1 AJ |
3072 | --p->slots[0]; |
3073 | } | |
3074 | } | |
3075 | return 0; | |
3076 | } | |
3077 | ||
74123bd7 CM |
3078 | /* |
3079 | * adjust the pointers going up the tree, starting at level | |
3080 | * making sure the right key of each node is points to 'key'. | |
3081 | * This is used after shifting pointers to the left, so it stops | |
3082 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3083 | * higher levels | |
aa5d6bed | 3084 | * |
74123bd7 | 3085 | */ |
b167fa91 | 3086 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3087 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3088 | { |
3089 | int i; | |
5f39d397 | 3090 | struct extent_buffer *t; |
0e82bcfe | 3091 | int ret; |
5f39d397 | 3092 | |
234b63a0 | 3093 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3094 | int tslot = path->slots[i]; |
0e82bcfe | 3095 | |
eb60ceac | 3096 | if (!path->nodes[i]) |
be0e5c09 | 3097 | break; |
5f39d397 | 3098 | t = path->nodes[i]; |
0e82bcfe DS |
3099 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3100 | GFP_ATOMIC); | |
3101 | BUG_ON(ret < 0); | |
5f39d397 | 3102 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3103 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3104 | if (tslot != 0) |
3105 | break; | |
3106 | } | |
3107 | } | |
3108 | ||
31840ae1 ZY |
3109 | /* |
3110 | * update item key. | |
3111 | * | |
3112 | * This function isn't completely safe. It's the caller's responsibility | |
3113 | * that the new key won't break the order | |
3114 | */ | |
b7a0365e DD |
3115 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3116 | struct btrfs_path *path, | |
310712b2 | 3117 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3118 | { |
3119 | struct btrfs_disk_key disk_key; | |
3120 | struct extent_buffer *eb; | |
3121 | int slot; | |
3122 | ||
3123 | eb = path->nodes[0]; | |
3124 | slot = path->slots[0]; | |
3125 | if (slot > 0) { | |
3126 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3127 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3128 | } |
3129 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3130 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3131 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3132 | } |
3133 | ||
3134 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3135 | btrfs_set_item_key(eb, &disk_key, slot); | |
3136 | btrfs_mark_buffer_dirty(eb); | |
3137 | if (slot == 0) | |
b167fa91 | 3138 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3139 | } |
3140 | ||
74123bd7 CM |
3141 | /* |
3142 | * try to push data from one node into the next node left in the | |
79f95c82 | 3143 | * tree. |
aa5d6bed CM |
3144 | * |
3145 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3146 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3147 | */ |
98ed5174 | 3148 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
3149 | struct btrfs_fs_info *fs_info, |
3150 | struct extent_buffer *dst, | |
971a1f66 | 3151 | struct extent_buffer *src, int empty) |
be0e5c09 | 3152 | { |
be0e5c09 | 3153 | int push_items = 0; |
bb803951 CM |
3154 | int src_nritems; |
3155 | int dst_nritems; | |
aa5d6bed | 3156 | int ret = 0; |
be0e5c09 | 3157 | |
5f39d397 CM |
3158 | src_nritems = btrfs_header_nritems(src); |
3159 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3160 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3161 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3162 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3163 | |
bce4eae9 | 3164 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3165 | return 1; |
3166 | ||
d397712b | 3167 | if (push_items <= 0) |
be0e5c09 CM |
3168 | return 1; |
3169 | ||
bce4eae9 | 3170 | if (empty) { |
971a1f66 | 3171 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3172 | if (push_items < src_nritems) { |
3173 | /* leave at least 8 pointers in the node if | |
3174 | * we aren't going to empty it | |
3175 | */ | |
3176 | if (src_nritems - push_items < 8) { | |
3177 | if (push_items <= 8) | |
3178 | return 1; | |
3179 | push_items -= 8; | |
3180 | } | |
3181 | } | |
3182 | } else | |
3183 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3184 | |
0b246afa | 3185 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3186 | push_items); |
3187 | if (ret) { | |
66642832 | 3188 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3189 | return ret; |
3190 | } | |
5f39d397 CM |
3191 | copy_extent_buffer(dst, src, |
3192 | btrfs_node_key_ptr_offset(dst_nritems), | |
3193 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3194 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3195 | |
bb803951 | 3196 | if (push_items < src_nritems) { |
57911b8b | 3197 | /* |
bf1d3425 DS |
3198 | * Don't call tree_mod_log_insert_move here, key removal was |
3199 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3200 | */ |
5f39d397 CM |
3201 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3202 | btrfs_node_key_ptr_offset(push_items), | |
3203 | (src_nritems - push_items) * | |
3204 | sizeof(struct btrfs_key_ptr)); | |
3205 | } | |
3206 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3207 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3208 | btrfs_mark_buffer_dirty(src); | |
3209 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3210 | |
79f95c82 CM |
3211 | return ret; |
3212 | } | |
3213 | ||
3214 | /* | |
3215 | * try to push data from one node into the next node right in the | |
3216 | * tree. | |
3217 | * | |
3218 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3219 | * error, and > 0 if there was no room in the right hand block. | |
3220 | * | |
3221 | * this will only push up to 1/2 the contents of the left node over | |
3222 | */ | |
5f39d397 | 3223 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 3224 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
3225 | struct extent_buffer *dst, |
3226 | struct extent_buffer *src) | |
79f95c82 | 3227 | { |
79f95c82 CM |
3228 | int push_items = 0; |
3229 | int max_push; | |
3230 | int src_nritems; | |
3231 | int dst_nritems; | |
3232 | int ret = 0; | |
79f95c82 | 3233 | |
7bb86316 CM |
3234 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3235 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3236 | ||
5f39d397 CM |
3237 | src_nritems = btrfs_header_nritems(src); |
3238 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3239 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3240 | if (push_items <= 0) |
79f95c82 | 3241 | return 1; |
bce4eae9 | 3242 | |
d397712b | 3243 | if (src_nritems < 4) |
bce4eae9 | 3244 | return 1; |
79f95c82 CM |
3245 | |
3246 | max_push = src_nritems / 2 + 1; | |
3247 | /* don't try to empty the node */ | |
d397712b | 3248 | if (max_push >= src_nritems) |
79f95c82 | 3249 | return 1; |
252c38f0 | 3250 | |
79f95c82 CM |
3251 | if (max_push < push_items) |
3252 | push_items = max_push; | |
3253 | ||
bf1d3425 DS |
3254 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3255 | BUG_ON(ret < 0); | |
5f39d397 CM |
3256 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3257 | btrfs_node_key_ptr_offset(0), | |
3258 | (dst_nritems) * | |
3259 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3260 | |
0b246afa | 3261 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3262 | src_nritems - push_items, push_items); |
3263 | if (ret) { | |
66642832 | 3264 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3265 | return ret; |
3266 | } | |
5f39d397 CM |
3267 | copy_extent_buffer(dst, src, |
3268 | btrfs_node_key_ptr_offset(0), | |
3269 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3270 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3271 | |
5f39d397 CM |
3272 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3273 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3274 | |
5f39d397 CM |
3275 | btrfs_mark_buffer_dirty(src); |
3276 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3277 | |
aa5d6bed | 3278 | return ret; |
be0e5c09 CM |
3279 | } |
3280 | ||
97571fd0 CM |
3281 | /* |
3282 | * helper function to insert a new root level in the tree. | |
3283 | * A new node is allocated, and a single item is inserted to | |
3284 | * point to the existing root | |
aa5d6bed CM |
3285 | * |
3286 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3287 | */ |
d397712b | 3288 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3289 | struct btrfs_root *root, |
fdd99c72 | 3290 | struct btrfs_path *path, int level) |
5c680ed6 | 3291 | { |
0b246afa | 3292 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3293 | u64 lower_gen; |
5f39d397 CM |
3294 | struct extent_buffer *lower; |
3295 | struct extent_buffer *c; | |
925baedd | 3296 | struct extent_buffer *old; |
5f39d397 | 3297 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3298 | int ret; |
5c680ed6 CM |
3299 | |
3300 | BUG_ON(path->nodes[level]); | |
3301 | BUG_ON(path->nodes[level-1] != root->node); | |
3302 | ||
7bb86316 CM |
3303 | lower = path->nodes[level-1]; |
3304 | if (level == 1) | |
3305 | btrfs_item_key(lower, &lower_key, 0); | |
3306 | else | |
3307 | btrfs_node_key(lower, &lower_key, 0); | |
3308 | ||
4d75f8a9 DS |
3309 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3310 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3311 | if (IS_ERR(c)) |
3312 | return PTR_ERR(c); | |
925baedd | 3313 | |
0b246afa | 3314 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3315 | |
5f39d397 | 3316 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3317 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3318 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3319 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3320 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3321 | |
3322 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3323 | |
5f39d397 | 3324 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3325 | |
925baedd | 3326 | old = root->node; |
d9d19a01 DS |
3327 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3328 | BUG_ON(ret < 0); | |
240f62c8 | 3329 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3330 | |
3331 | /* the super has an extra ref to root->node */ | |
3332 | free_extent_buffer(old); | |
3333 | ||
0b86a832 | 3334 | add_root_to_dirty_list(root); |
5f39d397 CM |
3335 | extent_buffer_get(c); |
3336 | path->nodes[level] = c; | |
95449a16 | 3337 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3338 | path->slots[level] = 0; |
3339 | return 0; | |
3340 | } | |
3341 | ||
74123bd7 CM |
3342 | /* |
3343 | * worker function to insert a single pointer in a node. | |
3344 | * the node should have enough room for the pointer already | |
97571fd0 | 3345 | * |
74123bd7 CM |
3346 | * slot and level indicate where you want the key to go, and |
3347 | * blocknr is the block the key points to. | |
3348 | */ | |
143bede5 | 3349 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2ff7e61e | 3350 | struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 3351 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3352 | int slot, int level) |
74123bd7 | 3353 | { |
5f39d397 | 3354 | struct extent_buffer *lower; |
74123bd7 | 3355 | int nritems; |
f3ea38da | 3356 | int ret; |
5c680ed6 CM |
3357 | |
3358 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3359 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3360 | lower = path->nodes[level]; |
3361 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3362 | BUG_ON(slot > nritems); |
0b246afa | 3363 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3364 | if (slot != nritems) { |
bf1d3425 DS |
3365 | if (level) { |
3366 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3367 | nritems - slot); |
bf1d3425 DS |
3368 | BUG_ON(ret < 0); |
3369 | } | |
5f39d397 CM |
3370 | memmove_extent_buffer(lower, |
3371 | btrfs_node_key_ptr_offset(slot + 1), | |
3372 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3373 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3374 | } |
c3e06965 | 3375 | if (level) { |
e09c2efe DS |
3376 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3377 | GFP_NOFS); | |
f3ea38da JS |
3378 | BUG_ON(ret < 0); |
3379 | } | |
5f39d397 | 3380 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3381 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3382 | WARN_ON(trans->transid == 0); |
3383 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3384 | btrfs_set_header_nritems(lower, nritems + 1); |
3385 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3386 | } |
3387 | ||
97571fd0 CM |
3388 | /* |
3389 | * split the node at the specified level in path in two. | |
3390 | * The path is corrected to point to the appropriate node after the split | |
3391 | * | |
3392 | * Before splitting this tries to make some room in the node by pushing | |
3393 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3394 | * |
3395 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3396 | */ |
e02119d5 CM |
3397 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3398 | struct btrfs_root *root, | |
3399 | struct btrfs_path *path, int level) | |
be0e5c09 | 3400 | { |
0b246afa | 3401 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3402 | struct extent_buffer *c; |
3403 | struct extent_buffer *split; | |
3404 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3405 | int mid; |
5c680ed6 | 3406 | int ret; |
7518a238 | 3407 | u32 c_nritems; |
eb60ceac | 3408 | |
5f39d397 | 3409 | c = path->nodes[level]; |
7bb86316 | 3410 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3411 | if (c == root->node) { |
d9abbf1c | 3412 | /* |
90f8d62e JS |
