]>
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 | 24 | struct extent_buffer *dst, |
971a1f66 | 25 | struct extent_buffer *src, int empty); |
5f39d397 | 26 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
27 | struct extent_buffer *dst_buf, |
28 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
29 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
30 | int level, int slot); | |
d97e63b6 | 31 | |
af024ed2 JT |
32 | static const struct btrfs_csums { |
33 | u16 size; | |
34 | const char *name; | |
35 | } btrfs_csums[] = { | |
36 | [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, | |
37 | }; | |
38 | ||
39 | int btrfs_super_csum_size(const struct btrfs_super_block *s) | |
40 | { | |
41 | u16 t = btrfs_super_csum_type(s); | |
42 | /* | |
43 | * csum type is validated at mount time | |
44 | */ | |
45 | return btrfs_csums[t].size; | |
46 | } | |
47 | ||
48 | const char *btrfs_super_csum_name(u16 csum_type) | |
49 | { | |
50 | /* csum type is validated at mount time */ | |
51 | return btrfs_csums[csum_type].name; | |
52 | } | |
53 | ||
df24a2b9 | 54 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 55 | { |
e2c89907 | 56 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
57 | } |
58 | ||
b4ce94de CM |
59 | /* |
60 | * set all locked nodes in the path to blocking locks. This should | |
61 | * be done before scheduling | |
62 | */ | |
63 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
64 | { | |
65 | int i; | |
66 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
67 | if (!p->nodes[i] || !p->locks[i]) |
68 | continue; | |
766ece54 DS |
69 | /* |
70 | * If we currently have a spinning reader or writer lock this | |
71 | * will bump the count of blocking holders and drop the | |
72 | * spinlock. | |
73 | */ | |
74 | if (p->locks[i] == BTRFS_READ_LOCK) { | |
75 | btrfs_set_lock_blocking_read(p->nodes[i]); | |
bd681513 | 76 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; |
766ece54 DS |
77 | } else if (p->locks[i] == BTRFS_WRITE_LOCK) { |
78 | btrfs_set_lock_blocking_write(p->nodes[i]); | |
bd681513 | 79 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; |
766ece54 | 80 | } |
b4ce94de CM |
81 | } |
82 | } | |
83 | ||
d352ac68 | 84 | /* this also releases the path */ |
df24a2b9 | 85 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 86 | { |
ff175d57 JJ |
87 | if (!p) |
88 | return; | |
b3b4aa74 | 89 | btrfs_release_path(p); |
df24a2b9 | 90 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
91 | } |
92 | ||
d352ac68 CM |
93 | /* |
94 | * path release drops references on the extent buffers in the path | |
95 | * and it drops any locks held by this path | |
96 | * | |
97 | * It is safe to call this on paths that no locks or extent buffers held. | |
98 | */ | |
b3b4aa74 | 99 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
100 | { |
101 | int i; | |
a2135011 | 102 | |
234b63a0 | 103 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 104 | p->slots[i] = 0; |
eb60ceac | 105 | if (!p->nodes[i]) |
925baedd CM |
106 | continue; |
107 | if (p->locks[i]) { | |
bd681513 | 108 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
109 | p->locks[i] = 0; |
110 | } | |
5f39d397 | 111 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 112 | p->nodes[i] = NULL; |
eb60ceac CM |
113 | } |
114 | } | |
115 | ||
d352ac68 CM |
116 | /* |
117 | * safely gets a reference on the root node of a tree. A lock | |
118 | * is not taken, so a concurrent writer may put a different node | |
119 | * at the root of the tree. See btrfs_lock_root_node for the | |
120 | * looping required. | |
121 | * | |
122 | * The extent buffer returned by this has a reference taken, so | |
123 | * it won't disappear. It may stop being the root of the tree | |
124 | * at any time because there are no locks held. | |
125 | */ | |
925baedd CM |
126 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
127 | { | |
128 | struct extent_buffer *eb; | |
240f62c8 | 129 | |
3083ee2e JB |
130 | while (1) { |
131 | rcu_read_lock(); | |
132 | eb = rcu_dereference(root->node); | |
133 | ||
134 | /* | |
135 | * RCU really hurts here, we could free up the root node because | |
01327610 | 136 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
137 | * the inc_not_zero dance and if it doesn't work then |
138 | * synchronize_rcu and try again. | |
139 | */ | |
140 | if (atomic_inc_not_zero(&eb->refs)) { | |
141 | rcu_read_unlock(); | |
142 | break; | |
143 | } | |
144 | rcu_read_unlock(); | |
145 | synchronize_rcu(); | |
146 | } | |
925baedd CM |
147 | return eb; |
148 | } | |
149 | ||
d352ac68 CM |
150 | /* loop around taking references on and locking the root node of the |
151 | * tree until you end up with a lock on the root. A locked buffer | |
152 | * is returned, with a reference held. | |
