]>
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" |
de37aa51 | 15 | #include "volumes.h" |
f616f5cd | 16 | #include "qgroup.h" |
9a8dd150 | 17 | |
e089f05c CM |
18 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
19 | *root, struct btrfs_path *path, int level); | |
310712b2 OS |
20 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
21 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
22 | int data_size, int extend); | |
5f39d397 | 23 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
24 | struct btrfs_fs_info *fs_info, |
25 | struct extent_buffer *dst, | |
971a1f66 | 26 | struct extent_buffer *src, int empty); |
5f39d397 | 27 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 28 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
29 | struct extent_buffer *dst_buf, |
30 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
31 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
32 | int level, int slot); | |
d97e63b6 | 33 | |
df24a2b9 | 34 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 35 | { |
e2c89907 | 36 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
37 | } |
38 | ||
b4ce94de CM |
39 | /* |
40 | * set all locked nodes in the path to blocking locks. This should | |
41 | * be done before scheduling | |
42 | */ | |
43 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
44 | { | |
45 | int i; | |
46 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
47 | if (!p->nodes[i] || !p->locks[i]) |
48 | continue; | |
766ece54 DS |
49 | /* |
50 | * If we currently have a spinning reader or writer lock this | |
51 | * will bump the count of blocking holders and drop the | |
52 | * spinlock. | |
53 | */ | |
54 | if (p->locks[i] == BTRFS_READ_LOCK) { | |
55 | btrfs_set_lock_blocking_read(p->nodes[i]); | |
bd681513 | 56 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; |
766ece54 DS |
57 | } else if (p->locks[i] == BTRFS_WRITE_LOCK) { |
58 | btrfs_set_lock_blocking_write(p->nodes[i]); | |
bd681513 | 59 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; |
766ece54 | 60 | } |
b4ce94de CM |
61 | } |
62 | } | |
63 | ||
d352ac68 | 64 | /* this also releases the path */ |
df24a2b9 | 65 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 66 | { |
ff175d57 JJ |
67 | if (!p) |
68 | return; | |
b3b4aa74 | 69 | btrfs_release_path(p); |
df24a2b9 | 70 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
71 | } |
72 | ||
d352ac68 CM |
73 | /* |
74 | * path release drops references on the extent buffers in the path | |
75 | * and it drops any locks held by this path | |
76 | * | |
77 | * It is safe to call this on paths that no locks or extent buffers held. | |
78 | */ | |
b3b4aa74 | 79 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
80 | { |
81 | int i; | |
a2135011 | 82 | |
234b63a0 | 83 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 84 | p->slots[i] = 0; |
eb60ceac | 85 | if (!p->nodes[i]) |
925baedd CM |
86 | continue; |
87 | if (p->locks[i]) { | |
bd681513 | 88 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
89 | p->locks[i] = 0; |
90 | } | |
5f39d397 | 91 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 92 | p->nodes[i] = NULL; |
eb60ceac CM |
93 | } |
94 | } | |
95 | ||
d352ac68 CM |
96 | /* |
97 | * safely gets a reference on the root node of a tree. A lock | |
98 | * is not taken, so a concurrent writer may put a different node | |
99 | * at the root of the tree. See btrfs_lock_root_node for the | |
100 | * looping required. | |
101 | * | |
102 | * The extent buffer returned by this has a reference taken, so | |
103 | * it won't disappear. It may stop being the root of the tree | |
104 | * at any time because there are no locks held. | |
105 | */ | |
925baedd CM |
106 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
107 | { | |
108 | struct extent_buffer *eb; | |
240f62c8 | 109 | |
3083ee2e JB |
110 | while (1) { |
111 | rcu_read_lock(); | |
112 | eb = rcu_dereference(root->node); | |
113 | ||
114 | /* | |
115 | * RCU really hurts here, we could free up the root node because | |
01327610 | 116 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
117 | * the inc_not_zero dance and if it doesn't work then |
118 | * synchronize_rcu and try again. | |
119 | */ | |
120 | if (atomic_inc_not_zero(&eb->refs)) { | |
121 | rcu_read_unlock(); | |
122 | break; | |
123 | } | |
124 | rcu_read_unlock(); | |
125 | synchronize_rcu(); | |
126 | } | |
925baedd CM |
127 | return eb; |
128 | } | |
129 | ||
d352ac68 CM |
130 | /* loop around taking references on and locking the root node of the |
131 | * tree until you end up with a lock on the root. A locked buffer | |
132 | * is returned, with a reference held. | |
133 | */ | |
925baedd CM |
134 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
135 | { | |
136 | struct extent_buffer *eb; | |
137 | ||
d397712b | 138 | while (1) { |
925baedd CM |
139 | eb = btrfs_root_node(root); |
140 | btrfs_tree_lock(eb); | |
240f62c8 | 141 | if (eb == root->node) |
925baedd | 142 | break; |
925baedd CM |
143 | btrfs_tree_unlock(eb); |
144 | free_extent_buffer(eb); | |
145 | } | |
146 | return eb; | |
147 | } | |
148 | ||
bd681513 CM |
149 | /* loop around taking references on and locking the root node of the |
150 | * tree until you end up with a lock on the root. A locked buffer | |
151 | * is returned, with a reference held. | |
152 | */ | |
84f7d8e6 | 153 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
154 | { |
155 | struct extent_buffer *eb; | |
156 | ||
157 | while (1) { | |
158 | eb = btrfs_root_node(root); | |
159 | btrfs_tree_read_lock(eb); | |
160 | if (eb == root->node) | |
161 | break; | |
162 | btrfs_tree_read_unlock(eb); | |
163 | free_extent_buffer(eb); | |
164 | } | |
165 | return eb; | |
166 | } | |
167 | ||
d352ac68 CM |
168 | /* cowonly root (everything not a reference counted cow subvolume), just get |
169 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
170 | * get properly updated on disk. | |
171 | */ | |
0b86a832 CM |
172 | static void add_root_to_dirty_list(struct btrfs_root *root) |
173 | { | |
0b246afa JM |
174 | struct btrfs_fs_info *fs_info = root->fs_info; |
175 | ||
e7070be1 JB |
176 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
177 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
178 | return; | |
179 | ||
0b246afa | 180 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
181 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
182 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 183 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 184 | list_move_tail(&root->dirty_list, |
0b246afa | 185 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
186 | else |
187 | list_move(&root->dirty_list, | |
0b246afa | 188 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 189 | } |
0b246afa | 190 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
191 | } |
192 | ||
d352ac68 CM |
193 | /* |
194 | * used by snapshot creation to make a copy of a root for a tree with | |
195 | * a given objectid. The buffer with the new root node is returned in | |
196 | * cow_ret, and this func returns zero on success or a negative error code. | |
197 | */ | |
be20aa9d CM |
198 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
199 | struct btrfs_root *root, | |
200 | struct extent_buffer *buf, | |
201 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
202 | { | |
0b246afa | 203 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 204 | struct extent_buffer *cow; |
be20aa9d CM |
205 | int ret = 0; |
206 | int level; | |
5d4f98a2 | 207 | struct btrfs_disk_key disk_key; |
be20aa9d | 208 | |
27cdeb70 | 209 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 210 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
211 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
212 | trans->transid != root->last_trans); | |
be20aa9d CM |
213 | |
214 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
215 | if (level == 0) |
216 | btrfs_item_key(buf, &disk_key, 0); | |
217 | else | |
218 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 219 | |
4d75f8a9 DS |
220 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
221 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 222 | if (IS_ERR(cow)) |
be20aa9d CM |
223 | return PTR_ERR(cow); |
224 | ||
58e8012c | 225 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
226 | btrfs_set_header_bytenr(cow, cow->start); |
227 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
228 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
229 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
230 | BTRFS_HEADER_FLAG_RELOC); | |
231 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
232 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
233 | else | |
234 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 235 | |
de37aa51 | 236 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 237 | |
be20aa9d | 238 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 239 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 240 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 241 | else |
e339a6b0 | 242 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 243 | |
be20aa9d CM |
244 | if (ret) |
245 | return ret; | |
246 | ||
247 | btrfs_mark_buffer_dirty(cow); | |
248 | *cow_ret = cow; | |
249 | return 0; | |
250 | } | |
251 | ||
bd989ba3 JS |
252 | enum mod_log_op { |
253 | MOD_LOG_KEY_REPLACE, | |
254 | MOD_LOG_KEY_ADD, | |
255 | MOD_LOG_KEY_REMOVE, | |
256 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
257 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
258 | MOD_LOG_MOVE_KEYS, | |
259 | MOD_LOG_ROOT_REPLACE, | |
260 | }; | |
261 | ||
bd989ba3 JS |
262 | struct tree_mod_root { |
263 | u64 logical; | |
264 | u8 level; | |
265 | }; | |
266 | ||
267 | struct tree_mod_elem { | |
268 | struct rb_node node; | |
298cfd36 | 269 | u64 logical; |
097b8a7c | 270 | u64 seq; |
bd989ba3 JS |
271 | enum mod_log_op op; |
272 | ||
273 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
274 | int slot; | |
275 | ||
276 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
277 | u64 generation; | |
278 | ||
279 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
280 | struct btrfs_disk_key key; | |
281 | u64 blockptr; | |
282 | ||
283 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
284 | struct { |
285 | int dst_slot; | |
286 | int nr_items; | |
287 | } move; | |
bd989ba3 JS |
288 | |
289 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
290 | struct tree_mod_root old_root; | |
291 | }; | |
292 | ||
fc36ed7e | 293 | /* |
fcebe456 | 294 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 295 | */ |
fcebe456 | 296 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
297 | { |
298 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
299 | } | |
300 | ||
097b8a7c JS |
301 | /* |
302 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
303 | * passed does not already have a sequence number set. So when a caller expects | |
304 | * to record tree modifications, it should ensure to set elem->seq to zero | |
305 | * before calling btrfs_get_tree_mod_seq. | |
306 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
307 | * blocker was added. | |
308 | */ | |
309 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
310 | struct seq_list *elem) | |
bd989ba3 | 311 | { |
b1a09f1e | 312 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 313 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 314 | if (!elem->seq) { |
fcebe456 | 315 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
316 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
317 | } | |
bd989ba3 | 318 | spin_unlock(&fs_info->tree_mod_seq_lock); |
b1a09f1e | 319 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 320 | |
fcebe456 | 321 | return elem->seq; |
bd989ba3 JS |
322 | } |
323 | ||
324 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
325 | struct seq_list *elem) | |
326 | { | |
327 | struct rb_root *tm_root; | |
328 | struct rb_node *node; | |
329 | struct rb_node *next; | |
330 | struct seq_list *cur_elem; | |
331 | struct tree_mod_elem *tm; | |
332 | u64 min_seq = (u64)-1; | |
333 | u64 seq_putting = elem->seq; | |
334 | ||
335 | if (!seq_putting) | |
336 | return; | |
337 | ||
bd989ba3 JS |
338 | spin_lock(&fs_info->tree_mod_seq_lock); |
339 | list_del(&elem->list); | |
097b8a7c | 340 | elem->seq = 0; |
bd989ba3 JS |
341 | |
342 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 343 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
344 | if (seq_putting > cur_elem->seq) { |
345 | /* | |
346 | * blocker with lower sequence number exists, we | |
347 | * cannot remove anything from the log | |
348 | */ | |
097b8a7c JS |
349 | spin_unlock(&fs_info->tree_mod_seq_lock); |
350 | return; | |
bd989ba3 JS |
351 | } |
352 | min_seq = cur_elem->seq; | |
353 | } | |
354 | } | |
097b8a7c JS |
355 | spin_unlock(&fs_info->tree_mod_seq_lock); |
356 | ||
bd989ba3 JS |
357 | /* |
358 | * anything that's lower than the lowest existing (read: blocked) | |
359 | * sequence number can be removed from the tree. | |
360 | */ | |
b1a09f1e | 361 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
362 | tm_root = &fs_info->tree_mod_log; |
363 | for (node = rb_first(tm_root); node; node = next) { | |
364 | next = rb_next(node); | |
6b4df8b6 | 365 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 366 | if (tm->seq > min_seq) |
bd989ba3 JS |
367 | continue; |
368 | rb_erase(node, tm_root); | |
bd989ba3 JS |
369 | kfree(tm); |
370 | } | |
b1a09f1e | 371 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
372 | } |
373 | ||
374 | /* | |
375 | * key order of the log: | |
298cfd36 | 376 | * node/leaf start address -> sequence |
bd989ba3 | 377 | * |
298cfd36 CR |
378 | * The 'start address' is the logical address of the *new* root node |
379 | * for root replace operations, or the logical address of the affected | |
380 | * block for all other operations. | |
5de865ee | 381 | * |
b1a09f1e | 382 | * Note: must be called with write lock for fs_info::tree_mod_log_lock. |
bd989ba3 JS |
383 | */ |
384 | static noinline int | |
385 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
386 | { | |
387 | struct rb_root *tm_root; | |
388 | struct rb_node **new; | |
389 | struct rb_node *parent = NULL; | |
390 | struct tree_mod_elem *cur; | |
c8cc6341 | 391 | |
fcebe456 | 392 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 393 | |
bd989ba3 JS |
394 | tm_root = &fs_info->tree_mod_log; |
395 | new = &tm_root->rb_node; | |
396 | while (*new) { | |
6b4df8b6 | 397 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 398 | parent = *new; |
298cfd36 | 399 | if (cur->logical < tm->logical) |
bd989ba3 | 400 | new = &((*new)->rb_left); |
298cfd36 | 401 | else if (cur->logical > tm->logical) |
bd989ba3 | 402 | new = &((*new)->rb_right); |
097b8a7c | 403 | else if (cur->seq < tm->seq) |
bd989ba3 | 404 | new = &((*new)->rb_left); |
097b8a7c | 405 | else if (cur->seq > tm->seq) |
bd989ba3 | 406 | new = &((*new)->rb_right); |
5de865ee FDBM |
407 | else |
408 | return -EEXIST; | |
bd989ba3 JS |
409 | } |
410 | ||
411 | rb_link_node(&tm->node, parent, new); | |
412 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 413 | return 0; |
bd989ba3 JS |
414 | } |
415 | ||
097b8a7c JS |
416 | /* |
417 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
418 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
419 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 420 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 421 | */ |
e9b7fd4d JS |
422 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
423 | struct extent_buffer *eb) { | |
424 | smp_mb(); | |
425 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
426 | return 1; | |
097b8a7c JS |
427 | if (eb && btrfs_header_level(eb) == 0) |
428 | return 1; | |
5de865ee | 429 | |
b1a09f1e | 430 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 431 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 432 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
433 | return 1; |
434 | } | |
435 | ||
e9b7fd4d JS |
436 | return 0; |
437 | } | |
438 | ||
5de865ee FDBM |
439 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
440 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
441 | struct extent_buffer *eb) | |
442 | { | |
443 | smp_mb(); | |
444 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
445 | return 0; | |
446 | if (eb && btrfs_header_level(eb) == 0) | |
447 | return 0; | |
448 | ||
449 | return 1; | |
450 | } | |
451 | ||
452 | static struct tree_mod_elem * | |
453 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
454 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 455 | { |
097b8a7c | 456 | struct tree_mod_elem *tm; |
bd989ba3 | 457 | |
c8cc6341 JB |
458 | tm = kzalloc(sizeof(*tm), flags); |
459 | if (!tm) | |
5de865ee | 460 | return NULL; |
bd989ba3 | 461 | |
298cfd36 | 462 | tm->logical = eb->start; |
bd989ba3 JS |
463 | if (op != MOD_LOG_KEY_ADD) { |
464 | btrfs_node_key(eb, &tm->key, slot); | |
465 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
466 | } | |
467 | tm->op = op; | |
468 | tm->slot = slot; | |
469 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 470 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 471 | |
5de865ee | 472 | return tm; |
097b8a7c JS |
473 | } |
474 | ||
e09c2efe DS |
475 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
476 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 477 | { |
5de865ee FDBM |
478 | struct tree_mod_elem *tm; |
479 | int ret; | |
480 | ||
e09c2efe | 481 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
482 | return 0; |
483 | ||
484 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
485 | if (!tm) | |
486 | return -ENOMEM; | |
487 | ||
e09c2efe | 488 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 489 | kfree(tm); |
097b8a7c | 490 | return 0; |
5de865ee FDBM |
491 | } |
492 | ||
e09c2efe | 493 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 494 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
495 | if (ret) |
496 | kfree(tm); | |
097b8a7c | 497 | |
5de865ee | 498 | return ret; |
097b8a7c JS |
499 | } |
500 | ||
6074d45f DS |
501 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
502 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 503 | { |
5de865ee FDBM |
504 | struct tree_mod_elem *tm = NULL; |
505 | struct tree_mod_elem **tm_list = NULL; | |
506 | int ret = 0; | |
bd989ba3 | 507 | int i; |
5de865ee | 508 | int locked = 0; |
bd989ba3 | 509 | |
6074d45f | 510 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 511 | return 0; |
bd989ba3 | 512 | |
176ef8f5 | 513 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
514 | if (!tm_list) |
515 | return -ENOMEM; | |
516 | ||
176ef8f5 | 517 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
518 | if (!tm) { |
519 | ret = -ENOMEM; | |
520 | goto free_tms; | |
521 | } | |
522 | ||
298cfd36 | 523 | tm->logical = eb->start; |
5de865ee FDBM |
524 | tm->slot = src_slot; |
525 | tm->move.dst_slot = dst_slot; | |
526 | tm->move.nr_items = nr_items; | |
527 | tm->op = MOD_LOG_MOVE_KEYS; | |
528 | ||
529 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
530 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 531 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
532 | if (!tm_list[i]) { |
533 | ret = -ENOMEM; | |
534 | goto free_tms; | |
535 | } | |
536 | } | |
537 | ||
6074d45f | 538 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
539 | goto free_tms; |
540 | locked = 1; | |
541 | ||
01763a2e JS |
542 | /* |
543 | * When we override something during the move, we log these removals. | |
544 | * This can only happen when we move towards the beginning of the | |
545 | * buffer, i.e. dst_slot < src_slot. | |
546 | */ | |
bd989ba3 | 547 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 548 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
549 | if (ret) |
550 | goto free_tms; | |
bd989ba3 JS |
551 | } |
552 | ||
6074d45f | 553 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
554 | if (ret) |
555 | goto free_tms; | |
b1a09f1e | 556 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 557 | kfree(tm_list); |
f395694c | 558 | |
5de865ee FDBM |
559 | return 0; |
560 | free_tms: | |
561 | for (i = 0; i < nr_items; i++) { | |
562 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 563 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
564 | kfree(tm_list[i]); |
565 | } | |
566 | if (locked) | |
b1a09f1e | 567 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
568 | kfree(tm_list); |
569 | kfree(tm); | |
bd989ba3 | 570 | |
5de865ee | 571 | return ret; |
bd989ba3 JS |
572 | } |
573 | ||
5de865ee FDBM |
574 | static inline int |
575 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
576 | struct tree_mod_elem **tm_list, | |
577 | int nritems) | |
097b8a7c | 578 | { |
5de865ee | 579 | int i, j; |
097b8a7c JS |
580 | int ret; |
581 | ||
097b8a7c | 582 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
583 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
584 | if (ret) { | |
585 | for (j = nritems - 1; j > i; j--) | |
586 | rb_erase(&tm_list[j]->node, | |
587 | &fs_info->tree_mod_log); | |
588 | return ret; | |
589 | } | |
097b8a7c | 590 | } |
5de865ee FDBM |
591 | |
592 | return 0; | |
097b8a7c JS |
593 | } |
594 | ||
95b757c1 DS |
595 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
596 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 597 | { |
95b757c1 | 598 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
599 | struct tree_mod_elem *tm = NULL; |
600 | struct tree_mod_elem **tm_list = NULL; | |
601 | int nritems = 0; | |
602 | int ret = 0; | |
603 | int i; | |
bd989ba3 | 604 | |
5de865ee | 605 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
606 | return 0; |
607 | ||
5de865ee FDBM |
608 | if (log_removal && btrfs_header_level(old_root) > 0) { |
609 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 610 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 611 | GFP_NOFS); |
5de865ee FDBM |
612 | if (!tm_list) { |
613 | ret = -ENOMEM; | |
614 | goto free_tms; | |
615 | } | |
616 | for (i = 0; i < nritems; i++) { | |
617 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 618 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
619 | if (!tm_list[i]) { |
620 | ret = -ENOMEM; | |
621 | goto free_tms; | |
622 | } | |
623 | } | |
624 | } | |
d9abbf1c | 625 | |
bcc8e07f | 626 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
627 | if (!tm) { |
628 | ret = -ENOMEM; | |
629 | goto free_tms; | |
630 | } | |
bd989ba3 | 631 | |
298cfd36 | 632 | tm->logical = new_root->start; |
bd989ba3 JS |
633 | tm->old_root.logical = old_root->start; |
634 | tm->old_root.level = btrfs_header_level(old_root); | |
635 | tm->generation = btrfs_header_generation(old_root); | |
636 | tm->op = MOD_LOG_ROOT_REPLACE; | |
637 | ||
5de865ee FDBM |
638 | if (tree_mod_dont_log(fs_info, NULL)) |
639 | goto free_tms; | |
640 | ||
641 | if (tm_list) | |
642 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
643 | if (!ret) | |
644 | ret = __tree_mod_log_insert(fs_info, tm); | |
645 | ||
b1a09f1e | 646 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
647 | if (ret) |
648 | goto free_tms; | |
649 | kfree(tm_list); | |
650 | ||
651 | return ret; | |
652 | ||
653 | free_tms: | |
654 | if (tm_list) { | |
655 | for (i = 0; i < nritems; i++) | |
656 | kfree(tm_list[i]); | |
657 | kfree(tm_list); | |
658 | } | |
659 | kfree(tm); | |
660 | ||
661 | return ret; | |
bd989ba3 JS |
662 | } |
663 | ||
664 | static struct tree_mod_elem * | |
665 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
666 | int smallest) | |
667 | { | |
668 | struct rb_root *tm_root; | |
669 | struct rb_node *node; | |
670 | struct tree_mod_elem *cur = NULL; | |
671 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 672 | |
b1a09f1e | 673 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
674 | tm_root = &fs_info->tree_mod_log; |
675 | node = tm_root->rb_node; | |
676 | while (node) { | |
6b4df8b6 | 677 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 678 | if (cur->logical < start) { |
bd989ba3 | 679 | node = node->rb_left; |
298cfd36 | 680 | } else if (cur->logical > start) { |
bd989ba3 | 681 | node = node->rb_right; |
097b8a7c | 682 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
683 | node = node->rb_left; |
684 | } else if (!smallest) { | |
685 | /* we want the node with the highest seq */ | |
686 | if (found) | |
097b8a7c | 687 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
688 | found = cur; |
689 | node = node->rb_left; | |
097b8a7c | 690 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
691 | /* we want the node with the smallest seq */ |
692 | if (found) | |
097b8a7c | 693 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
694 | found = cur; |
695 | node = node->rb_right; | |
696 | } else { | |
697 | found = cur; | |
698 | break; | |
699 | } | |
700 | } | |
b1a09f1e | 701 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
702 | |
703 | return found; | |
704 | } | |
705 | ||
706 | /* | |
707 | * this returns the element from the log with the smallest time sequence | |
708 | * value that's in the log (the oldest log item). any element with a time | |
709 | * sequence lower than min_seq will be ignored. | |
710 | */ | |
711 | static struct tree_mod_elem * | |
712 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
713 | u64 min_seq) | |
714 | { | |
715 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
716 | } | |
717 | ||
718 | /* | |
719 | * this returns the element from the log with the largest time sequence | |
720 | * value that's in the log (the most recent log item). any element with | |
721 | * a time sequence lower than min_seq will be ignored. | |
722 | */ | |
723 | static struct tree_mod_elem * | |
724 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
725 | { | |
726 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
727 | } | |
728 | ||
5de865ee | 729 | static noinline int |
bd989ba3 JS |
730 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
731 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 732 | unsigned long src_offset, int nr_items) |
bd989ba3 | 733 | { |
5de865ee FDBM |
734 | int ret = 0; |
735 | struct tree_mod_elem **tm_list = NULL; | |
736 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 737 | int i; |
5de865ee | 738 | int locked = 0; |
bd989ba3 | 739 | |
5de865ee FDBM |
740 | if (!tree_mod_need_log(fs_info, NULL)) |
741 | return 0; | |
bd989ba3 | 742 | |
c8cc6341 | 743 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
744 | return 0; |
745 | ||
31e818fe | 746 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
747 | GFP_NOFS); |
748 | if (!tm_list) | |
749 | return -ENOMEM; | |
bd989ba3 | 750 | |
5de865ee FDBM |
751 | tm_list_add = tm_list; |
752 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 753 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
754 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
755 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
756 | if (!tm_list_rem[i]) { | |
757 | ret = -ENOMEM; | |
758 | goto free_tms; | |
759 | } | |
760 | ||
761 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
762 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
763 | if (!tm_list_add[i]) { | |
764 | ret = -ENOMEM; | |
765 | goto free_tms; | |
766 | } | |
767 | } | |
768 | ||
769 | if (tree_mod_dont_log(fs_info, NULL)) | |
770 | goto free_tms; | |
771 | locked = 1; | |
772 | ||
773 | for (i = 0; i < nr_items; i++) { | |
774 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
775 | if (ret) | |
776 | goto free_tms; | |
777 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
778 | if (ret) | |
779 | goto free_tms; | |
bd989ba3 | 780 | } |
5de865ee | 781 | |
b1a09f1e | 782 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
783 | kfree(tm_list); |
784 | ||
785 | return 0; | |
786 | ||
787 | free_tms: | |
788 | for (i = 0; i < nr_items * 2; i++) { | |
789 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
790 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
791 | kfree(tm_list[i]); | |
792 | } | |
793 | if (locked) | |