3413 | * trying to split the root, lets make a new one |
3414 | * | |
fdd99c72 | 3415 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3416 | * insert_new_root, because that root buffer will be kept as a |
3417 | * normal node. We are going to log removal of half of the | |
3418 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3419 | * tree lock on the buffer, which is why we cannot race with | |
3420 | * other tree_mod_log users. | |
d9abbf1c | 3421 | */ |
fdd99c72 | 3422 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3423 | if (ret) |
3424 | return ret; | |
b3612421 | 3425 | } else { |
e66f709b | 3426 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3427 | c = path->nodes[level]; |
3428 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3429 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3430 | return 0; |
54aa1f4d CM |
3431 | if (ret < 0) |
3432 | return ret; | |
be0e5c09 | 3433 | } |
e66f709b | 3434 | |
5f39d397 | 3435 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3436 | mid = (c_nritems + 1) / 2; |
3437 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3438 | |
4d75f8a9 DS |
3439 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3440 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3441 | if (IS_ERR(split)) |
3442 | return PTR_ERR(split); | |
3443 | ||
0b246afa | 3444 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3445 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3446 | |
0b246afa | 3447 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3448 | if (ret) { |
66642832 | 3449 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3450 | return ret; |
3451 | } | |
5f39d397 CM |
3452 | copy_extent_buffer(split, c, |
3453 | btrfs_node_key_ptr_offset(0), | |
3454 | btrfs_node_key_ptr_offset(mid), | |
3455 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3456 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3457 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3458 | ret = 0; |
3459 | ||
5f39d397 CM |
3460 | btrfs_mark_buffer_dirty(c); |
3461 | btrfs_mark_buffer_dirty(split); | |
3462 | ||
2ff7e61e | 3463 | insert_ptr(trans, fs_info, path, &disk_key, split->start, |
c3e06965 | 3464 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3465 | |
5de08d7d | 3466 | if (path->slots[level] >= mid) { |
5c680ed6 | 3467 | path->slots[level] -= mid; |
925baedd | 3468 | btrfs_tree_unlock(c); |
5f39d397 CM |
3469 | free_extent_buffer(c); |
3470 | path->nodes[level] = split; | |
5c680ed6 CM |
3471 | path->slots[level + 1] += 1; |
3472 | } else { | |
925baedd | 3473 | btrfs_tree_unlock(split); |
5f39d397 | 3474 | free_extent_buffer(split); |
be0e5c09 | 3475 | } |
aa5d6bed | 3476 | return ret; |
be0e5c09 CM |
3477 | } |
3478 | ||
74123bd7 CM |
3479 | /* |
3480 | * how many bytes are required to store the items in a leaf. start | |
3481 | * and nr indicate which items in the leaf to check. This totals up the | |
3482 | * space used both by the item structs and the item data | |
3483 | */ | |
5f39d397 | 3484 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3485 | { |
41be1f3b JB |
3486 | struct btrfs_item *start_item; |
3487 | struct btrfs_item *end_item; | |
3488 | struct btrfs_map_token token; | |
be0e5c09 | 3489 | int data_len; |
5f39d397 | 3490 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3491 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3492 | |
3493 | if (!nr) | |
3494 | return 0; | |
41be1f3b | 3495 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3496 | start_item = btrfs_item_nr(start); |
3497 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3498 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3499 | btrfs_token_item_size(l, start_item, &token); | |
3500 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3501 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3502 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3503 | return data_len; |
3504 | } | |
3505 | ||
d4dbff95 CM |
3506 | /* |
3507 | * The space between the end of the leaf items and | |
3508 | * the start of the leaf data. IOW, how much room | |
3509 | * the leaf has left for both items and data | |
3510 | */ | |
2ff7e61e | 3511 | noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info, |
e02119d5 | 3512 | struct extent_buffer *leaf) |
d4dbff95 | 3513 | { |
5f39d397 CM |
3514 | int nritems = btrfs_header_nritems(leaf); |
3515 | int ret; | |
0b246afa JM |
3516 | |
3517 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3518 | if (ret < 0) { |
0b246afa JM |
3519 | btrfs_crit(fs_info, |
3520 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3521 | ret, | |
3522 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3523 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3524 | } |
3525 | return ret; | |
d4dbff95 CM |
3526 | } |
3527 | ||
99d8f83c CM |
3528 | /* |
3529 | * min slot controls the lowest index we're willing to push to the | |
3530 | * right. We'll push up to and including min_slot, but no lower | |
3531 | */ | |
1e47eef2 | 3532 | static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3533 | struct btrfs_path *path, |
3534 | int data_size, int empty, | |
3535 | struct extent_buffer *right, | |
99d8f83c CM |
3536 | int free_space, u32 left_nritems, |
3537 | u32 min_slot) | |
00ec4c51 | 3538 | { |
5f39d397 | 3539 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3540 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3541 | struct btrfs_map_token token; |
5f39d397 | 3542 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3543 | int slot; |
34a38218 | 3544 | u32 i; |
00ec4c51 CM |
3545 | int push_space = 0; |
3546 | int push_items = 0; | |
0783fcfc | 3547 | struct btrfs_item *item; |
34a38218 | 3548 | u32 nr; |
7518a238 | 3549 | u32 right_nritems; |
5f39d397 | 3550 | u32 data_end; |
db94535d | 3551 | u32 this_item_size; |
00ec4c51 | 3552 | |
cfed81a0 CM |
3553 | btrfs_init_map_token(&token); |
3554 | ||
34a38218 CM |
3555 | if (empty) |
3556 | nr = 0; | |
3557 | else | |
99d8f83c | 3558 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3559 | |
31840ae1 | 3560 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3561 | push_space += data_size; |
31840ae1 | 3562 | |
44871b1b | 3563 | slot = path->slots[1]; |
34a38218 CM |
3564 | i = left_nritems - 1; |
3565 | while (i >= nr) { | |
dd3cc16b | 3566 | item = btrfs_item_nr(i); |
db94535d | 3567 | |
31840ae1 ZY |
3568 | if (!empty && push_items > 0) { |
3569 | if (path->slots[0] > i) | |
3570 | break; | |
3571 | if (path->slots[0] == i) { | |
2ff7e61e | 3572 | int space = btrfs_leaf_free_space(fs_info, left); |
31840ae1 ZY |
3573 | if (space + push_space * 2 > free_space) |
3574 | break; | |
3575 | } | |
3576 | } | |
3577 | ||
00ec4c51 | 3578 | if (path->slots[0] == i) |
87b29b20 | 3579 | push_space += data_size; |
db94535d | 3580 | |
db94535d CM |
3581 | this_item_size = btrfs_item_size(left, item); |
3582 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3583 | break; |
31840ae1 | 3584 | |
00ec4c51 | 3585 | push_items++; |
db94535d | 3586 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3587 | if (i == 0) |
3588 | break; | |
3589 | i--; | |
db94535d | 3590 | } |
5f39d397 | 3591 | |
925baedd CM |
3592 | if (push_items == 0) |
3593 | goto out_unlock; | |
5f39d397 | 3594 | |
6c1500f2 | 3595 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3596 | |
00ec4c51 | 3597 | /* push left to right */ |
5f39d397 | 3598 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3599 | |
5f39d397 | 3600 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
2ff7e61e | 3601 | push_space -= leaf_data_end(fs_info, left); |
5f39d397 | 3602 | |
00ec4c51 | 3603 | /* make room in the right data area */ |
2ff7e61e | 3604 | data_end = leaf_data_end(fs_info, right); |
5f39d397 | 3605 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3606 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3607 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3608 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3609 | |
00ec4c51 | 3610 | /* copy from the left data area */ |
3d9ec8c4 | 3611 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3612 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3613 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left), |
d6025579 | 3614 | push_space); |
5f39d397 CM |
3615 | |
3616 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3617 | btrfs_item_nr_offset(0), | |
3618 | right_nritems * sizeof(struct btrfs_item)); | |
3619 | ||
00ec4c51 | 3620 | /* copy the items from left to right */ |
5f39d397 CM |
3621 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3622 | btrfs_item_nr_offset(left_nritems - push_items), | |
3623 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3624 | |
3625 | /* update the item pointers */ | |
7518a238 | 3626 | right_nritems += push_items; |
5f39d397 | 3627 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3628 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3629 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3630 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3631 | push_space -= btrfs_token_item_size(right, item, &token); |
3632 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3633 | } |
3634 | ||
7518a238 | 3635 | left_nritems -= push_items; |
5f39d397 | 3636 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3637 | |
34a38218 CM |
3638 | if (left_nritems) |
3639 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3640 | else |
7c302b49 | 3641 | clean_tree_block(fs_info, left); |
f0486c68 | 3642 | |
5f39d397 | 3643 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3644 | |
5f39d397 CM |
3645 | btrfs_item_key(right, &disk_key, 0); |
3646 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3647 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3648 | |
00ec4c51 | 3649 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3650 | if (path->slots[0] >= left_nritems) { |
3651 | path->slots[0] -= left_nritems; | |
925baedd | 3652 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
7c302b49 | 3653 | clean_tree_block(fs_info, path->nodes[0]); |
925baedd | 3654 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3655 | free_extent_buffer(path->nodes[0]); |
3656 | path->nodes[0] = right; | |
00ec4c51 CM |
3657 | path->slots[1] += 1; |
3658 | } else { | |
925baedd | 3659 | btrfs_tree_unlock(right); |
5f39d397 | 3660 | free_extent_buffer(right); |
00ec4c51 CM |
3661 | } |
3662 | return 0; | |
925baedd CM |
3663 | |
3664 | out_unlock: | |
3665 | btrfs_tree_unlock(right); | |
3666 | free_extent_buffer(right); | |
3667 | return 1; | |
00ec4c51 | 3668 | } |
925baedd | 3669 | |
44871b1b CM |
3670 | /* |
3671 | * push some data in the path leaf to the right, trying to free up at | |
3672 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3673 | * | |
3674 | * returns 1 if the push failed because the other node didn't have enough | |
3675 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3676 | * |
3677 | * this will push starting from min_slot to the end of the leaf. It won't | |
3678 | * push any slot lower than min_slot | |
44871b1b CM |
3679 | */ |
3680 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3681 | *root, struct btrfs_path *path, |
3682 | int min_data_size, int data_size, | |
3683 | int empty, u32 min_slot) | |
44871b1b | 3684 | { |
2ff7e61e | 3685 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3686 | struct extent_buffer *left = path->nodes[0]; |
3687 | struct extent_buffer *right; | |
3688 | struct extent_buffer *upper; | |
3689 | int slot; | |
3690 | int free_space; | |
3691 | u32 left_nritems; | |
3692 | int ret; | |
3693 | ||
3694 | if (!path->nodes[1]) | |
3695 | return 1; | |
3696 | ||
3697 | slot = path->slots[1]; | |
3698 | upper = path->nodes[1]; | |
3699 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3700 | return 1; | |
3701 | ||
3702 | btrfs_assert_tree_locked(path->nodes[1]); | |
3703 | ||
2ff7e61e | 3704 | right = read_node_slot(fs_info, upper, slot + 1); |
fb770ae4 LB |
3705 | /* |
3706 | * slot + 1 is not valid or we fail to read the right node, | |
3707 | * no big deal, just return. | |
3708 | */ | |
3709 | if (IS_ERR(right)) | |
91ca338d TI |
3710 | return 1; |
3711 | ||
44871b1b CM |
3712 | btrfs_tree_lock(right); |
3713 | btrfs_set_lock_blocking(right); | |
3714 | ||
2ff7e61e | 3715 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3716 | if (free_space < data_size) |
3717 | goto out_unlock; | |
3718 | ||
3719 | /* cow and double check */ | |
3720 | ret = btrfs_cow_block(trans, root, right, upper, | |
3721 | slot + 1, &right); | |
3722 | if (ret) | |
3723 | goto out_unlock; | |
3724 | ||
2ff7e61e | 3725 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3726 | if (free_space < data_size) |
3727 | goto out_unlock; | |
3728 | ||
3729 | left_nritems = btrfs_header_nritems(left); | |
3730 | if (left_nritems == 0) | |
3731 | goto out_unlock; | |
3732 | ||
2ef1fed2 FDBM |
3733 | if (path->slots[0] == left_nritems && !empty) { |
3734 | /* Key greater than all keys in the leaf, right neighbor has | |
3735 | * enough room for it and we're not emptying our leaf to delete | |
3736 | * it, therefore use right neighbor to insert the new item and | |
3737 | * no need to touch/dirty our left leaft. */ | |
3738 | btrfs_tree_unlock(left); | |
3739 | free_extent_buffer(left); | |
3740 | path->nodes[0] = right; | |
3741 | path->slots[0] = 0; | |
3742 | path->slots[1]++; | |
3743 | return 0; | |
3744 | } | |
3745 | ||
1e47eef2 | 3746 | return __push_leaf_right(fs_info, path, min_data_size, empty, |