153 | */ | |
925baedd CM |
154 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
155 | { | |
156 | struct extent_buffer *eb; | |
157 | ||
d397712b | 158 | while (1) { |
925baedd CM |
159 | eb = btrfs_root_node(root); |
160 | btrfs_tree_lock(eb); | |
240f62c8 | 161 | if (eb == root->node) |
925baedd | 162 | break; |
925baedd CM |
163 | btrfs_tree_unlock(eb); |
164 | free_extent_buffer(eb); | |
165 | } | |
166 | return eb; | |
167 | } | |
168 | ||
bd681513 CM |
169 | /* loop around taking references on and locking the root node of the |
170 | * tree until you end up with a lock on the root. A locked buffer | |
171 | * is returned, with a reference held. | |
172 | */ | |
84f7d8e6 | 173 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
174 | { |
175 | struct extent_buffer *eb; | |
176 | ||
177 | while (1) { | |
178 | eb = btrfs_root_node(root); | |
179 | btrfs_tree_read_lock(eb); | |
180 | if (eb == root->node) | |
181 | break; | |
182 | btrfs_tree_read_unlock(eb); | |
183 | free_extent_buffer(eb); | |
184 | } | |
185 | return eb; | |
186 | } | |
187 | ||
d352ac68 CM |
188 | /* cowonly root (everything not a reference counted cow subvolume), just get |
189 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
190 | * get properly updated on disk. | |
191 | */ | |
0b86a832 CM |
192 | static void add_root_to_dirty_list(struct btrfs_root *root) |
193 | { | |
0b246afa JM |
194 | struct btrfs_fs_info *fs_info = root->fs_info; |
195 | ||
e7070be1 JB |
196 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
197 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
198 | return; | |
199 | ||
0b246afa | 200 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
201 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
202 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 203 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 204 | list_move_tail(&root->dirty_list, |
0b246afa | 205 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
206 | else |
207 | list_move(&root->dirty_list, | |
0b246afa | 208 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 209 | } |
0b246afa | 210 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
211 | } |
212 | ||
d352ac68 CM |
213 | /* |
214 | * used by snapshot creation to make a copy of a root for a tree with | |
215 | * a given objectid. The buffer with the new root node is returned in | |
216 | * cow_ret, and this func returns zero on success or a negative error code. | |
217 | */ | |
be20aa9d CM |
218 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
219 | struct btrfs_root *root, | |
220 | struct extent_buffer *buf, | |
221 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
222 | { | |
0b246afa | 223 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 224 | struct extent_buffer *cow; |
be20aa9d CM |
225 | int ret = 0; |
226 | int level; | |
5d4f98a2 | 227 | struct btrfs_disk_key disk_key; |
be20aa9d | 228 | |
27cdeb70 | 229 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 230 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
231 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
232 | trans->transid != root->last_trans); | |
be20aa9d CM |
233 | |
234 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
235 | if (level == 0) |
236 | btrfs_item_key(buf, &disk_key, 0); | |
237 | else | |
238 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 239 | |
4d75f8a9 DS |
240 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
241 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 242 | if (IS_ERR(cow)) |
be20aa9d CM |
243 | return PTR_ERR(cow); |
244 | ||
58e8012c | 245 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
246 | btrfs_set_header_bytenr(cow, cow->start); |
247 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
248 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
249 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
250 | BTRFS_HEADER_FLAG_RELOC); | |
251 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
252 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
253 | else | |
254 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 255 | |
de37aa51 | 256 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 257 | |
be20aa9d | 258 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 259 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 260 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 261 | else |
e339a6b0 | 262 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 263 | |
be20aa9d CM |
264 | if (ret) |
265 | return ret; | |
266 | ||
267 | btrfs_mark_buffer_dirty(cow); | |
268 | *cow_ret = cow; | |
269 | return 0; | |
270 | } | |
271 | ||
bd989ba3 JS |
272 | enum mod_log_op { |
273 | MOD_LOG_KEY_REPLACE, | |
274 | MOD_LOG_KEY_ADD, | |