b1a09f1e | 794 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
795 | kfree(tm_list); |
796 | ||
797 | return ret; | |
bd989ba3 JS |
798 | } |
799 | ||
db7279a2 | 800 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 801 | { |
5de865ee FDBM |
802 | struct tree_mod_elem **tm_list = NULL; |
803 | int nritems = 0; | |
804 | int i; | |
805 | int ret = 0; | |
806 | ||
807 | if (btrfs_header_level(eb) == 0) | |
808 | return 0; | |
809 | ||
db7279a2 | 810 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
811 | return 0; |
812 | ||
813 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 814 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
815 | if (!tm_list) |
816 | return -ENOMEM; | |
817 | ||
818 | for (i = 0; i < nritems; i++) { | |
819 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
820 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
821 | if (!tm_list[i]) { | |
822 | ret = -ENOMEM; | |
823 | goto free_tms; | |
824 | } | |
825 | } | |
826 | ||
db7279a2 | 827 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
828 | goto free_tms; |
829 | ||
db7279a2 | 830 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 831 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
832 | if (ret) |
833 | goto free_tms; | |
834 | kfree(tm_list); | |
835 | ||
836 | return 0; | |
837 | ||
838 | free_tms: | |
839 | for (i = 0; i < nritems; i++) | |
840 | kfree(tm_list[i]); | |
841 | kfree(tm_list); | |
842 | ||
843 | return ret; | |
bd989ba3 JS |
844 | } |
845 | ||
5d4f98a2 YZ |
846 | /* |
847 | * check if the tree block can be shared by multiple trees | |
848 | */ | |
849 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
850 | struct extent_buffer *buf) | |
851 | { | |
852 | /* | |
01327610 | 853 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
854 | * are never shared. If a block was allocated after the last |
855 | * snapshot and the block was not allocated by tree relocation, | |
856 | * we know the block is not shared. | |
857 | */ | |
27cdeb70 | 858 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
859 | buf != root->node && buf != root->commit_root && |
860 | (btrfs_header_generation(buf) <= | |
861 | btrfs_root_last_snapshot(&root->root_item) || | |
862 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
863 | return 1; | |
a79865c6 | 864 | |
5d4f98a2 YZ |
865 | return 0; |
866 | } | |
867 | ||
868 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
869 | struct btrfs_root *root, | |
870 | struct extent_buffer *buf, | |
f0486c68 YZ |
871 | struct extent_buffer *cow, |
872 | int *last_ref) | |
5d4f98a2 | 873 | { |
0b246afa | 874 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
875 | u64 refs; |
876 | u64 owner; | |
877 | u64 flags; | |
878 | u64 new_flags = 0; | |
879 | int ret; | |
880 | ||
881 | /* | |
882 | * Backrefs update rules: | |
883 | * | |
884 | * Always use full backrefs for extent pointers in tree block | |
885 | * allocated by tree relocation. | |
886 | * | |
887 | * If a shared tree block is no longer referenced by its owner | |
888 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
889 | * use full backrefs for extent pointers in tree block. | |
890 | * | |
891 | * If a tree block is been relocating | |
892 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
893 | * use full backrefs for extent pointers in tree block. | |
894 | * The reason for this is some operations (such as drop tree) | |
895 | * are only allowed for blocks use full backrefs. | |
896 | */ | |
897 | ||
898 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 899 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
900 | btrfs_header_level(buf), 1, |
901 | &refs, &flags); | |
be1a5564 MF |
902 | if (ret) |
903 | return ret; | |
e5df9573 MF |
904 | if (refs == 0) { |
905 | ret = -EROFS; | |
0b246afa | 906 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
907 | return ret; |
908 | } | |
5d4f98a2 YZ |
909 | } else { |
910 | refs = 1; | |
911 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
912 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
913 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
914 | else | |
915 | flags = 0; | |
916 | } | |
917 | ||
918 | owner = btrfs_header_owner(buf); | |
919 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
920 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
921 | ||
922 | if (refs > 1) { | |
923 | if ((owner == root->root_key.objectid || | |
924 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
925 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 926 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
927 | if (ret) |
928 | return ret; | |
5d4f98a2 YZ |
929 | |
930 | if (root->root_key.objectid == | |
931 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 932 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
933 | if (ret) |
934 | return ret; | |
e339a6b0 | 935 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
936 | if (ret) |
937 | return ret; | |
5d4f98a2 YZ |
938 | } |
939 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
940 | } else { | |
941 | ||
942 | if (root->root_key.objectid == | |
943 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 944 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 945 | else |
e339a6b0 | 946 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
947 | if (ret) |
948 | return ret; | |
5d4f98a2 YZ |
949 | } |
950 | if (new_flags != 0) { | |
b1c79e09 JB |
951 | int level = btrfs_header_level(buf); |
952 | ||
2ff7e61e | 953 | ret = btrfs_set_disk_extent_flags(trans, fs_info, |
5d4f98a2 YZ |
954 | buf->start, |
955 | buf->len, | |
b1c79e09 | 956 | new_flags, level, 0); |
be1a5564 MF |
957 | if (ret) |
958 | return ret; | |
5d4f98a2 YZ |
959 | } |
960 | } else { | |
961 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
962 | if (root->root_key.objectid == | |
963 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 964 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 965 | else |
e339a6b0 | 966 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
967 | if (ret) |
968 | return ret; | |
e339a6b0 | 969 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
970 | if (ret) |
971 | return ret; | |
5d4f98a2 | 972 | } |
7c302b49 | 973 | clean_tree_block(fs_info, buf); |
f0486c68 | 974 | *last_ref = 1; |
5d4f98a2 YZ |
975 | } |
976 | return 0; | |
977 | } | |
978 | ||
a6279470 FM |
979 | static struct extent_buffer *alloc_tree_block_no_bg_flush( |
980 | struct btrfs_trans_handle *trans, | |
981 | struct btrfs_root *root, | |
982 | u64 parent_start, | |
983 | const struct btrfs_disk_key *disk_key, | |
984 | int level, | |
985 | u64 hint, | |
986 | u64 empty_size) | |
987 | { | |
988 | struct btrfs_fs_info *fs_info = root->fs_info; | |
989 | struct extent_buffer *ret; | |
990 | ||
991 | /* | |
992 | * If we are COWing a node/leaf from the extent, chunk, device or free | |
993 | * space trees, make sure that we do not finish block group creation of | |
994 | * pending block groups. We do this to avoid a deadlock. | |
995 | * COWing can result in allocation of a new chunk, and flushing pending | |
996 | * block groups (btrfs_create_pending_block_groups()) can be triggered | |
997 | * when finishing allocation of a new chunk. Creation of a pending block | |
998 | * group modifies the extent, chunk, device and free space trees, | |
999 | * therefore we could deadlock with ourselves since we are holding a | |
1000 | * lock on an extent buffer that btrfs_create_pending_block_groups() may | |
1001 | * try to COW later. | |
1002 | * For similar reasons, we also need to delay flushing pending block | |
1003 | * groups when splitting a leaf or node, from one of those trees, since | |
1004 | * we are holding a write lock on it and its parent or when inserting a | |
1005 | * new root node for one of those trees. | |
1006 | */ | |
1007 | if (root == fs_info->extent_root || | |
1008 | root == fs_info->chunk_root || | |
1009 | root == fs_info->dev_root || | |
1010 | root == fs_info->free_space_root) | |
1011 | trans->can_flush_pending_bgs = false; | |
1012 | ||
1013 | ret = btrfs_alloc_tree_block(trans, root, parent_start, | |
1014 | root->root_key.objectid, disk_key, level, | |
1015 | hint, empty_size); | |
1016 | trans->can_flush_pending_bgs = true; | |
1017 | ||
1018 | return ret; | |
1019 | } | |
1020 | ||
d352ac68 | 1021 | /* |
d397712b CM |
1022 | * does the dirty work in cow of a single block. The parent block (if |
1023 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1024 | * dirty and returned locked. If you modify the block it needs to be marked | |
1025 | * dirty again. | |
d352ac68 CM |
1026 | * |
1027 | * search_start -- an allocation hint for the new block | |
1028 | * | |
d397712b CM |
1029 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1030 | * bytes the allocator should try to find free next to the block it returns. | |
1031 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1032 | */ |
d397712b | 1033 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1034 | struct btrfs_root *root, |
1035 | struct extent_buffer *buf, | |
1036 | struct extent_buffer *parent, int parent_slot, | |
1037 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1038 | u64 search_start, u64 empty_size) |
02217ed2 | 1039 | { |
0b246afa | 1040 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1041 | struct btrfs_disk_key disk_key; |
5f39d397 | 1042 | struct extent_buffer *cow; |
be1a5564 | 1043 | int level, ret; |
f0486c68 | 1044 | int last_ref = 0; |
925baedd | 1045 | int unlock_orig = 0; |
0f5053eb | 1046 | u64 parent_start = 0; |
7bb86316 | 1047 | |
925baedd CM |
1048 | if (*cow_ret == buf) |
1049 | unlock_orig = 1; | |
1050 | ||
b9447ef8 | 1051 | btrfs_assert_tree_locked(buf); |
925baedd | 1052 | |
27cdeb70 | 1053 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1054 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1055 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1056 | trans->transid != root->last_trans); | |
5f39d397 | 1057 | |
7bb86316 | 1058 | level = btrfs_header_level(buf); |
31840ae1 | 1059 | |
5d4f98a2 YZ |
1060 | if (level == 0) |
1061 | btrfs_item_key(buf, &disk_key, 0); | |
1062 | else | |
1063 | btrfs_node_key(buf, &disk_key, 0); | |
1064 | ||
0f5053eb GR |
1065 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1066 | parent_start = parent->start; | |
5d4f98a2 | 1067 | |
a6279470 FM |
1068 | cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, |
1069 | level, search_start, empty_size); | |
54aa1f4d CM |
1070 | if (IS_ERR(cow)) |
1071 | return PTR_ERR(cow); | |
6702ed49 | 1072 | |
b4ce94de CM |
1073 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1074 | ||
58e8012c | 1075 | copy_extent_buffer_full(cow, buf); |
db94535d | 1076 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1077 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1078 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1079 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1080 | BTRFS_HEADER_FLAG_RELOC); | |
1081 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1082 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1083 | else | |
1084 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1085 | |
de37aa51 | 1086 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 1087 | |
be1a5564 | 1088 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1089 | if (ret) { |
66642832 | 1090 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1091 | return ret; |
1092 | } | |
1a40e23b | 1093 | |
27cdeb70 | 1094 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1095 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1096 | if (ret) { |
66642832 | 1097 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1098 | return ret; |
93314e3b | 1099 | } |
83d4cfd4 | 1100 | } |
3fd0a558 | 1101 | |
02217ed2 | 1102 | if (buf == root->node) { |
925baedd | 1103 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1104 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1105 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1106 | parent_start = buf->start; | |
925baedd | 1107 | |
5f39d397 | 1108 | extent_buffer_get(cow); |
d9d19a01 DS |
1109 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1110 | BUG_ON(ret < 0); | |
240f62c8 | 1111 | rcu_assign_pointer(root->node, cow); |
925baedd | 1112 | |
f0486c68 | 1113 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1114 | last_ref); |
5f39d397 | 1115 | free_extent_buffer(buf); |
0b86a832 | 1116 | add_root_to_dirty_list(root); |
02217ed2 | 1117 | } else { |
5d4f98a2 | 1118 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1119 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1120 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1121 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1122 | cow->start); |
74493f7a CM |
1123 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1124 | trans->transid); | |
d6025579 | 1125 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1126 | if (last_ref) { |
db7279a2 | 1127 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1128 | if (ret) { |
66642832 | 1129 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1130 | return ret; |
1131 | } | |
1132 | } | |
f0486c68 | 1133 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1134 | last_ref); |
02217ed2 | 1135 | } |
925baedd CM |
1136 | if (unlock_orig) |
1137 | btrfs_tree_unlock(buf); | |
3083ee2e | 1138 | free_extent_buffer_stale(buf); |
ccd467d6 | 1139 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1140 | *cow_ret = cow; |
02217ed2 CM |
1141 | return 0; |
1142 | } | |
1143 | ||
5d9e75c4 JS |
1144 | /* |
1145 | * returns the logical address of the oldest predecessor of the given root. | |
1146 | * entries older than time_seq are ignored. | |
1147 | */ | |
bcd24dab DS |
1148 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1149 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1150 | { |
1151 | struct tree_mod_elem *tm; | |
1152 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1153 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1154 | int looped = 0; |
1155 | ||
1156 | if (!time_seq) | |
35a3621b | 1157 | return NULL; |
5d9e75c4 JS |
1158 | |
1159 | /* | |
298cfd36 CR |
1160 | * the very last operation that's logged for a root is the |
1161 | * replacement operation (if it is replaced at all). this has | |
1162 | * the logical address of the *new* root, making it the very | |
1163 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1164 | */ |
1165 | while (1) { | |
bcd24dab | 1166 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1167 | time_seq); |
1168 | if (!looped && !tm) | |
35a3621b | 1169 | return NULL; |
5d9e75c4 | 1170 | /* |
28da9fb4 JS |
1171 | * if there are no tree operation for the oldest root, we simply |
1172 | * return it. this should only happen if that (old) root is at | |
1173 | * level 0. | |
5d9e75c4 | 1174 | */ |
28da9fb4 JS |
1175 | if (!tm) |
1176 | break; | |
5d9e75c4 | 1177 | |
28da9fb4 JS |
1178 | /* |
1179 | * if there's an operation that's not a root replacement, we | |
1180 | * found the oldest version of our root. normally, we'll find a | |
1181 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1182 | */ | |
5d9e75c4 JS |
1183 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1184 | break; | |
1185 | ||
1186 | found = tm; | |
1187 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1188 | looped = 1; |
1189 | } | |
1190 | ||
a95236d9 JS |
1191 | /* if there's no old root to return, return what we found instead */ |
1192 | if (!found) | |
1193 | found = tm; | |
1194 | ||
5d9e75c4 JS |
1195 | return found; |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1200 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1201 | * time_seq). |
1202 | */ | |
1203 | static void | |
f1ca7e98 JB |
1204 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1205 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1206 | { |
1207 | u32 n; | |
1208 | struct rb_node *next; | |
1209 | struct tree_mod_elem *tm = first_tm; | |
1210 | unsigned long o_dst; | |
1211 | unsigned long o_src; | |
1212 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1213 | ||
1214 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1215 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1216 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1217 | /* |
1218 | * all the operations are recorded with the operator used for | |
1219 | * the modification. as we're going backwards, we do the | |
1220 | * opposite of each operation here. | |
1221 | */ | |
1222 | switch (tm->op) { | |
1223 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1224 | BUG_ON(tm->slot < n); | |
1c697d4a | 1225 | /* Fallthrough */ |
95c80bb1 | 1226 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1227 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1228 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1229 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1230 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1231 | tm->generation); | |
4c3e6969 | 1232 | n++; |
5d9e75c4 JS |
1233 | break; |
1234 | case MOD_LOG_KEY_REPLACE: | |
1235 | BUG_ON(tm->slot >= n); | |
1236 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1237 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1238 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1239 | tm->generation); | |
1240 | break; | |
1241 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1242 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1243 | n--; |
1244 | break; | |
1245 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1246 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1247 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1248 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1249 | tm->move.nr_items * p_size); |
1250 | break; | |
1251 | case MOD_LOG_ROOT_REPLACE: | |
1252 | /* | |
1253 | * this operation is special. for roots, this must be | |
1254 | * handled explicitly before rewinding. | |
1255 | * for non-roots, this operation may exist if the node | |
1256 | * was a root: root A -> child B; then A gets empty and | |
1257 | * B is promoted to the new root. in the mod log, we'll | |
1258 | * have a root-replace operation for B, a tree block | |
1259 | * that is no root. we simply ignore that operation. | |
1260 | */ | |
1261 | break; | |
1262 | } | |
1263 | next = rb_next(&tm->node); | |
1264 | if (!next) | |
1265 | break; | |
6b4df8b6 | 1266 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1267 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1268 | break; |
1269 | } | |
b1a09f1e | 1270 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1271 | btrfs_set_header_nritems(eb, n); |
1272 | } | |
1273 | ||
47fb091f | 1274 | /* |
01327610 | 1275 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1276 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1277 | * returned buffer is always read-locked. If the returned buffer is not the | |
1278 | * input buffer, the lock on the input buffer is released and the input buffer | |
1279 | * is freed (its refcount is decremented). | |
1280 | */ | |
5d9e75c4 | 1281 | static struct extent_buffer * |
9ec72677 JB |
1282 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1283 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1284 | { |
1285 | struct extent_buffer *eb_rewin; | |
1286 | struct tree_mod_elem *tm; | |
1287 | ||
1288 | if (!time_seq) | |
1289 | return eb; | |
1290 | ||
1291 | if (btrfs_header_level(eb) == 0) | |
1292 | return eb; | |
1293 | ||
1294 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1295 | if (!tm) | |
1296 | return eb; | |
1297 | ||
9ec72677 | 1298 | btrfs_set_path_blocking(path); |
300aa896 | 1299 | btrfs_set_lock_blocking_read(eb); |
9ec72677 | 1300 | |
5d9e75c4 JS |
1301 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1302 | BUG_ON(tm->slot != 0); | |
da17066c | 1303 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1304 | if (!eb_rewin) { |
9ec72677 | 1305 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1306 | free_extent_buffer(eb); |
1307 | return NULL; | |
1308 | } | |
5d9e75c4 JS |
1309 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1310 | btrfs_set_header_backref_rev(eb_rewin, | |
1311 | btrfs_header_backref_rev(eb)); | |
1312 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1313 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1314 | } else { |
1315 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1316 | if (!eb_rewin) { |
9ec72677 | 1317 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1318 | free_extent_buffer(eb); |
1319 | return NULL; | |
1320 | } | |
5d9e75c4 JS |
1321 | } |
1322 | ||
9ec72677 | 1323 | btrfs_tree_read_unlock_blocking(eb); |
5d9e75c4 JS |
1324 | free_extent_buffer(eb); |
1325 | ||
47fb091f | 1326 | btrfs_tree_read_lock(eb_rewin); |
f1ca7e98 | 1327 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1328 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1329 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1330 | |
1331 | return eb_rewin; | |
1332 | } | |
1333 | ||
8ba97a15 JS |
1334 | /* |
1335 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1336 | * value. If there are no changes, the current root->root_node is returned. If | |
1337 | * anything changed in between, there's a fresh buffer allocated on which the | |
1338 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1339 | * Returns NULL on error (with no locks held). | |
1340 | */ | |
5d9e75c4 JS |
1341 | static inline struct extent_buffer * |
1342 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1343 | { | |
0b246afa | 1344 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1345 | struct tree_mod_elem *tm; |
30b0463a JS |
1346 | struct extent_buffer *eb = NULL; |
1347 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1348 | struct extent_buffer *old; |
a95236d9 | 1349 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1350 | u64 old_generation = 0; |
a95236d9 | 1351 | u64 logical; |
581c1760 | 1352 | int level; |
5d9e75c4 | 1353 | |
30b0463a | 1354 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1355 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1356 | if (!tm) |
30b0463a | 1357 | return eb_root; |
5d9e75c4 | 1358 | |
a95236d9 JS |
1359 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1360 | old_root = &tm->old_root; | |
1361 | old_generation = tm->generation; | |
1362 | logical = old_root->logical; | |
581c1760 | 1363 | level = old_root->level; |
a95236d9 | 1364 | } else { |
30b0463a | 1365 | logical = eb_root->start; |
581c1760 | 1366 | level = btrfs_header_level(eb_root); |
a95236d9 | 1367 | } |
5d9e75c4 | 1368 | |
0b246afa | 1369 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1370 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1371 | btrfs_tree_read_unlock(eb_root); |
1372 | free_extent_buffer(eb_root); | |
581c1760 | 1373 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1374 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1375 | if (!IS_ERR(old)) | |
1376 | free_extent_buffer(old); | |
0b246afa JM |
1377 | btrfs_warn(fs_info, |
1378 | "failed to read tree block %llu from get_old_root", | |
1379 | logical); | |
834328a8 | 1380 | } else { |
7bfdcf7f LB |
1381 | eb = btrfs_clone_extent_buffer(old); |
1382 | free_extent_buffer(old); | |
834328a8 JS |
1383 | } |
1384 | } else if (old_root) { | |
30b0463a JS |
1385 | btrfs_tree_read_unlock(eb_root); |
1386 | free_extent_buffer(eb_root); | |
0b246afa | 1387 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1388 | } else { |
300aa896 | 1389 | btrfs_set_lock_blocking_read(eb_root); |
30b0463a | 1390 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1391 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1392 | free_extent_buffer(eb_root); |
834328a8 JS |
1393 | } |
1394 | ||
8ba97a15 JS |
1395 | if (!eb) |
1396 | return NULL; | |
1397 | btrfs_tree_read_lock(eb); | |
a95236d9 | 1398 | if (old_root) { |
5d9e75c4 JS |
1399 | btrfs_set_header_bytenr(eb, eb->start); |
1400 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1401 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1402 | btrfs_set_header_level(eb, old_root->level); |
1403 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1404 | } |
28da9fb4 | 1405 | if (tm) |
0b246afa | 1406 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1407 | else |
1408 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1409 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1410 | |
1411 | return eb; | |
1412 | } | |
1413 | ||
5b6602e7 JS |
1414 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1415 | { | |
1416 | struct tree_mod_elem *tm; | |
1417 | int level; | |
30b0463a | 1418 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1419 | |
bcd24dab | 1420 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1421 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1422 | level = tm->old_root.level; | |
1423 | } else { | |
30b0463a | 1424 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1425 | } |
30b0463a | 1426 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1427 | |
1428 | return level; | |
1429 | } | |
1430 | ||
5d4f98a2 YZ |
1431 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1432 | struct btrfs_root *root, | |
1433 | struct extent_buffer *buf) | |
1434 | { | |
f5ee5c9a | 1435 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1436 | return 0; |
fccb84c9 | 1437 | |
d1980131 DS |
1438 | /* Ensure we can see the FORCE_COW bit */ |
1439 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1440 | |
1441 | /* | |
1442 | * We do not need to cow a block if | |
1443 | * 1) this block is not created or changed in this transaction; | |
1444 | * 2) this block does not belong to TREE_RELOC tree; | |
1445 | * 3) the root is not forced COW. | |
1446 | * | |
1447 | * What is forced COW: | |
01327610 | 1448 | * when we create snapshot during committing the transaction, |
52042d8e | 1449 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
1450 | * block to ensure the metadata consistency. |
1451 | */ | |
5d4f98a2 YZ |
1452 | if (btrfs_header_generation(buf) == trans->transid && |
1453 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1454 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1455 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1456 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1457 | return 0; |
1458 | return 1; | |
1459 | } | |
1460 | ||
d352ac68 CM |
1461 | /* |
1462 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1463 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1464 | * once per transaction, as long as it hasn't been written yet |
1465 | */ | |
d397712b | 1466 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1467 | struct btrfs_root *root, struct extent_buffer *buf, |
1468 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1469 | struct extent_buffer **cow_ret) |
6702ed49 | 1470 | { |
0b246afa | 1471 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1472 | u64 search_start; |
f510cfec | 1473 | int ret; |
dc17ff8f | 1474 | |
83354f07 JB |
1475 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
1476 | btrfs_err(fs_info, | |
1477 | "COW'ing blocks on a fs root that's being dropped"); | |
1478 | ||
0b246afa | 1479 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1480 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1481 | trans->transid, |
0b246afa | 1482 | fs_info->running_transaction->transid); |
31b1a2bd | 1483 | |
0b246afa | 1484 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1485 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1486 | trans->transid, fs_info->generation); |
dc17ff8f | 1487 | |
5d4f98a2 | 1488 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1489 | trans->dirty = true; |
6702ed49 CM |
1490 | *cow_ret = buf; |
1491 | return 0; | |
1492 | } | |
c487685d | 1493 | |
ee22184b | 1494 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1495 | |
1496 | if (parent) | |
8bead258 DS |
1497 | btrfs_set_lock_blocking_write(parent); |
1498 | btrfs_set_lock_blocking_write(buf); | |
b4ce94de | 1499 | |
f616f5cd QW |
1500 | /* |