99d8f83c | 3747 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3748 | out_unlock: |
3749 | btrfs_tree_unlock(right); | |
3750 | free_extent_buffer(right); | |
3751 | return 1; | |
3752 | } | |
3753 | ||
74123bd7 CM |
3754 | /* |
3755 | * push some data in the path leaf to the left, trying to free up at | |
3756 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3757 | * |
3758 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3759 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3760 | * items | |
74123bd7 | 3761 | */ |
66cb7ddb | 3762 | static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3763 | struct btrfs_path *path, int data_size, |
3764 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3765 | int free_space, u32 right_nritems, |
3766 | u32 max_slot) | |
be0e5c09 | 3767 | { |
5f39d397 CM |
3768 | struct btrfs_disk_key disk_key; |
3769 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3770 | int i; |
be0e5c09 CM |
3771 | int push_space = 0; |
3772 | int push_items = 0; | |
0783fcfc | 3773 | struct btrfs_item *item; |
7518a238 | 3774 | u32 old_left_nritems; |
34a38218 | 3775 | u32 nr; |
aa5d6bed | 3776 | int ret = 0; |
db94535d CM |
3777 | u32 this_item_size; |
3778 | u32 old_left_item_size; | |
cfed81a0 CM |
3779 | struct btrfs_map_token token; |
3780 | ||
3781 | btrfs_init_map_token(&token); | |
be0e5c09 | 3782 | |
34a38218 | 3783 | if (empty) |
99d8f83c | 3784 | nr = min(right_nritems, max_slot); |
34a38218 | 3785 | else |
99d8f83c | 3786 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3787 | |
3788 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3789 | item = btrfs_item_nr(i); |
db94535d | 3790 | |
31840ae1 ZY |
3791 | if (!empty && push_items > 0) { |
3792 | if (path->slots[0] < i) | |
3793 | break; | |
3794 | if (path->slots[0] == i) { | |
2ff7e61e | 3795 | int space = btrfs_leaf_free_space(fs_info, right); |
31840ae1 ZY |
3796 | if (space + push_space * 2 > free_space) |
3797 | break; | |
3798 | } | |
3799 | } | |
3800 | ||
be0e5c09 | 3801 | if (path->slots[0] == i) |
87b29b20 | 3802 | push_space += data_size; |
db94535d CM |
3803 | |
3804 | this_item_size = btrfs_item_size(right, item); | |
3805 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3806 | break; |
db94535d | 3807 | |
be0e5c09 | 3808 | push_items++; |
db94535d CM |
3809 | push_space += this_item_size + sizeof(*item); |
3810 | } | |
3811 | ||
be0e5c09 | 3812 | if (push_items == 0) { |
925baedd CM |
3813 | ret = 1; |
3814 | goto out; | |
be0e5c09 | 3815 | } |
fae7f21c | 3816 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3817 | |
be0e5c09 | 3818 | /* push data from right to left */ |
5f39d397 CM |
3819 | copy_extent_buffer(left, right, |
3820 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3821 | btrfs_item_nr_offset(0), | |
3822 | push_items * sizeof(struct btrfs_item)); | |
3823 | ||
0b246afa | 3824 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3825 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3826 | |
3d9ec8c4 | 3827 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3828 | leaf_data_end(fs_info, left) - push_space, |
3d9ec8c4 | 3829 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3830 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3831 | push_space); |
5f39d397 | 3832 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3833 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3834 | |
db94535d | 3835 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3836 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3837 | u32 ioff; |
db94535d | 3838 | |
dd3cc16b | 3839 | item = btrfs_item_nr(i); |
db94535d | 3840 | |
cfed81a0 CM |
3841 | ioff = btrfs_token_item_offset(left, item, &token); |
3842 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3843 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3844 | &token); |
be0e5c09 | 3845 | } |
5f39d397 | 3846 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3847 | |
3848 | /* fixup right node */ | |
31b1a2bd JL |
3849 | if (push_items > right_nritems) |
3850 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3851 | right_nritems); |
34a38218 CM |
3852 | |
3853 | if (push_items < right_nritems) { | |
3854 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2ff7e61e | 3855 | leaf_data_end(fs_info, right); |
3d9ec8c4 | 3856 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3857 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3858 | BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3859 | leaf_data_end(fs_info, right), push_space); |
34a38218 CM |
3860 | |
3861 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3862 | btrfs_item_nr_offset(push_items), |
3863 | (btrfs_header_nritems(right) - push_items) * | |
3864 | sizeof(struct btrfs_item)); | |
34a38218 | 3865 | } |
eef1c494 Y |
3866 | right_nritems -= push_items; |
3867 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3868 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3869 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3870 | item = btrfs_item_nr(i); |
db94535d | 3871 | |
cfed81a0 CM |
3872 | push_space = push_space - btrfs_token_item_size(right, |
3873 | item, &token); | |
3874 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3875 | } |
eb60ceac | 3876 | |
5f39d397 | 3877 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3878 | if (right_nritems) |
3879 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3880 | else |
7c302b49 | 3881 | clean_tree_block(fs_info, right); |
098f59c2 | 3882 | |
5f39d397 | 3883 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3884 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3885 | |
3886 | /* then fixup the leaf pointer in the path */ | |
3887 | if (path->slots[0] < push_items) { | |
3888 | path->slots[0] += old_left_nritems; | |
925baedd | 3889 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3890 | free_extent_buffer(path->nodes[0]); |
3891 | path->nodes[0] = left; | |
be0e5c09 CM |
3892 | path->slots[1] -= 1; |
3893 | } else { | |
925baedd | 3894 | btrfs_tree_unlock(left); |
5f39d397 | 3895 | free_extent_buffer(left); |
be0e5c09 CM |
3896 | path->slots[0] -= push_items; |
3897 | } | |
eb60ceac | 3898 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3899 | return ret; |
925baedd CM |
3900 | out: |
3901 | btrfs_tree_unlock(left); | |
3902 | free_extent_buffer(left); | |
3903 | return ret; | |
be0e5c09 CM |
3904 | } |
3905 | ||
44871b1b CM |
3906 | /* |
3907 | * push some data in the path leaf to the left, trying to free up at | |
3908 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3909 | * |
3910 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3911 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3912 | * items | |
44871b1b CM |
3913 | */ |
3914 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3915 | *root, struct btrfs_path *path, int min_data_size, |
3916 | int data_size, int empty, u32 max_slot) | |
44871b1b | 3917 | { |
2ff7e61e | 3918 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3919 | struct extent_buffer *right = path->nodes[0]; |
3920 | struct extent_buffer *left; | |
3921 | int slot; | |
3922 | int free_space; | |
3923 | u32 right_nritems; | |
3924 | int ret = 0; | |
3925 | ||
3926 | slot = path->slots[1]; | |
3927 | if (slot == 0) | |
3928 | return 1; | |
3929 | if (!path->nodes[1]) | |
3930 | return 1; | |
3931 | ||
3932 | right_nritems = btrfs_header_nritems(right); | |
3933 | if (right_nritems == 0) | |
3934 | return 1; | |
3935 | ||
3936 | btrfs_assert_tree_locked(path->nodes[1]); | |
3937 | ||
2ff7e61e | 3938 | left = read_node_slot(fs_info, path->nodes[1], slot - 1); |
fb770ae4 LB |
3939 | /* |
3940 | * slot - 1 is not valid or we fail to read the left node, | |
3941 | * no big deal, just return. | |
3942 | */ | |
3943 | if (IS_ERR(left)) | |
91ca338d TI |
3944 | return 1; |
3945 | ||
44871b1b CM |
3946 | btrfs_tree_lock(left); |
3947 | btrfs_set_lock_blocking(left); | |
3948 | ||
2ff7e61e | 3949 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
3950 | if (free_space < data_size) { |
3951 | ret = 1; | |
3952 | goto out; | |
3953 | } | |
3954 | ||
3955 | /* cow and double check */ | |
3956 | ret = btrfs_cow_block(trans, root, left, | |
3957 | path->nodes[1], slot - 1, &left); | |
3958 | if (ret) { | |
3959 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3960 | if (ret == -ENOSPC) |
3961 | ret = 1; | |
44871b1b CM |
3962 | goto out; |
3963 | } | |
3964 | ||
2ff7e61e | 3965 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
3966 | if (free_space < data_size) { |
3967 | ret = 1; | |
3968 | goto out; | |
3969 | } | |
3970 | ||
66cb7ddb | 3971 | return __push_leaf_left(fs_info, path, min_data_size, |
99d8f83c CM |
3972 | empty, left, free_space, right_nritems, |
3973 | max_slot); | |
44871b1b CM |
3974 | out: |
3975 | btrfs_tree_unlock(left); | |
3976 | free_extent_buffer(left); | |
3977 | return ret; | |
3978 | } | |
3979 | ||
3980 | /* | |
3981 | * split the path's leaf in two, making sure there is at least data_size | |
3982 | * available for the resulting leaf level of the path. | |
44871b1b | 3983 | */ |
143bede5 | 3984 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2ff7e61e | 3985 | struct btrfs_fs_info *fs_info, |
143bede5 JM |
3986 | struct btrfs_path *path, |
3987 | struct extent_buffer *l, | |
3988 | struct extent_buffer *right, | |
3989 | int slot, int mid, int nritems) | |
44871b1b CM |
3990 | { |
3991 | int data_copy_size; | |
3992 | int rt_data_off; | |
3993 | int i; | |
44871b1b | 3994 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3995 | struct btrfs_map_token token; |
3996 | ||
3997 | btrfs_init_map_token(&token); | |
44871b1b CM |
3998 | |
3999 | nritems = nritems - mid; | |
4000 | btrfs_set_header_nritems(right, nritems); | |
2ff7e61e | 4001 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l); |
44871b1b CM |
4002 | |
4003 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4004 | btrfs_item_nr_offset(mid), | |
4005 | nritems * sizeof(struct btrfs_item)); | |
4006 | ||
4007 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4008 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4009 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
2ff7e61e | 4010 | leaf_data_end(fs_info, l), data_copy_size); |
44871b1b | 4011 | |
0b246afa | 4012 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4013 | |
4014 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4015 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4016 | u32 ioff; |
4017 | ||
cfed81a0 CM |
4018 | ioff = btrfs_token_item_offset(right, item, &token); |
4019 | btrfs_set_token_item_offset(right, item, | |
4020 | ioff + rt_data_off, &token); | |
44871b1b CM |
4021 | } |
4022 | ||
44871b1b | 4023 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4024 | btrfs_item_key(right, &disk_key, 0); |
2ff7e61e | 4025 | insert_ptr(trans, fs_info, path, &disk_key, right->start, |
c3e06965 | 4026 | path->slots[1] + 1, 1); |
44871b1b CM |
4027 | |
4028 | btrfs_mark_buffer_dirty(right); | |
4029 | btrfs_mark_buffer_dirty(l); | |
4030 | BUG_ON(path->slots[0] != slot); | |
4031 | ||
44871b1b CM |
4032 | if (mid <= slot) { |
4033 | btrfs_tree_unlock(path->nodes[0]); | |
4034 | free_extent_buffer(path->nodes[0]); | |
4035 | path->nodes[0] = right; | |
4036 | path->slots[0] -= mid; | |
4037 | path->slots[1] += 1; | |
4038 | } else { | |
4039 | btrfs_tree_unlock(right); | |
4040 | free_extent_buffer(right); | |
4041 | } | |
4042 | ||
4043 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4044 | } |
4045 | ||
99d8f83c CM |
4046 | /* |
4047 | * double splits happen when we need to insert a big item in the middle | |
4048 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4049 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4050 | * A B C | |
4051 | * | |
4052 | * We avoid this by trying to push the items on either side of our target | |
4053 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4054 | * completely. | |
4055 | */ | |
4056 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4057 | struct btrfs_root *root, | |
4058 | struct btrfs_path *path, | |
4059 | int data_size) | |
4060 | { | |
2ff7e61e | 4061 | struct btrfs_fs_info *fs_info = root->fs_info; |
99d8f83c CM |
4062 | int ret; |
4063 | int progress = 0; | |
4064 | int slot; | |
4065 | u32 nritems; | |
5a4267ca | 4066 | int space_needed = data_size; |
99d8f83c CM |
4067 | |
4068 | slot = path->slots[0]; | |
5a4267ca | 4069 | if (slot < btrfs_header_nritems(path->nodes[0])) |
2ff7e61e | 4070 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); |
99d8f83c CM |
4071 | |
4072 | /* | |
4073 | * try to push all the items after our slot into the | |
4074 | * right leaf | |
4075 | */ | |
5a4267ca | 4076 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4077 | if (ret < 0) |
4078 | return ret; | |
4079 | ||
4080 | if (ret == 0) | |
4081 | progress++; | |
4082 | ||
4083 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4084 | /* | |
4085 | * our goal is to get our slot at the start or end of a leaf. If | |
4086 | * we've done so we're done | |
4087 | */ | |
4088 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4089 | return 0; | |
4090 | ||
2ff7e61e | 4091 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4092 | return 0; |
4093 | ||
4094 | /* try to push all the items before our slot into the next leaf */ | |
4095 | slot = path->slots[0]; | |
263d3995 FM |
4096 | space_needed = data_size; |
4097 | if (slot > 0) | |
4098 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); | |
5a4267ca | 4099 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4100 | if (ret < 0) |
4101 | return ret; | |
4102 | ||
4103 | if (ret == 0) | |
4104 | progress++; | |
4105 | ||
4106 | if (progress) | |
4107 | return 0; | |
4108 | return 1; | |
4109 | } | |
4110 | ||
74123bd7 CM |
4111 | /* |
4112 | * split the path's leaf in two, making sure there is at least data_size | |
4113 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4114 | * |
4115 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4116 | */ |
e02119d5 CM |
4117 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4118 | struct btrfs_root *root, | |
310712b2 | 4119 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4120 | struct btrfs_path *path, int data_size, |
4121 | int extend) | |
be0e5c09 | 4122 | { |
5d4f98a2 | 4123 | struct btrfs_disk_key disk_key; |
5f39d397 | 4124 | struct extent_buffer *l; |
7518a238 | 4125 | u32 nritems; |
eb60ceac CM |
4126 | int mid; |
4127 | int slot; | |
5f39d397 | 4128 | struct extent_buffer *right; |
b7a0365e | 4129 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4130 | int ret = 0; |
aa5d6bed | 4131 | int wret; |
5d4f98a2 | 4132 | int split; |
cc0c5538 | 4133 | int num_doubles = 0; |
99d8f83c | 4134 | int tried_avoid_double = 0; |
aa5d6bed | 4135 | |
a5719521 YZ |
4136 | l = path->nodes[0]; |
4137 | slot = path->slots[0]; | |
4138 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4139 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4140 | return -EOVERFLOW; |
4141 | ||
40689478 | 4142 | /* first try to make some room by pushing left and right */ |
33157e05 | 4143 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4144 | int space_needed = data_size; |
4145 | ||
4146 | if (slot < btrfs_header_nritems(l)) | |
2ff7e61e | 4147 | space_needed -= btrfs_leaf_free_space(fs_info, l); |
5a4267ca FDBM |
4148 | |
4149 | wret = push_leaf_right(trans, root, path, space_needed, | |
4150 | space_needed, 0, 0); | |
d397712b | 4151 | if (wret < 0) |
eaee50e8 | 4152 | return wret; |
3685f791 | 4153 | if (wret) { |
263d3995 FM |
4154 | space_needed = data_size; |
4155 | if (slot > 0) | |
4156 | space_needed -= btrfs_leaf_free_space(fs_info, | |
4157 | l); | |
5a4267ca FDBM |
4158 | wret = push_leaf_left(trans, root, path, space_needed, |
4159 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4160 | if (wret < 0) |
4161 | return wret; | |
4162 | } | |
4163 | l = path->nodes[0]; | |
aa5d6bed | 4164 | |
3685f791 | 4165 | /* did the pushes work? */ |
2ff7e61e | 4166 | if (btrfs_leaf_free_space(fs_info, l) >= data_size) |
3685f791 | 4167 | return 0; |
3326d1b0 | 4168 | } |
aa5d6bed | 4169 | |
5c680ed6 | 4170 | if (!path->nodes[1]) { |
fdd99c72 | 4171 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4172 | if (ret) |
4173 | return ret; | |
4174 | } | |
cc0c5538 | 4175 | again: |
5d4f98a2 | 4176 | split = 1; |
cc0c5538 | 4177 | l = path->nodes[0]; |
eb60ceac | 4178 | slot = path->slots[0]; |
5f39d397 | 4179 | nritems = btrfs_header_nritems(l); |
d397712b | 4180 | mid = (nritems + 1) / 2; |
54aa1f4d | 4181 | |
5d4f98a2 YZ |
4182 | if (mid <= slot) { |
4183 | if (nritems == 1 || | |
4184 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4185 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4186 | if (slot >= nritems) { |
4187 | split = 0; | |
4188 | } else { | |
4189 | mid = slot; | |
4190 | if (mid != nritems && | |
4191 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4192 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4193 | if (data_size && !tried_avoid_double) |
4194 | goto push_for_double; | |
5d4f98a2 YZ |
4195 | split = 2; |
4196 | } | |
4197 | } | |
4198 | } | |
4199 | } else { | |
4200 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4201 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4202 | if (!extend && data_size && slot == 0) { |
4203 | split = 0; | |
4204 | } else if ((extend || !data_size) && slot == 0) { | |
4205 | mid = 1; | |
4206 | } else { | |
4207 | mid = slot; | |
4208 | if (mid != nritems && | |
4209 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4210 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4211 | if (data_size && !tried_avoid_double) |
4212 | goto push_for_double; | |
67871254 | 4213 | split = 2; |
5d4f98a2 YZ |
4214 | } |
4215 | } | |
4216 | } | |
4217 | } | |
4218 | ||
4219 | if (split == 0) | |
4220 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4221 | else | |
4222 | btrfs_item_key(l, &disk_key, mid); | |
4223 | ||
4d75f8a9 DS |
4224 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4225 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4226 | if (IS_ERR(right)) |
5f39d397 | 4227 | return PTR_ERR(right); |
f0486c68 | 4228 | |
0b246afa | 4229 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4230 | |
5d4f98a2 YZ |
4231 | if (split == 0) { |
4232 | if (mid <= slot) { | |
4233 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4234 | insert_ptr(trans, fs_info, path, &disk_key, |
4235 | right->start, path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
4236 | btrfs_tree_unlock(path->nodes[0]); |
4237 | free_extent_buffer(path->nodes[0]); | |
4238 | path->nodes[0] = right; | |
4239 | path->slots[0] = 0; | |
4240 | path->slots[1] += 1; | |
4241 | } else { | |
4242 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4243 | insert_ptr(trans, fs_info, path, &disk_key, |
4244 | right->start, path->slots[1], 1); | |
5d4f98a2 YZ |
4245 | btrfs_tree_unlock(path->nodes[0]); |
4246 | free_extent_buffer(path->nodes[0]); | |
4247 | path->nodes[0] = right; | |
4248 | path->slots[0] = 0; | |
143bede5 | 4249 | if (path->slots[1] == 0) |
b167fa91 | 4250 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4251 | } |
196e0249 LB |
4252 | /* |
4253 | * We create a new leaf 'right' for the required ins_len and | |
4254 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4255 | * the content of ins_len to 'right'. | |
4256 | */ | |
5d4f98a2 | 4257 | return ret; |
d4dbff95 | 4258 | } |
74123bd7 | 4259 | |
2ff7e61e | 4260 | copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems); |
31840ae1 | 4261 | |
5d4f98a2 | 4262 | if (split == 2) { |
cc0c5538 CM |
4263 | BUG_ON(num_doubles != 0); |
4264 | num_doubles++; | |
4265 | goto again; | |
a429e513 | 4266 | } |
44871b1b | 4267 | |
143bede5 | 4268 | return 0; |
99d8f83c CM |
4269 | |
4270 | push_for_double: | |
4271 | push_for_double_split(trans, root, path, data_size); | |
4272 | tried_avoid_double = 1; | |
2ff7e61e | 4273 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4274 | return 0; |
4275 | goto again; | |
be0e5c09 CM |
4276 | } |
4277 | ||
ad48fd75 YZ |
4278 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4279 | struct btrfs_root *root, | |
4280 | struct btrfs_path *path, int ins_len) | |
459931ec | 4281 | { |
2ff7e61e | 4282 | struct btrfs_fs_info *fs_info = root->fs_info; |
ad48fd75 | 4283 | struct btrfs_key key; |
459931ec | 4284 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4285 | struct btrfs_file_extent_item *fi; |
4286 | u64 extent_len = 0; | |
4287 | u32 item_size; | |
4288 | int ret; | |
459931ec CM |
4289 | |
4290 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4291 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4292 | ||
4293 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4294 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4295 | ||
2ff7e61e | 4296 | if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len) |
ad48fd75 | 4297 | return 0; |
459931ec CM |
4298 | |
4299 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4300 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4301 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4302 | struct btrfs_file_extent_item); | |
4303 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4304 | } | |
b3b4aa74 | 4305 | btrfs_release_path(path); |
459931ec | 4306 | |
459931ec | 4307 | path->keep_locks = 1; |
ad48fd75 YZ |
4308 | path->search_for_split = 1; |
4309 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4310 | path->search_for_split = 0; |
a8df6fe6 FM |
4311 | if (ret > 0) |
4312 | ret = -EAGAIN; | |
ad48fd75 YZ |
4313 | if (ret < 0) |
4314 | goto err; | |
459931ec | 4315 | |
ad48fd75 YZ |
4316 | ret = -EAGAIN; |
4317 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4318 | /* if our item isn't there, return now */ |
4319 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4320 | goto err; |
4321 | ||
109f6aef | 4322 | /* the leaf has changed, it now has room. return now */ |
2ff7e61e | 4323 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len) |
109f6aef CM |
4324 | goto err; |
4325 | ||
ad48fd75 YZ |
4326 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4327 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4328 | struct btrfs_file_extent_item); | |
4329 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4330 | goto err; | |
459931ec CM |
4331 | } |
4332 | ||
b9473439 | 4333 | btrfs_set_path_blocking(path); |
ad48fd75 | 4334 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4335 | if (ret) |
4336 | goto err; | |
459931ec | 4337 | |
ad48fd75 | 4338 | path->keep_locks = 0; |
b9473439 | 4339 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4340 | return 0; |
4341 | err: | |
4342 | path->keep_locks = 0; | |
4343 | return ret; | |
4344 | } | |
4345 | ||
4961e293 | 4346 | static noinline int split_item(struct btrfs_fs_info *fs_info, |
ad48fd75 | 4347 | struct btrfs_path *path, |
310712b2 | 4348 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4349 | unsigned long split_offset) |
4350 | { | |
4351 | struct extent_buffer *leaf; | |
4352 | struct btrfs_item *item; | |
4353 | struct btrfs_item *new_item; | |
4354 | int slot; | |
4355 | char *buf; | |
4356 | u32 nritems; | |
4357 | u32 item_size; | |
4358 | u32 orig_offset; | |
4359 | struct btrfs_disk_key disk_key; | |
4360 | ||
b9473439 | 4361 | leaf = path->nodes[0]; |
2ff7e61e | 4362 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4363 | |
b4ce94de CM |
4364 | btrfs_set_path_blocking(path); |
4365 | ||
dd3cc16b | 4366 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4367 | orig_offset = btrfs_item_offset(leaf, item); |
4368 | item_size = btrfs_item_size(leaf, item); | |
4369 | ||
459931ec | 4370 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4371 | if (!buf) |
4372 | return -ENOMEM; | |
4373 | ||
459931ec CM |
4374 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4375 | path->slots[0]), item_size); | |
459931ec | 4376 | |
ad48fd75 | 4377 | slot = path->slots[0] + 1; |
459931ec | 4378 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4379 | if (slot != nritems) { |
4380 | /* shift the items */ | |
4381 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4382 | btrfs_item_nr_offset(slot), |
4383 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4384 | } |
4385 | ||
4386 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4387 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4388 | ||
dd3cc16b | 4389 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4390 | |
4391 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4392 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4393 | ||
4394 | btrfs_set_item_offset(leaf, item, | |
4395 | orig_offset + item_size - split_offset); | |
4396 | btrfs_set_item_size(leaf, item, split_offset); | |
4397 | ||
4398 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4399 | ||
4400 | /* write the data for the start of the original item */ | |
4401 | write_extent_buffer(leaf, buf, | |
4402 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4403 | split_offset); | |
4404 | ||
4405 | /* write the data for the new item */ | |
4406 | write_extent_buffer(leaf, buf + split_offset, | |
4407 | btrfs_item_ptr_offset(leaf, slot), | |
4408 | item_size - split_offset); | |
4409 | btrfs_mark_buffer_dirty(leaf); | |
4410 | ||
2ff7e61e | 4411 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0); |
459931ec | 4412 | kfree(buf); |
ad48fd75 YZ |
4413 | return 0; |
4414 | } | |
4415 | ||
4416 | /* | |
4417 | * This function splits a single item into two items, | |
4418 | * giving 'new_key' to the new item and splitting the | |
4419 | * old one at split_offset (from the start of the item). | |
4420 | * | |
4421 | * The path may be released by this operation. After | |
4422 | * the split, the path is pointing to the old item. The | |
4423 | * new item is going to be in the same node as the old one. | |
4424 | * | |
4425 | * Note, the item being split must be smaller enough to live alone on | |
4426 | * a tree block with room for one extra struct btrfs_item | |
4427 | * | |
4428 | * This allows us to split the item in place, keeping a lock on the | |
4429 | * leaf the entire time. | |
4430 | */ | |
4431 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4432 | struct btrfs_root *root, | |
4433 | struct btrfs_path *path, | |
310712b2 | 4434 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4435 | unsigned long split_offset) |
4436 | { | |
4437 | int ret; | |
4438 | ret = setup_leaf_for_split(trans, root, path, | |
4439 | sizeof(struct btrfs_item)); | |
4440 | if (ret) | |
4441 | return ret; | |
4442 | ||
4961e293 | 4443 | ret = split_item(root->fs_info, path, new_key, split_offset); |
459931ec CM |
4444 | return ret; |
4445 | } | |
4446 | ||
ad48fd75 YZ |
4447 | /* |
4448 | * This function duplicate a item, giving 'new_key' to the new item. | |
4449 | * It guarantees both items live in the same tree leaf and the new item | |
4450 | * is contiguous with the original item. | |