1501 | * Before CoWing this block for later modification, check if it's | |
1502 | * the subtree root and do the delayed subtree trace if needed. | |
1503 | * | |
1504 | * Also We don't care about the error, as it's handled internally. | |
1505 | */ | |
1506 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 1507 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1508 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1509 | |
1510 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1511 | ||
f510cfec | 1512 | return ret; |
6702ed49 CM |
1513 | } |
1514 | ||
d352ac68 CM |
1515 | /* |
1516 | * helper function for defrag to decide if two blocks pointed to by a | |
1517 | * node are actually close by | |
1518 | */ | |
6b80053d | 1519 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1520 | { |
6b80053d | 1521 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1522 | return 1; |
6b80053d | 1523 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1524 | return 1; |
1525 | return 0; | |
1526 | } | |
1527 | ||
081e9573 CM |
1528 | /* |
1529 | * compare two keys in a memcmp fashion | |
1530 | */ | |
310712b2 OS |
1531 | static int comp_keys(const struct btrfs_disk_key *disk, |
1532 | const struct btrfs_key *k2) | |
081e9573 CM |
1533 | { |
1534 | struct btrfs_key k1; | |
1535 | ||
1536 | btrfs_disk_key_to_cpu(&k1, disk); | |
1537 | ||
20736aba | 1538 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1539 | } |
1540 | ||
f3465ca4 JB |
1541 | /* |
1542 | * same as comp_keys only with two btrfs_key's | |
1543 | */ | |
310712b2 | 1544 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1545 | { |
1546 | if (k1->objectid > k2->objectid) | |
1547 | return 1; | |
1548 | if (k1->objectid < k2->objectid) | |
1549 | return -1; | |
1550 | if (k1->type > k2->type) | |
1551 | return 1; | |
1552 | if (k1->type < k2->type) | |
1553 | return -1; | |
1554 | if (k1->offset > k2->offset) | |
1555 | return 1; | |
1556 | if (k1->offset < k2->offset) | |
1557 | return -1; | |
1558 | return 0; | |
1559 | } | |
081e9573 | 1560 | |
d352ac68 CM |
1561 | /* |
1562 | * this is used by the defrag code to go through all the | |
1563 | * leaves pointed to by a node and reallocate them so that | |
1564 | * disk order is close to key order | |
1565 | */ | |
6702ed49 | 1566 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1567 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1568 | int start_slot, u64 *last_ret, |
a6b6e75e | 1569 | struct btrfs_key *progress) |
6702ed49 | 1570 | { |
0b246afa | 1571 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1572 | struct extent_buffer *cur; |
6702ed49 | 1573 | u64 blocknr; |
ca7a79ad | 1574 | u64 gen; |
e9d0b13b CM |
1575 | u64 search_start = *last_ret; |
1576 | u64 last_block = 0; | |
6702ed49 CM |
1577 | u64 other; |
1578 | u32 parent_nritems; | |
6702ed49 CM |
1579 | int end_slot; |
1580 | int i; | |
1581 | int err = 0; | |
f2183bde | 1582 | int parent_level; |
6b80053d CM |
1583 | int uptodate; |
1584 | u32 blocksize; | |
081e9573 CM |
1585 | int progress_passed = 0; |
1586 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1587 | |
5708b959 | 1588 | parent_level = btrfs_header_level(parent); |
5708b959 | 1589 | |
0b246afa JM |
1590 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1591 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1592 | |
6b80053d | 1593 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1594 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1595 | end_slot = parent_nritems - 1; |
6702ed49 | 1596 | |
5dfe2be7 | 1597 | if (parent_nritems <= 1) |
6702ed49 CM |
1598 | return 0; |
1599 | ||
8bead258 | 1600 | btrfs_set_lock_blocking_write(parent); |
b4ce94de | 1601 | |
5dfe2be7 | 1602 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1603 | struct btrfs_key first_key; |
6702ed49 | 1604 | int close = 1; |
a6b6e75e | 1605 | |
081e9573 CM |
1606 | btrfs_node_key(parent, &disk_key, i); |
1607 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1608 | continue; | |
1609 | ||
1610 | progress_passed = 1; | |
6b80053d | 1611 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1612 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1613 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1614 | if (last_block == 0) |
1615 | last_block = blocknr; | |
5708b959 | 1616 | |
6702ed49 | 1617 | if (i > 0) { |
6b80053d CM |
1618 | other = btrfs_node_blockptr(parent, i - 1); |
1619 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1620 | } |
5dfe2be7 | 1621 | if (!close && i < end_slot) { |
6b80053d CM |
1622 | other = btrfs_node_blockptr(parent, i + 1); |
1623 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1624 | } |
e9d0b13b CM |
1625 | if (close) { |
1626 | last_block = blocknr; | |
6702ed49 | 1627 | continue; |
e9d0b13b | 1628 | } |
6702ed49 | 1629 | |
0b246afa | 1630 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1631 | if (cur) |
b9fab919 | 1632 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1633 | else |
1634 | uptodate = 0; | |
5708b959 | 1635 | if (!cur || !uptodate) { |
6b80053d | 1636 | if (!cur) { |
581c1760 QW |
1637 | cur = read_tree_block(fs_info, blocknr, gen, |
1638 | parent_level - 1, | |
1639 | &first_key); | |
64c043de LB |
1640 | if (IS_ERR(cur)) { |
1641 | return PTR_ERR(cur); | |
1642 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1643 | free_extent_buffer(cur); |
97d9a8a4 | 1644 | return -EIO; |
416bc658 | 1645 | } |
6b80053d | 1646 | } else if (!uptodate) { |
581c1760 QW |
1647 | err = btrfs_read_buffer(cur, gen, |
1648 | parent_level - 1,&first_key); | |
018642a1 TI |
1649 | if (err) { |
1650 | free_extent_buffer(cur); | |
1651 | return err; | |
1652 | } | |
f2183bde | 1653 | } |
6702ed49 | 1654 | } |
e9d0b13b | 1655 | if (search_start == 0) |
6b80053d | 1656 | search_start = last_block; |
e9d0b13b | 1657 | |
e7a84565 | 1658 | btrfs_tree_lock(cur); |
8bead258 | 1659 | btrfs_set_lock_blocking_write(cur); |
6b80053d | 1660 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1661 | &cur, search_start, |
6b80053d | 1662 | min(16 * blocksize, |
9fa8cfe7 | 1663 | (end_slot - i) * blocksize)); |
252c38f0 | 1664 | if (err) { |
e7a84565 | 1665 | btrfs_tree_unlock(cur); |
6b80053d | 1666 | free_extent_buffer(cur); |
6702ed49 | 1667 | break; |
252c38f0 | 1668 | } |
e7a84565 CM |
1669 | search_start = cur->start; |
1670 | last_block = cur->start; | |
f2183bde | 1671 | *last_ret = search_start; |
e7a84565 CM |
1672 | btrfs_tree_unlock(cur); |
1673 | free_extent_buffer(cur); | |
6702ed49 CM |
1674 | } |
1675 | return err; | |
1676 | } | |
1677 | ||
74123bd7 | 1678 | /* |
5f39d397 CM |
1679 | * search for key in the extent_buffer. The items start at offset p, |
1680 | * and they are item_size apart. There are 'max' items in p. | |
1681 | * | |
74123bd7 CM |
1682 | * the slot in the array is returned via slot, and it points to |
1683 | * the place where you would insert key if it is not found in | |
1684 | * the array. | |
1685 | * | |
1686 | * slot may point to max if the key is bigger than all of the keys | |
1687 | */ | |
e02119d5 | 1688 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1689 | unsigned long p, int item_size, |
1690 | const struct btrfs_key *key, | |
e02119d5 | 1691 | int max, int *slot) |
be0e5c09 CM |
1692 | { |
1693 | int low = 0; | |
1694 | int high = max; | |
1695 | int mid; | |
1696 | int ret; | |
479965d6 | 1697 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1698 | struct btrfs_disk_key unaligned; |
1699 | unsigned long offset; | |
5f39d397 CM |
1700 | char *kaddr = NULL; |
1701 | unsigned long map_start = 0; | |
1702 | unsigned long map_len = 0; | |
479965d6 | 1703 | int err; |
be0e5c09 | 1704 | |
5e24e9af LB |
1705 | if (low > high) { |
1706 | btrfs_err(eb->fs_info, | |
1707 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1708 | __func__, low, high, eb->start, | |
1709 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1710 | return -EINVAL; | |
1711 | } | |
1712 | ||
d397712b | 1713 | while (low < high) { |
be0e5c09 | 1714 | mid = (low + high) / 2; |
5f39d397 CM |
1715 | offset = p + mid * item_size; |
1716 | ||
a6591715 | 1717 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1718 | (offset + sizeof(struct btrfs_disk_key)) > |
1719 | map_start + map_len) { | |
934d375b CM |
1720 | |
1721 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1722 | sizeof(struct btrfs_disk_key), |
a6591715 | 1723 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1724 | |
1725 | if (!err) { | |
1726 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1727 | map_start); | |
415b35a5 | 1728 | } else if (err == 1) { |
479965d6 CM |
1729 | read_extent_buffer(eb, &unaligned, |
1730 | offset, sizeof(unaligned)); | |
1731 | tmp = &unaligned; | |
415b35a5 LB |
1732 | } else { |
1733 | return err; | |
479965d6 | 1734 | } |
5f39d397 | 1735 | |
5f39d397 CM |
1736 | } else { |
1737 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1738 | map_start); | |
1739 | } | |
be0e5c09 CM |
1740 | ret = comp_keys(tmp, key); |
1741 | ||
1742 | if (ret < 0) | |
1743 | low = mid + 1; | |
1744 | else if (ret > 0) | |
1745 | high = mid; | |
1746 | else { | |
1747 | *slot = mid; | |
1748 | return 0; | |
1749 | } | |
1750 | } | |
1751 | *slot = low; | |
1752 | return 1; | |
1753 | } | |
1754 | ||
97571fd0 CM |
1755 | /* |
1756 | * simple bin_search frontend that does the right thing for | |
1757 | * leaves vs nodes | |
1758 | */ | |
a74b35ec NB |
1759 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1760 | int level, int *slot) | |
be0e5c09 | 1761 | { |
f775738f | 1762 | if (level == 0) |
5f39d397 CM |
1763 | return generic_bin_search(eb, |
1764 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1765 | sizeof(struct btrfs_item), |
5f39d397 | 1766 | key, btrfs_header_nritems(eb), |
7518a238 | 1767 | slot); |
f775738f | 1768 | else |
5f39d397 CM |
1769 | return generic_bin_search(eb, |
1770 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1771 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1772 | key, btrfs_header_nritems(eb), |
7518a238 | 1773 | slot); |
be0e5c09 CM |
1774 | } |
1775 | ||
f0486c68 YZ |
1776 | static void root_add_used(struct btrfs_root *root, u32 size) |
1777 | { | |
1778 | spin_lock(&root->accounting_lock); | |
1779 | btrfs_set_root_used(&root->root_item, | |
1780 | btrfs_root_used(&root->root_item) + size); | |
1781 | spin_unlock(&root->accounting_lock); | |
1782 | } | |
1783 | ||
1784 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1785 | { | |
1786 | spin_lock(&root->accounting_lock); | |
1787 | btrfs_set_root_used(&root->root_item, | |
1788 | btrfs_root_used(&root->root_item) - size); | |
1789 | spin_unlock(&root->accounting_lock); | |
1790 | } | |
1791 | ||
d352ac68 CM |
1792 | /* given a node and slot number, this reads the blocks it points to. The |
1793 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1794 | */ |
2ff7e61e JM |
1795 | static noinline struct extent_buffer * |
1796 | read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent, | |
1797 | int slot) | |
bb803951 | 1798 | { |
ca7a79ad | 1799 | int level = btrfs_header_level(parent); |
416bc658 | 1800 | struct extent_buffer *eb; |
581c1760 | 1801 | struct btrfs_key first_key; |
416bc658 | 1802 | |
fb770ae4 LB |
1803 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1804 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1805 | |
1806 | BUG_ON(level == 0); | |
1807 | ||
581c1760 | 1808 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
2ff7e61e | 1809 | eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1810 | btrfs_node_ptr_generation(parent, slot), |
1811 | level - 1, &first_key); | |
fb770ae4 LB |
1812 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1813 | free_extent_buffer(eb); | |
1814 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1815 | } |
1816 | ||
1817 | return eb; | |
bb803951 CM |
1818 | } |
1819 | ||
d352ac68 CM |
1820 | /* |
1821 | * node level balancing, used to make sure nodes are in proper order for | |
1822 | * item deletion. We balance from the top down, so we have to make sure | |
1823 | * that a deletion won't leave an node completely empty later on. | |
1824 | */ | |
e02119d5 | 1825 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1826 | struct btrfs_root *root, |
1827 | struct btrfs_path *path, int level) | |
bb803951 | 1828 | { |
0b246afa | 1829 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1830 | struct extent_buffer *right = NULL; |
1831 | struct extent_buffer *mid; | |
1832 | struct extent_buffer *left = NULL; | |
1833 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1834 | int ret = 0; |
1835 | int wret; | |
1836 | int pslot; | |
bb803951 | 1837 | int orig_slot = path->slots[level]; |
79f95c82 | 1838 | u64 orig_ptr; |
bb803951 | 1839 | |
98e6b1eb | 1840 | ASSERT(level > 0); |
bb803951 | 1841 | |
5f39d397 | 1842 | mid = path->nodes[level]; |
b4ce94de | 1843 | |
bd681513 CM |
1844 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1845 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1846 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1847 | ||
1d4f8a0c | 1848 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1849 | |
a05a9bb1 | 1850 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1851 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1852 | pslot = path->slots[level + 1]; |
1853 | } | |
bb803951 | 1854 | |
40689478 CM |
1855 | /* |
1856 | * deal with the case where there is only one pointer in the root | |
1857 | * by promoting the node below to a root | |
1858 | */ | |
5f39d397 CM |
1859 | if (!parent) { |
1860 | struct extent_buffer *child; | |
bb803951 | 1861 | |
5f39d397 | 1862 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1863 | return 0; |
1864 | ||
1865 | /* promote the child to a root */ | |
2ff7e61e | 1866 | child = read_node_slot(fs_info, mid, 0); |
fb770ae4 LB |
1867 | if (IS_ERR(child)) { |
1868 | ret = PTR_ERR(child); | |
0b246afa | 1869 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1870 | goto enospc; |
1871 | } | |
1872 | ||
925baedd | 1873 | btrfs_tree_lock(child); |
8bead258 | 1874 | btrfs_set_lock_blocking_write(child); |
9fa8cfe7 | 1875 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1876 | if (ret) { |
1877 | btrfs_tree_unlock(child); | |
1878 | free_extent_buffer(child); | |
1879 | goto enospc; | |
1880 | } | |
2f375ab9 | 1881 | |
d9d19a01 DS |
1882 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1883 | BUG_ON(ret < 0); | |
240f62c8 | 1884 | rcu_assign_pointer(root->node, child); |
925baedd | 1885 | |
0b86a832 | 1886 | add_root_to_dirty_list(root); |
925baedd | 1887 | btrfs_tree_unlock(child); |
b4ce94de | 1888 | |
925baedd | 1889 | path->locks[level] = 0; |
bb803951 | 1890 | path->nodes[level] = NULL; |
7c302b49 | 1891 | clean_tree_block(fs_info, mid); |
925baedd | 1892 | btrfs_tree_unlock(mid); |
bb803951 | 1893 | /* once for the path */ |
5f39d397 | 1894 | free_extent_buffer(mid); |
f0486c68 YZ |
1895 | |
1896 | root_sub_used(root, mid->len); | |
5581a51a | 1897 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1898 | /* once for the root ptr */ |
3083ee2e | 1899 | free_extent_buffer_stale(mid); |
f0486c68 | 1900 | return 0; |
bb803951 | 1901 | } |
5f39d397 | 1902 | if (btrfs_header_nritems(mid) > |
0b246afa | 1903 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1904 | return 0; |
1905 | ||
2ff7e61e | 1906 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
1907 | if (IS_ERR(left)) |
1908 | left = NULL; | |
1909 | ||
5f39d397 | 1910 | if (left) { |
925baedd | 1911 | btrfs_tree_lock(left); |
8bead258 | 1912 | btrfs_set_lock_blocking_write(left); |
5f39d397 | 1913 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1914 | parent, pslot - 1, &left); |
54aa1f4d CM |
1915 | if (wret) { |
1916 | ret = wret; | |
1917 | goto enospc; | |
1918 | } | |
2cc58cf2 | 1919 | } |
fb770ae4 | 1920 | |
2ff7e61e | 1921 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
1922 | if (IS_ERR(right)) |
1923 | right = NULL; | |
1924 | ||
5f39d397 | 1925 | if (right) { |
925baedd | 1926 | btrfs_tree_lock(right); |
8bead258 | 1927 | btrfs_set_lock_blocking_write(right); |
5f39d397 | 1928 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1929 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1930 | if (wret) { |
1931 | ret = wret; | |
1932 | goto enospc; | |
1933 | } | |
1934 | } | |
1935 | ||
1936 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1937 | if (left) { |
1938 | orig_slot += btrfs_header_nritems(left); | |
2ff7e61e | 1939 | wret = push_node_left(trans, fs_info, left, mid, 1); |
79f95c82 CM |
1940 | if (wret < 0) |
1941 | ret = wret; | |
bb803951 | 1942 | } |
79f95c82 CM |
1943 | |
1944 | /* | |
1945 | * then try to empty the right most buffer into the middle | |
1946 | */ | |
5f39d397 | 1947 | if (right) { |
2ff7e61e | 1948 | wret = push_node_left(trans, fs_info, mid, right, 1); |
54aa1f4d | 1949 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1950 | ret = wret; |
5f39d397 | 1951 | if (btrfs_header_nritems(right) == 0) { |
7c302b49 | 1952 | clean_tree_block(fs_info, right); |
925baedd | 1953 | btrfs_tree_unlock(right); |
afe5fea7 | 1954 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1955 | root_sub_used(root, right->len); |
5581a51a | 1956 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1957 | free_extent_buffer_stale(right); |
f0486c68 | 1958 | right = NULL; |
bb803951 | 1959 | } else { |
5f39d397 CM |
1960 | struct btrfs_disk_key right_key; |
1961 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1962 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1963 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1964 | BUG_ON(ret < 0); | |
5f39d397 CM |
1965 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1966 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1967 | } |
1968 | } | |
5f39d397 | 1969 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1970 | /* |
1971 | * we're not allowed to leave a node with one item in the | |
1972 | * tree during a delete. A deletion from lower in the tree | |
1973 | * could try to delete the only pointer in this node. | |
1974 | * So, pull some keys from the left. | |
1975 | * There has to be a left pointer at this point because | |
1976 | * otherwise we would have pulled some pointers from the | |
1977 | * right | |
1978 | */ | |
305a26af MF |
1979 | if (!left) { |
1980 | ret = -EROFS; | |
0b246afa | 1981 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1982 | goto enospc; |
1983 | } | |
2ff7e61e | 1984 | wret = balance_node_right(trans, fs_info, mid, left); |
54aa1f4d | 1985 | if (wret < 0) { |
79f95c82 | 1986 | ret = wret; |
54aa1f4d CM |
1987 | goto enospc; |
1988 | } | |
bce4eae9 | 1989 | if (wret == 1) { |
2ff7e61e | 1990 | wret = push_node_left(trans, fs_info, left, mid, 1); |
bce4eae9 CM |
1991 | if (wret < 0) |
1992 | ret = wret; | |
1993 | } | |
79f95c82 CM |
1994 | BUG_ON(wret == 1); |
1995 | } | |
5f39d397 | 1996 | if (btrfs_header_nritems(mid) == 0) { |
7c302b49 | 1997 | clean_tree_block(fs_info, mid); |
925baedd | 1998 | btrfs_tree_unlock(mid); |
afe5fea7 | 1999 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2000 | root_sub_used(root, mid->len); |
5581a51a | 2001 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2002 | free_extent_buffer_stale(mid); |
f0486c68 | 2003 | mid = NULL; |
79f95c82 CM |
2004 | } else { |
2005 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2006 | struct btrfs_disk_key mid_key; |
2007 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
2008 | ret = tree_mod_log_insert_key(parent, pslot, |
2009 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2010 | BUG_ON(ret < 0); | |
5f39d397 CM |
2011 | btrfs_set_node_key(parent, &mid_key, pslot); |
2012 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2013 | } |
bb803951 | 2014 | |
79f95c82 | 2015 | /* update the path */ |
5f39d397 CM |
2016 | if (left) { |
2017 | if (btrfs_header_nritems(left) > orig_slot) { | |
2018 | extent_buffer_get(left); | |
925baedd | 2019 | /* left was locked after cow */ |
5f39d397 | 2020 | path->nodes[level] = left; |
bb803951 CM |
2021 | path->slots[level + 1] -= 1; |
2022 | path->slots[level] = orig_slot; | |
925baedd CM |
2023 | if (mid) { |
2024 | btrfs_tree_unlock(mid); | |
5f39d397 | 2025 | free_extent_buffer(mid); |
925baedd | 2026 | } |
bb803951 | 2027 | } else { |
5f39d397 | 2028 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2029 | path->slots[level] = orig_slot; |
2030 | } | |
2031 | } | |
79f95c82 | 2032 | /* double check we haven't messed things up */ |
e20d96d6 | 2033 | if (orig_ptr != |
5f39d397 | 2034 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2035 | BUG(); |
54aa1f4d | 2036 | enospc: |
925baedd CM |
2037 | if (right) { |
2038 | btrfs_tree_unlock(right); | |
5f39d397 | 2039 | free_extent_buffer(right); |
925baedd CM |
2040 | } |
2041 | if (left) { | |
2042 | if (path->nodes[level] != left) | |
2043 | btrfs_tree_unlock(left); | |
5f39d397 | 2044 | free_extent_buffer(left); |
925baedd | 2045 | } |
bb803951 CM |
2046 | return ret; |
2047 | } | |
2048 | ||
d352ac68 CM |
2049 | /* Node balancing for insertion. Here we only split or push nodes around |
2050 | * when they are completely full. This is also done top down, so we | |
2051 | * have to be pessimistic. | |
2052 | */ | |
d397712b | 2053 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2054 | struct btrfs_root *root, |
2055 | struct btrfs_path *path, int level) | |
e66f709b | 2056 | { |
0b246afa | 2057 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2058 | struct extent_buffer *right = NULL; |
2059 | struct extent_buffer *mid; | |
2060 | struct extent_buffer *left = NULL; | |
2061 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2062 | int ret = 0; |
2063 | int wret; | |
2064 | int pslot; | |
2065 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2066 | |
2067 | if (level == 0) | |
2068 | return 1; | |
2069 | ||
5f39d397 | 2070 | mid = path->nodes[level]; |
7bb86316 | 2071 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2072 | |
a05a9bb1 | 2073 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2074 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2075 | pslot = path->slots[level + 1]; |
2076 | } | |
e66f709b | 2077 | |
5f39d397 | 2078 | if (!parent) |
e66f709b | 2079 | return 1; |
e66f709b | 2080 | |
2ff7e61e | 2081 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
2082 | if (IS_ERR(left)) |
2083 | left = NULL; | |
e66f709b CM |
2084 | |
2085 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2086 | if (left) { |
e66f709b | 2087 | u32 left_nr; |
925baedd CM |
2088 | |
2089 | btrfs_tree_lock(left); | |
8bead258 | 2090 | btrfs_set_lock_blocking_write(left); |
b4ce94de | 2091 | |
5f39d397 | 2092 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2093 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2094 | wret = 1; |
2095 | } else { | |
5f39d397 | 2096 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2097 | pslot - 1, &left); |
54aa1f4d CM |
2098 | if (ret) |
2099 | wret = 1; | |
2100 | else { | |
2ff7e61e | 2101 | wret = push_node_left(trans, fs_info, |
971a1f66 | 2102 | left, mid, 0); |
54aa1f4d | 2103 | } |
33ade1f8 | 2104 | } |
e66f709b CM |
2105 | if (wret < 0) |
2106 | ret = wret; | |
2107 | if (wret == 0) { | |
5f39d397 | 2108 | struct btrfs_disk_key disk_key; |
e66f709b | 2109 | orig_slot += left_nr; |
5f39d397 | 2110 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2111 | ret = tree_mod_log_insert_key(parent, pslot, |
2112 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2113 | BUG_ON(ret < 0); | |
5f39d397 CM |
2114 | btrfs_set_node_key(parent, &disk_key, pslot); |
2115 | btrfs_mark_buffer_dirty(parent); | |
2116 | if (btrfs_header_nritems(left) > orig_slot) { | |
2117 | path->nodes[level] = left; | |
e66f709b CM |
2118 | path->slots[level + 1] -= 1; |
2119 | path->slots[level] = orig_slot; | |
925baedd | 2120 | btrfs_tree_unlock(mid); |
5f39d397 | 2121 | free_extent_buffer(mid); |
e66f709b CM |
2122 | } else { |
2123 | orig_slot -= | |
5f39d397 | 2124 | btrfs_header_nritems(left); |
e66f709b | 2125 | path->slots[level] = orig_slot; |
925baedd | 2126 | btrfs_tree_unlock(left); |
5f39d397 | 2127 | free_extent_buffer(left); |
e66f709b | 2128 | } |
e66f709b CM |
2129 | return 0; |
2130 | } | |
925baedd | 2131 | btrfs_tree_unlock(left); |
5f39d397 | 2132 | free_extent_buffer(left); |
e66f709b | 2133 | } |
2ff7e61e | 2134 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
2135 | if (IS_ERR(right)) |
2136 | right = NULL; | |
e66f709b CM |
2137 | |
2138 | /* | |
2139 | * then try to empty the right most buffer into the middle | |
2140 | */ | |
5f39d397 | 2141 | if (right) { |
33ade1f8 | 2142 | u32 right_nr; |
b4ce94de | 2143 | |
925baedd | 2144 | btrfs_tree_lock(right); |
8bead258 | 2145 | btrfs_set_lock_blocking_write(right); |
b4ce94de | 2146 | |
5f39d397 | 2147 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2148 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2149 | wret = 1; |
2150 | } else { | |
5f39d397 CM |
2151 | ret = btrfs_cow_block(trans, root, right, |
2152 | parent, pslot + 1, | |
9fa8cfe7 | 2153 | &right); |
54aa1f4d CM |
2154 | if (ret) |
2155 | wret = 1; | |
2156 | else { | |
2ff7e61e | 2157 | wret = balance_node_right(trans, fs_info, |
5f39d397 | 2158 | right, mid); |
54aa1f4d | 2159 | } |
33ade1f8 | 2160 | } |
e66f709b CM |
2161 | if (wret < 0) |
2162 | ret = wret; | |
2163 | if (wret == 0) { | |
5f39d397 CM |
2164 | struct btrfs_disk_key disk_key; |
2165 | ||
2166 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2167 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2168 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2169 | BUG_ON(ret < 0); | |
5f39d397 CM |
2170 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2171 | btrfs_mark_buffer_dirty(parent); | |
2172 | ||
2173 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2174 | path->nodes[level] = right; | |
e66f709b CM |
2175 | path->slots[level + 1] += 1; |
2176 | path->slots[level] = orig_slot - | |
5f39d397 | 2177 | btrfs_header_nritems(mid); |
925baedd | 2178 | btrfs_tree_unlock(mid); |
5f39d397 | 2179 | free_extent_buffer(mid); |
e66f709b | 2180 | } else { |
925baedd | 2181 | btrfs_tree_unlock(right); |
5f39d397 | 2182 | free_extent_buffer(right); |
e66f709b | 2183 | } |
e66f709b CM |
2184 | return 0; |
2185 | } | |
925baedd | 2186 | btrfs_tree_unlock(right); |
5f39d397 | 2187 | free_extent_buffer(right); |
e66f709b | 2188 | } |
e66f709b CM |
2189 | return 1; |
2190 | } | |
2191 | ||
3c69faec | 2192 | /* |
d352ac68 CM |
2193 | * readahead one full node of leaves, finding things that are close |
2194 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2195 | */ |
2ff7e61e | 2196 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2197 | struct btrfs_path *path, |
2198 | int level, int slot, u64 objectid) | |
3c69faec | 2199 | { |
5f39d397 | 2200 | struct extent_buffer *node; |
01f46658 | 2201 | struct btrfs_disk_key disk_key; |
3c69faec | 2202 | u32 nritems; |
3c69faec | 2203 | u64 search; |
a7175319 | 2204 | u64 target; |
6b80053d | 2205 | u64 nread = 0; |
5f39d397 | 2206 | struct extent_buffer *eb; |
6b80053d CM |
2207 | u32 nr; |
2208 | u32 blocksize; | |
2209 | u32 nscan = 0; | |
db94535d | 2210 | |
a6b6e75e | 2211 | if (level != 1) |
6702ed49 CM |
2212 | return; |
2213 | ||
2214 | if (!path->nodes[level]) | |
3c69faec CM |
2215 | return; |
2216 | ||
5f39d397 | 2217 | node = path->nodes[level]; |
925baedd | 2218 | |
3c69faec | 2219 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2220 | blocksize = fs_info->nodesize; |
2221 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2222 | if (eb) { |
2223 | free_extent_buffer(eb); | |
3c69faec CM |
2224 | return; |
2225 | } | |
2226 | ||
a7175319 | 2227 | target = search; |
6b80053d | 2228 | |
5f39d397 | 2229 | nritems = btrfs_header_nritems(node); |
6b80053d | 2230 | nr = slot; |
25b8b936 | 2231 | |
d397712b | 2232 | while (1) { |
e4058b54 | 2233 | if (path->reada == READA_BACK) { |
6b80053d CM |
2234 | if (nr == 0) |
2235 | break; | |
2236 | nr--; | |