4451 | * | |
4452 | * This allows us to split file extent in place, keeping a lock on the | |
4453 | * leaf the entire time. | |
4454 | */ | |
4455 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4456 | struct btrfs_root *root, | |
4457 | struct btrfs_path *path, | |
310712b2 | 4458 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4459 | { |
4460 | struct extent_buffer *leaf; | |
4461 | int ret; | |
4462 | u32 item_size; | |
4463 | ||
4464 | leaf = path->nodes[0]; | |
4465 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4466 | ret = setup_leaf_for_split(trans, root, path, | |
4467 | item_size + sizeof(struct btrfs_item)); | |
4468 | if (ret) | |
4469 | return ret; | |
4470 | ||
4471 | path->slots[0]++; | |
afe5fea7 | 4472 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4473 | item_size, item_size + |
4474 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4475 | leaf = path->nodes[0]; |
4476 | memcpy_extent_buffer(leaf, | |
4477 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4478 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4479 | item_size); | |
4480 | return 0; | |
4481 | } | |
4482 | ||
d352ac68 CM |
4483 | /* |
4484 | * make the item pointed to by the path smaller. new_size indicates | |
4485 | * how small to make it, and from_end tells us if we just chop bytes | |
4486 | * off the end of the item or if we shift the item to chop bytes off | |
4487 | * the front. | |
4488 | */ | |
2ff7e61e JM |
4489 | void btrfs_truncate_item(struct btrfs_fs_info *fs_info, |
4490 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 4491 | { |
b18c6685 | 4492 | int slot; |
5f39d397 CM |
4493 | struct extent_buffer *leaf; |
4494 | struct btrfs_item *item; | |
b18c6685 CM |
4495 | u32 nritems; |
4496 | unsigned int data_end; | |
4497 | unsigned int old_data_start; | |
4498 | unsigned int old_size; | |
4499 | unsigned int size_diff; | |
4500 | int i; | |
cfed81a0 CM |
4501 | struct btrfs_map_token token; |
4502 | ||
4503 | btrfs_init_map_token(&token); | |
b18c6685 | 4504 | |
5f39d397 | 4505 | leaf = path->nodes[0]; |
179e29e4 CM |
4506 | slot = path->slots[0]; |
4507 | ||
4508 | old_size = btrfs_item_size_nr(leaf, slot); | |
4509 | if (old_size == new_size) | |
143bede5 | 4510 | return; |
b18c6685 | 4511 | |
5f39d397 | 4512 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4513 | data_end = leaf_data_end(fs_info, leaf); |
b18c6685 | 4514 | |
5f39d397 | 4515 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4516 | |
b18c6685 CM |
4517 | size_diff = old_size - new_size; |
4518 | ||
4519 | BUG_ON(slot < 0); | |
4520 | BUG_ON(slot >= nritems); | |
4521 | ||
4522 | /* | |
4523 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4524 | */ | |
4525 | /* first correct the data pointers */ | |
4526 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4527 | u32 ioff; |
dd3cc16b | 4528 | item = btrfs_item_nr(i); |
db94535d | 4529 | |
cfed81a0 CM |
4530 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4531 | btrfs_set_token_item_offset(leaf, item, | |
4532 | ioff + size_diff, &token); | |
b18c6685 | 4533 | } |
db94535d | 4534 | |
b18c6685 | 4535 | /* shift the data */ |
179e29e4 | 4536 | if (from_end) { |
3d9ec8c4 NB |
4537 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4538 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4539 | data_end, old_data_start + new_size - data_end); |
4540 | } else { | |
4541 | struct btrfs_disk_key disk_key; | |
4542 | u64 offset; | |
4543 | ||
4544 | btrfs_item_key(leaf, &disk_key, slot); | |
4545 | ||
4546 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4547 | unsigned long ptr; | |
4548 | struct btrfs_file_extent_item *fi; | |
4549 | ||
4550 | fi = btrfs_item_ptr(leaf, slot, | |
4551 | struct btrfs_file_extent_item); | |
4552 | fi = (struct btrfs_file_extent_item *)( | |
4553 | (unsigned long)fi - size_diff); | |
4554 | ||
4555 | if (btrfs_file_extent_type(leaf, fi) == | |
4556 | BTRFS_FILE_EXTENT_INLINE) { | |
4557 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4558 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4559 | (unsigned long)fi, |
7ec20afb | 4560 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4561 | } |
4562 | } | |
4563 | ||
3d9ec8c4 NB |
4564 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4565 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4566 | data_end, old_data_start - data_end); |
4567 | ||
4568 | offset = btrfs_disk_key_offset(&disk_key); | |
4569 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4570 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4571 | if (slot == 0) | |
b167fa91 | 4572 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4573 | } |
5f39d397 | 4574 | |
dd3cc16b | 4575 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4576 | btrfs_set_item_size(leaf, item, new_size); |
4577 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4578 | |
2ff7e61e | 4579 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4580 | btrfs_print_leaf(leaf); |
b18c6685 | 4581 | BUG(); |
5f39d397 | 4582 | } |
b18c6685 CM |
4583 | } |
4584 | ||
d352ac68 | 4585 | /* |
8f69dbd2 | 4586 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4587 | */ |
2ff7e61e | 4588 | void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 4589 | u32 data_size) |
6567e837 | 4590 | { |
6567e837 | 4591 | int slot; |
5f39d397 CM |
4592 | struct extent_buffer *leaf; |
4593 | struct btrfs_item *item; | |
6567e837 CM |
4594 | u32 nritems; |
4595 | unsigned int data_end; | |
4596 | unsigned int old_data; | |
4597 | unsigned int old_size; | |
4598 | int i; | |
cfed81a0 CM |
4599 | struct btrfs_map_token token; |
4600 | ||
4601 | btrfs_init_map_token(&token); | |
6567e837 | 4602 | |
5f39d397 | 4603 | leaf = path->nodes[0]; |
6567e837 | 4604 | |
5f39d397 | 4605 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4606 | data_end = leaf_data_end(fs_info, leaf); |
6567e837 | 4607 | |
2ff7e61e | 4608 | if (btrfs_leaf_free_space(fs_info, leaf) < data_size) { |
a4f78750 | 4609 | btrfs_print_leaf(leaf); |
6567e837 | 4610 | BUG(); |
5f39d397 | 4611 | } |
6567e837 | 4612 | slot = path->slots[0]; |
5f39d397 | 4613 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4614 | |
4615 | BUG_ON(slot < 0); | |
3326d1b0 | 4616 | if (slot >= nritems) { |
a4f78750 | 4617 | btrfs_print_leaf(leaf); |
0b246afa JM |
4618 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4619 | slot, nritems); | |
3326d1b0 CM |
4620 | BUG_ON(1); |
4621 | } | |
6567e837 CM |
4622 | |
4623 | /* | |
4624 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4625 | */ | |
4626 | /* first correct the data pointers */ | |
4627 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4628 | u32 ioff; |
dd3cc16b | 4629 | item = btrfs_item_nr(i); |
db94535d | 4630 | |
cfed81a0 CM |
4631 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4632 | btrfs_set_token_item_offset(leaf, item, | |
4633 | ioff - data_size, &token); | |
6567e837 | 4634 | } |
5f39d397 | 4635 | |
6567e837 | 4636 | /* shift the data */ |
3d9ec8c4 NB |
4637 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4638 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4639 | data_end, old_data - data_end); |
5f39d397 | 4640 | |
6567e837 | 4641 | data_end = old_data; |
5f39d397 | 4642 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4643 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4644 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4645 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4646 | |
2ff7e61e | 4647 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4648 | btrfs_print_leaf(leaf); |
6567e837 | 4649 | BUG(); |
5f39d397 | 4650 | } |
6567e837 CM |
4651 | } |
4652 | ||
74123bd7 | 4653 | /* |
44871b1b CM |
4654 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4655 | * to save stack depth by doing the bulk of the work in a function | |
4656 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4657 | */ |
afe5fea7 | 4658 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4659 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4660 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4661 | { |
0b246afa | 4662 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4663 | struct btrfs_item *item; |
9c58309d | 4664 | int i; |
7518a238 | 4665 | u32 nritems; |
be0e5c09 | 4666 | unsigned int data_end; |
e2fa7227 | 4667 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4668 | struct extent_buffer *leaf; |
4669 | int slot; | |
cfed81a0 CM |
4670 | struct btrfs_map_token token; |
4671 | ||
24cdc847 FM |
4672 | if (path->slots[0] == 0) { |
4673 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4674 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4675 | } |
4676 | btrfs_unlock_up_safe(path, 1); | |
4677 | ||
cfed81a0 | 4678 | btrfs_init_map_token(&token); |
e2fa7227 | 4679 | |
5f39d397 | 4680 | leaf = path->nodes[0]; |
44871b1b | 4681 | slot = path->slots[0]; |
74123bd7 | 4682 | |
5f39d397 | 4683 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4684 | data_end = leaf_data_end(fs_info, leaf); |
eb60ceac | 4685 | |
2ff7e61e | 4686 | if (btrfs_leaf_free_space(fs_info, leaf) < total_size) { |
a4f78750 | 4687 | btrfs_print_leaf(leaf); |
0b246afa | 4688 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
2ff7e61e | 4689 | total_size, btrfs_leaf_free_space(fs_info, leaf)); |
be0e5c09 | 4690 | BUG(); |
d4dbff95 | 4691 | } |
5f39d397 | 4692 | |
be0e5c09 | 4693 | if (slot != nritems) { |
5f39d397 | 4694 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4695 | |
5f39d397 | 4696 | if (old_data < data_end) { |
a4f78750 | 4697 | btrfs_print_leaf(leaf); |
0b246afa | 4698 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4699 | slot, old_data, data_end); |
5f39d397 CM |
4700 | BUG_ON(1); |
4701 | } | |
be0e5c09 CM |
4702 | /* |
4703 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4704 | */ | |
4705 | /* first correct the data pointers */ | |
0783fcfc | 4706 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4707 | u32 ioff; |
db94535d | 4708 | |
62e85577 | 4709 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4710 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4711 | btrfs_set_token_item_offset(leaf, item, | |
4712 | ioff - total_data, &token); | |
0783fcfc | 4713 | } |
be0e5c09 | 4714 | /* shift the items */ |
9c58309d | 4715 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4716 | btrfs_item_nr_offset(slot), |
d6025579 | 4717 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4718 | |
4719 | /* shift the data */ | |
3d9ec8c4 NB |
4720 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4721 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4722 | data_end, old_data - data_end); |
be0e5c09 CM |
4723 | data_end = old_data; |
4724 | } | |
5f39d397 | 4725 | |
62e2749e | 4726 | /* setup the item for the new data */ |
9c58309d CM |
4727 | for (i = 0; i < nr; i++) { |
4728 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4729 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4730 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4731 | btrfs_set_token_item_offset(leaf, item, |
4732 | data_end - data_size[i], &token); | |
9c58309d | 4733 | data_end -= data_size[i]; |
cfed81a0 | 4734 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4735 | } |
44871b1b | 4736 | |
9c58309d | 4737 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4738 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4739 | |
2ff7e61e | 4740 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4741 | btrfs_print_leaf(leaf); |
be0e5c09 | 4742 | BUG(); |
5f39d397 | 4743 | } |
44871b1b CM |
4744 | } |
4745 | ||
4746 | /* | |
4747 | * Given a key and some data, insert items into the tree. | |
4748 | * This does all the path init required, making room in the tree if needed. | |
4749 | */ | |
4750 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4751 | struct btrfs_root *root, | |
4752 | struct btrfs_path *path, | |
310712b2 | 4753 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4754 | int nr) |
4755 | { | |
44871b1b CM |
4756 | int ret = 0; |
4757 | int slot; | |
4758 | int i; | |
4759 | u32 total_size = 0; | |
4760 | u32 total_data = 0; | |
4761 | ||
4762 | for (i = 0; i < nr; i++) | |
4763 | total_data += data_size[i]; | |
4764 | ||
4765 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4766 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4767 | if (ret == 0) | |
4768 | return -EEXIST; | |
4769 | if (ret < 0) | |
143bede5 | 4770 | return ret; |
44871b1b | 4771 | |
44871b1b CM |
4772 | slot = path->slots[0]; |
4773 | BUG_ON(slot < 0); | |
4774 | ||
afe5fea7 | 4775 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4776 | total_data, total_size, nr); |
143bede5 | 4777 | return 0; |
62e2749e CM |
4778 | } |
4779 | ||
4780 | /* | |
4781 | * Given a key and some data, insert an item into the tree. | |
4782 | * This does all the path init required, making room in the tree if needed. | |
4783 | */ | |
310712b2 OS |
4784 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4785 | const struct btrfs_key *cpu_key, void *data, | |
4786 | u32 data_size) | |
62e2749e CM |
4787 | { |
4788 | int ret = 0; | |
2c90e5d6 | 4789 | struct btrfs_path *path; |
5f39d397 CM |
4790 | struct extent_buffer *leaf; |
4791 | unsigned long ptr; | |
62e2749e | 4792 | |
2c90e5d6 | 4793 | path = btrfs_alloc_path(); |
db5b493a TI |
4794 | if (!path) |
4795 | return -ENOMEM; | |
2c90e5d6 | 4796 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4797 | if (!ret) { |
5f39d397 CM |
4798 | leaf = path->nodes[0]; |
4799 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4800 | write_extent_buffer(leaf, data, ptr, data_size); | |
4801 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4802 | } |
2c90e5d6 | 4803 | btrfs_free_path(path); |
aa5d6bed | 4804 | return ret; |
be0e5c09 CM |
4805 | } |
4806 | ||
74123bd7 | 4807 | /* |
5de08d7d | 4808 | * delete the pointer from a given node. |
74123bd7 | 4809 | * |
d352ac68 CM |
4810 | * the tree should have been previously balanced so the deletion does not |
4811 | * empty a node. | |
74123bd7 | 4812 | */ |
afe5fea7 TI |
4813 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4814 | int level, int slot) | |
be0e5c09 | 4815 | { |
5f39d397 | 4816 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4817 | u32 nritems; |
f3ea38da | 4818 | int ret; |
be0e5c09 | 4819 | |
5f39d397 | 4820 | nritems = btrfs_header_nritems(parent); |
d397712b | 4821 | if (slot != nritems - 1) { |
bf1d3425 DS |
4822 | if (level) { |
4823 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4824 | nritems - slot - 1); |
bf1d3425 DS |
4825 | BUG_ON(ret < 0); |
4826 | } | |
5f39d397 CM |
4827 | memmove_extent_buffer(parent, |
4828 | btrfs_node_key_ptr_offset(slot), | |
4829 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4830 | sizeof(struct btrfs_key_ptr) * |
4831 | (nritems - slot - 1)); | |
57ba86c0 | 4832 | } else if (level) { |
e09c2efe DS |
4833 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4834 | GFP_NOFS); | |
57ba86c0 | 4835 | BUG_ON(ret < 0); |
bb803951 | 4836 | } |
f3ea38da | 4837 | |
7518a238 | 4838 | nritems--; |
5f39d397 | 4839 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4840 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4841 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4842 | /* just turn the root into a leaf and break */ |
5f39d397 | 4843 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4844 | } else if (slot == 0) { |
5f39d397 CM |
4845 | struct btrfs_disk_key disk_key; |
4846 | ||
4847 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4848 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4849 | } |
d6025579 | 4850 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4851 | } |
4852 | ||
323ac95b CM |
4853 | /* |
4854 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4855 | * path->nodes[1]. |
323ac95b CM |
4856 | * |
4857 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4858 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4859 | * | |
4860 | * The path must have already been setup for deleting the leaf, including | |
4861 | * all the proper balancing. path->nodes[1] must be locked. | |
4862 | */ | |
143bede5 JM |
4863 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4864 | struct btrfs_root *root, | |
4865 | struct btrfs_path *path, | |
4866 | struct extent_buffer *leaf) | |
323ac95b | 4867 | { |
5d4f98a2 | 4868 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4869 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4870 | |
4d081c41 CM |
4871 | /* |
4872 | * btrfs_free_extent is expensive, we want to make sure we | |
4873 | * aren't holding any locks when we call it | |
4874 | */ | |
4875 | btrfs_unlock_up_safe(path, 0); | |
4876 | ||
f0486c68 YZ |
4877 | root_sub_used(root, leaf->len); |
4878 | ||
3083ee2e | 4879 | extent_buffer_get(leaf); |
5581a51a | 4880 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4881 | free_extent_buffer_stale(leaf); |
323ac95b | 4882 | } |
74123bd7 CM |
4883 | /* |
4884 | * delete the item at the leaf level in path. If that empties | |
4885 | * the leaf, remove it from the tree | |
4886 | */ | |
85e21bac CM |
4887 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4888 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4889 | { |
0b246afa | 4890 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4891 | struct extent_buffer *leaf; |
4892 | struct btrfs_item *item; | |
ce0eac2a AM |
4893 | u32 last_off; |
4894 | u32 dsize = 0; | |
aa5d6bed CM |
4895 | int ret = 0; |
4896 | int wret; | |
85e21bac | 4897 | int i; |
7518a238 | 4898 | u32 nritems; |
cfed81a0 CM |
4899 | struct btrfs_map_token token; |
4900 | ||
4901 | btrfs_init_map_token(&token); | |
be0e5c09 | 4902 | |
5f39d397 | 4903 | leaf = path->nodes[0]; |
85e21bac CM |
4904 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4905 | ||
4906 | for (i = 0; i < nr; i++) | |
4907 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4908 | ||
5f39d397 | 4909 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4910 | |
85e21bac | 4911 | if (slot + nr != nritems) { |
2ff7e61e | 4912 | int data_end = leaf_data_end(fs_info, leaf); |
5f39d397 | 4913 | |
3d9ec8c4 | 4914 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4915 | data_end + dsize, |
3d9ec8c4 | 4916 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4917 | last_off - data_end); |
5f39d397 | 4918 | |
85e21bac | 4919 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4920 | u32 ioff; |
db94535d | 4921 | |
dd3cc16b | 4922 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4923 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4924 | btrfs_set_token_item_offset(leaf, item, | |
4925 | ioff + dsize, &token); | |
0783fcfc | 4926 | } |
db94535d | 4927 | |
5f39d397 | 4928 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4929 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4930 | sizeof(struct btrfs_item) * |
85e21bac | 4931 | (nritems - slot - nr)); |
be0e5c09 | 4932 | } |
85e21bac CM |
4933 | btrfs_set_header_nritems(leaf, nritems - nr); |
4934 | nritems -= nr; | |
5f39d397 | 4935 | |
74123bd7 | 4936 | /* delete the leaf if we've emptied it */ |
7518a238 | 4937 | if (nritems == 0) { |
5f39d397 CM |
4938 | if (leaf == root->node) { |
4939 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4940 | } else { |
f0486c68 | 4941 | btrfs_set_path_blocking(path); |
7c302b49 | 4942 | clean_tree_block(fs_info, leaf); |
143bede5 | 4943 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4944 | } |
be0e5c09 | 4945 | } else { |
7518a238 | 4946 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4947 | if (slot == 0) { |
5f39d397 CM |
4948 | struct btrfs_disk_key disk_key; |
4949 | ||
4950 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4951 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 4952 | } |
aa5d6bed | 4953 | |
74123bd7 | 4954 | /* delete the leaf if it is mostly empty */ |
0b246afa | 4955 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
4956 | /* push_leaf_left fixes the path. |
4957 | * make sure the path still points to our leaf | |
4958 | * for possible call to del_ptr below | |
4959 | */ | |
4920c9ac | 4960 | slot = path->slots[1]; |
5f39d397 CM |
4961 | extent_buffer_get(leaf); |
4962 | ||
b9473439 | 4963 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4964 | wret = push_leaf_left(trans, root, path, 1, 1, |
4965 | 1, (u32)-1); | |
54aa1f4d | 4966 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4967 | ret = wret; |
5f39d397 CM |
4968 | |
4969 | if (path->nodes[0] == leaf && | |
4970 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4971 | wret = push_leaf_right(trans, root, path, 1, |
4972 | 1, 1, 0); | |
54aa1f4d | 4973 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4974 | ret = wret; |
4975 | } | |
5f39d397 CM |
4976 | |
4977 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4978 | path->slots[1] = slot; |
143bede5 | 4979 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4980 | free_extent_buffer(leaf); |
143bede5 | 4981 | ret = 0; |
5de08d7d | 4982 | } else { |
925baedd CM |
4983 | /* if we're still in the path, make sure |
4984 | * we're dirty. Otherwise, one of the | |
4985 | * push_leaf functions must have already | |
4986 | * dirtied this buffer | |
4987 | */ | |
4988 | if (path->nodes[0] == leaf) | |
4989 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4990 | free_extent_buffer(leaf); |
be0e5c09 | 4991 | } |
d5719762 | 4992 | } else { |
5f39d397 | 4993 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4994 | } |
4995 | } | |
aa5d6bed | 4996 | return ret; |
be0e5c09 CM |
4997 | } |
4998 | ||
7bb86316 | 4999 | /* |
925baedd | 5000 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5001 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5002 | * returns < 0 on io errors. | |
d352ac68 CM |
5003 | * |
5004 | * This may release the path, and so you may lose any locks held at the | |
5005 | * time you call it. | |
7bb86316 | 5006 | */ |
16e7549f | 5007 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5008 | { |
925baedd CM |
5009 | struct btrfs_key key; |
5010 | struct btrfs_disk_key found_key; | |
5011 | int ret; | |
7bb86316 | 5012 | |
925baedd | 5013 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5014 | |
e8b0d724 | 5015 | if (key.offset > 0) { |
925baedd | 5016 | key.offset--; |
e8b0d724 | 5017 | } else if (key.type > 0) { |
925baedd | 5018 | key.type--; |
e8b0d724 FDBM |
5019 | key.offset = (u64)-1; |
5020 | } else if (key.objectid > 0) { | |
925baedd | 5021 | key.objectid--; |
e8b0d724 FDBM |
5022 | key.type = (u8)-1; |
5023 | key.offset = (u64)-1; | |
5024 | } else { | |
925baedd | 5025 | return 1; |
e8b0d724 | 5026 | } |
7bb86316 | 5027 | |
b3b4aa74 | 5028 | btrfs_release_path(path); |
925baedd CM |
5029 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5030 | if (ret < 0) | |
5031 | return ret; | |
5032 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5033 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5034 | /* |
5035 | * We might have had an item with the previous key in the tree right | |
5036 | * before we released our path. And after we released our path, that | |
5037 | * item might have been pushed to the first slot (0) of the leaf we | |
5038 | * were holding due to a tree balance. Alternatively, an item with the | |
5039 | * previous key can exist as the only element of a leaf (big fat item). | |
5040 | * Therefore account for these 2 cases, so that our callers (like | |
5041 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5042 | * the previous key we computed above. | |
5043 | */ | |
5044 | if (ret <= 0) | |
925baedd CM |
5045 | return 0; |
5046 | return 1; | |
7bb86316 CM |
5047 | } |
5048 | ||
3f157a2f CM |
5049 | /* |
5050 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5051 | * for nodes or leaves that are have a minimum transaction id. |
5052 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5053 | * |
5054 | * This does not cow, but it does stuff the starting key it finds back | |
5055 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5056 | * key and get a writable path. | |
5057 | * | |
3f157a2f CM |
5058 | * This honors path->lowest_level to prevent descent past a given level |
5059 | * of the tree. | |
5060 | * | |
d352ac68 CM |
5061 | * min_trans indicates the oldest transaction that you are interested |
5062 | * in walking through. Any nodes or leaves older than min_trans are | |
5063 | * skipped over (without reading them). | |
5064 | * | |
3f157a2f CM |
5065 | * returns zero if something useful was found, < 0 on error and 1 if there |
5066 | * was nothing in the tree that matched the search criteria. | |
5067 | */ | |
5068 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5069 | struct btrfs_path *path, |
3f157a2f CM |
5070 | u64 min_trans) |
5071 | { | |
2ff7e61e | 5072 | struct btrfs_fs_info *fs_info = root->fs_info; |
3f157a2f CM |
5073 | struct extent_buffer *cur; |
5074 | struct btrfs_key found_key; | |
5075 | int slot; | |
9652480b | 5076 | int sret; |
3f157a2f CM |
5077 | u32 nritems; |
5078 | int level; | |
5079 | int ret = 1; | |
f98de9b9 | 5080 | int keep_locks = path->keep_locks; |
3f157a2f | 5081 | |
f98de9b9 | 5082 | path->keep_locks = 1; |
3f157a2f | 5083 | again: |
bd681513 | 5084 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5085 | level = btrfs_header_level(cur); |
e02119d5 | 5086 | WARN_ON(path->nodes[level]); |
3f157a2f | 5087 | path->nodes[level] = cur; |
bd681513 | 5088 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5089 | |
5090 | if (btrfs_header_generation(cur) < min_trans) { | |
5091 | ret = 1; | |
5092 | goto out; | |
5093 | } | |
d397712b | 5094 | while (1) { |
3f157a2f CM |
5095 | nritems = btrfs_header_nritems(cur); |
5096 | level = btrfs_header_level(cur); | |
a74b35ec | 5097 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
3f157a2f | 5098 | |
323ac95b CM |
5099 | /* at the lowest level, we're done, setup the path and exit */ |
5100 | if (level == path->lowest_level) { | |
e02119d5 CM |
5101 | if (slot >= nritems) |
5102 | goto find_next_key; | |
3f157a2f CM |
5103 | ret = 0; |
5104 | path->slots[level] = slot; | |
5105 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5106 | goto out; | |
5107 | } | |
9652480b Y |
5108 | if (sret && slot > 0) |
5109 | slot--; | |
3f157a2f | 5110 | /* |
de78b51a ES |
5111 | * check this node pointer against the min_trans parameters. |
5112 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5113 | */ |
d397712b | 5114 | while (slot < nritems) { |
3f157a2f | 5115 | u64 gen; |
e02119d5 | 5116 | |
3f157a2f CM |
5117 | gen = btrfs_node_ptr_generation(cur, slot); |
5118 | if (gen < min_trans) { | |
5119 | slot++; | |
5120 | continue; | |
5121 | } | |
de78b51a | 5122 | break; |
3f157a2f | 5123 | } |
e02119d5 | 5124 | find_next_key: |
3f157a2f CM |
5125 | /* |
5126 | * we didn't find a candidate key in this node, walk forward | |
5127 | * and find another one | |
5128 | */ | |
5129 | if (slot >= nritems) { | |
e02119d5 | 5130 | path->slots[level] = slot; |