e4058b54 | 2237 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2238 | nr++; |
2239 | if (nr >= nritems) | |
2240 | break; | |
3c69faec | 2241 | } |
e4058b54 | 2242 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2243 | btrfs_node_key(node, &disk_key, nr); |
2244 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2245 | break; | |
2246 | } | |
6b80053d | 2247 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2248 | if ((search <= target && target - search <= 65536) || |
2249 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2250 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2251 | nread += blocksize; |
2252 | } | |
2253 | nscan++; | |
a7175319 | 2254 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2255 | break; |
3c69faec CM |
2256 | } |
2257 | } | |
925baedd | 2258 | |
2ff7e61e | 2259 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2260 | struct btrfs_path *path, int level) |
b4ce94de CM |
2261 | { |
2262 | int slot; | |
2263 | int nritems; | |
2264 | struct extent_buffer *parent; | |
2265 | struct extent_buffer *eb; | |
2266 | u64 gen; | |
2267 | u64 block1 = 0; | |
2268 | u64 block2 = 0; | |
b4ce94de | 2269 | |
8c594ea8 | 2270 | parent = path->nodes[level + 1]; |
b4ce94de | 2271 | if (!parent) |
0b08851f | 2272 | return; |
b4ce94de CM |
2273 | |
2274 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2275 | slot = path->slots[level + 1]; |
b4ce94de CM |
2276 | |
2277 | if (slot > 0) { | |
2278 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2279 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2280 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2281 | /* |
2282 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2283 | * don't want to return eagain here. That will loop | |
2284 | * forever | |
2285 | */ | |
2286 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2287 | block1 = 0; |
2288 | free_extent_buffer(eb); | |
2289 | } | |
8c594ea8 | 2290 | if (slot + 1 < nritems) { |
b4ce94de CM |
2291 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2292 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2293 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2294 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2295 | block2 = 0; |
2296 | free_extent_buffer(eb); | |
2297 | } | |
8c594ea8 | 2298 | |
0b08851f | 2299 | if (block1) |
2ff7e61e | 2300 | readahead_tree_block(fs_info, block1); |
0b08851f | 2301 | if (block2) |
2ff7e61e | 2302 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2303 | } |
2304 | ||
2305 | ||
d352ac68 | 2306 | /* |
d397712b CM |
2307 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2308 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2309 | * operations on the tree might require changing key pointers higher up in the | |
2310 | * tree. | |
d352ac68 | 2311 | * |
d397712b CM |
2312 | * callers might also have set path->keep_locks, which tells this code to keep |
2313 | * the lock if the path points to the last slot in the block. This is part of | |
2314 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2315 | * |
d397712b CM |
2316 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2317 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2318 | */ |
e02119d5 | 2319 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2320 | int lowest_unlock, int min_write_lock_level, |
2321 | int *write_lock_level) | |
925baedd CM |
2322 | { |
2323 | int i; | |
2324 | int skip_level = level; | |
051e1b9f | 2325 | int no_skips = 0; |
925baedd CM |
2326 | struct extent_buffer *t; |
2327 | ||
2328 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2329 | if (!path->nodes[i]) | |
2330 | break; | |
2331 | if (!path->locks[i]) | |
2332 | break; | |
051e1b9f | 2333 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2334 | skip_level = i + 1; |
2335 | continue; | |
2336 | } | |
051e1b9f | 2337 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2338 | u32 nritems; |
2339 | t = path->nodes[i]; | |
2340 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2341 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2342 | skip_level = i + 1; |
2343 | continue; | |
2344 | } | |
2345 | } | |
051e1b9f CM |
2346 | if (skip_level < i && i >= lowest_unlock) |
2347 | no_skips = 1; | |
2348 | ||
925baedd | 2349 | t = path->nodes[i]; |
d80bb3f9 | 2350 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2351 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2352 | path->locks[i] = 0; |
f7c79f30 CM |
2353 | if (write_lock_level && |
2354 | i > min_write_lock_level && | |
2355 | i <= *write_lock_level) { | |
2356 | *write_lock_level = i - 1; | |
2357 | } | |
925baedd CM |
2358 | } |
2359 | } | |
2360 | } | |
2361 | ||
b4ce94de CM |
2362 | /* |
2363 | * This releases any locks held in the path starting at level and | |
2364 | * going all the way up to the root. | |
2365 | * | |
2366 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2367 | * corner cases, such as COW of the block at slot zero in the node. This | |
2368 | * ignores those rules, and it should only be called when there are no | |
2369 | * more updates to be done higher up in the tree. | |
2370 | */ | |
2371 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2372 | { | |
2373 | int i; | |
2374 | ||
09a2a8f9 | 2375 | if (path->keep_locks) |
b4ce94de CM |
2376 | return; |
2377 | ||
2378 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2379 | if (!path->nodes[i]) | |
12f4dacc | 2380 | continue; |
b4ce94de | 2381 | if (!path->locks[i]) |
12f4dacc | 2382 | continue; |
bd681513 | 2383 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2384 | path->locks[i] = 0; |
2385 | } | |
2386 | } | |
2387 | ||
c8c42864 CM |
2388 | /* |
2389 | * helper function for btrfs_search_slot. The goal is to find a block | |
2390 | * in cache without setting the path to blocking. If we find the block | |
2391 | * we return zero and the path is unchanged. | |
2392 | * | |
2393 | * If we can't find the block, we set the path blocking and do some | |
2394 | * reada. -EAGAIN is returned and the search must be repeated. | |
2395 | */ | |
2396 | static int | |
d07b8528 LB |
2397 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2398 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2399 | const struct btrfs_key *key) |
c8c42864 | 2400 | { |
0b246afa | 2401 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2402 | u64 blocknr; |
2403 | u64 gen; | |
c8c42864 CM |
2404 | struct extent_buffer *b = *eb_ret; |
2405 | struct extent_buffer *tmp; | |
581c1760 | 2406 | struct btrfs_key first_key; |
76a05b35 | 2407 | int ret; |
581c1760 | 2408 | int parent_level; |
c8c42864 CM |
2409 | |
2410 | blocknr = btrfs_node_blockptr(b, slot); | |
2411 | gen = btrfs_node_ptr_generation(b, slot); | |
581c1760 QW |
2412 | parent_level = btrfs_header_level(b); |
2413 | btrfs_node_key_to_cpu(b, &first_key, slot); | |
c8c42864 | 2414 | |
0b246afa | 2415 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2416 | if (tmp) { |
b9fab919 | 2417 | /* first we do an atomic uptodate check */ |
bdf7c00e | 2418 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
2419 | /* |
2420 | * Do extra check for first_key, eb can be stale due to | |
2421 | * being cached, read from scrub, or have multiple | |
2422 | * parents (shared tree blocks). | |
2423 | */ | |
2424 | if (btrfs_verify_level_key(fs_info, tmp, | |
2425 | parent_level - 1, &first_key, gen)) { | |
2426 | free_extent_buffer(tmp); | |
2427 | return -EUCLEAN; | |
2428 | } | |
bdf7c00e JB |
2429 | *eb_ret = tmp; |
2430 | return 0; | |
2431 | } | |
2432 | ||
2433 | /* the pages were up to date, but we failed | |
2434 | * the generation number check. Do a full | |
2435 | * read for the generation number that is correct. | |
2436 | * We must do this without dropping locks so | |
2437 | * we can trust our generation number | |
2438 | */ | |
2439 | btrfs_set_path_blocking(p); | |
2440 | ||
2441 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2442 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2443 | if (!ret) { |
2444 | *eb_ret = tmp; | |
2445 | return 0; | |
cb44921a | 2446 | } |
bdf7c00e JB |
2447 | free_extent_buffer(tmp); |
2448 | btrfs_release_path(p); | |
2449 | return -EIO; | |
c8c42864 CM |
2450 | } |
2451 | ||
2452 | /* | |
2453 | * reduce lock contention at high levels | |
2454 | * of the btree by dropping locks before | |
76a05b35 CM |
2455 | * we read. Don't release the lock on the current |
2456 | * level because we need to walk this node to figure | |
2457 | * out which blocks to read. | |
c8c42864 | 2458 | */ |
8c594ea8 CM |
2459 | btrfs_unlock_up_safe(p, level + 1); |
2460 | btrfs_set_path_blocking(p); | |
2461 | ||
e4058b54 | 2462 | if (p->reada != READA_NONE) |
2ff7e61e | 2463 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2464 | |
76a05b35 | 2465 | ret = -EAGAIN; |
02a3307a | 2466 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2467 | &first_key); |
64c043de | 2468 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2469 | /* |
2470 | * If the read above didn't mark this buffer up to date, | |
2471 | * it will never end up being up to date. Set ret to EIO now | |
2472 | * and give up so that our caller doesn't loop forever | |
2473 | * on our EAGAINs. | |
2474 | */ | |
e6a1d6fd | 2475 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2476 | ret = -EIO; |
c8c42864 | 2477 | free_extent_buffer(tmp); |
c871b0f2 LB |
2478 | } else { |
2479 | ret = PTR_ERR(tmp); | |
76a05b35 | 2480 | } |
02a3307a LB |
2481 | |
2482 | btrfs_release_path(p); | |
76a05b35 | 2483 | return ret; |
c8c42864 CM |
2484 | } |
2485 | ||
2486 | /* | |
2487 | * helper function for btrfs_search_slot. This does all of the checks | |
2488 | * for node-level blocks and does any balancing required based on | |
2489 | * the ins_len. | |
2490 | * | |
2491 | * If no extra work was required, zero is returned. If we had to | |
2492 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2493 | * start over | |
2494 | */ | |
2495 | static int | |
2496 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2497 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2498 | struct extent_buffer *b, int level, int ins_len, |
2499 | int *write_lock_level) | |
c8c42864 | 2500 | { |
0b246afa | 2501 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2502 | int ret; |
0b246afa | 2503 | |
c8c42864 | 2504 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2505 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2506 | int sret; |
2507 | ||
bd681513 CM |
2508 | if (*write_lock_level < level + 1) { |
2509 | *write_lock_level = level + 1; | |
2510 | btrfs_release_path(p); | |
2511 | goto again; | |
2512 | } | |
2513 | ||
c8c42864 | 2514 | btrfs_set_path_blocking(p); |
2ff7e61e | 2515 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2516 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2517 | |
2518 | BUG_ON(sret > 0); | |
2519 | if (sret) { | |
2520 | ret = sret; | |
2521 | goto done; | |
2522 | } | |
2523 | b = p->nodes[level]; | |
2524 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2525 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2526 | int sret; |
2527 | ||
bd681513 CM |
2528 | if (*write_lock_level < level + 1) { |
2529 | *write_lock_level = level + 1; | |
2530 | btrfs_release_path(p); | |
2531 | goto again; | |
2532 | } | |
2533 | ||
c8c42864 | 2534 | btrfs_set_path_blocking(p); |
2ff7e61e | 2535 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2536 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2537 | |
2538 | if (sret) { | |
2539 | ret = sret; | |
2540 | goto done; | |
2541 | } | |
2542 | b = p->nodes[level]; | |
2543 | if (!b) { | |
b3b4aa74 | 2544 | btrfs_release_path(p); |
c8c42864 CM |
2545 | goto again; |
2546 | } | |
2547 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2548 | } | |
2549 | return 0; | |
2550 | ||
2551 | again: | |
2552 | ret = -EAGAIN; | |
2553 | done: | |
2554 | return ret; | |
2555 | } | |
2556 | ||
310712b2 | 2557 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2558 | int level, int *prev_cmp, int *slot) |
2559 | { | |
2560 | if (*prev_cmp != 0) { | |
a74b35ec | 2561 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2562 | return *prev_cmp; |
2563 | } | |
2564 | ||
d7396f07 FDBM |
2565 | *slot = 0; |
2566 | ||
2567 | return 0; | |
2568 | } | |
2569 | ||
381cf658 | 2570 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2571 | u64 iobjectid, u64 ioff, u8 key_type, |
2572 | struct btrfs_key *found_key) | |
2573 | { | |
2574 | int ret; | |
2575 | struct btrfs_key key; | |
2576 | struct extent_buffer *eb; | |
381cf658 DS |
2577 | |
2578 | ASSERT(path); | |
1d4c08e0 | 2579 | ASSERT(found_key); |
e33d5c3d KN |
2580 | |
2581 | key.type = key_type; | |
2582 | key.objectid = iobjectid; | |
2583 | key.offset = ioff; | |
2584 | ||
2585 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2586 | if (ret < 0) |
e33d5c3d KN |
2587 | return ret; |
2588 | ||
2589 | eb = path->nodes[0]; | |
2590 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2591 | ret = btrfs_next_leaf(fs_root, path); | |
2592 | if (ret) | |
2593 | return ret; | |
2594 | eb = path->nodes[0]; | |
2595 | } | |
2596 | ||
2597 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2598 | if (found_key->type != key.type || | |
2599 | found_key->objectid != key.objectid) | |
2600 | return 1; | |
2601 | ||
2602 | return 0; | |
2603 | } | |
2604 | ||
1fc28d8e LB |
2605 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2606 | struct btrfs_path *p, | |
2607 | int write_lock_level) | |
2608 | { | |
2609 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2610 | struct extent_buffer *b; | |
2611 | int root_lock; | |
2612 | int level = 0; | |
2613 | ||
2614 | /* We try very hard to do read locks on the root */ | |
2615 | root_lock = BTRFS_READ_LOCK; | |
2616 | ||
2617 | if (p->search_commit_root) { | |
be6821f8 FM |
2618 | /* |
2619 | * The commit roots are read only so we always do read locks, | |
2620 | * and we always must hold the commit_root_sem when doing | |
2621 | * searches on them, the only exception is send where we don't | |
2622 | * want to block transaction commits for a long time, so | |
2623 | * we need to clone the commit root in order to avoid races | |
2624 | * with transaction commits that create a snapshot of one of | |
2625 | * the roots used by a send operation. | |
2626 | */ | |
2627 | if (p->need_commit_sem) { | |
1fc28d8e | 2628 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 2629 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 2630 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
2631 | if (!b) |
2632 | return ERR_PTR(-ENOMEM); | |
2633 | ||
2634 | } else { | |
2635 | b = root->commit_root; | |
2636 | extent_buffer_get(b); | |
2637 | } | |
2638 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
2639 | /* |
2640 | * Ensure that all callers have set skip_locking when | |
2641 | * p->search_commit_root = 1. | |
2642 | */ | |
2643 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2644 | |
2645 | goto out; | |
2646 | } | |
2647 | ||
2648 | if (p->skip_locking) { | |
2649 | b = btrfs_root_node(root); | |
2650 | level = btrfs_header_level(b); | |
2651 | goto out; | |
2652 | } | |
2653 | ||
2654 | /* | |
662c653b LB |
2655 | * If the level is set to maximum, we can skip trying to get the read |
2656 | * lock. | |
1fc28d8e | 2657 | */ |
662c653b LB |
2658 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2659 | /* | |
2660 | * We don't know the level of the root node until we actually | |
2661 | * have it read locked | |
2662 | */ | |
2663 | b = btrfs_read_lock_root_node(root); | |
2664 | level = btrfs_header_level(b); | |
2665 | if (level > write_lock_level) | |
2666 | goto out; | |
2667 | ||
2668 | /* Whoops, must trade for write lock */ | |
2669 | btrfs_tree_read_unlock(b); | |
2670 | free_extent_buffer(b); | |
2671 | } | |
1fc28d8e | 2672 | |
1fc28d8e LB |
2673 | b = btrfs_lock_root_node(root); |
2674 | root_lock = BTRFS_WRITE_LOCK; | |
2675 | ||
2676 | /* The level might have changed, check again */ | |
2677 | level = btrfs_header_level(b); | |
2678 | ||
2679 | out: | |
2680 | p->nodes[level] = b; | |
2681 | if (!p->skip_locking) | |
2682 | p->locks[level] = root_lock; | |
2683 | /* | |
2684 | * Callers are responsible for dropping b's references. | |
2685 | */ | |
2686 | return b; | |
2687 | } | |
2688 | ||
2689 | ||
74123bd7 | 2690 | /* |
4271ecea NB |
2691 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2692 | * modifications to preserve tree invariants. | |
74123bd7 | 2693 | * |
4271ecea NB |
2694 | * @trans: Handle of transaction, used when modifying the tree |
2695 | * @p: Holds all btree nodes along the search path | |
2696 | * @root: The root node of the tree | |
2697 | * @key: The key we are looking for | |
2698 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2699 | * deletions it's -1. 0 for plain searches | |
2700 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2701 | * when modifying the tree. | |
97571fd0 | 2702 | * |
4271ecea NB |
2703 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2704 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2705 | * | |
2706 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2707 | * of the path (level 0) | |
2708 | * | |
2709 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2710 | * points to the slot where it should be inserted | |
2711 | * | |
2712 | * If an error is encountered while searching the tree a negative error number | |
2713 | * is returned | |
74123bd7 | 2714 | */ |
310712b2 OS |
2715 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2716 | const struct btrfs_key *key, struct btrfs_path *p, | |
2717 | int ins_len, int cow) | |
be0e5c09 | 2718 | { |
0b246afa | 2719 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2720 | struct extent_buffer *b; |
be0e5c09 CM |
2721 | int slot; |
2722 | int ret; | |
33c66f43 | 2723 | int err; |
be0e5c09 | 2724 | int level; |
925baedd | 2725 | int lowest_unlock = 1; |
bd681513 CM |
2726 | /* everything at write_lock_level or lower must be write locked */ |
2727 | int write_lock_level = 0; | |
9f3a7427 | 2728 | u8 lowest_level = 0; |
f7c79f30 | 2729 | int min_write_lock_level; |
d7396f07 | 2730 | int prev_cmp; |
9f3a7427 | 2731 | |
6702ed49 | 2732 | lowest_level = p->lowest_level; |
323ac95b | 2733 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2734 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2735 | BUG_ON(!cow && ins_len); |
25179201 | 2736 | |
bd681513 | 2737 | if (ins_len < 0) { |
925baedd | 2738 | lowest_unlock = 2; |
65b51a00 | 2739 | |
bd681513 CM |
2740 | /* when we are removing items, we might have to go up to level |
2741 | * two as we update tree pointers Make sure we keep write | |
2742 | * for those levels as well | |
2743 | */ | |
2744 | write_lock_level = 2; | |
2745 | } else if (ins_len > 0) { | |
2746 | /* | |
2747 | * for inserting items, make sure we have a write lock on | |
2748 | * level 1 so we can update keys | |
2749 | */ | |
2750 | write_lock_level = 1; | |
2751 | } | |
2752 | ||
2753 | if (!cow) | |
2754 | write_lock_level = -1; | |
2755 | ||
09a2a8f9 | 2756 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2757 | write_lock_level = BTRFS_MAX_LEVEL; |
2758 | ||
f7c79f30 CM |
2759 | min_write_lock_level = write_lock_level; |
2760 | ||
bb803951 | 2761 | again: |
d7396f07 | 2762 | prev_cmp = -1; |
1fc28d8e | 2763 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2764 | if (IS_ERR(b)) { |
2765 | ret = PTR_ERR(b); | |
2766 | goto done; | |
2767 | } | |
925baedd | 2768 | |
eb60ceac | 2769 | while (b) { |
5f39d397 | 2770 | level = btrfs_header_level(b); |
65b51a00 CM |
2771 | |
2772 | /* | |
2773 | * setup the path here so we can release it under lock | |
2774 | * contention with the cow code | |
2775 | */ | |
02217ed2 | 2776 | if (cow) { |
9ea2c7c9 NB |
2777 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2778 | ||
c8c42864 CM |
2779 | /* |
2780 | * if we don't really need to cow this block | |
2781 | * then we don't want to set the path blocking, | |
2782 | * so we test it here | |
2783 | */ | |
64c12921 JM |
2784 | if (!should_cow_block(trans, root, b)) { |
2785 | trans->dirty = true; | |
65b51a00 | 2786 | goto cow_done; |
64c12921 | 2787 | } |
5d4f98a2 | 2788 | |
bd681513 CM |
2789 | /* |
2790 | * must have write locks on this node and the | |
2791 | * parent | |
2792 | */ | |
5124e00e JB |
2793 | if (level > write_lock_level || |
2794 | (level + 1 > write_lock_level && | |
2795 | level + 1 < BTRFS_MAX_LEVEL && | |
2796 | p->nodes[level + 1])) { | |
bd681513 CM |
2797 | write_lock_level = level + 1; |
2798 | btrfs_release_path(p); | |
2799 | goto again; | |
2800 | } | |
2801 | ||
160f4089 | 2802 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2803 | if (last_level) |
2804 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2805 | &b); | |
2806 | else | |
2807 | err = btrfs_cow_block(trans, root, b, | |
2808 | p->nodes[level + 1], | |
2809 | p->slots[level + 1], &b); | |
33c66f43 | 2810 | if (err) { |
33c66f43 | 2811 | ret = err; |
65b51a00 | 2812 | goto done; |
54aa1f4d | 2813 | } |
02217ed2 | 2814 | } |
65b51a00 | 2815 | cow_done: |
eb60ceac | 2816 | p->nodes[level] = b; |
52398340 LB |
2817 | /* |
2818 | * Leave path with blocking locks to avoid massive | |
2819 | * lock context switch, this is made on purpose. | |
2820 | */ | |
b4ce94de CM |
2821 | |
2822 | /* | |
2823 | * we have a lock on b and as long as we aren't changing | |
2824 | * the tree, there is no way to for the items in b to change. | |
2825 | * It is safe to drop the lock on our parent before we | |
2826 | * go through the expensive btree search on b. | |
2827 | * | |
eb653de1 FDBM |
2828 | * If we're inserting or deleting (ins_len != 0), then we might |
2829 | * be changing slot zero, which may require changing the parent. | |
2830 | * So, we can't drop the lock until after we know which slot | |
2831 | * we're operating on. | |
b4ce94de | 2832 | */ |
eb653de1 FDBM |
2833 | if (!ins_len && !p->keep_locks) { |
2834 | int u = level + 1; | |
2835 | ||
2836 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2837 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2838 | p->locks[u] = 0; | |
2839 | } | |
2840 | } | |
b4ce94de | 2841 | |
d7396f07 | 2842 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2843 | if (ret < 0) |
2844 | goto done; | |
b4ce94de | 2845 | |
5f39d397 | 2846 | if (level != 0) { |
33c66f43 YZ |
2847 | int dec = 0; |
2848 | if (ret && slot > 0) { | |
2849 | dec = 1; | |
be0e5c09 | 2850 | slot -= 1; |
33c66f43 | 2851 | } |
be0e5c09 | 2852 | p->slots[level] = slot; |
33c66f43 | 2853 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2854 | ins_len, &write_lock_level); |
33c66f43 | 2855 | if (err == -EAGAIN) |
c8c42864 | 2856 | goto again; |
33c66f43 YZ |
2857 | if (err) { |
2858 | ret = err; | |
c8c42864 | 2859 | goto done; |
33c66f43 | 2860 | } |
c8c42864 CM |
2861 | b = p->nodes[level]; |
2862 | slot = p->slots[level]; | |
b4ce94de | 2863 | |
bd681513 CM |
2864 | /* |
2865 | * slot 0 is special, if we change the key | |
2866 | * we have to update the parent pointer | |
2867 | * which means we must have a write lock | |
2868 | * on the parent | |
2869 | */ | |
eb653de1 | 2870 | if (slot == 0 && ins_len && |
bd681513 CM |
2871 | write_lock_level < level + 1) { |
2872 | write_lock_level = level + 1; | |
2873 | btrfs_release_path(p); | |
2874 | goto again; | |
2875 | } | |
2876 | ||
f7c79f30 CM |
2877 | unlock_up(p, level, lowest_unlock, |
2878 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2879 | |
925baedd | 2880 | if (level == lowest_level) { |
33c66f43 YZ |
2881 | if (dec) |
2882 | p->slots[level]++; | |
5b21f2ed | 2883 | goto done; |
925baedd | 2884 | } |
ca7a79ad | 2885 | |
d07b8528 | 2886 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2887 | slot, key); |
33c66f43 | 2888 | if (err == -EAGAIN) |
c8c42864 | 2889 | goto again; |
33c66f43 YZ |
2890 | if (err) { |
2891 | ret = err; | |
76a05b35 | 2892 | goto done; |
33c66f43 | 2893 | } |
76a05b35 | 2894 | |
b4ce94de | 2895 | if (!p->skip_locking) { |
bd681513 CM |
2896 | level = btrfs_header_level(b); |
2897 | if (level <= write_lock_level) { | |
2898 | err = btrfs_try_tree_write_lock(b); | |
2899 | if (!err) { | |
2900 | btrfs_set_path_blocking(p); | |
2901 | btrfs_tree_lock(b); | |
bd681513 CM |
2902 | } |
2903 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2904 | } else { | |
f82c458a | 2905 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2906 | if (!err) { |
2907 | btrfs_set_path_blocking(p); | |
2908 | btrfs_tree_read_lock(b); | |
bd681513 CM |
2909 | } |
2910 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2911 | } |
bd681513 | 2912 | p->nodes[level] = b; |
b4ce94de | 2913 | } |
be0e5c09 CM |
2914 | } else { |
2915 | p->slots[level] = slot; | |
87b29b20 | 2916 | if (ins_len > 0 && |
2ff7e61e | 2917 | btrfs_leaf_free_space(fs_info, b) < ins_len) { |
bd681513 CM |
2918 | if (write_lock_level < 1) { |
2919 | write_lock_level = 1; | |
2920 | btrfs_release_path(p); | |
2921 | goto again; | |
2922 | } | |
2923 | ||
b4ce94de | 2924 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2925 | err = split_leaf(trans, root, key, |
2926 | p, ins_len, ret == 0); | |
b4ce94de | 2927 | |
33c66f43 YZ |
2928 | BUG_ON(err > 0); |
2929 | if (err) { | |
2930 | ret = err; | |
65b51a00 CM |
2931 | goto done; |
2932 | } | |
5c680ed6 | 2933 | } |
459931ec | 2934 | if (!p->search_for_split) |
f7c79f30 | 2935 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2936 | min_write_lock_level, NULL); |
65b51a00 | 2937 | goto done; |
be0e5c09 CM |
2938 | } |
2939 | } | |
65b51a00 CM |
2940 | ret = 1; |
2941 | done: | |
b4ce94de CM |
2942 | /* |
2943 | * we don't really know what they plan on doing with the path | |
2944 | * from here on, so for now just mark it as blocking | |
2945 | */ | |
b9473439 CM |
2946 | if (!p->leave_spinning) |
2947 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2948 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2949 | btrfs_release_path(p); |
65b51a00 | 2950 | return ret; |
be0e5c09 CM |
2951 | } |
2952 | ||
5d9e75c4 JS |
2953 | /* |
2954 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2955 | * current state of the tree together with the operations recorded in the tree | |
2956 | * modification log to search for the key in a previous version of this tree, as | |
2957 | * denoted by the time_seq parameter. | |
2958 | * | |
2959 | * Naturally, there is no support for insert, delete or cow operations. | |
2960 | * | |
2961 | * The resulting path and return value will be set up as if we called | |
2962 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2963 | */ | |
310712b2 | 2964 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2965 | struct btrfs_path *p, u64 time_seq) |
2966 | { | |
0b246afa | 2967 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2968 | struct extent_buffer *b; |
2969 | int slot; | |
2970 | int ret; | |
2971 | int err; | |
2972 | int level; | |
2973 | int lowest_unlock = 1; | |
2974 | u8 lowest_level = 0; | |
d4b4087c | 2975 | int prev_cmp = -1; |
5d9e75c4 JS |
2976 | |
2977 | lowest_level = p->lowest_level; | |
2978 | WARN_ON(p->nodes[0] != NULL); | |
2979 | ||
2980 | if (p->search_commit_root) { | |
2981 | BUG_ON(time_seq); | |
2982 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2983 | } | |
2984 | ||
2985 | again: | |
5d9e75c4 | 2986 | b = get_old_root(root, time_seq); |
315bed43 NB |
2987 | if (!b) { |
2988 | ret = -EIO; | |
2989 | goto done; | |
2990 | } | |
5d9e75c4 | 2991 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2992 | p->locks[level] = BTRFS_READ_LOCK; |
2993 | ||
2994 | while (b) { | |
2995 | level = btrfs_header_level(b); | |
2996 | p->nodes[level] = b; | |
5d9e75c4 JS |
2997 | |
2998 | /* | |
2999 | * we have a lock on b and as long as we aren't changing | |
3000 | * the tree, there is no way to for the items in b to change. | |
3001 | * It is safe to drop the lock on our parent before we | |
3002 | * go through the expensive btree search on b. | |
3003 | */ | |
3004 | btrfs_unlock_up_safe(p, level + 1); | |
3005 | ||