b4ce94de | 5131 | btrfs_set_path_blocking(path); |
e02119d5 | 5132 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5133 | min_trans); |
e02119d5 | 5134 | if (sret == 0) { |
b3b4aa74 | 5135 | btrfs_release_path(path); |
3f157a2f CM |
5136 | goto again; |
5137 | } else { | |
5138 | goto out; | |
5139 | } | |
5140 | } | |
5141 | /* save our key for returning back */ | |
5142 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5143 | path->slots[level] = slot; | |
5144 | if (level == path->lowest_level) { | |
5145 | ret = 0; | |
3f157a2f CM |
5146 | goto out; |
5147 | } | |
b4ce94de | 5148 | btrfs_set_path_blocking(path); |
2ff7e61e | 5149 | cur = read_node_slot(fs_info, cur, slot); |
fb770ae4 LB |
5150 | if (IS_ERR(cur)) { |
5151 | ret = PTR_ERR(cur); | |
5152 | goto out; | |
5153 | } | |
3f157a2f | 5154 | |
bd681513 | 5155 | btrfs_tree_read_lock(cur); |
b4ce94de | 5156 | |
bd681513 | 5157 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5158 | path->nodes[level - 1] = cur; |
f7c79f30 | 5159 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5160 | } |
5161 | out: | |
f98de9b9 FM |
5162 | path->keep_locks = keep_locks; |
5163 | if (ret == 0) { | |
5164 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5165 | btrfs_set_path_blocking(path); | |
3f157a2f | 5166 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5167 | } |
3f157a2f CM |
5168 | return ret; |
5169 | } | |
5170 | ||
2ff7e61e | 5171 | static int tree_move_down(struct btrfs_fs_info *fs_info, |
7069830a | 5172 | struct btrfs_path *path, |
ab6a43e1 | 5173 | int *level) |
7069830a | 5174 | { |
fb770ae4 LB |
5175 | struct extent_buffer *eb; |
5176 | ||
74dd17fb | 5177 | BUG_ON(*level == 0); |
2ff7e61e | 5178 | eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]); |
fb770ae4 LB |
5179 | if (IS_ERR(eb)) |
5180 | return PTR_ERR(eb); | |
5181 | ||
5182 | path->nodes[*level - 1] = eb; | |
7069830a AB |
5183 | path->slots[*level - 1] = 0; |
5184 | (*level)--; | |
fb770ae4 | 5185 | return 0; |
7069830a AB |
5186 | } |
5187 | ||
f1e30261 | 5188 | static int tree_move_next_or_upnext(struct btrfs_path *path, |
7069830a AB |
5189 | int *level, int root_level) |
5190 | { | |
5191 | int ret = 0; | |
5192 | int nritems; | |
5193 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5194 | ||
5195 | path->slots[*level]++; | |
5196 | ||
74dd17fb | 5197 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5198 | if (*level == root_level) |
5199 | return -1; | |
5200 | ||
5201 | /* move upnext */ | |
5202 | path->slots[*level] = 0; | |
5203 | free_extent_buffer(path->nodes[*level]); | |
5204 | path->nodes[*level] = NULL; | |
5205 | (*level)++; | |
5206 | path->slots[*level]++; | |
5207 | ||
5208 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5209 | ret = 1; | |
5210 | } | |
5211 | return ret; | |
5212 | } | |
5213 | ||
5214 | /* | |
5215 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5216 | * or down. | |
5217 | */ | |
2ff7e61e | 5218 | static int tree_advance(struct btrfs_fs_info *fs_info, |
7069830a AB |
5219 | struct btrfs_path *path, |
5220 | int *level, int root_level, | |
5221 | int allow_down, | |
5222 | struct btrfs_key *key) | |
5223 | { | |
5224 | int ret; | |
5225 | ||
5226 | if (*level == 0 || !allow_down) { | |
f1e30261 | 5227 | ret = tree_move_next_or_upnext(path, level, root_level); |
7069830a | 5228 | } else { |
ab6a43e1 | 5229 | ret = tree_move_down(fs_info, path, level); |
7069830a AB |
5230 | } |
5231 | if (ret >= 0) { | |
5232 | if (*level == 0) | |
5233 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5234 | path->slots[*level]); | |
5235 | else | |
5236 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5237 | path->slots[*level]); | |
5238 | } | |
5239 | return ret; | |
5240 | } | |
5241 | ||
2ff7e61e | 5242 | static int tree_compare_item(struct btrfs_path *left_path, |
7069830a AB |
5243 | struct btrfs_path *right_path, |
5244 | char *tmp_buf) | |
5245 | { | |
5246 | int cmp; | |
5247 | int len1, len2; | |
5248 | unsigned long off1, off2; | |
5249 | ||
5250 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5251 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5252 | if (len1 != len2) | |
5253 | return 1; | |
5254 | ||
5255 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5256 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5257 | right_path->slots[0]); | |
5258 | ||
5259 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5260 | ||
5261 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5262 | if (cmp) | |
5263 | return 1; | |
5264 | return 0; | |
5265 | } | |
5266 | ||
5267 | #define ADVANCE 1 | |
5268 | #define ADVANCE_ONLY_NEXT -1 | |
5269 | ||
5270 | /* | |
5271 | * This function compares two trees and calls the provided callback for | |
5272 | * every changed/new/deleted item it finds. | |
5273 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5274 | * the compare pretty fast on snapshotted subvolumes. | |
5275 | * | |
5276 | * This currently works on commit roots only. As commit roots are read only, | |
5277 | * we don't do any locking. The commit roots are protected with transactions. | |
5278 | * Transactions are ended and rejoined when a commit is tried in between. | |
5279 | * | |
5280 | * This function checks for modifications done to the trees while comparing. | |
5281 | * If it detects a change, it aborts immediately. | |
5282 | */ | |
5283 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5284 | struct btrfs_root *right_root, | |
5285 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5286 | { | |
0b246afa | 5287 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5288 | int ret; |
5289 | int cmp; | |
7069830a AB |
5290 | struct btrfs_path *left_path = NULL; |
5291 | struct btrfs_path *right_path = NULL; | |
5292 | struct btrfs_key left_key; | |
5293 | struct btrfs_key right_key; | |
5294 | char *tmp_buf = NULL; | |
5295 | int left_root_level; | |
5296 | int right_root_level; | |
5297 | int left_level; | |
5298 | int right_level; | |
5299 | int left_end_reached; | |
5300 | int right_end_reached; | |
5301 | int advance_left; | |
5302 | int advance_right; | |
5303 | u64 left_blockptr; | |
5304 | u64 right_blockptr; | |
6baa4293 FM |
5305 | u64 left_gen; |
5306 | u64 right_gen; | |
7069830a AB |
5307 | |
5308 | left_path = btrfs_alloc_path(); | |
5309 | if (!left_path) { | |
5310 | ret = -ENOMEM; | |
5311 | goto out; | |
5312 | } | |
5313 | right_path = btrfs_alloc_path(); | |
5314 | if (!right_path) { | |
5315 | ret = -ENOMEM; | |
5316 | goto out; | |
5317 | } | |
5318 | ||
752ade68 | 5319 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); |
7069830a | 5320 | if (!tmp_buf) { |
752ade68 MH |
5321 | ret = -ENOMEM; |
5322 | goto out; | |
7069830a AB |
5323 | } |
5324 | ||
5325 | left_path->search_commit_root = 1; | |
5326 | left_path->skip_locking = 1; | |
5327 | right_path->search_commit_root = 1; | |
5328 | right_path->skip_locking = 1; | |
5329 | ||
7069830a AB |
5330 | /* |
5331 | * Strategy: Go to the first items of both trees. Then do | |
5332 | * | |
5333 | * If both trees are at level 0 | |
5334 | * Compare keys of current items | |
5335 | * If left < right treat left item as new, advance left tree | |
5336 | * and repeat | |
5337 | * If left > right treat right item as deleted, advance right tree | |
5338 | * and repeat | |
5339 | * If left == right do deep compare of items, treat as changed if | |
5340 | * needed, advance both trees and repeat | |
5341 | * If both trees are at the same level but not at level 0 | |
5342 | * Compare keys of current nodes/leafs | |
5343 | * If left < right advance left tree and repeat | |
5344 | * If left > right advance right tree and repeat | |
5345 | * If left == right compare blockptrs of the next nodes/leafs | |
5346 | * If they match advance both trees but stay at the same level | |
5347 | * and repeat | |
5348 | * If they don't match advance both trees while allowing to go | |
5349 | * deeper and repeat | |
5350 | * If tree levels are different | |
5351 | * Advance the tree that needs it and repeat | |
5352 | * | |
5353 | * Advancing a tree means: | |
5354 | * If we are at level 0, try to go to the next slot. If that's not | |
5355 | * possible, go one level up and repeat. Stop when we found a level | |
5356 | * where we could go to the next slot. We may at this point be on a | |
5357 | * node or a leaf. | |
5358 | * | |
5359 | * If we are not at level 0 and not on shared tree blocks, go one | |
5360 | * level deeper. | |
5361 | * | |
5362 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5363 | * the right if possible or go up and right. | |
5364 | */ | |
5365 | ||
0b246afa | 5366 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5367 | left_level = btrfs_header_level(left_root->commit_root); |
5368 | left_root_level = left_level; | |
6f2f0b39 RK |
5369 | left_path->nodes[left_level] = |
5370 | btrfs_clone_extent_buffer(left_root->commit_root); | |
5371 | if (!left_path->nodes[left_level]) { | |
5372 | up_read(&fs_info->commit_root_sem); | |
5373 | ret = -ENOMEM; | |
5374 | goto out; | |
5375 | } | |
7069830a AB |
5376 | extent_buffer_get(left_path->nodes[left_level]); |
5377 | ||
5378 | right_level = btrfs_header_level(right_root->commit_root); | |
5379 | right_root_level = right_level; | |
6f2f0b39 RK |
5380 | right_path->nodes[right_level] = |
5381 | btrfs_clone_extent_buffer(right_root->commit_root); | |
5382 | if (!right_path->nodes[right_level]) { | |
5383 | up_read(&fs_info->commit_root_sem); | |
5384 | ret = -ENOMEM; | |
5385 | goto out; | |
5386 | } | |
7069830a | 5387 | extent_buffer_get(right_path->nodes[right_level]); |
0b246afa | 5388 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5389 | |
5390 | if (left_level == 0) | |
5391 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5392 | &left_key, left_path->slots[left_level]); | |
5393 | else | |
5394 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5395 | &left_key, left_path->slots[left_level]); | |
5396 | if (right_level == 0) | |
5397 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5398 | &right_key, right_path->slots[right_level]); | |
5399 | else | |
5400 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5401 | &right_key, right_path->slots[right_level]); | |
5402 | ||
5403 | left_end_reached = right_end_reached = 0; | |
5404 | advance_left = advance_right = 0; | |
5405 | ||
5406 | while (1) { | |
7069830a | 5407 | if (advance_left && !left_end_reached) { |
2ff7e61e | 5408 | ret = tree_advance(fs_info, left_path, &left_level, |
7069830a AB |
5409 | left_root_level, |
5410 | advance_left != ADVANCE_ONLY_NEXT, | |
5411 | &left_key); | |
fb770ae4 | 5412 | if (ret == -1) |
7069830a | 5413 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5414 | else if (ret < 0) |
5415 | goto out; | |
7069830a AB |
5416 | advance_left = 0; |
5417 | } | |
5418 | if (advance_right && !right_end_reached) { | |
2ff7e61e | 5419 | ret = tree_advance(fs_info, right_path, &right_level, |
7069830a AB |
5420 | right_root_level, |
5421 | advance_right != ADVANCE_ONLY_NEXT, | |
5422 | &right_key); | |
fb770ae4 | 5423 | if (ret == -1) |
7069830a | 5424 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5425 | else if (ret < 0) |
5426 | goto out; | |
7069830a AB |
5427 | advance_right = 0; |
5428 | } | |
5429 | ||
5430 | if (left_end_reached && right_end_reached) { | |
5431 | ret = 0; | |
5432 | goto out; | |
5433 | } else if (left_end_reached) { | |
5434 | if (right_level == 0) { | |
ee8c494f | 5435 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5436 | &right_key, |
5437 | BTRFS_COMPARE_TREE_DELETED, | |
5438 | ctx); | |
5439 | if (ret < 0) | |
5440 | goto out; | |
5441 | } | |
5442 | advance_right = ADVANCE; | |
5443 | continue; | |
5444 | } else if (right_end_reached) { | |
5445 | if (left_level == 0) { | |
ee8c494f | 5446 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5447 | &left_key, |
5448 | BTRFS_COMPARE_TREE_NEW, | |
5449 | ctx); | |
5450 | if (ret < 0) | |
5451 | goto out; | |
5452 | } | |
5453 | advance_left = ADVANCE; | |
5454 | continue; | |
5455 | } | |
5456 | ||
5457 | if (left_level == 0 && right_level == 0) { | |
5458 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5459 | if (cmp < 0) { | |
ee8c494f | 5460 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5461 | &left_key, |
5462 | BTRFS_COMPARE_TREE_NEW, | |
5463 | ctx); | |
5464 | if (ret < 0) | |
5465 | goto out; | |
5466 | advance_left = ADVANCE; | |
5467 | } else if (cmp > 0) { | |
ee8c494f | 5468 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5469 | &right_key, |
5470 | BTRFS_COMPARE_TREE_DELETED, | |
5471 | ctx); | |
5472 | if (ret < 0) | |
5473 | goto out; | |
5474 | advance_right = ADVANCE; | |
5475 | } else { | |
b99d9a6a | 5476 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5477 | |
74dd17fb | 5478 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
2ff7e61e JM |
5479 | ret = tree_compare_item(left_path, right_path, |
5480 | tmp_buf); | |
ba5e8f2e | 5481 | if (ret) |
b99d9a6a | 5482 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5483 | else |
b99d9a6a | 5484 | result = BTRFS_COMPARE_TREE_SAME; |
ee8c494f | 5485 | ret = changed_cb(left_path, right_path, |
b99d9a6a | 5486 | &left_key, result, ctx); |
ba5e8f2e JB |
5487 | if (ret < 0) |
5488 | goto out; | |
7069830a AB |
5489 | advance_left = ADVANCE; |
5490 | advance_right = ADVANCE; | |
5491 | } | |
5492 | } else if (left_level == right_level) { | |
5493 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5494 | if (cmp < 0) { | |