d4b4087c | 3006 | /* |
01327610 | 3007 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
3008 | * time. |
3009 | */ | |
3010 | prev_cmp = -1; | |
d7396f07 | 3011 | ret = key_search(b, key, level, &prev_cmp, &slot); |
cbca7d59 FM |
3012 | if (ret < 0) |
3013 | goto done; | |
5d9e75c4 JS |
3014 | |
3015 | if (level != 0) { | |
3016 | int dec = 0; | |
3017 | if (ret && slot > 0) { | |
3018 | dec = 1; | |
3019 | slot -= 1; | |
3020 | } | |
3021 | p->slots[level] = slot; | |
3022 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3023 | ||
3024 | if (level == lowest_level) { | |
3025 | if (dec) | |
3026 | p->slots[level]++; | |
3027 | goto done; | |
3028 | } | |
3029 | ||
d07b8528 | 3030 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 3031 | slot, key); |
5d9e75c4 JS |
3032 | if (err == -EAGAIN) |
3033 | goto again; | |
3034 | if (err) { | |
3035 | ret = err; | |
3036 | goto done; | |
3037 | } | |
3038 | ||
3039 | level = btrfs_header_level(b); | |
f82c458a | 3040 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3041 | if (!err) { |
3042 | btrfs_set_path_blocking(p); | |
3043 | btrfs_tree_read_lock(b); | |
5d9e75c4 | 3044 | } |
0b246afa | 3045 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3046 | if (!b) { |
3047 | ret = -ENOMEM; | |
3048 | goto done; | |
3049 | } | |
5d9e75c4 JS |
3050 | p->locks[level] = BTRFS_READ_LOCK; |
3051 | p->nodes[level] = b; | |
5d9e75c4 JS |
3052 | } else { |
3053 | p->slots[level] = slot; | |
3054 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3055 | goto done; | |
3056 | } | |
3057 | } | |
3058 | ret = 1; | |
3059 | done: | |
3060 | if (!p->leave_spinning) | |
3061 | btrfs_set_path_blocking(p); | |
3062 | if (ret < 0) | |
3063 | btrfs_release_path(p); | |
3064 | ||
3065 | return ret; | |
3066 | } | |
3067 | ||
2f38b3e1 AJ |
3068 | /* |
3069 | * helper to use instead of search slot if no exact match is needed but | |
3070 | * instead the next or previous item should be returned. | |
3071 | * When find_higher is true, the next higher item is returned, the next lower | |
3072 | * otherwise. | |
3073 | * When return_any and find_higher are both true, and no higher item is found, | |
3074 | * return the next lower instead. | |
3075 | * When return_any is true and find_higher is false, and no lower item is found, | |
3076 | * return the next higher instead. | |
3077 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3078 | * < 0 on error | |
3079 | */ | |
3080 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3081 | const struct btrfs_key *key, |
3082 | struct btrfs_path *p, int find_higher, | |
3083 | int return_any) | |
2f38b3e1 AJ |
3084 | { |
3085 | int ret; | |
3086 | struct extent_buffer *leaf; | |
3087 | ||
3088 | again: | |
3089 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3090 | if (ret <= 0) | |
3091 | return ret; | |
3092 | /* | |
3093 | * a return value of 1 means the path is at the position where the | |
3094 | * item should be inserted. Normally this is the next bigger item, | |
3095 | * but in case the previous item is the last in a leaf, path points | |
3096 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3097 | * item. | |
3098 | */ | |
3099 | leaf = p->nodes[0]; | |
3100 | ||
3101 | if (find_higher) { | |
3102 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3103 | ret = btrfs_next_leaf(root, p); | |
3104 | if (ret <= 0) | |
3105 | return ret; | |
3106 | if (!return_any) | |
3107 | return 1; | |
3108 | /* | |
3109 | * no higher item found, return the next | |
3110 | * lower instead | |
3111 | */ | |
3112 | return_any = 0; | |
3113 | find_higher = 0; | |
3114 | btrfs_release_path(p); | |
3115 | goto again; | |
3116 | } | |
3117 | } else { | |
e6793769 AJ |
3118 | if (p->slots[0] == 0) { |
3119 | ret = btrfs_prev_leaf(root, p); | |
3120 | if (ret < 0) | |
3121 | return ret; | |
3122 | if (!ret) { | |
23c6bf6a FDBM |
3123 | leaf = p->nodes[0]; |
3124 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3125 | p->slots[0]--; | |
e6793769 | 3126 | return 0; |
2f38b3e1 | 3127 | } |
e6793769 AJ |
3128 | if (!return_any) |
3129 | return 1; | |
3130 | /* | |
3131 | * no lower item found, return the next | |
3132 | * higher instead | |
3133 | */ | |
3134 | return_any = 0; | |
3135 | find_higher = 1; | |
3136 | btrfs_release_path(p); | |
3137 | goto again; | |
3138 | } else { | |
2f38b3e1 AJ |
3139 | --p->slots[0]; |
3140 | } | |
3141 | } | |
3142 | return 0; | |
3143 | } | |
3144 | ||
74123bd7 CM |
3145 | /* |
3146 | * adjust the pointers going up the tree, starting at level | |
3147 | * making sure the right key of each node is points to 'key'. | |
3148 | * This is used after shifting pointers to the left, so it stops | |
3149 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3150 | * higher levels | |
aa5d6bed | 3151 | * |
74123bd7 | 3152 | */ |
b167fa91 | 3153 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3154 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3155 | { |
3156 | int i; | |
5f39d397 | 3157 | struct extent_buffer *t; |
0e82bcfe | 3158 | int ret; |
5f39d397 | 3159 | |
234b63a0 | 3160 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3161 | int tslot = path->slots[i]; |
0e82bcfe | 3162 | |
eb60ceac | 3163 | if (!path->nodes[i]) |
be0e5c09 | 3164 | break; |
5f39d397 | 3165 | t = path->nodes[i]; |
0e82bcfe DS |
3166 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3167 | GFP_ATOMIC); | |
3168 | BUG_ON(ret < 0); | |
5f39d397 | 3169 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3170 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3171 | if (tslot != 0) |
3172 | break; | |
3173 | } | |
3174 | } | |
3175 | ||
31840ae1 ZY |
3176 | /* |
3177 | * update item key. | |
3178 | * | |
3179 | * This function isn't completely safe. It's the caller's responsibility | |
3180 | * that the new key won't break the order | |
3181 | */ | |
b7a0365e DD |
3182 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3183 | struct btrfs_path *path, | |
310712b2 | 3184 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3185 | { |
3186 | struct btrfs_disk_key disk_key; | |
3187 | struct extent_buffer *eb; | |
3188 | int slot; | |
3189 | ||
3190 | eb = path->nodes[0]; | |
3191 | slot = path->slots[0]; | |
3192 | if (slot > 0) { | |
3193 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3194 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3195 | } |
3196 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3197 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3198 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3199 | } |
3200 | ||
3201 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3202 | btrfs_set_item_key(eb, &disk_key, slot); | |
3203 | btrfs_mark_buffer_dirty(eb); | |
3204 | if (slot == 0) | |
b167fa91 | 3205 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3206 | } |
3207 | ||
74123bd7 CM |
3208 | /* |
3209 | * try to push data from one node into the next node left in the | |
79f95c82 | 3210 | * tree. |
aa5d6bed CM |
3211 | * |
3212 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3213 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3214 | */ |
98ed5174 | 3215 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
3216 | struct btrfs_fs_info *fs_info, |
3217 | struct extent_buffer *dst, | |
971a1f66 | 3218 | struct extent_buffer *src, int empty) |
be0e5c09 | 3219 | { |
be0e5c09 | 3220 | int push_items = 0; |
bb803951 CM |
3221 | int src_nritems; |
3222 | int dst_nritems; | |
aa5d6bed | 3223 | int ret = 0; |
be0e5c09 | 3224 | |
5f39d397 CM |
3225 | src_nritems = btrfs_header_nritems(src); |
3226 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3227 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3228 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3229 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3230 | |
bce4eae9 | 3231 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3232 | return 1; |
3233 | ||
d397712b | 3234 | if (push_items <= 0) |
be0e5c09 CM |
3235 | return 1; |
3236 | ||
bce4eae9 | 3237 | if (empty) { |
971a1f66 | 3238 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3239 | if (push_items < src_nritems) { |
3240 | /* leave at least 8 pointers in the node if | |
3241 | * we aren't going to empty it | |
3242 | */ | |
3243 | if (src_nritems - push_items < 8) { | |
3244 | if (push_items <= 8) | |
3245 | return 1; | |
3246 | push_items -= 8; | |
3247 | } | |
3248 | } | |
3249 | } else | |
3250 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3251 | |
0b246afa | 3252 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3253 | push_items); |
3254 | if (ret) { | |
66642832 | 3255 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3256 | return ret; |
3257 | } | |
5f39d397 CM |
3258 | copy_extent_buffer(dst, src, |
3259 | btrfs_node_key_ptr_offset(dst_nritems), | |
3260 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3261 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3262 | |
bb803951 | 3263 | if (push_items < src_nritems) { |
57911b8b | 3264 | /* |
bf1d3425 DS |
3265 | * Don't call tree_mod_log_insert_move here, key removal was |
3266 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3267 | */ |
5f39d397 CM |
3268 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3269 | btrfs_node_key_ptr_offset(push_items), | |
3270 | (src_nritems - push_items) * | |
3271 | sizeof(struct btrfs_key_ptr)); | |
3272 | } | |
3273 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3274 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3275 | btrfs_mark_buffer_dirty(src); | |
3276 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3277 | |
79f95c82 CM |
3278 | return ret; |
3279 | } | |
3280 | ||
3281 | /* | |
3282 | * try to push data from one node into the next node right in the | |
3283 | * tree. | |
3284 | * | |
3285 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3286 | * error, and > 0 if there was no room in the right hand block. | |
3287 | * | |
3288 | * this will only push up to 1/2 the contents of the left node over | |
3289 | */ | |
5f39d397 | 3290 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 3291 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
3292 | struct extent_buffer *dst, |
3293 | struct extent_buffer *src) | |
79f95c82 | 3294 | { |
79f95c82 CM |
3295 | int push_items = 0; |
3296 | int max_push; | |
3297 | int src_nritems; | |
3298 | int dst_nritems; | |
3299 | int ret = 0; | |
79f95c82 | 3300 | |
7bb86316 CM |
3301 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3302 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3303 | ||
5f39d397 CM |
3304 | src_nritems = btrfs_header_nritems(src); |
3305 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3306 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3307 | if (push_items <= 0) |
79f95c82 | 3308 | return 1; |
bce4eae9 | 3309 | |
d397712b | 3310 | if (src_nritems < 4) |
bce4eae9 | 3311 | return 1; |
79f95c82 CM |
3312 | |
3313 | max_push = src_nritems / 2 + 1; | |
3314 | /* don't try to empty the node */ | |
d397712b | 3315 | if (max_push >= src_nritems) |
79f95c82 | 3316 | return 1; |
252c38f0 | 3317 | |
79f95c82 CM |
3318 | if (max_push < push_items) |
3319 | push_items = max_push; | |
3320 | ||
bf1d3425 DS |
3321 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3322 | BUG_ON(ret < 0); | |
5f39d397 CM |
3323 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3324 | btrfs_node_key_ptr_offset(0), | |
3325 | (dst_nritems) * | |
3326 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3327 | |
0b246afa | 3328 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3329 | src_nritems - push_items, push_items); |
3330 | if (ret) { | |
66642832 | 3331 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3332 | return ret; |
3333 | } | |
5f39d397 CM |
3334 | copy_extent_buffer(dst, src, |
3335 | btrfs_node_key_ptr_offset(0), | |
3336 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3337 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3338 | |
5f39d397 CM |
3339 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3340 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3341 | |
5f39d397 CM |
3342 | btrfs_mark_buffer_dirty(src); |
3343 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3344 | |
aa5d6bed | 3345 | return ret; |
be0e5c09 CM |
3346 | } |
3347 | ||
97571fd0 CM |
3348 | /* |
3349 | * helper function to insert a new root level in the tree. | |
3350 | * A new node is allocated, and a single item is inserted to | |
3351 | * point to the existing root | |
aa5d6bed CM |
3352 | * |
3353 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3354 | */ |
d397712b | 3355 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3356 | struct btrfs_root *root, |
fdd99c72 | 3357 | struct btrfs_path *path, int level) |
5c680ed6 | 3358 | { |
0b246afa | 3359 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3360 | u64 lower_gen; |
5f39d397 CM |
3361 | struct extent_buffer *lower; |
3362 | struct extent_buffer *c; | |
925baedd | 3363 | struct extent_buffer *old; |
5f39d397 | 3364 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3365 | int ret; |
5c680ed6 CM |
3366 | |
3367 | BUG_ON(path->nodes[level]); | |
3368 | BUG_ON(path->nodes[level-1] != root->node); | |
3369 | ||
7bb86316 CM |
3370 | lower = path->nodes[level-1]; |
3371 | if (level == 1) | |
3372 | btrfs_item_key(lower, &lower_key, 0); | |
3373 | else | |
3374 | btrfs_node_key(lower, &lower_key, 0); | |
3375 | ||
a6279470 FM |
3376 | c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, |
3377 | root->node->start, 0); | |
5f39d397 CM |
3378 | if (IS_ERR(c)) |
3379 | return PTR_ERR(c); | |
925baedd | 3380 | |
0b246afa | 3381 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3382 | |
5f39d397 | 3383 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3384 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3385 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3386 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3387 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3388 | |
3389 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3390 | |
5f39d397 | 3391 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3392 | |
925baedd | 3393 | old = root->node; |
d9d19a01 DS |
3394 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3395 | BUG_ON(ret < 0); | |
240f62c8 | 3396 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3397 | |
3398 | /* the super has an extra ref to root->node */ | |
3399 | free_extent_buffer(old); | |
3400 | ||
0b86a832 | 3401 | add_root_to_dirty_list(root); |
5f39d397 CM |
3402 | extent_buffer_get(c); |
3403 | path->nodes[level] = c; | |
95449a16 | 3404 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3405 | path->slots[level] = 0; |
3406 | return 0; | |
3407 | } | |
3408 | ||
74123bd7 CM |
3409 | /* |
3410 | * worker function to insert a single pointer in a node. | |
3411 | * the node should have enough room for the pointer already | |
97571fd0 | 3412 | * |
74123bd7 CM |
3413 | * slot and level indicate where you want the key to go, and |
3414 | * blocknr is the block the key points to. | |
3415 | */ | |
143bede5 | 3416 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2ff7e61e | 3417 | struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 3418 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3419 | int slot, int level) |
74123bd7 | 3420 | { |
5f39d397 | 3421 | struct extent_buffer *lower; |
74123bd7 | 3422 | int nritems; |
f3ea38da | 3423 | int ret; |
5c680ed6 CM |
3424 | |
3425 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3426 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3427 | lower = path->nodes[level]; |
3428 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3429 | BUG_ON(slot > nritems); |
0b246afa | 3430 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3431 | if (slot != nritems) { |
bf1d3425 DS |
3432 | if (level) { |
3433 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3434 | nritems - slot); |
bf1d3425 DS |
3435 | BUG_ON(ret < 0); |
3436 | } | |
5f39d397 CM |
3437 | memmove_extent_buffer(lower, |
3438 | btrfs_node_key_ptr_offset(slot + 1), | |
3439 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3440 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3441 | } |
c3e06965 | 3442 | if (level) { |
e09c2efe DS |
3443 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3444 | GFP_NOFS); | |
f3ea38da JS |
3445 | BUG_ON(ret < 0); |
3446 | } | |
5f39d397 | 3447 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3448 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3449 | WARN_ON(trans->transid == 0); |
3450 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3451 | btrfs_set_header_nritems(lower, nritems + 1); |
3452 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3453 | } |
3454 | ||
97571fd0 CM |
3455 | /* |
3456 | * split the node at the specified level in path in two. | |
3457 | * The path is corrected to point to the appropriate node after the split | |
3458 | * | |
3459 | * Before splitting this tries to make some room in the node by pushing | |
3460 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3461 | * |
3462 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3463 | */ |
e02119d5 CM |
3464 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3465 | struct btrfs_root *root, | |
3466 | struct btrfs_path *path, int level) | |
be0e5c09 | 3467 | { |
0b246afa | 3468 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3469 | struct extent_buffer *c; |
3470 | struct extent_buffer *split; | |
3471 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3472 | int mid; |
5c680ed6 | 3473 | int ret; |
7518a238 | 3474 | u32 c_nritems; |
eb60ceac | 3475 | |
5f39d397 | 3476 | c = path->nodes[level]; |
7bb86316 | 3477 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3478 | if (c == root->node) { |
d9abbf1c | 3479 | /* |
90f8d62e JS |
3480 | * trying to split the root, lets make a new one |
3481 | * | |
fdd99c72 | 3482 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3483 | * insert_new_root, because that root buffer will be kept as a |
3484 | * normal node. We are going to log removal of half of the | |
3485 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3486 | * tree lock on the buffer, which is why we cannot race with | |
3487 | * other tree_mod_log users. | |
d9abbf1c | 3488 | */ |
fdd99c72 | 3489 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3490 | if (ret) |
3491 | return ret; | |
b3612421 | 3492 | } else { |
e66f709b | 3493 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3494 | c = path->nodes[level]; |
3495 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3496 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3497 | return 0; |
54aa1f4d CM |
3498 | if (ret < 0) |
3499 | return ret; | |
be0e5c09 | 3500 | } |
e66f709b | 3501 | |
5f39d397 | 3502 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3503 | mid = (c_nritems + 1) / 2; |
3504 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3505 | |
a6279470 FM |
3506 | split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, |
3507 | c->start, 0); | |
5f39d397 CM |
3508 | if (IS_ERR(split)) |
3509 | return PTR_ERR(split); | |
3510 | ||
0b246afa | 3511 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3512 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3513 | |
0b246afa | 3514 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3515 | if (ret) { |
66642832 | 3516 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3517 | return ret; |
3518 | } | |
5f39d397 CM |
3519 | copy_extent_buffer(split, c, |
3520 | btrfs_node_key_ptr_offset(0), | |
3521 | btrfs_node_key_ptr_offset(mid), | |
3522 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3523 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3524 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3525 | ret = 0; |
3526 | ||
5f39d397 CM |
3527 | btrfs_mark_buffer_dirty(c); |
3528 | btrfs_mark_buffer_dirty(split); | |
3529 | ||
2ff7e61e | 3530 | insert_ptr(trans, fs_info, path, &disk_key, split->start, |
c3e06965 | 3531 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3532 | |
5de08d7d | 3533 | if (path->slots[level] >= mid) { |
5c680ed6 | 3534 | path->slots[level] -= mid; |
925baedd | 3535 | btrfs_tree_unlock(c); |
5f39d397 CM |
3536 | free_extent_buffer(c); |
3537 | path->nodes[level] = split; | |
5c680ed6 CM |
3538 | path->slots[level + 1] += 1; |
3539 | } else { | |
925baedd | 3540 | btrfs_tree_unlock(split); |
5f39d397 | 3541 | free_extent_buffer(split); |
be0e5c09 | 3542 | } |
aa5d6bed | 3543 | return ret; |
be0e5c09 CM |
3544 | } |
3545 | ||
74123bd7 CM |
3546 | /* |
3547 | * how many bytes are required to store the items in a leaf. start | |
3548 | * and nr indicate which items in the leaf to check. This totals up the | |
3549 | * space used both by the item structs and the item data | |
3550 | */ | |
5f39d397 | 3551 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3552 | { |
41be1f3b JB |
3553 | struct btrfs_item *start_item; |
3554 | struct btrfs_item *end_item; | |
3555 | struct btrfs_map_token token; | |
be0e5c09 | 3556 | int data_len; |
5f39d397 | 3557 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3558 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3559 | |
3560 | if (!nr) | |
3561 | return 0; | |
41be1f3b | 3562 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3563 | start_item = btrfs_item_nr(start); |
3564 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3565 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3566 | btrfs_token_item_size(l, start_item, &token); | |
3567 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3568 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3569 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3570 | return data_len; |
3571 | } | |
3572 | ||
d4dbff95 CM |
3573 | /* |
3574 | * The space between the end of the leaf items and | |
3575 | * the start of the leaf data. IOW, how much room | |
3576 | * the leaf has left for both items and data | |
3577 | */ | |
2ff7e61e | 3578 | noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info, |
e02119d5 | 3579 | struct extent_buffer *leaf) |
d4dbff95 | 3580 | { |
5f39d397 CM |
3581 | int nritems = btrfs_header_nritems(leaf); |
3582 | int ret; | |
0b246afa JM |
3583 | |
3584 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3585 | if (ret < 0) { |
0b246afa JM |
3586 | btrfs_crit(fs_info, |
3587 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3588 | ret, | |
3589 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3590 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3591 | } |
3592 | return ret; | |
d4dbff95 CM |
3593 | } |
3594 | ||
99d8f83c CM |
3595 | /* |
3596 | * min slot controls the lowest index we're willing to push to the | |
3597 | * right. We'll push up to and including min_slot, but no lower | |
3598 | */ | |
1e47eef2 | 3599 | static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3600 | struct btrfs_path *path, |
3601 | int data_size, int empty, | |
3602 | struct extent_buffer *right, | |
99d8f83c CM |
3603 | int free_space, u32 left_nritems, |
3604 | u32 min_slot) | |
00ec4c51 | 3605 | { |
5f39d397 | 3606 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3607 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3608 | struct btrfs_map_token token; |
5f39d397 | 3609 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3610 | int slot; |
34a38218 | 3611 | u32 i; |
00ec4c51 CM |
3612 | int push_space = 0; |
3613 | int push_items = 0; | |
0783fcfc | 3614 | struct btrfs_item *item; |
34a38218 | 3615 | u32 nr; |
7518a238 | 3616 | u32 right_nritems; |
5f39d397 | 3617 | u32 data_end; |
db94535d | 3618 | u32 this_item_size; |
00ec4c51 | 3619 | |
cfed81a0 CM |
3620 | btrfs_init_map_token(&token); |
3621 | ||
34a38218 CM |
3622 | if (empty) |
3623 | nr = 0; | |
3624 | else | |
99d8f83c | 3625 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3626 | |
31840ae1 | 3627 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3628 | push_space += data_size; |
31840ae1 | 3629 | |
44871b1b | 3630 | slot = path->slots[1]; |
34a38218 CM |
3631 | i = left_nritems - 1; |
3632 | while (i >= nr) { | |
dd3cc16b | 3633 | item = btrfs_item_nr(i); |
db94535d | 3634 | |
31840ae1 ZY |
3635 | if (!empty && push_items > 0) { |
3636 | if (path->slots[0] > i) | |
3637 | break; | |
3638 | if (path->slots[0] == i) { | |
2ff7e61e | 3639 | int space = btrfs_leaf_free_space(fs_info, left); |
31840ae1 ZY |
3640 | if (space + push_space * 2 > free_space) |
3641 | break; | |
3642 | } | |
3643 | } | |
3644 | ||
00ec4c51 | 3645 | if (path->slots[0] == i) |
87b29b20 | 3646 | push_space += data_size; |
db94535d | 3647 | |
db94535d CM |
3648 | this_item_size = btrfs_item_size(left, item); |
3649 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3650 | break; |
31840ae1 | 3651 | |
00ec4c51 | 3652 | push_items++; |
db94535d | 3653 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3654 | if (i == 0) |
3655 | break; | |
3656 | i--; | |
db94535d | 3657 | } |
5f39d397 | 3658 | |
925baedd CM |
3659 | if (push_items == 0) |
3660 | goto out_unlock; | |
5f39d397 | 3661 | |
6c1500f2 | 3662 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3663 | |
00ec4c51 | 3664 | /* push left to right */ |
5f39d397 | 3665 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3666 | |
5f39d397 | 3667 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
2ff7e61e | 3668 | push_space -= leaf_data_end(fs_info, left); |
5f39d397 | 3669 | |
00ec4c51 | 3670 | /* make room in the right data area */ |
2ff7e61e | 3671 | data_end = leaf_data_end(fs_info, right); |
5f39d397 | 3672 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3673 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3674 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3675 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3676 | |
00ec4c51 | 3677 | /* copy from the left data area */ |
3d9ec8c4 | 3678 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3679 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3680 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left), |
d6025579 | 3681 | push_space); |
5f39d397 CM |
3682 | |
3683 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3684 | btrfs_item_nr_offset(0), | |
3685 | right_nritems * sizeof(struct btrfs_item)); | |
3686 | ||
00ec4c51 | 3687 | /* copy the items from left to right */ |
5f39d397 CM |
3688 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3689 | btrfs_item_nr_offset(left_nritems - push_items), | |
3690 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3691 | |
3692 | /* update the item pointers */ | |
7518a238 | 3693 | right_nritems += push_items; |
5f39d397 | 3694 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3695 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3696 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3697 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3698 | push_space -= btrfs_token_item_size(right, item, &token); |
3699 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3700 | } |
3701 | ||
7518a238 | 3702 | left_nritems -= push_items; |
5f39d397 | 3703 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3704 | |
34a38218 CM |
3705 | if (left_nritems) |
3706 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3707 | else |
7c302b49 | 3708 | clean_tree_block(fs_info, left); |
f0486c68 | 3709 | |
5f39d397 | 3710 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3711 | |
5f39d397 CM |
3712 | btrfs_item_key(right, &disk_key, 0); |
3713 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3714 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3715 | |
00ec4c51 | 3716 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3717 | if (path->slots[0] >= left_nritems) { |
3718 | path->slots[0] -= left_nritems; | |
925baedd | 3719 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
7c302b49 | 3720 | clean_tree_block(fs_info, path->nodes[0]); |
925baedd | 3721 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3722 | free_extent_buffer(path->nodes[0]); |
3723 | path->nodes[0] = right; | |
00ec4c51 CM |
3724 | path->slots[1] += 1; |
3725 | } else { | |
925baedd | 3726 | btrfs_tree_unlock(right); |
5f39d397 | 3727 | free_extent_buffer(right); |
00ec4c51 CM |
3728 | } |
3729 | return 0; | |
925baedd CM |
3730 | |
3731 | out_unlock: | |
3732 | btrfs_tree_unlock(right); | |
3733 | free_extent_buffer(right); | |
3734 | return 1; | |
00ec4c51 | 3735 | } |
925baedd | 3736 | |
44871b1b CM |
3737 | /* |
3738 | * push some data in the path leaf to the right, trying to free up at | |
3739 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3740 | * | |
3741 | * returns 1 if the push failed because the other node didn't have enough | |
3742 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3743 | * |
3744 | * this will push starting from min_slot to the end of the leaf. It won't | |
3745 | * push any slot lower than min_slot | |
44871b1b CM |
3746 | */ |
3747 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3748 | *root, struct btrfs_path *path, |
3749 | int min_data_size, int data_size, | |
3750 | int empty, u32 min_slot) | |
44871b1b | 3751 | { |
2ff7e61e | 3752 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3753 | struct extent_buffer *left = path->nodes[0]; |
3754 | struct extent_buffer *right; | |
3755 | struct extent_buffer *upper; | |
3756 | int slot; | |
3757 | int free_space; | |
3758 | u32 left_nritems; | |
3759 | int ret; | |
3760 | ||
3761 | if (!path->nodes[1]) | |
3762 | return 1; | |
3763 | ||
3764 | slot = path->slots[1]; | |
3765 | upper = path->nodes[1]; | |
3766 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3767 | return 1; | |
3768 | ||
3769 | btrfs_assert_tree_locked(path->nodes[1]); | |
3770 | ||
2ff7e61e | 3771 | right = read_node_slot(fs_info, upper, slot + 1); |
fb770ae4 LB |
3772 | /* |
3773 | * slot + 1 is not valid or we fail to read the right node, | |
3774 | * no big deal, just return. | |
3775 | */ | |
3776 | if (IS_ERR(right)) | |
91ca338d TI |
3777 | return 1; |
3778 | ||
44871b1b | 3779 | btrfs_tree_lock(right); |
8bead258 | 3780 | btrfs_set_lock_blocking_write(right); |
44871b1b | 3781 | |
2ff7e61e | 3782 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3783 | if (free_space < data_size) |
3784 | goto out_unlock; | |
3785 | ||
3786 | /* cow and double check */ | |
3787 | ret = btrfs_cow_block(trans, root, right, upper, | |
3788 | slot + 1, &right); | |
3789 | if (ret) | |
3790 | goto out_unlock; | |
3791 | ||
2ff7e61e | 3792 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3793 | if (free_space < data_size) |
3794 | goto out_unlock; | |
3795 | ||
3796 | left_nritems = btrfs_header_nritems(left); | |
3797 | if (left_nritems == 0) | |
3798 | goto out_unlock; | |
3799 | ||
2ef1fed2 FDBM |
3800 | if (path->slots[0] == left_nritems && !empty) { |
3801 | /* Key greater than all keys in the leaf, right neighbor has | |
3802 | * enough room for it and we're not emptying our leaf to delete | |
3803 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3804 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3805 | btrfs_tree_unlock(left); |
3806 | free_extent_buffer(left); | |
3807 | path->nodes[0] = right; | |
3808 | path->slots[0] = 0; | |
3809 | path->slots[1]++; | |
3810 | return 0; | |
3811 | } | |
3812 | ||
1e47eef2 | 3813 | return __push_leaf_right(fs_info, path, min_data_size, empty, |
99d8f83c | 3814 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3815 | out_unlock: |
3816 | btrfs_tree_unlock(right); | |
3817 | free_extent_buffer(right); | |
3818 | return 1; | |
3819 | } | |
3820 | ||
74123bd7 CM |
3821 | /* |
3822 | * push some data in the path leaf to the left, trying to free up at | |
3823 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3824 | * |
3825 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3826 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3827 | * items | |
74123bd7 | 3828 | */ |
66cb7ddb | 3829 | static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3830 | struct btrfs_path *path, int data_size, |
3831 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3832 | int free_space, u32 right_nritems, |
3833 | u32 max_slot) | |
be0e5c09 | 3834 | { |
5f39d397 CM |
3835 | struct btrfs_disk_key disk_key; |
3836 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3837 | int i; |
be0e5c09 CM |
3838 | int push_space = 0; |
3839 | int push_items = 0; | |
0783fcfc | 3840 | struct btrfs_item *item; |
7518a238 | 3841 | u32 old_left_nritems; |
34a38218 | 3842 | u32 nr; |
aa5d6bed | 3843 | int ret = 0; |
db94535d CM |
3844 | u32 this_item_size; |
3845 | u32 old_left_item_size; | |
cfed81a0 CM |
3846 | struct btrfs_map_token token; |
3847 | ||
3848 | btrfs_init_map_token(&token); | |
be0e5c09 | 3849 | |
34a38218 | 3850 | if (empty) |
99d8f83c | 3851 | nr = min(right_nritems, max_slot); |
34a38218 | 3852 | else |
99d8f83c | 3853 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3854 | |
3855 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3856 | item = btrfs_item_nr(i); |
db94535d | 3857 | |
31840ae1 ZY |
3858 | if (!empty && push_items > 0) { |
3859 | if (path->slots[0] < i) | |
3860 | break; | |
3861 | if (path->slots[0] == i) { | |
2ff7e61e | 3862 | int space = btrfs_leaf_free_space(fs_info, right); |
31840ae1 ZY |
3863 | if (space + push_space * 2 > free_space) |
3864 | break; | |
3865 | } | |
3866 | } | |
3867 | ||
be0e5c09 | 3868 | if (path->slots[0] == i) |
87b29b20 | 3869 | push_space += data_size; |
db94535d CM |
3870 | |
3871 | this_item_size = btrfs_item_size(right, item); | |
3872 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3873 | break; |
db94535d | 3874 | |
be0e5c09 | 3875 | push_items++; |
db94535d CM |
3876 | push_space += this_item_size + sizeof(*item); |
3877 | } | |
3878 | ||
be0e5c09 | 3879 | if (push_items == 0) { |
925baedd CM |
3880 | ret = 1; |
3881 | goto out; | |
be0e5c09 | 3882 | } |
fae7f21c | 3883 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3884 | |
be0e5c09 | 3885 | /* push data from right to left */ |
5f39d397 CM |
3886 | copy_extent_buffer(left, right, |
3887 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3888 | btrfs_item_nr_offset(0), | |
3889 | push_items * sizeof(struct btrfs_item)); | |
3890 | ||
0b246afa | 3891 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3892 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3893 | |
3d9ec8c4 | 3894 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3895 | leaf_data_end(fs_info, left) - push_space, |
3d9ec8c4 | 3896 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3897 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3898 | push_space); |
5f39d397 | 3899 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3900 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3901 | |
db94535d | 3902 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3903 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3904 | u32 ioff; |
db94535d | 3905 | |
dd3cc16b | 3906 | item = btrfs_item_nr(i); |
db94535d | 3907 | |
cfed81a0 CM |
3908 | ioff = btrfs_token_item_offset(left, item, &token); |
3909 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3910 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3911 | &token); |
be0e5c09 | 3912 | } |
5f39d397 | 3913 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3914 | |
3915 | /* fixup right node */ | |
31b1a2bd JL |
3916 | if (push_items > right_nritems) |
3917 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3918 | right_nritems); |
34a38218 CM |
3919 | |
3920 | if (push_items < right_nritems) { | |
3921 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2ff7e61e | 3922 | leaf_data_end(fs_info, right); |
3d9ec8c4 | 3923 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3924 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3925 | BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3926 | leaf_data_end(fs_info, right), push_space); |
34a38218 CM |
3927 | |
3928 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3929 | btrfs_item_nr_offset(push_items), |
3930 | (btrfs_header_nritems(right) - push_items) * | |
3931 | sizeof(struct btrfs_item)); | |
34a38218 | 3932 | } |
eef1c494 Y |
3933 | right_nritems -= push_items; |
3934 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3935 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3936 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3937 | item = btrfs_item_nr(i); |
db94535d | 3938 | |
cfed81a0 CM |
3939 | push_space = push_space - btrfs_token_item_size(right, |
3940 | item, &token); | |
3941 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3942 | } |
eb60ceac | 3943 | |
5f39d397 | 3944 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3945 | if (right_nritems) |
3946 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3947 | else |
7c302b49 | 3948 | clean_tree_block(fs_info, right); |
098f59c2 | 3949 | |
5f39d397 | 3950 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3951 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3952 | |
3953 | /* then fixup the leaf pointer in the path */ | |
3954 | if (path->slots[0] < push_items) { | |
3955 | path->slots[0] += old_left_nritems; | |
925baedd | 3956 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3957 | free_extent_buffer(path->nodes[0]); |
3958 | path->nodes[0] = left; | |
be0e5c09 CM |
3959 | path->slots[1] -= 1; |
3960 | } else { | |
925baedd | 3961 | btrfs_tree_unlock(left); |
5f39d397 | 3962 | free_extent_buffer(left); |
be0e5c09 CM |
3963 | path->slots[0] -= push_items; |
3964 | } | |
eb60ceac | 3965 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3966 | return ret; |
925baedd CM |
3967 | out: |
3968 | btrfs_tree_unlock(left); | |
3969 | free_extent_buffer(left); | |
3970 | return ret; | |
be0e5c09 CM |
3971 | } |
3972 | ||
44871b1b CM |
3973 | /* |
3974 | * push some data in the path leaf to the left, trying to free up at | |
3975 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3976 | * |
3977 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3978 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3979 | * items | |
44871b1b CM |
3980 | */ |
3981 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3982 | *root, struct btrfs_path *path, int min_data_size, |
3983 | int data_size, int empty, u32 max_slot) | |
44871b1b | 3984 | { |
2ff7e61e | 3985 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3986 | struct extent_buffer *right = path->nodes[0]; |
3987 | struct extent_buffer *left; | |
3988 | int slot; | |
3989 | int free_space; | |
3990 | u32 right_nritems; | |
3991 | int ret = 0; | |
3992 | ||
3993 | slot = path->slots[1]; | |
3994 | if (slot == 0) | |
3995 | return 1; | |
3996 | if (!path->nodes[1]) | |
3997 | return 1; | |
3998 | ||
3999 | right_nritems = btrfs_header_nritems(right); | |
4000 | if (right_nritems == 0) | |
4001 | return 1; | |
4002 | ||
4003 | btrfs_assert_tree_locked(path->nodes[1]); | |
4004 | ||
2ff7e61e | 4005 | left = read_node_slot(fs_info, path->nodes[1], slot - 1); |
fb770ae4 LB |
4006 | /* |
4007 | * slot - 1 is not valid or we fail to read the left node, | |
4008 | * no big deal, just return. | |
4009 | */ | |
4010 | if (IS_ERR(left)) | |
91ca338d TI |
4011 | return 1; |
4012 | ||
44871b1b | 4013 | btrfs_tree_lock(left); |
8bead258 | 4014 | btrfs_set_lock_blocking_write(left); |
44871b1b | 4015 | |
2ff7e61e | 4016 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4017 | if (free_space < data_size) { |
4018 | ret = 1; | |
4019 | goto out; | |
4020 | } | |
4021 | ||
4022 | /* cow and double check */ | |
4023 | ret = btrfs_cow_block(trans, root, left, | |
4024 | path->nodes[1], slot - 1, &left); | |
4025 | if (ret) { | |
4026 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4027 | if (ret == -ENOSPC) |
4028 | ret = 1; | |
44871b1b CM |
4029 | goto out; |
4030 | } | |
4031 | ||
2ff7e61e | 4032 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4033 | if (free_space < data_size) { |
4034 | ret = 1; | |
4035 | goto out; | |
4036 | } | |
4037 | ||
66cb7ddb | 4038 | return __push_leaf_left(fs_info, path, min_data_size, |
99d8f83c CM |
4039 | empty, left, free_space, right_nritems, |
4040 | max_slot); | |
44871b1b CM |
4041 | out: |
4042 | btrfs_tree_unlock(left); | |
4043 | free_extent_buffer(left); | |
4044 | return ret; | |
4045 | } | |
4046 | ||
4047 | /* | |
4048 | * split the path's leaf in two, making sure there is at least data_size | |
4049 | * available for the resulting leaf level of the path. | |
44871b1b | 4050 | */ |
143bede5 | 4051 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2ff7e61e | 4052 | struct btrfs_fs_info *fs_info, |
143bede5 JM |
4053 | struct btrfs_path *path, |
4054 | struct extent_buffer *l, | |
4055 | struct extent_buffer *right, | |
4056 | int slot, int mid, int nritems) | |
44871b1b CM |
4057 | { |
4058 | int data_copy_size; | |
4059 | int rt_data_off; | |
4060 | int i; | |
44871b1b | 4061 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4062 | struct btrfs_map_token token; |
4063 | ||
4064 | btrfs_init_map_token(&token); | |
44871b1b CM |
4065 | |
4066 | nritems = nritems - mid; | |
4067 | btrfs_set_header_nritems(right, nritems); | |
2ff7e61e | 4068 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l); |
44871b1b CM |
4069 | |
4070 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4071 | btrfs_item_nr_offset(mid), | |
4072 | nritems * sizeof(struct btrfs_item)); | |
4073 | ||
4074 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4075 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4076 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
2ff7e61e | 4077 | leaf_data_end(fs_info, l), data_copy_size); |
44871b1b | 4078 | |
0b246afa | 4079 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4080 | |
4081 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4082 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4083 | u32 ioff; |
4084 | ||
cfed81a0 CM |
4085 | ioff = btrfs_token_item_offset(right, item, &token); |
4086 | btrfs_set_token_item_offset(right, item, | |
4087 | ioff + rt_data_off, &token); | |
44871b1b CM |
4088 | } |
4089 | ||
44871b1b | 4090 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4091 | btrfs_item_key(right, &disk_key, 0); |
2ff7e61e | 4092 | insert_ptr(trans, fs_info, path, &disk_key, right->start, |
c3e06965 | 4093 | path->slots[1] + 1, 1); |
44871b1b CM |
4094 | |
4095 | btrfs_mark_buffer_dirty(right); | |
4096 | btrfs_mark_buffer_dirty(l); | |
4097 | BUG_ON(path->slots[0] != slot); | |
4098 | ||
44871b1b CM |
4099 | if (mid <= slot) { |
4100 | btrfs_tree_unlock(path->nodes[0]); | |
4101 | free_extent_buffer(path->nodes[0]); | |
4102 | path->nodes[0] = right; | |
4103 | path->slots[0] -= mid; | |
4104 | path->slots[1] += 1; | |
4105 | } else { | |
4106 | btrfs_tree_unlock(right); | |
4107 | free_extent_buffer(right); | |
4108 | } | |
4109 | ||
4110 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4111 | } |
4112 | ||
99d8f83c CM |
4113 | /* |
4114 | * double splits happen when we need to insert a big item in the middle | |
4115 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4116 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4117 | * A B C | |
4118 | * | |
4119 | * We avoid this by trying to push the items on either side of our target | |
4120 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4121 | * completely. | |
4122 | */ | |
4123 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4124 | struct btrfs_root *root, | |
4125 | struct btrfs_path *path, | |
4126 | int data_size) | |
4127 | { | |
2ff7e61e | 4128 | struct btrfs_fs_info *fs_info = root->fs_info; |
99d8f83c CM |
4129 | int ret; |
4130 | int progress = 0; | |
4131 | int slot; | |
4132 | u32 nritems; | |
5a4267ca | 4133 | int space_needed = data_size; |
99d8f83c CM |
4134 | |
4135 | slot = path->slots[0]; | |
5a4267ca | 4136 | if (slot < btrfs_header_nritems(path->nodes[0])) |
2ff7e61e | 4137 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); |
99d8f83c CM |
4138 | |
4139 | /* | |
4140 | * try to push all the items after our slot into the | |
4141 | * right leaf | |
4142 | */ | |
5a4267ca | 4143 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4144 | if (ret < 0) |
4145 | return ret; | |
4146 | ||
4147 | if (ret == 0) | |
4148 | progress++; | |
4149 | ||
4150 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4151 | /* | |
4152 | * our goal is to get our slot at the start or end of a leaf. If | |
4153 | * we've done so we're done | |
4154 | */ | |
4155 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4156 | return 0; | |
4157 | ||
2ff7e61e | 4158 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4159 | return 0; |
4160 | ||
4161 | /* try to push all the items before our slot into the next leaf */ | |
4162 | slot = path->slots[0]; | |
263d3995 FM |
4163 | space_needed = data_size; |
4164 | if (slot > 0) | |
4165 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); | |
5a4267ca | 4166 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4167 | if (ret < 0) |
4168 | return ret; | |
4169 | ||
4170 | if (ret == 0) | |
4171 | progress++; | |
4172 | ||
4173 | if (progress) | |
4174 | return 0; | |
4175 | return 1; | |
4176 | } | |
4177 | ||
74123bd7 CM |
4178 | /* |
4179 | * split the path's leaf in two, making sure there is at least data_size | |
4180 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4181 | * |
4182 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4183 | */ |
e02119d5 CM |
4184 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4185 | struct btrfs_root *root, | |
310712b2 | 4186 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4187 | struct btrfs_path *path, int data_size, |
4188 | int extend) | |
be0e5c09 | 4189 | { |
5d4f98a2 | 4190 | struct btrfs_disk_key disk_key; |
5f39d397 | 4191 | struct extent_buffer *l; |
7518a238 | 4192 | u32 nritems; |
eb60ceac CM |
4193 | int mid; |
4194 | int slot; | |
5f39d397 | 4195 | struct extent_buffer *right; |
b7a0365e | 4196 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4197 | int ret = 0; |
aa5d6bed | 4198 | int wret; |
5d4f98a2 | 4199 | int split; |
cc0c5538 | 4200 | int num_doubles = 0; |
99d8f83c | 4201 | int tried_avoid_double = 0; |
aa5d6bed | 4202 | |
a5719521 YZ |
4203 | l = path->nodes[0]; |
4204 | slot = path->slots[0]; | |
4205 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4206 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4207 | return -EOVERFLOW; |
4208 | ||
40689478 | 4209 | /* first try to make some room by pushing left and right */ |
33157e05 | 4210 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4211 | int space_needed = data_size; |
4212 | ||
4213 | if (slot < btrfs_header_nritems(l)) | |
2ff7e61e | 4214 | space_needed -= btrfs_leaf_free_space(fs_info, l); |
5a4267ca FDBM |
4215 | |
4216 | wret = push_leaf_right(trans, root, path, space_needed, | |
4217 | space_needed, 0, 0); | |
d397712b | 4218 | if (wret < 0) |
eaee50e8 | 4219 | return wret; |
3685f791 | 4220 | if (wret) { |
263d3995 FM |
4221 | space_needed = data_size; |
4222 | if (slot > 0) | |
4223 | space_needed -= btrfs_leaf_free_space(fs_info, | |
4224 | l); | |
5a4267ca FDBM |
4225 | wret = push_leaf_left(trans, root, path, space_needed, |
4226 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4227 | if (wret < 0) |
4228 | return wret; | |
4229 | } | |
4230 | l = path->nodes[0]; | |
aa5d6bed | 4231 | |
3685f791 | 4232 | /* did the pushes work? */ |
2ff7e61e | 4233 | if (btrfs_leaf_free_space(fs_info, l) >= data_size) |
3685f791 | 4234 | return 0; |
3326d1b0 | 4235 | } |
aa5d6bed | 4236 | |
5c680ed6 | 4237 | if (!path->nodes[1]) { |
fdd99c72 | 4238 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4239 | if (ret) |
4240 | return ret; | |
4241 | } | |
cc0c5538 | 4242 | again: |
5d4f98a2 | 4243 | split = 1; |
cc0c5538 | 4244 | l = path->nodes[0]; |
eb60ceac | 4245 | slot = path->slots[0]; |
5f39d397 | 4246 | nritems = btrfs_header_nritems(l); |
d397712b | 4247 | mid = (nritems + 1) / 2; |
54aa1f4d | 4248 | |
5d4f98a2 YZ |
4249 | if (mid <= slot) { |
4250 | if (nritems == 1 || | |
4251 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4252 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4253 | if (slot >= nritems) { |
4254 | split = 0; | |
4255 | } else { | |
4256 | mid = slot; | |
4257 | if (mid != nritems && | |
4258 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4259 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4260 | if (data_size && !tried_avoid_double) |
4261 | goto push_for_double; | |
5d4f98a2 YZ |
4262 | split = 2; |
4263 | } | |
4264 | } | |
4265 | } | |
4266 | } else { | |
4267 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4268 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4269 | if (!extend && data_size && slot == 0) { |
4270 | split = 0; | |
4271 | } else if ((extend || !data_size) && slot == 0) { | |
4272 | mid = 1; | |
4273 | } else { | |
4274 | mid = slot; | |
4275 | if (mid != nritems && | |
4276 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4277 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4278 | if (data_size && !tried_avoid_double) |
4279 | goto push_for_double; | |
67871254 | 4280 | split = 2; |
5d4f98a2 YZ |
4281 | } |
4282 | } | |
4283 | } | |
4284 | } | |
4285 | ||
4286 | if (split == 0) | |
4287 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4288 | else | |
4289 | btrfs_item_key(l, &disk_key, mid); | |
4290 | ||
a6279470 FM |
4291 | right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, |
4292 | l->start, 0); | |
f0486c68 | 4293 | if (IS_ERR(right)) |
5f39d397 | 4294 | return PTR_ERR(right); |
f0486c68 | 4295 | |
0b246afa | 4296 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4297 | |
5d4f98a2 YZ |
4298 | if (split == 0) { |
4299 | if (mid <= slot) { | |
4300 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4301 | insert_ptr(trans, fs_info, path, &disk_key, |
4302 | right->start, path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
4303 | btrfs_tree_unlock(path->nodes[0]); |
4304 | free_extent_buffer(path->nodes[0]); | |
4305 | path->nodes[0] = right; | |
4306 | path->slots[0] = 0; | |
4307 | path->slots[1] += 1; | |
4308 | } else { | |
4309 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4310 | insert_ptr(trans, fs_info, path, &disk_key, |
4311 | right->start, path->slots[1], 1); | |
5d4f98a2 YZ |
4312 | btrfs_tree_unlock(path->nodes[0]); |
4313 | free_extent_buffer(path->nodes[0]); | |
4314 | path->nodes[0] = right; | |
4315 | path->slots[0] = 0; | |
143bede5 | 4316 | if (path->slots[1] == 0) |
b167fa91 | 4317 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4318 | } |
196e0249 LB |
4319 | /* |
4320 | * We create a new leaf 'right' for the required ins_len and | |
4321 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4322 | * the content of ins_len to 'right'. | |
4323 | */ | |
5d4f98a2 | 4324 | return ret; |
d4dbff95 | 4325 | } |
74123bd7 | 4326 | |
2ff7e61e | 4327 | copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems); |
31840ae1 | 4328 | |
5d4f98a2 | 4329 | if (split == 2) { |
cc0c5538 CM |
4330 | BUG_ON(num_doubles != 0); |
4331 | num_doubles++; | |
4332 | goto again; | |
a429e513 | 4333 | } |
44871b1b | 4334 | |
143bede5 | 4335 | return 0; |
99d8f83c CM |
4336 | |
4337 | push_for_double: | |
4338 | push_for_double_split(trans, root, path, data_size); | |
4339 | tried_avoid_double = 1; | |
2ff7e61e | 4340 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4341 | return 0; |
4342 | goto again; | |
be0e5c09 CM |
4343 | } |
4344 | ||
ad48fd75 YZ |
4345 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4346 | struct btrfs_root *root, | |
4347 | struct btrfs_path *path, int ins_len) | |
459931ec | 4348 | { |
2ff7e61e | 4349 | struct btrfs_fs_info *fs_info = root->fs_info; |
ad48fd75 | 4350 | struct btrfs_key key; |
459931ec | 4351 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4352 | struct btrfs_file_extent_item *fi; |
4353 | u64 extent_len = 0; | |
4354 | u32 item_size; | |
4355 | int ret; | |
459931ec CM |
4356 | |
4357 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4358 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4359 | ||
4360 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4361 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4362 | ||
2ff7e61e | 4363 | if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len) |
ad48fd75 | 4364 | return 0; |
459931ec CM |
4365 | |
4366 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4367 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4368 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4369 | struct btrfs_file_extent_item); | |
4370 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4371 | } | |
b3b4aa74 | 4372 | btrfs_release_path(path); |
459931ec | 4373 | |
459931ec | 4374 | path->keep_locks = 1; |
ad48fd75 YZ |
4375 | path->search_for_split = 1; |
4376 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4377 | path->search_for_split = 0; |
a8df6fe6 FM |
4378 | if (ret > 0) |
4379 | ret = -EAGAIN; | |
ad48fd75 YZ |
4380 | if (ret < 0) |
4381 | goto err; | |
459931ec | 4382 | |
ad48fd75 YZ |
4383 | ret = -EAGAIN; |
4384 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4385 | /* if our item isn't there, return now */ |
4386 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4387 | goto err; |
4388 | ||
109f6aef | 4389 | /* the leaf has changed, it now has room. return now */ |
2ff7e61e | 4390 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len) |
109f6aef CM |
4391 | goto err; |
4392 | ||
ad48fd75 YZ |
4393 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4394 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4395 | struct btrfs_file_extent_item); | |
4396 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4397 | goto err; | |
459931ec CM |
4398 | } |
4399 | ||
b9473439 | 4400 | btrfs_set_path_blocking(path); |
ad48fd75 | 4401 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4402 | if (ret) |
4403 | goto err; | |
459931ec | 4404 | |
ad48fd75 | 4405 | path->keep_locks = 0; |
b9473439 | 4406 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4407 | return 0; |
4408 | err: | |
4409 | path->keep_locks = 0; | |
4410 | return ret; | |
4411 | } | |
4412 | ||
4961e293 | 4413 | static noinline int split_item(struct btrfs_fs_info *fs_info, |
ad48fd75 | 4414 | struct btrfs_path *path, |
310712b2 | 4415 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4416 | unsigned long split_offset) |
4417 | { | |
4418 | struct extent_buffer *leaf; | |
4419 | struct btrfs_item *item; | |
4420 | struct btrfs_item *new_item; | |
4421 | int slot; | |
4422 | char *buf; | |
4423 | u32 nritems; | |
4424 | u32 item_size; | |
4425 | u32 orig_offset; | |
4426 | struct btrfs_disk_key disk_key; | |
4427 | ||
b9473439 | 4428 | leaf = path->nodes[0]; |
2ff7e61e | 4429 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4430 | |
b4ce94de CM |
4431 | btrfs_set_path_blocking(path); |
4432 | ||
dd3cc16b | 4433 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4434 | orig_offset = btrfs_item_offset(leaf, item); |
4435 | item_size = btrfs_item_size(leaf, item); | |
4436 | ||
459931ec | 4437 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4438 | if (!buf) |
4439 | return -ENOMEM; | |
4440 | ||
459931ec CM |
4441 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4442 | path->slots[0]), item_size); | |
459931ec | 4443 | |
ad48fd75 | 4444 | slot = path->slots[0] + 1; |
459931ec | 4445 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4446 | if (slot != nritems) { |
4447 | /* shift the items */ | |
4448 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4449 | btrfs_item_nr_offset(slot), |
4450 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4451 | } |
4452 | ||
4453 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4454 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4455 | ||
dd3cc16b | 4456 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4457 | |
4458 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4459 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4460 | ||
4461 | btrfs_set_item_offset(leaf, item, | |
4462 | orig_offset + item_size - split_offset); | |
4463 | btrfs_set_item_size(leaf, item, split_offset); | |