5495 | advance_left = ADVANCE; | |
5496 | } else if (cmp > 0) { | |
5497 | advance_right = ADVANCE; | |
5498 | } else { | |
5499 | left_blockptr = btrfs_node_blockptr( | |
5500 | left_path->nodes[left_level], | |
5501 | left_path->slots[left_level]); | |
5502 | right_blockptr = btrfs_node_blockptr( | |
5503 | right_path->nodes[right_level], | |
5504 | right_path->slots[right_level]); | |
6baa4293 FM |
5505 | left_gen = btrfs_node_ptr_generation( |
5506 | left_path->nodes[left_level], | |
5507 | left_path->slots[left_level]); | |
5508 | right_gen = btrfs_node_ptr_generation( | |
5509 | right_path->nodes[right_level], | |
5510 | right_path->slots[right_level]); | |
5511 | if (left_blockptr == right_blockptr && | |
5512 | left_gen == right_gen) { | |
7069830a AB |
5513 | /* |
5514 | * As we're on a shared block, don't | |
5515 | * allow to go deeper. | |
5516 | */ | |
5517 | advance_left = ADVANCE_ONLY_NEXT; | |
5518 | advance_right = ADVANCE_ONLY_NEXT; | |
5519 | } else { | |
5520 | advance_left = ADVANCE; | |
5521 | advance_right = ADVANCE; | |
5522 | } | |
5523 | } | |
5524 | } else if (left_level < right_level) { | |
5525 | advance_right = ADVANCE; | |
5526 | } else { | |
5527 | advance_left = ADVANCE; | |
5528 | } | |
5529 | } | |
5530 | ||
5531 | out: | |
5532 | btrfs_free_path(left_path); | |
5533 | btrfs_free_path(right_path); | |
8f282f71 | 5534 | kvfree(tmp_buf); |
7069830a AB |
5535 | return ret; |
5536 | } | |
5537 | ||
3f157a2f CM |
5538 | /* |
5539 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5540 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5541 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5542 | * |
5543 | * 0 is returned if another key is found, < 0 if there are any errors | |
5544 | * and 1 is returned if there are no higher keys in the tree | |
5545 | * | |
5546 | * path->keep_locks should be set to 1 on the search made before | |
5547 | * calling this function. | |
5548 | */ | |
e7a84565 | 5549 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5550 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5551 | { |
e7a84565 CM |
5552 | int slot; |
5553 | struct extent_buffer *c; | |
5554 | ||
934d375b | 5555 | WARN_ON(!path->keep_locks); |
d397712b | 5556 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5557 | if (!path->nodes[level]) |
5558 | return 1; | |
5559 | ||
5560 | slot = path->slots[level] + 1; | |
5561 | c = path->nodes[level]; | |
3f157a2f | 5562 | next: |
e7a84565 | 5563 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5564 | int ret; |
5565 | int orig_lowest; | |
5566 | struct btrfs_key cur_key; | |
5567 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5568 | !path->nodes[level + 1]) | |
e7a84565 | 5569 | return 1; |
33c66f43 YZ |
5570 | |
5571 | if (path->locks[level + 1]) { | |
5572 | level++; | |
5573 | continue; | |
5574 | } | |
5575 | ||
5576 | slot = btrfs_header_nritems(c) - 1; | |
5577 | if (level == 0) | |
5578 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5579 | else | |
5580 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5581 | ||
5582 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5583 | btrfs_release_path(path); |
33c66f43 YZ |
5584 | path->lowest_level = level; |
5585 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5586 | 0, 0); | |
5587 | path->lowest_level = orig_lowest; | |
5588 | if (ret < 0) | |
5589 | return ret; | |
5590 | ||
5591 | c = path->nodes[level]; | |
5592 | slot = path->slots[level]; | |
5593 | if (ret == 0) | |
5594 | slot++; | |
5595 | goto next; | |
e7a84565 | 5596 | } |
33c66f43 | 5597 | |
e7a84565 CM |
5598 | if (level == 0) |
5599 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5600 | else { |
3f157a2f CM |
5601 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5602 | ||
3f157a2f CM |
5603 | if (gen < min_trans) { |
5604 | slot++; | |
5605 | goto next; | |
5606 | } | |
e7a84565 | 5607 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5608 | } |
e7a84565 CM |
5609 | return 0; |
5610 | } | |
5611 | return 1; | |
5612 | } | |
5613 | ||
97571fd0 | 5614 | /* |
925baedd | 5615 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5616 | * returns 0 if it found something or 1 if there are no greater leaves. |
5617 | * returns < 0 on io errors. | |
97571fd0 | 5618 | */ |
234b63a0 | 5619 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5620 | { |
5621 | return btrfs_next_old_leaf(root, path, 0); | |
5622 | } | |
5623 | ||
5624 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5625 | u64 time_seq) | |
d97e63b6 CM |
5626 | { |
5627 | int slot; | |
8e73f275 | 5628 | int level; |
5f39d397 | 5629 | struct extent_buffer *c; |
8e73f275 | 5630 | struct extent_buffer *next; |
925baedd CM |
5631 | struct btrfs_key key; |
5632 | u32 nritems; | |
5633 | int ret; | |
8e73f275 | 5634 | int old_spinning = path->leave_spinning; |
bd681513 | 5635 | int next_rw_lock = 0; |
925baedd CM |
5636 | |
5637 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5638 | if (nritems == 0) |
925baedd | 5639 | return 1; |
925baedd | 5640 | |
8e73f275 CM |
5641 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5642 | again: | |
5643 | level = 1; | |
5644 | next = NULL; | |
bd681513 | 5645 | next_rw_lock = 0; |
b3b4aa74 | 5646 | btrfs_release_path(path); |
8e73f275 | 5647 | |
a2135011 | 5648 | path->keep_locks = 1; |
31533fb2 | 5649 | path->leave_spinning = 1; |
8e73f275 | 5650 | |
3d7806ec JS |
5651 | if (time_seq) |
5652 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5653 | else | |
5654 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5655 | path->keep_locks = 0; |
5656 | ||
5657 | if (ret < 0) | |
5658 | return ret; | |
5659 | ||
a2135011 | 5660 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5661 | /* |
5662 | * by releasing the path above we dropped all our locks. A balance | |
5663 | * could have added more items next to the key that used to be | |
5664 | * at the very end of the block. So, check again here and | |
5665 | * advance the path if there are now more items available. | |
5666 | */ | |
a2135011 | 5667 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5668 | if (ret == 0) |
5669 | path->slots[0]++; | |
8e73f275 | 5670 | ret = 0; |
925baedd CM |
5671 | goto done; |
5672 | } | |
0b43e04f LB |
5673 | /* |
5674 | * So the above check misses one case: | |
5675 | * - after releasing the path above, someone has removed the item that | |
5676 | * used to be at the very end of the block, and balance between leafs | |
5677 | * gets another one with bigger key.offset to replace it. | |
5678 | * | |
5679 | * This one should be returned as well, or we can get leaf corruption | |
5680 | * later(esp. in __btrfs_drop_extents()). | |
5681 | * | |
5682 | * And a bit more explanation about this check, | |
5683 | * with ret > 0, the key isn't found, the path points to the slot | |
5684 | * where it should be inserted, so the path->slots[0] item must be the | |
5685 | * bigger one. | |
5686 | */ | |
5687 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5688 | ret = 0; | |
5689 | goto done; | |
5690 | } | |
d97e63b6 | 5691 | |
d397712b | 5692 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5693 | if (!path->nodes[level]) { |
5694 | ret = 1; | |
5695 | goto done; | |
5696 | } | |
5f39d397 | 5697 | |
d97e63b6 CM |
5698 | slot = path->slots[level] + 1; |
5699 | c = path->nodes[level]; | |
5f39d397 | 5700 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5701 | level++; |
8e73f275 CM |
5702 | if (level == BTRFS_MAX_LEVEL) { |
5703 | ret = 1; | |
5704 | goto done; | |
5705 | } | |
d97e63b6 CM |
5706 | continue; |
5707 | } | |
5f39d397 | 5708 | |
925baedd | 5709 | if (next) { |
bd681513 | 5710 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5711 | free_extent_buffer(next); |
925baedd | 5712 | } |
5f39d397 | 5713 | |
8e73f275 | 5714 | next = c; |
bd681513 | 5715 | next_rw_lock = path->locks[level]; |
d07b8528 | 5716 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5717 | slot, &key); |
8e73f275 CM |
5718 | if (ret == -EAGAIN) |
5719 | goto again; | |
5f39d397 | 5720 | |
76a05b35 | 5721 | if (ret < 0) { |
b3b4aa74 | 5722 | btrfs_release_path(path); |
76a05b35 CM |
5723 | goto done; |
5724 | } | |
5725 | ||
5cd57b2c | 5726 | if (!path->skip_locking) { |
bd681513 | 5727 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5728 | if (!ret && time_seq) { |
5729 | /* | |
5730 | * If we don't get the lock, we may be racing | |
5731 | * with push_leaf_left, holding that lock while | |
5732 | * itself waiting for the leaf we've currently | |
5733 | * locked. To solve this situation, we give up | |
5734 | * on our lock and cycle. | |
5735 | */ | |
cf538830 | 5736 | free_extent_buffer(next); |
d42244a0 JS |
5737 | btrfs_release_path(path); |
5738 | cond_resched(); | |
5739 | goto again; | |
5740 | } | |
8e73f275 CM |
5741 | if (!ret) { |
5742 | btrfs_set_path_blocking(path); | |
bd681513 | 5743 | btrfs_tree_read_lock(next); |
8e73f275 | 5744 | } |
31533fb2 | 5745 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5746 | } |
d97e63b6 CM |
5747 | break; |
5748 | } | |
5749 | path->slots[level] = slot; | |
d397712b | 5750 | while (1) { |
d97e63b6 CM |
5751 | level--; |
5752 | c = path->nodes[level]; | |
925baedd | 5753 | if (path->locks[level]) |
bd681513 | 5754 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5755 | |
5f39d397 | 5756 | free_extent_buffer(c); |
d97e63b6 CM |
5757 | path->nodes[level] = next; |
5758 | path->slots[level] = 0; | |
a74a4b97 | 5759 | if (!path->skip_locking) |
bd681513 | 5760 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5761 | if (!level) |
5762 | break; | |
b4ce94de | 5763 | |
d07b8528 | 5764 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5765 | 0, &key); |
8e73f275 CM |
5766 | if (ret == -EAGAIN) |
5767 | goto again; | |
5768 | ||
76a05b35 | 5769 | if (ret < 0) { |
b3b4aa74 | 5770 | btrfs_release_path(path); |
76a05b35 CM |
5771 | goto done; |
5772 | } | |
5773 | ||
5cd57b2c | 5774 | if (!path->skip_locking) { |
bd681513 | 5775 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5776 | if (!ret) { |
5777 | btrfs_set_path_blocking(path); | |
bd681513 | 5778 | btrfs_tree_read_lock(next); |
bd681513 | 5779 | } |
31533fb2 | 5780 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5781 | } |
d97e63b6 | 5782 | } |
8e73f275 | 5783 | ret = 0; |
925baedd | 5784 | done: |
f7c79f30 | 5785 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5786 | path->leave_spinning = old_spinning; |
5787 | if (!old_spinning) | |
5788 | btrfs_set_path_blocking(path); | |
5789 | ||
5790 | return ret; | |
d97e63b6 | 5791 | } |
0b86a832 | 5792 | |
3f157a2f CM |
5793 | /* |
5794 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5795 | * searching until it gets past min_objectid or finds an item of 'type' | |
5796 | * | |
5797 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5798 | */ | |
0b86a832 CM |
5799 | int btrfs_previous_item(struct btrfs_root *root, |
5800 | struct btrfs_path *path, u64 min_objectid, | |
5801 | int type) | |
5802 | { | |
5803 | struct btrfs_key found_key; | |
5804 | struct extent_buffer *leaf; | |
e02119d5 | 5805 | u32 nritems; |
0b86a832 CM |
5806 | int ret; |
5807 | ||
d397712b | 5808 | while (1) { |
0b86a832 | 5809 | if (path->slots[0] == 0) { |
b4ce94de | 5810 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5811 | ret = btrfs_prev_leaf(root, path); |
5812 | if (ret != 0) | |
5813 | return ret; | |
5814 | } else { | |
5815 | path->slots[0]--; | |
5816 | } | |
5817 | leaf = path->nodes[0]; | |
e02119d5 CM |
5818 | nritems = btrfs_header_nritems(leaf); |
5819 | if (nritems == 0) | |
5820 | return 1; | |
5821 | if (path->slots[0] == nritems) | |
5822 | path->slots[0]--; | |
5823 | ||
0b86a832 | 5824 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5825 | if (found_key.objectid < min_objectid) |
5826 | break; | |
0a4eefbb YZ |
5827 | if (found_key.type == type) |
5828 | return 0; | |
e02119d5 CM |
5829 | if (found_key.objectid == min_objectid && |
5830 | found_key.type < type) | |
5831 | break; | |
0b86a832 CM |
5832 | } |
5833 | return 1; | |
5834 | } | |
ade2e0b3 WS |
5835 | |
5836 | /* | |
5837 | * search in extent tree to find a previous Metadata/Data extent item with | |
5838 | * min objecitd. | |
5839 | * | |
5840 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5841 | */ | |
5842 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5843 | struct btrfs_path *path, u64 min_objectid) | |
5844 | { | |
5845 | struct btrfs_key found_key; | |
5846 | struct extent_buffer *leaf; | |
5847 | u32 nritems; | |
5848 | int ret; | |
5849 | ||
5850 | while (1) { | |
5851 | if (path->slots[0] == 0) { | |
5852 | btrfs_set_path_blocking(path); | |
5853 | ret = btrfs_prev_leaf(root, path); | |
5854 | if (ret != 0) | |
5855 | return ret; | |
5856 | } else { | |
5857 | path->slots[0]--; | |
5858 | } | |
5859 | leaf = path->nodes[0]; | |
5860 | nritems = btrfs_header_nritems(leaf); | |
5861 | if (nritems == 0) | |
5862 | return 1; | |
5863 | if (path->slots[0] == nritems) | |
5864 | path->slots[0]--; | |
5865 | ||
5866 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5867 | if (found_key.objectid < min_objectid) | |
5868 | break; | |
5869 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5870 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5871 | return 0; | |
5872 | if (found_key.objectid == min_objectid && | |
5873 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5874 | break; | |
5875 | } | |
5876 | return 1; | |
5877 | } |