4464 | ||
4465 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4466 | ||
4467 | /* write the data for the start of the original item */ | |
4468 | write_extent_buffer(leaf, buf, | |
4469 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4470 | split_offset); | |
4471 | ||
4472 | /* write the data for the new item */ | |
4473 | write_extent_buffer(leaf, buf + split_offset, | |
4474 | btrfs_item_ptr_offset(leaf, slot), | |
4475 | item_size - split_offset); | |
4476 | btrfs_mark_buffer_dirty(leaf); | |
4477 | ||
2ff7e61e | 4478 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0); |
459931ec | 4479 | kfree(buf); |
ad48fd75 YZ |
4480 | return 0; |
4481 | } | |
4482 | ||
4483 | /* | |
4484 | * This function splits a single item into two items, | |
4485 | * giving 'new_key' to the new item and splitting the | |
4486 | * old one at split_offset (from the start of the item). | |
4487 | * | |
4488 | * The path may be released by this operation. After | |
4489 | * the split, the path is pointing to the old item. The | |
4490 | * new item is going to be in the same node as the old one. | |
4491 | * | |
4492 | * Note, the item being split must be smaller enough to live alone on | |
4493 | * a tree block with room for one extra struct btrfs_item | |
4494 | * | |
4495 | * This allows us to split the item in place, keeping a lock on the | |
4496 | * leaf the entire time. | |
4497 | */ | |
4498 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4499 | struct btrfs_root *root, | |
4500 | struct btrfs_path *path, | |
310712b2 | 4501 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4502 | unsigned long split_offset) |
4503 | { | |
4504 | int ret; | |
4505 | ret = setup_leaf_for_split(trans, root, path, | |
4506 | sizeof(struct btrfs_item)); | |
4507 | if (ret) | |
4508 | return ret; | |
4509 | ||
4961e293 | 4510 | ret = split_item(root->fs_info, path, new_key, split_offset); |
459931ec CM |
4511 | return ret; |
4512 | } | |
4513 | ||
ad48fd75 YZ |
4514 | /* |
4515 | * This function duplicate a item, giving 'new_key' to the new item. | |
4516 | * It guarantees both items live in the same tree leaf and the new item | |
4517 | * is contiguous with the original item. | |
4518 | * | |
4519 | * This allows us to split file extent in place, keeping a lock on the | |
4520 | * leaf the entire time. | |
4521 | */ | |
4522 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4523 | struct btrfs_root *root, | |
4524 | struct btrfs_path *path, | |
310712b2 | 4525 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4526 | { |
4527 | struct extent_buffer *leaf; | |
4528 | int ret; | |
4529 | u32 item_size; | |
4530 | ||
4531 | leaf = path->nodes[0]; | |
4532 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4533 | ret = setup_leaf_for_split(trans, root, path, | |
4534 | item_size + sizeof(struct btrfs_item)); | |
4535 | if (ret) | |
4536 | return ret; | |
4537 | ||
4538 | path->slots[0]++; | |
afe5fea7 | 4539 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4540 | item_size, item_size + |
4541 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4542 | leaf = path->nodes[0]; |
4543 | memcpy_extent_buffer(leaf, | |
4544 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4545 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4546 | item_size); | |
4547 | return 0; | |
4548 | } | |
4549 | ||
d352ac68 CM |
4550 | /* |
4551 | * make the item pointed to by the path smaller. new_size indicates | |
4552 | * how small to make it, and from_end tells us if we just chop bytes | |
4553 | * off the end of the item or if we shift the item to chop bytes off | |
4554 | * the front. | |
4555 | */ | |
2ff7e61e JM |
4556 | void btrfs_truncate_item(struct btrfs_fs_info *fs_info, |
4557 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 4558 | { |
b18c6685 | 4559 | int slot; |
5f39d397 CM |
4560 | struct extent_buffer *leaf; |
4561 | struct btrfs_item *item; | |
b18c6685 CM |
4562 | u32 nritems; |
4563 | unsigned int data_end; | |
4564 | unsigned int old_data_start; | |
4565 | unsigned int old_size; | |
4566 | unsigned int size_diff; | |
4567 | int i; | |
cfed81a0 CM |
4568 | struct btrfs_map_token token; |
4569 | ||
4570 | btrfs_init_map_token(&token); | |
b18c6685 | 4571 | |
5f39d397 | 4572 | leaf = path->nodes[0]; |
179e29e4 CM |
4573 | slot = path->slots[0]; |
4574 | ||
4575 | old_size = btrfs_item_size_nr(leaf, slot); | |
4576 | if (old_size == new_size) | |
143bede5 | 4577 | return; |
b18c6685 | 4578 | |
5f39d397 | 4579 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4580 | data_end = leaf_data_end(fs_info, leaf); |
b18c6685 | 4581 | |
5f39d397 | 4582 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4583 | |
b18c6685 CM |
4584 | size_diff = old_size - new_size; |
4585 | ||
4586 | BUG_ON(slot < 0); | |
4587 | BUG_ON(slot >= nritems); | |
4588 | ||
4589 | /* | |
4590 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4591 | */ | |
4592 | /* first correct the data pointers */ | |
4593 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4594 | u32 ioff; |
dd3cc16b | 4595 | item = btrfs_item_nr(i); |
db94535d | 4596 | |
cfed81a0 CM |
4597 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4598 | btrfs_set_token_item_offset(leaf, item, | |
4599 | ioff + size_diff, &token); | |
b18c6685 | 4600 | } |
db94535d | 4601 | |
b18c6685 | 4602 | /* shift the data */ |
179e29e4 | 4603 | if (from_end) { |
3d9ec8c4 NB |
4604 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4605 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4606 | data_end, old_data_start + new_size - data_end); |
4607 | } else { | |
4608 | struct btrfs_disk_key disk_key; | |
4609 | u64 offset; | |
4610 | ||
4611 | btrfs_item_key(leaf, &disk_key, slot); | |
4612 | ||
4613 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4614 | unsigned long ptr; | |
4615 | struct btrfs_file_extent_item *fi; | |
4616 | ||
4617 | fi = btrfs_item_ptr(leaf, slot, | |
4618 | struct btrfs_file_extent_item); | |
4619 | fi = (struct btrfs_file_extent_item *)( | |
4620 | (unsigned long)fi - size_diff); | |
4621 | ||
4622 | if (btrfs_file_extent_type(leaf, fi) == | |
4623 | BTRFS_FILE_EXTENT_INLINE) { | |
4624 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4625 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4626 | (unsigned long)fi, |
7ec20afb | 4627 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4628 | } |
4629 | } | |
4630 | ||
3d9ec8c4 NB |
4631 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4632 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4633 | data_end, old_data_start - data_end); |
4634 | ||
4635 | offset = btrfs_disk_key_offset(&disk_key); | |
4636 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4637 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4638 | if (slot == 0) | |
b167fa91 | 4639 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4640 | } |
5f39d397 | 4641 | |
dd3cc16b | 4642 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4643 | btrfs_set_item_size(leaf, item, new_size); |
4644 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4645 | |
2ff7e61e | 4646 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4647 | btrfs_print_leaf(leaf); |
b18c6685 | 4648 | BUG(); |
5f39d397 | 4649 | } |
b18c6685 CM |
4650 | } |
4651 | ||
d352ac68 | 4652 | /* |
8f69dbd2 | 4653 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4654 | */ |
2ff7e61e | 4655 | void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 4656 | u32 data_size) |
6567e837 | 4657 | { |
6567e837 | 4658 | int slot; |
5f39d397 CM |
4659 | struct extent_buffer *leaf; |
4660 | struct btrfs_item *item; | |
6567e837 CM |
4661 | u32 nritems; |
4662 | unsigned int data_end; | |
4663 | unsigned int old_data; | |
4664 | unsigned int old_size; | |
4665 | int i; | |
cfed81a0 CM |
4666 | struct btrfs_map_token token; |
4667 | ||
4668 | btrfs_init_map_token(&token); | |
6567e837 | 4669 | |
5f39d397 | 4670 | leaf = path->nodes[0]; |
6567e837 | 4671 | |
5f39d397 | 4672 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4673 | data_end = leaf_data_end(fs_info, leaf); |
6567e837 | 4674 | |
2ff7e61e | 4675 | if (btrfs_leaf_free_space(fs_info, leaf) < data_size) { |
a4f78750 | 4676 | btrfs_print_leaf(leaf); |
6567e837 | 4677 | BUG(); |
5f39d397 | 4678 | } |
6567e837 | 4679 | slot = path->slots[0]; |
5f39d397 | 4680 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4681 | |
4682 | BUG_ON(slot < 0); | |
3326d1b0 | 4683 | if (slot >= nritems) { |
a4f78750 | 4684 | btrfs_print_leaf(leaf); |
0b246afa JM |
4685 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4686 | slot, nritems); | |
290342f6 | 4687 | BUG(); |
3326d1b0 | 4688 | } |
6567e837 CM |
4689 | |
4690 | /* | |
4691 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4692 | */ | |
4693 | /* first correct the data pointers */ | |
4694 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4695 | u32 ioff; |
dd3cc16b | 4696 | item = btrfs_item_nr(i); |
db94535d | 4697 | |
cfed81a0 CM |
4698 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4699 | btrfs_set_token_item_offset(leaf, item, | |
4700 | ioff - data_size, &token); | |
6567e837 | 4701 | } |
5f39d397 | 4702 | |
6567e837 | 4703 | /* shift the data */ |
3d9ec8c4 NB |
4704 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4705 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4706 | data_end, old_data - data_end); |
5f39d397 | 4707 | |
6567e837 | 4708 | data_end = old_data; |
5f39d397 | 4709 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4710 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4711 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4712 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4713 | |
2ff7e61e | 4714 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4715 | btrfs_print_leaf(leaf); |
6567e837 | 4716 | BUG(); |
5f39d397 | 4717 | } |
6567e837 CM |
4718 | } |
4719 | ||
74123bd7 | 4720 | /* |
44871b1b CM |
4721 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4722 | * to save stack depth by doing the bulk of the work in a function | |
4723 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4724 | */ |
afe5fea7 | 4725 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4726 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4727 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4728 | { |
0b246afa | 4729 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4730 | struct btrfs_item *item; |
9c58309d | 4731 | int i; |
7518a238 | 4732 | u32 nritems; |
be0e5c09 | 4733 | unsigned int data_end; |
e2fa7227 | 4734 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4735 | struct extent_buffer *leaf; |
4736 | int slot; | |
cfed81a0 CM |
4737 | struct btrfs_map_token token; |
4738 | ||
24cdc847 FM |
4739 | if (path->slots[0] == 0) { |
4740 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4741 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4742 | } |
4743 | btrfs_unlock_up_safe(path, 1); | |
4744 | ||
cfed81a0 | 4745 | btrfs_init_map_token(&token); |
e2fa7227 | 4746 | |
5f39d397 | 4747 | leaf = path->nodes[0]; |
44871b1b | 4748 | slot = path->slots[0]; |
74123bd7 | 4749 | |
5f39d397 | 4750 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4751 | data_end = leaf_data_end(fs_info, leaf); |
eb60ceac | 4752 | |
2ff7e61e | 4753 | if (btrfs_leaf_free_space(fs_info, leaf) < total_size) { |
a4f78750 | 4754 | btrfs_print_leaf(leaf); |
0b246afa | 4755 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
2ff7e61e | 4756 | total_size, btrfs_leaf_free_space(fs_info, leaf)); |
be0e5c09 | 4757 | BUG(); |
d4dbff95 | 4758 | } |
5f39d397 | 4759 | |
be0e5c09 | 4760 | if (slot != nritems) { |
5f39d397 | 4761 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4762 | |
5f39d397 | 4763 | if (old_data < data_end) { |
a4f78750 | 4764 | btrfs_print_leaf(leaf); |
0b246afa | 4765 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4766 | slot, old_data, data_end); |
290342f6 | 4767 | BUG(); |
5f39d397 | 4768 | } |
be0e5c09 CM |
4769 | /* |
4770 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4771 | */ | |
4772 | /* first correct the data pointers */ | |
0783fcfc | 4773 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4774 | u32 ioff; |
db94535d | 4775 | |
62e85577 | 4776 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4777 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4778 | btrfs_set_token_item_offset(leaf, item, | |
4779 | ioff - total_data, &token); | |
0783fcfc | 4780 | } |
be0e5c09 | 4781 | /* shift the items */ |
9c58309d | 4782 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4783 | btrfs_item_nr_offset(slot), |
d6025579 | 4784 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4785 | |
4786 | /* shift the data */ | |
3d9ec8c4 NB |
4787 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4788 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4789 | data_end, old_data - data_end); |
be0e5c09 CM |
4790 | data_end = old_data; |
4791 | } | |
5f39d397 | 4792 | |
62e2749e | 4793 | /* setup the item for the new data */ |
9c58309d CM |
4794 | for (i = 0; i < nr; i++) { |
4795 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4796 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4797 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4798 | btrfs_set_token_item_offset(leaf, item, |
4799 | data_end - data_size[i], &token); | |
9c58309d | 4800 | data_end -= data_size[i]; |
cfed81a0 | 4801 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4802 | } |
44871b1b | 4803 | |
9c58309d | 4804 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4805 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4806 | |
2ff7e61e | 4807 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4808 | btrfs_print_leaf(leaf); |
be0e5c09 | 4809 | BUG(); |
5f39d397 | 4810 | } |
44871b1b CM |
4811 | } |
4812 | ||
4813 | /* | |
4814 | * Given a key and some data, insert items into the tree. | |
4815 | * This does all the path init required, making room in the tree if needed. | |
4816 | */ | |
4817 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4818 | struct btrfs_root *root, | |
4819 | struct btrfs_path *path, | |
310712b2 | 4820 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4821 | int nr) |
4822 | { | |
44871b1b CM |
4823 | int ret = 0; |
4824 | int slot; | |
4825 | int i; | |
4826 | u32 total_size = 0; | |
4827 | u32 total_data = 0; | |
4828 | ||
4829 | for (i = 0; i < nr; i++) | |
4830 | total_data += data_size[i]; | |
4831 | ||
4832 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4833 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4834 | if (ret == 0) | |
4835 | return -EEXIST; | |
4836 | if (ret < 0) | |
143bede5 | 4837 | return ret; |
44871b1b | 4838 | |
44871b1b CM |
4839 | slot = path->slots[0]; |
4840 | BUG_ON(slot < 0); | |
4841 | ||
afe5fea7 | 4842 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4843 | total_data, total_size, nr); |
143bede5 | 4844 | return 0; |
62e2749e CM |
4845 | } |
4846 | ||
4847 | /* | |
4848 | * Given a key and some data, insert an item into the tree. | |
4849 | * This does all the path init required, making room in the tree if needed. | |
4850 | */ | |
310712b2 OS |
4851 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4852 | const struct btrfs_key *cpu_key, void *data, | |
4853 | u32 data_size) | |
62e2749e CM |
4854 | { |
4855 | int ret = 0; | |
2c90e5d6 | 4856 | struct btrfs_path *path; |
5f39d397 CM |
4857 | struct extent_buffer *leaf; |
4858 | unsigned long ptr; | |
62e2749e | 4859 | |
2c90e5d6 | 4860 | path = btrfs_alloc_path(); |
db5b493a TI |
4861 | if (!path) |
4862 | return -ENOMEM; | |
2c90e5d6 | 4863 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4864 | if (!ret) { |
5f39d397 CM |
4865 | leaf = path->nodes[0]; |
4866 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4867 | write_extent_buffer(leaf, data, ptr, data_size); | |
4868 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4869 | } |
2c90e5d6 | 4870 | btrfs_free_path(path); |
aa5d6bed | 4871 | return ret; |
be0e5c09 CM |
4872 | } |
4873 | ||
74123bd7 | 4874 | /* |
5de08d7d | 4875 | * delete the pointer from a given node. |
74123bd7 | 4876 | * |
d352ac68 CM |
4877 | * the tree should have been previously balanced so the deletion does not |
4878 | * empty a node. | |
74123bd7 | 4879 | */ |
afe5fea7 TI |
4880 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4881 | int level, int slot) | |
be0e5c09 | 4882 | { |
5f39d397 | 4883 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4884 | u32 nritems; |
f3ea38da | 4885 | int ret; |
be0e5c09 | 4886 | |
5f39d397 | 4887 | nritems = btrfs_header_nritems(parent); |
d397712b | 4888 | if (slot != nritems - 1) { |
bf1d3425 DS |
4889 | if (level) { |
4890 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4891 | nritems - slot - 1); |
bf1d3425 DS |
4892 | BUG_ON(ret < 0); |
4893 | } | |
5f39d397 CM |
4894 | memmove_extent_buffer(parent, |
4895 | btrfs_node_key_ptr_offset(slot), | |
4896 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4897 | sizeof(struct btrfs_key_ptr) * |
4898 | (nritems - slot - 1)); | |
57ba86c0 | 4899 | } else if (level) { |
e09c2efe DS |
4900 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4901 | GFP_NOFS); | |
57ba86c0 | 4902 | BUG_ON(ret < 0); |
bb803951 | 4903 | } |
f3ea38da | 4904 | |
7518a238 | 4905 | nritems--; |
5f39d397 | 4906 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4907 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4908 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4909 | /* just turn the root into a leaf and break */ |
5f39d397 | 4910 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4911 | } else if (slot == 0) { |
5f39d397 CM |
4912 | struct btrfs_disk_key disk_key; |
4913 | ||
4914 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4915 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4916 | } |
d6025579 | 4917 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4918 | } |
4919 | ||
323ac95b CM |
4920 | /* |
4921 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4922 | * path->nodes[1]. |
323ac95b CM |
4923 | * |
4924 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4925 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4926 | * | |
4927 | * The path must have already been setup for deleting the leaf, including | |
4928 | * all the proper balancing. path->nodes[1] must be locked. | |
4929 | */ | |
143bede5 JM |
4930 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4931 | struct btrfs_root *root, | |
4932 | struct btrfs_path *path, | |
4933 | struct extent_buffer *leaf) | |
323ac95b | 4934 | { |
5d4f98a2 | 4935 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4936 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4937 | |
4d081c41 CM |
4938 | /* |
4939 | * btrfs_free_extent is expensive, we want to make sure we | |
4940 | * aren't holding any locks when we call it | |
4941 | */ | |
4942 | btrfs_unlock_up_safe(path, 0); | |
4943 | ||
f0486c68 YZ |
4944 | root_sub_used(root, leaf->len); |
4945 | ||
3083ee2e | 4946 | extent_buffer_get(leaf); |
5581a51a | 4947 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4948 | free_extent_buffer_stale(leaf); |
323ac95b | 4949 | } |
74123bd7 CM |
4950 | /* |
4951 | * delete the item at the leaf level in path. If that empties | |
4952 | * the leaf, remove it from the tree | |
4953 | */ | |
85e21bac CM |
4954 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4955 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4956 | { |
0b246afa | 4957 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4958 | struct extent_buffer *leaf; |
4959 | struct btrfs_item *item; | |
ce0eac2a AM |
4960 | u32 last_off; |
4961 | u32 dsize = 0; | |
aa5d6bed CM |
4962 | int ret = 0; |
4963 | int wret; | |
85e21bac | 4964 | int i; |
7518a238 | 4965 | u32 nritems; |
cfed81a0 CM |
4966 | struct btrfs_map_token token; |
4967 | ||
4968 | btrfs_init_map_token(&token); | |
be0e5c09 | 4969 | |
5f39d397 | 4970 | leaf = path->nodes[0]; |
85e21bac CM |
4971 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4972 | ||
4973 | for (i = 0; i < nr; i++) | |
4974 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4975 | ||
5f39d397 | 4976 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4977 | |
85e21bac | 4978 | if (slot + nr != nritems) { |
2ff7e61e | 4979 | int data_end = leaf_data_end(fs_info, leaf); |
5f39d397 | 4980 | |
3d9ec8c4 | 4981 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4982 | data_end + dsize, |
3d9ec8c4 | 4983 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4984 | last_off - data_end); |
5f39d397 | 4985 | |
85e21bac | 4986 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4987 | u32 ioff; |
db94535d | 4988 | |
dd3cc16b | 4989 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4990 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4991 | btrfs_set_token_item_offset(leaf, item, | |
4992 | ioff + dsize, &token); | |
0783fcfc | 4993 | } |
db94535d | 4994 | |
5f39d397 | 4995 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4996 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4997 | sizeof(struct btrfs_item) * |
85e21bac | 4998 | (nritems - slot - nr)); |
be0e5c09 | 4999 | } |
85e21bac CM |
5000 | btrfs_set_header_nritems(leaf, nritems - nr); |
5001 | nritems -= nr; | |
5f39d397 | 5002 | |
74123bd7 | 5003 | /* delete the leaf if we've emptied it */ |
7518a238 | 5004 | if (nritems == 0) { |
5f39d397 CM |
5005 | if (leaf == root->node) { |
5006 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5007 | } else { |
f0486c68 | 5008 | btrfs_set_path_blocking(path); |
7c302b49 | 5009 | clean_tree_block(fs_info, leaf); |
143bede5 | 5010 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5011 | } |
be0e5c09 | 5012 | } else { |
7518a238 | 5013 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5014 | if (slot == 0) { |
5f39d397 CM |
5015 | struct btrfs_disk_key disk_key; |
5016 | ||
5017 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 5018 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 5019 | } |
aa5d6bed | 5020 | |
74123bd7 | 5021 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5022 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5023 | /* push_leaf_left fixes the path. |
5024 | * make sure the path still points to our leaf | |
5025 | * for possible call to del_ptr below | |
5026 | */ | |
4920c9ac | 5027 | slot = path->slots[1]; |
5f39d397 CM |
5028 | extent_buffer_get(leaf); |
5029 | ||
b9473439 | 5030 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5031 | wret = push_leaf_left(trans, root, path, 1, 1, |
5032 | 1, (u32)-1); | |
54aa1f4d | 5033 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5034 | ret = wret; |
5f39d397 CM |
5035 | |
5036 | if (path->nodes[0] == leaf && | |
5037 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5038 | wret = push_leaf_right(trans, root, path, 1, |
5039 | 1, 1, 0); | |
54aa1f4d | 5040 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5041 | ret = wret; |
5042 | } | |
5f39d397 CM |
5043 | |
5044 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5045 | path->slots[1] = slot; |
143bede5 | 5046 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5047 | free_extent_buffer(leaf); |
143bede5 | 5048 | ret = 0; |
5de08d7d | 5049 | } else { |
925baedd CM |
5050 | /* if we're still in the path, make sure |
5051 | * we're dirty. Otherwise, one of the | |
5052 | * push_leaf functions must have already | |
5053 | * dirtied this buffer | |
5054 | */ | |
5055 | if (path->nodes[0] == leaf) | |
5056 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5057 | free_extent_buffer(leaf); |
be0e5c09 | 5058 | } |
d5719762 | 5059 | } else { |
5f39d397 | 5060 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5061 | } |
5062 | } | |
aa5d6bed | 5063 | return ret; |
be0e5c09 CM |
5064 | } |
5065 | ||
7bb86316 | 5066 | /* |
925baedd | 5067 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5068 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5069 | * returns < 0 on io errors. | |
d352ac68 CM |
5070 | * |
5071 | * This may release the path, and so you may lose any locks held at the | |
5072 | * time you call it. | |
7bb86316 | 5073 | */ |
16e7549f | 5074 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5075 | { |
925baedd CM |
5076 | struct btrfs_key key; |
5077 | struct btrfs_disk_key found_key; | |
5078 | int ret; | |
7bb86316 | 5079 | |
925baedd | 5080 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5081 | |
e8b0d724 | 5082 | if (key.offset > 0) { |
925baedd | 5083 | key.offset--; |
e8b0d724 | 5084 | } else if (key.type > 0) { |
925baedd | 5085 | key.type--; |
e8b0d724 FDBM |
5086 | key.offset = (u64)-1; |
5087 | } else if (key.objectid > 0) { | |
925baedd | 5088 | key.objectid--; |
e8b0d724 FDBM |
5089 | key.type = (u8)-1; |
5090 | key.offset = (u64)-1; | |
5091 | } else { | |
925baedd | 5092 | return 1; |
e8b0d724 | 5093 | } |
7bb86316 | 5094 | |
b3b4aa74 | 5095 | btrfs_release_path(path); |
925baedd CM |
5096 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5097 | if (ret < 0) | |
5098 | return ret; | |
5099 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5100 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5101 | /* |
5102 | * We might have had an item with the previous key in the tree right | |
5103 | * before we released our path. And after we released our path, that | |
5104 | * item might have been pushed to the first slot (0) of the leaf we | |
5105 | * were holding due to a tree balance. Alternatively, an item with the | |
5106 | * previous key can exist as the only element of a leaf (big fat item). | |
5107 | * Therefore account for these 2 cases, so that our callers (like | |
5108 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5109 | * the previous key we computed above. | |
5110 | */ | |
5111 | if (ret <= 0) | |
925baedd CM |
5112 | return 0; |
5113 | return 1; | |
7bb86316 CM |
5114 | } |
5115 | ||
3f157a2f CM |
5116 | /* |
5117 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5118 | * for nodes or leaves that are have a minimum transaction id. |
5119 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5120 | * |
5121 | * This does not cow, but it does stuff the starting key it finds back | |
5122 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5123 | * key and get a writable path. | |
5124 | * | |
3f157a2f CM |
5125 | * This honors path->lowest_level to prevent descent past a given level |
5126 | * of the tree. | |
5127 | * | |
d352ac68 CM |
5128 | * min_trans indicates the oldest transaction that you are interested |
5129 | * in walking through. Any nodes or leaves older than min_trans are | |
5130 | * skipped over (without reading them). | |
5131 | * | |
3f157a2f CM |
5132 | * returns zero if something useful was found, < 0 on error and 1 if there |
5133 | * was nothing in the tree that matched the search criteria. | |
5134 | */ | |
5135 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5136 | struct btrfs_path *path, |
3f157a2f CM |
5137 | u64 min_trans) |
5138 | { | |
2ff7e61e | 5139 | struct btrfs_fs_info *fs_info = root->fs_info; |
3f157a2f CM |
5140 | struct extent_buffer *cur; |
5141 | struct btrfs_key found_key; | |
5142 | int slot; | |
9652480b | 5143 | int sret; |
3f157a2f CM |
5144 | u32 nritems; |
5145 | int level; | |
5146 | int ret = 1; | |
f98de9b9 | 5147 | int keep_locks = path->keep_locks; |
3f157a2f | 5148 | |
f98de9b9 | 5149 | path->keep_locks = 1; |
3f157a2f | 5150 | again: |
bd681513 | 5151 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5152 | level = btrfs_header_level(cur); |
e02119d5 | 5153 | WARN_ON(path->nodes[level]); |
3f157a2f | 5154 | path->nodes[level] = cur; |
bd681513 | 5155 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5156 | |
5157 | if (btrfs_header_generation(cur) < min_trans) { | |
5158 | ret = 1; | |
5159 | goto out; | |
5160 | } | |
d397712b | 5161 | while (1) { |
3f157a2f CM |
5162 | nritems = btrfs_header_nritems(cur); |
5163 | level = btrfs_header_level(cur); | |
a74b35ec | 5164 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
cbca7d59 FM |
5165 | if (sret < 0) { |
5166 | ret = sret; | |
5167 | goto out; | |
5168 | } | |
3f157a2f | 5169 | |
323ac95b CM |
5170 | /* at the lowest level, we're done, setup the path and exit */ |
5171 | if (level == path->lowest_level) { | |
e02119d5 CM |
5172 | if (slot >= nritems) |
5173 | goto find_next_key; | |
3f157a2f CM |
5174 | ret = 0; |
5175 | path->slots[level] = slot; | |
5176 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5177 | goto out; | |
5178 | } | |
9652480b Y |
5179 | if (sret && slot > 0) |
5180 | slot--; | |
3f157a2f | 5181 | /* |
de78b51a ES |
5182 | * check this node pointer against the min_trans parameters. |
5183 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5184 | */ |
d397712b | 5185 | while (slot < nritems) { |
3f157a2f | 5186 | u64 gen; |
e02119d5 | 5187 | |
3f157a2f CM |
5188 | gen = btrfs_node_ptr_generation(cur, slot); |
5189 | if (gen < min_trans) { | |
5190 | slot++; | |
5191 | continue; | |
5192 | } | |
de78b51a | 5193 | break; |
3f157a2f | 5194 | } |
e02119d5 | 5195 | find_next_key: |
3f157a2f CM |
5196 | /* |
5197 | * we didn't find a candidate key in this node, walk forward | |
5198 | * and find another one | |
5199 | */ | |
5200 | if (slot >= nritems) { | |
e02119d5 | 5201 | path->slots[level] = slot; |
b4ce94de | 5202 | btrfs_set_path_blocking(path); |
e02119d5 | 5203 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5204 | min_trans); |
e02119d5 | 5205 | if (sret == 0) { |
b3b4aa74 | 5206 | btrfs_release_path(path); |
3f157a2f CM |
5207 | goto again; |
5208 | } else { | |
5209 | goto out; | |
5210 | } | |
5211 | } | |
5212 | /* save our key for returning back */ | |
5213 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5214 | path->slots[level] = slot; | |
5215 | if (level == path->lowest_level) { | |
5216 | ret = 0; | |
3f157a2f CM |
5217 | goto out; |
5218 | } | |
b4ce94de | 5219 | btrfs_set_path_blocking(path); |
2ff7e61e | 5220 | cur = read_node_slot(fs_info, cur, slot); |
fb770ae4 LB |
5221 | if (IS_ERR(cur)) { |
5222 | ret = PTR_ERR(cur); | |
5223 | goto out; | |
5224 | } | |
3f157a2f | 5225 | |
bd681513 | 5226 | btrfs_tree_read_lock(cur); |
b4ce94de | 5227 | |
bd681513 | 5228 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5229 | path->nodes[level - 1] = cur; |
f7c79f30 | 5230 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5231 | } |
5232 | out: | |
f98de9b9 FM |
5233 | path->keep_locks = keep_locks; |
5234 | if (ret == 0) { | |
5235 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5236 | btrfs_set_path_blocking(path); | |
3f157a2f | 5237 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5238 | } |
3f157a2f CM |
5239 | return ret; |
5240 | } | |
5241 | ||
2ff7e61e | 5242 | static int tree_move_down(struct btrfs_fs_info *fs_info, |
7069830a | 5243 | struct btrfs_path *path, |
ab6a43e1 | 5244 | int *level) |
7069830a | 5245 | { |
fb770ae4 LB |
5246 | struct extent_buffer *eb; |
5247 | ||
74dd17fb | 5248 | BUG_ON(*level == 0); |
2ff7e61e | 5249 | eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]); |
fb770ae4 LB |
5250 | if (IS_ERR(eb)) |
5251 | return PTR_ERR(eb); | |
5252 | ||
5253 | path->nodes[*level - 1] = eb; | |
7069830a AB |
5254 | path->slots[*level - 1] = 0; |
5255 | (*level)--; | |
fb770ae4 | 5256 | return 0; |
7069830a AB |
5257 | } |
5258 | ||
f1e30261 | 5259 | static int tree_move_next_or_upnext(struct btrfs_path *path, |
7069830a AB |
5260 | int *level, int root_level) |
5261 | { | |
5262 | int ret = 0; | |
5263 | int nritems; | |
5264 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5265 | ||
5266 | path->slots[*level]++; | |
5267 | ||
74dd17fb | 5268 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5269 | if (*level == root_level) |
5270 | return -1; | |
5271 | ||
5272 | /* move upnext */ | |
5273 | path->slots[*level] = 0; | |
5274 | free_extent_buffer(path->nodes[*level]); | |
5275 | path->nodes[*level] = NULL; | |
5276 | (*level)++; | |
5277 | path->slots[*level]++; | |
5278 | ||
5279 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5280 | ret = 1; | |
5281 | } | |
5282 | return ret; | |
5283 | } | |
5284 | ||
5285 | /* | |
5286 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5287 | * or down. | |
5288 | */ | |
2ff7e61e | 5289 | static int tree_advance(struct btrfs_fs_info *fs_info, |
7069830a AB |
5290 | struct btrfs_path *path, |
5291 | int *level, int root_level, | |
5292 | int allow_down, | |
5293 | struct btrfs_key *key) | |
5294 | { | |
5295 | int ret; | |
5296 | ||
5297 | if (*level == 0 || !allow_down) { | |
f1e30261 | 5298 | ret = tree_move_next_or_upnext(path, level, root_level); |
7069830a | 5299 | } else { |
ab6a43e1 | 5300 | ret = tree_move_down(fs_info, path, level); |
7069830a AB |
5301 | } |
5302 | if (ret >= 0) { | |
5303 | if (*level == 0) | |
5304 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5305 | path->slots[*level]); | |
5306 | else | |
5307 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5308 | path->slots[*level]); | |
5309 | } | |
5310 | return ret; | |
5311 | } | |
5312 | ||
2ff7e61e | 5313 | static int tree_compare_item(struct btrfs_path *left_path, |
7069830a AB |
5314 | struct btrfs_path *right_path, |
5315 | char *tmp_buf) | |
5316 | { | |
5317 | int cmp; | |
5318 | int len1, len2; | |
5319 | unsigned long off1, off2; | |
5320 | ||
5321 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5322 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5323 | if (len1 != len2) | |
5324 | return 1; | |
5325 | ||
5326 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5327 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5328 | right_path->slots[0]); | |
5329 | ||
5330 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5331 | ||
5332 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5333 | if (cmp) | |
5334 | return 1; | |
5335 | return 0; | |
5336 | } | |
5337 | ||
5338 | #define ADVANCE 1 | |
5339 | #define ADVANCE_ONLY_NEXT -1 | |
5340 | ||
5341 | /* | |
5342 | * This function compares two trees and calls the provided callback for | |
5343 | * every changed/new/deleted item it finds. | |
5344 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5345 | * the compare pretty fast on snapshotted subvolumes. | |
5346 | * | |
5347 | * This currently works on commit roots only. As commit roots are read only, | |
5348 | * we don't do any locking. The commit roots are protected with transactions. | |
5349 | * Transactions are ended and rejoined when a commit is tried in between. | |
5350 | * | |
5351 | * This function checks for modifications done to the trees while comparing. | |
5352 | * If it detects a change, it aborts immediately. | |
5353 | */ | |
5354 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5355 | struct btrfs_root *right_root, | |
5356 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5357 | { | |
0b246afa | 5358 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5359 | int ret; |
5360 | int cmp; | |
7069830a AB |
5361 | struct btrfs_path *left_path = NULL; |
5362 | struct btrfs_path *right_path = NULL; | |
5363 | struct btrfs_key left_key; | |
5364 | struct btrfs_key right_key; | |
5365 | char *tmp_buf = NULL; | |
5366 | int left_root_level; | |
5367 | int right_root_level; | |
5368 | int left_level; | |
5369 | int right_level; | |
5370 | int left_end_reached; | |
5371 | int right_end_reached; | |
5372 | int advance_left; | |
5373 | int advance_right; | |
5374 | u64 left_blockptr; | |
5375 | u64 right_blockptr; | |
6baa4293 FM |
5376 | u64 left_gen; |
5377 | u64 right_gen; | |
7069830a AB |
5378 | |
5379 | left_path = btrfs_alloc_path(); | |
5380 | if (!left_path) { | |
5381 | ret = -ENOMEM; | |
5382 | goto out; | |
5383 | } | |
5384 | right_path = btrfs_alloc_path(); | |
5385 | if (!right_path) { | |
5386 | ret = -ENOMEM; | |
5387 | goto out; | |
5388 | } | |
5389 | ||
752ade68 | 5390 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); |
7069830a | 5391 | if (!tmp_buf) { |
752ade68 MH |
5392 | ret = -ENOMEM; |
5393 | goto out; | |
7069830a AB |
5394 | } |
5395 | ||
5396 | left_path->search_commit_root = 1; | |
5397 | left_path->skip_locking = 1; | |
5398 | right_path->search_commit_root = 1; | |
5399 | right_path->skip_locking = 1; | |
5400 | ||
7069830a AB |
5401 | /* |
5402 | * Strategy: Go to the first items of both trees. Then do | |
5403 | * | |
5404 | * If both trees are at level 0 | |
5405 | * Compare keys of current items | |
5406 | * If left < right treat left item as new, advance left tree | |
5407 | * and repeat | |
5408 | * If left > right treat right item as deleted, advance right tree | |
5409 | * and repeat | |
5410 | * If left == right do deep compare of items, treat as changed if | |
5411 | * needed, advance both trees and repeat | |
5412 | * If both trees are at the same level but not at level 0 | |
5413 | * Compare keys of current nodes/leafs | |
5414 | * If left < right advance left tree and repeat | |
5415 | * If left > right advance right tree and repeat | |
5416 | * If left == right compare blockptrs of the next nodes/leafs | |
5417 | * If they match advance both trees but stay at the same level | |
5418 | * and repeat | |
5419 | * If they don't match advance both trees while allowing to go | |
5420 | * deeper and repeat | |
5421 | * If tree levels are different | |
5422 | * Advance the tree that needs it and repeat | |
5423 | * | |
5424 | * Advancing a tree means: | |
5425 | * If we are at level 0, try to go to the next slot. If that's not | |
5426 | * possible, go one level up and repeat. Stop when we found a level | |
5427 | * where we could go to the next slot. We may at this point be on a | |
5428 | * node or a leaf. | |
5429 | * | |
5430 | * If we are not at level 0 and not on shared tree blocks, go one | |
5431 | * level deeper. | |
5432 | * | |
5433 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5434 | * the right if possible or go up and right. | |
5435 | */ | |
5436 | ||
0b246afa | 5437 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5438 | left_level = btrfs_header_level(left_root->commit_root); |
5439 | left_root_level = left_level; | |
6f2f0b39 RK |
5440 | left_path->nodes[left_level] = |
5441 | btrfs_clone_extent_buffer(left_root->commit_root); | |
5442 | if (!left_path->nodes[left_level]) { | |
5443 | up_read(&fs_info->commit_root_sem); | |
5444 | ret = -ENOMEM; | |
5445 | goto out; | |
5446 | } | |
7069830a AB |
5447 | |
5448 | right_level = btrfs_header_level(right_root->commit_root); | |
5449 | right_root_level = right_level; | |
6f2f0b39 RK |
5450 | right_path->nodes[right_level] = |
5451 | btrfs_clone_extent_buffer(right_root->commit_root); | |
5452 | if (!right_path->nodes[right_level]) { | |
5453 | up_read(&fs_info->commit_root_sem); | |
5454 | ret = -ENOMEM; | |
5455 | goto out; | |
5456 | } | |
0b246afa | 5457 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5458 | |
5459 | if (left_level == 0) | |
5460 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5461 | &left_key, left_path->slots[left_level]); | |
5462 | else | |
5463 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5464 | &left_key, left_path->slots[left_level]); | |
5465 | if (right_level == 0) | |
5466 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5467 | &right_key, right_path->slots[right_level]); | |
5468 | else | |
5469 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5470 | &right_key, right_path->slots[right_level]); | |
5471 | ||
5472 | left_end_reached = right_end_reached = 0; | |
5473 | advance_left = advance_right = 0; | |
5474 | ||
5475 | while (1) { | |
7069830a | 5476 | if (advance_left && !left_end_reached) { |
2ff7e61e | 5477 | ret = tree_advance(fs_info, left_path, &left_level, |
7069830a AB |
5478 | left_root_level, |
5479 | advance_left != ADVANCE_ONLY_NEXT, | |
5480 | &left_key); | |
fb770ae4 | 5481 | if (ret == -1) |
7069830a | 5482 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5483 | else if (ret < 0) |
5484 | goto out; | |
7069830a AB |
5485 | advance_left = 0; |
5486 | } | |
5487 | if (advance_right && !right_end_reached) { | |
2ff7e61e | 5488 | ret = tree_advance(fs_info, right_path, &right_level, |
7069830a AB |
5489 | right_root_level, |
5490 | advance_right != ADVANCE_ONLY_NEXT, | |
5491 | &right_key); | |
fb770ae4 | 5492 | if (ret == -1) |
7069830a | 5493 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5494 | else if (ret < 0) |
5495 | goto out; | |
7069830a AB |
5496 | advance_right = 0; |
5497 | } | |
5498 | ||
5499 | if (left_end_reached && right_end_reached) { | |
5500 | ret = 0; | |
5501 | goto out; | |
5502 | } else if (left_end_reached) { | |
5503 | if (right_level == 0) { | |
ee8c494f | 5504 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5505 | &right_key, |
5506 | BTRFS_COMPARE_TREE_DELETED, | |
5507 | ctx); | |
5508 | if (ret < 0) | |
5509 | goto out; | |
5510 | } | |
5511 | advance_right = ADVANCE; | |
5512 | continue; | |
5513 | } else if (right_end_reached) { | |
5514 | if (left_level == 0) { | |
ee8c494f | 5515 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5516 | &left_key, |
5517 | BTRFS_COMPARE_TREE_NEW, | |
5518 | ctx); | |
5519 | if (ret < 0) | |
5520 | goto out; | |
5521 | } | |
5522 | advance_left = ADVANCE; | |
5523 | continue; | |
5524 | } | |
5525 | ||
5526 | if (left_level == 0 && right_level == 0) { | |
5527 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5528 | if (cmp < 0) { | |
ee8c494f | 5529 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5530 | &left_key, |
5531 | BTRFS_COMPARE_TREE_NEW, | |
5532 | ctx); | |
5533 | if (ret < 0) | |
5534 | goto out; | |
5535 | advance_left = ADVANCE; | |
5536 | } else if (cmp > 0) { | |
ee8c494f | 5537 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5538 | &right_key, |
5539 | BTRFS_COMPARE_TREE_DELETED, | |
5540 | ctx); | |
5541 | if (ret < 0) | |
5542 | goto out; | |
5543 | advance_right = ADVANCE; | |
5544 | } else { | |
b99d9a6a | 5545 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5546 | |
74dd17fb | 5547 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
2ff7e61e JM |
5548 | ret = tree_compare_item(left_path, right_path, |
5549 | tmp_buf); | |
ba5e8f2e | 5550 | if (ret) |
b99d9a6a | 5551 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5552 | else |
b99d9a6a | 5553 | result = BTRFS_COMPARE_TREE_SAME; |
ee8c494f | 5554 | ret = changed_cb(left_path, right_path, |
b99d9a6a | 5555 | &left_key, result, ctx); |
ba5e8f2e JB |
5556 | if (ret < 0) |
5557 | goto out; | |
7069830a AB |
5558 | advance_left = ADVANCE; |
5559 | advance_right = ADVANCE; | |
5560 | } | |
5561 | } else if (left_level == right_level) { | |
5562 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5563 | if (cmp < 0) { | |
5564 | advance_left = ADVANCE; | |
5565 | } else if (cmp > 0) { | |
5566 | advance_right = ADVANCE; | |
5567 | } else { | |
5568 | left_blockptr = btrfs_node_blockptr( | |
5569 | left_path->nodes[left_level], | |
5570 | left_path->slots[left_level]); | |
5571 | right_blockptr = btrfs_node_blockptr( | |
5572 | right_path->nodes[right_level], | |
5573 | right_path->slots[right_level]); | |
6baa4293 FM |
5574 | left_gen = btrfs_node_ptr_generation( |
5575 | left_path->nodes[left_level], | |
5576 | left_path->slots[left_level]); | |
5577 | right_gen = btrfs_node_ptr_generation( | |
5578 | right_path->nodes[right_level], | |
5579 | right_path->slots[right_level]); | |
5580 | if (left_blockptr == right_blockptr && | |
5581 | left_gen == right_gen) { | |
7069830a AB |
5582 | /* |
5583 | * As we're on a shared block, don't | |
5584 | * allow to go deeper. | |
5585 | */ | |
5586 | advance_left = ADVANCE_ONLY_NEXT; | |
5587 | advance_right = ADVANCE_ONLY_NEXT; | |
5588 | } else { | |
5589 | advance_left = ADVANCE; | |
5590 | advance_right = ADVANCE; | |
5591 | } | |
5592 | } | |
5593 | } else if (left_level < right_level) { | |
5594 | advance_right = ADVANCE; | |
5595 | } else { | |
5596 | advance_left = ADVANCE; | |
5597 | } | |
5598 | } | |
5599 | ||
5600 | out: | |
5601 | btrfs_free_path(left_path); | |
5602 | btrfs_free_path(right_path); | |
8f282f71 | 5603 | kvfree(tmp_buf); |
7069830a AB |
5604 | return ret; |
5605 | } | |
5606 | ||
3f157a2f CM |
5607 | /* |
5608 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5609 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5610 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5611 | * |
5612 | * 0 is returned if another key is found, < 0 if there are any errors | |
5613 | * and 1 is returned if there are no higher keys in the tree | |
5614 | * | |
5615 | * path->keep_locks should be set to 1 on the search made before | |
5616 | * calling this function. | |
5617 | */ | |
e7a84565 | 5618 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5619 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5620 | { |
e7a84565 CM |
5621 | int slot; |
5622 | struct extent_buffer *c; | |
5623 | ||
934d375b | 5624 | WARN_ON(!path->keep_locks); |
d397712b | 5625 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5626 | if (!path->nodes[level]) |
5627 | return 1; | |
5628 | ||
5629 | slot = path->slots[level] + 1; | |
5630 | c = path->nodes[level]; | |
3f157a2f | 5631 | next: |
e7a84565 | 5632 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5633 | int ret; |
5634 | int orig_lowest; | |
5635 | struct btrfs_key cur_key; | |
5636 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5637 | !path->nodes[level + 1]) | |
e7a84565 | 5638 | return 1; |
33c66f43 YZ |
5639 | |
5640 | if (path->locks[level + 1]) { | |
5641 | level++; | |
5642 | continue; | |
5643 | } | |
5644 | ||
5645 | slot = btrfs_header_nritems(c) - 1; | |
5646 | if (level == 0) | |
5647 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5648 | else | |
5649 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5650 | ||
5651 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5652 | btrfs_release_path(path); |
33c66f43 YZ |
5653 | path->lowest_level = level; |
5654 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5655 | 0, 0); | |
5656 | path->lowest_level = orig_lowest; | |
5657 | if (ret < 0) | |
5658 | return ret; | |
5659 | ||
5660 | c = path->nodes[level]; | |
5661 | slot = path->slots[level]; | |
5662 | if (ret == 0) | |
5663 | slot++; | |
5664 | goto next; | |
e7a84565 | 5665 | } |
33c66f43 | 5666 | |
e7a84565 CM |
5667 | if (level == 0) |
5668 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5669 | else { |
3f157a2f CM |
5670 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5671 | ||
3f157a2f CM |
5672 | if (gen < min_trans) { |
5673 | slot++; | |
5674 | goto next; | |
5675 | } | |
e7a84565 | 5676 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5677 | } |
e7a84565 CM |
5678 | return 0; |
5679 | } | |
5680 | return 1; | |
5681 | } | |
5682 | ||
97571fd0 | 5683 | /* |
925baedd | 5684 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5685 | * returns 0 if it found something or 1 if there are no greater leaves. |
5686 | * returns < 0 on io errors. | |
97571fd0 | 5687 | */ |
234b63a0 | 5688 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5689 | { |
5690 | return btrfs_next_old_leaf(root, path, 0); | |
5691 | } | |
5692 | ||
5693 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5694 | u64 time_seq) | |
d97e63b6 CM |
5695 | { |
5696 | int slot; | |
8e73f275 | 5697 | int level; |
5f39d397 | 5698 | struct extent_buffer *c; |
8e73f275 | 5699 | struct extent_buffer *next; |
925baedd CM |
5700 | struct btrfs_key key; |
5701 | u32 nritems; | |
5702 | int ret; | |
8e73f275 | 5703 | int old_spinning = path->leave_spinning; |
bd681513 | 5704 | int next_rw_lock = 0; |
925baedd CM |
5705 | |
5706 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5707 | if (nritems == 0) |
925baedd | 5708 | return 1; |
925baedd | 5709 | |
8e73f275 CM |
5710 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5711 | again: | |
5712 | level = 1; | |
5713 | next = NULL; | |
bd681513 | 5714 | next_rw_lock = 0; |
b3b4aa74 | 5715 | btrfs_release_path(path); |
8e73f275 | 5716 | |
a2135011 | 5717 | path->keep_locks = 1; |
31533fb2 | 5718 | path->leave_spinning = 1; |
8e73f275 | 5719 | |
3d7806ec JS |
5720 | if (time_seq) |
5721 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5722 | else | |
5723 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5724 | path->keep_locks = 0; |
5725 | ||
5726 | if (ret < 0) | |
5727 | return ret; | |
5728 | ||
a2135011 | 5729 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5730 | /* |
5731 | * by releasing the path above we dropped all our locks. A balance | |
5732 | * could have added more items next to the key that used to be | |
5733 | * at the very end of the block. So, check again here and | |
5734 | * advance the path if there are now more items available. | |
5735 | */ | |
a2135011 | 5736 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5737 | if (ret == 0) |
5738 | path->slots[0]++; | |
8e73f275 | 5739 | ret = 0; |
925baedd CM |
5740 | goto done; |
5741 | } | |
0b43e04f LB |
5742 | /* |
5743 | * So the above check misses one case: | |
5744 | * - after releasing the path above, someone has removed the item that | |
5745 | * used to be at the very end of the block, and balance between leafs | |
5746 | * gets another one with bigger key.offset to replace it. | |
5747 | * | |
5748 | * This one should be returned as well, or we can get leaf corruption | |
5749 | * later(esp. in __btrfs_drop_extents()). | |
5750 | * | |
5751 | * And a bit more explanation about this check, | |
5752 | * with ret > 0, the key isn't found, the path points to the slot | |
5753 | * where it should be inserted, so the path->slots[0] item must be the | |
5754 | * bigger one. | |
5755 | */ | |
5756 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5757 | ret = 0; | |
5758 | goto done; | |
5759 | } | |
d97e63b6 | 5760 | |
d397712b | 5761 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5762 | if (!path->nodes[level]) { |
5763 | ret = 1; | |
5764 | goto done; | |
5765 | } | |
5f39d397 | 5766 | |
d97e63b6 CM |
5767 | slot = path->slots[level] + 1; |
5768 | c = path->nodes[level]; | |
5f39d397 | 5769 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5770 | level++; |
8e73f275 CM |
5771 | if (level == BTRFS_MAX_LEVEL) { |
5772 | ret = 1; | |
5773 | goto done; | |
5774 | } | |
d97e63b6 CM |
5775 | continue; |
5776 | } | |
5f39d397 | 5777 | |
925baedd | 5778 | if (next) { |
bd681513 | 5779 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5780 | free_extent_buffer(next); |
925baedd | 5781 | } |
5f39d397 | 5782 | |
8e73f275 | 5783 | next = c; |
bd681513 | 5784 | next_rw_lock = path->locks[level]; |
d07b8528 | 5785 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5786 | slot, &key); |
8e73f275 CM |
5787 | if (ret == -EAGAIN) |
5788 | goto again; | |
5f39d397 | 5789 | |
76a05b35 | 5790 | if (ret < 0) { |
b3b4aa74 | 5791 | btrfs_release_path(path); |
76a05b35 CM |
5792 | goto done; |
5793 | } | |
5794 | ||
5cd57b2c | 5795 | if (!path->skip_locking) { |
bd681513 | 5796 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5797 | if (!ret && time_seq) { |
5798 | /* | |
5799 | * If we don't get the lock, we may be racing | |
5800 | * with push_leaf_left, holding that lock while | |
5801 | * itself waiting for the leaf we've currently | |
5802 | * locked. To solve this situation, we give up | |
5803 | * on our lock and cycle. | |
5804 | */ | |
cf538830 | 5805 | free_extent_buffer(next); |
d42244a0 JS |
5806 | btrfs_release_path(path); |
5807 | cond_resched(); | |
5808 | goto again; | |
5809 | } | |
8e73f275 CM |
5810 | if (!ret) { |
5811 | btrfs_set_path_blocking(path); | |
bd681513 | 5812 | btrfs_tree_read_lock(next); |
8e73f275 | 5813 | } |
31533fb2 | 5814 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5815 | } |
d97e63b6 CM |
5816 | break; |
5817 | } | |
5818 | path->slots[level] = slot; | |
d397712b | 5819 | while (1) { |
d97e63b6 CM |
5820 | level--; |
5821 | c = path->nodes[level]; | |
925baedd | 5822 | if (path->locks[level]) |
bd681513 | 5823 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5824 | |
5f39d397 | 5825 | free_extent_buffer(c); |
d97e63b6 CM |
5826 | path->nodes[level] = next; |
5827 | path->slots[level] = 0; | |
a74a4b97 | 5828 | if (!path->skip_locking) |
bd681513 | 5829 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5830 | if (!level) |
5831 | break; | |
b4ce94de | 5832 | |
d07b8528 | 5833 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5834 | 0, &key); |
8e73f275 CM |
5835 | if (ret == -EAGAIN) |
5836 | goto again; | |
5837 | ||
76a05b35 | 5838 | if (ret < 0) { |
b3b4aa74 | 5839 | btrfs_release_path(path); |
76a05b35 CM |
5840 | goto done; |
5841 | } | |
5842 | ||
5cd57b2c | 5843 | if (!path->skip_locking) { |
bd681513 | 5844 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5845 | if (!ret) { |
5846 | btrfs_set_path_blocking(path); | |
bd681513 | 5847 | btrfs_tree_read_lock(next); |
bd681513 | 5848 | } |
31533fb2 | 5849 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5850 | } |
d97e63b6 | 5851 | } |
8e73f275 | 5852 | ret = 0; |
925baedd | 5853 | done: |
f7c79f30 | 5854 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5855 | path->leave_spinning = old_spinning; |
5856 | if (!old_spinning) | |
5857 | btrfs_set_path_blocking(path); | |
5858 | ||
5859 | return ret; | |
d97e63b6 | 5860 | } |
0b86a832 | 5861 | |
3f157a2f CM |
5862 | /* |
5863 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5864 | * searching until it gets past min_objectid or finds an item of 'type' | |
5865 | * | |
5866 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5867 | */ | |
0b86a832 CM |
5868 | int btrfs_previous_item(struct btrfs_root *root, |
5869 | struct btrfs_path *path, u64 min_objectid, | |
5870 | int type) | |
5871 | { | |
5872 | struct btrfs_key found_key; | |
5873 | struct extent_buffer *leaf; | |
e02119d5 | 5874 | u32 nritems; |
0b86a832 CM |
5875 | int ret; |
5876 | ||
d397712b | 5877 | while (1) { |
0b86a832 | 5878 | if (path->slots[0] == 0) { |
b4ce94de | 5879 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5880 | ret = btrfs_prev_leaf(root, path); |
5881 | if (ret != 0) | |
5882 | return ret; | |
5883 | } else { | |
5884 | path->slots[0]--; | |
5885 | } | |
5886 | leaf = path->nodes[0]; | |
e02119d5 CM |
5887 | nritems = btrfs_header_nritems(leaf); |
5888 | if (nritems == 0) | |
5889 | return 1; | |
5890 | if (path->slots[0] == nritems) | |
5891 | path->slots[0]--; | |
5892 | ||
0b86a832 | 5893 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5894 | if (found_key.objectid < min_objectid) |
5895 | break; | |
0a4eefbb YZ |
5896 | if (found_key.type == type) |
5897 | return 0; | |
e02119d5 CM |
5898 | if (found_key.objectid == min_objectid && |
5899 | found_key.type < type) | |
5900 | break; | |
0b86a832 CM |
5901 | } |
5902 | return 1; | |
5903 | } | |
ade2e0b3 WS |
5904 | |
5905 | /* | |
5906 | * search in extent tree to find a previous Metadata/Data extent item with | |
5907 | * min objecitd. | |
5908 | * | |
5909 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5910 | */ | |
5911 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5912 | struct btrfs_path *path, u64 min_objectid) | |
5913 | { | |
5914 | struct btrfs_key found_key; | |
5915 | struct extent_buffer *leaf; | |
5916 | u32 nritems; | |
5917 | int ret; | |
5918 | ||
5919 | while (1) { | |
5920 | if (path->slots[0] == 0) { | |
5921 | btrfs_set_path_blocking(path); | |
5922 | ret = btrfs_prev_leaf(root, path); | |
5923 | if (ret != 0) | |
5924 | return ret; | |
5925 | } else { | |
5926 | path->slots[0]--; | |
5927 | } | |
5928 | leaf = path->nodes[0]; | |
5929 | nritems = btrfs_header_nritems(leaf); | |
5930 | if (nritems == 0) | |
5931 | return 1; | |
5932 | if (path->slots[0] == nritems) | |
5933 | path->slots[0]--; | |
5934 | ||
5935 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5936 | if (found_key.objectid < min_objectid) | |
5937 | break; | |
5938 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5939 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5940 | return 0; | |
5941 | if (found_key.objectid == min_objectid && | |
5942 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5943 | break; | |
5944 | } | |
5945 | return 1; | |
5946 | } |