275 | MOD_LOG_KEY_REMOVE, | |
276 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
277 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
278 | MOD_LOG_MOVE_KEYS, | |
279 | MOD_LOG_ROOT_REPLACE, | |
280 | }; | |
281 | ||
bd989ba3 JS |
282 | struct tree_mod_root { |
283 | u64 logical; | |
284 | u8 level; | |
285 | }; | |
286 | ||
287 | struct tree_mod_elem { | |
288 | struct rb_node node; | |
298cfd36 | 289 | u64 logical; |
097b8a7c | 290 | u64 seq; |
bd989ba3 JS |
291 | enum mod_log_op op; |
292 | ||
293 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
294 | int slot; | |
295 | ||
296 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
297 | u64 generation; | |
298 | ||
299 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
300 | struct btrfs_disk_key key; | |
301 | u64 blockptr; | |
302 | ||
303 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
304 | struct { |
305 | int dst_slot; | |
306 | int nr_items; | |
307 | } move; | |
bd989ba3 JS |
308 | |
309 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
310 | struct tree_mod_root old_root; | |
311 | }; | |
312 | ||
fc36ed7e | 313 | /* |
fcebe456 | 314 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 315 | */ |
fcebe456 | 316 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
317 | { |
318 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
319 | } | |
320 | ||
097b8a7c JS |
321 | /* |
322 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
323 | * passed does not already have a sequence number set. So when a caller expects | |
324 | * to record tree modifications, it should ensure to set elem->seq to zero | |
325 | * before calling btrfs_get_tree_mod_seq. | |
326 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
327 | * blocker was added. | |
328 | */ | |
329 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
330 | struct seq_list *elem) | |
bd989ba3 | 331 | { |
b1a09f1e | 332 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 333 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 334 | if (!elem->seq) { |
fcebe456 | 335 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
336 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
337 | } | |
bd989ba3 | 338 | spin_unlock(&fs_info->tree_mod_seq_lock); |
b1a09f1e | 339 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 340 | |
fcebe456 | 341 | return elem->seq; |
bd989ba3 JS |
342 | } |
343 | ||
344 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
345 | struct seq_list *elem) | |
346 | { | |
347 | struct rb_root *tm_root; | |
348 | struct rb_node *node; | |
349 | struct rb_node *next; | |
350 | struct seq_list *cur_elem; | |
351 | struct tree_mod_elem *tm; | |
352 | u64 min_seq = (u64)-1; | |
353 | u64 seq_putting = elem->seq; | |
354 | ||
355 | if (!seq_putting) | |
356 | return; | |
357 | ||
bd989ba3 JS |
358 | spin_lock(&fs_info->tree_mod_seq_lock); |
359 | list_del(&elem->list); | |
097b8a7c | 360 | elem->seq = 0; |
bd989ba3 JS |
361 | |
362 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 363 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
364 | if (seq_putting > cur_elem->seq) { |
365 | /* | |
366 | * blocker with lower sequence number exists, we | |
367 | * cannot remove anything from the log | |
368 | */ | |
097b8a7c JS |
369 | spin_unlock(&fs_info->tree_mod_seq_lock); |
370 | return; | |
bd989ba3 JS |
371 | } |
372 | min_seq = cur_elem->seq; | |
373 | } | |
374 | } | |
097b8a7c JS |
375 | spin_unlock(&fs_info->tree_mod_seq_lock); |
376 | ||
bd989ba3 JS |
377 | /* |
378 | * anything that's lower than the lowest existing (read: blocked) | |
379 | * sequence number can be removed from the tree. | |
380 | */ | |
b1a09f1e | 381 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
382 | tm_root = &fs_info->tree_mod_log; |
383 | for (node = rb_first(tm_root); node; node = next) { | |
384 | next = rb_next(node); | |
6b4df8b6 | 385 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 386 | if (tm->seq > min_seq) |
bd989ba3 JS |
387 | continue; |
388 | rb_erase(node, tm_root); | |
bd989ba3 JS |
389 | kfree(tm); |
390 | } | |
b1a09f1e | 391 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
392 | } |
393 | ||
394 | /* | |
395 | * key order of the log: | |
298cfd36 | 396 | * node/leaf start address -> sequence |
bd989ba3 | 397 | * |
298cfd36 CR |
398 | * The 'start address' is the logical address of the *new* root node |
399 | * for root replace operations, or the logical address of the affected | |
400 | * block for all other operations. | |
bd989ba3 JS |
401 | */ |
402 | static noinline int | |
403 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
404 | { | |
405 | struct rb_root *tm_root; | |
406 | struct rb_node **new; | |
407 | struct rb_node *parent = NULL; | |
408 | struct tree_mod_elem *cur; | |
c8cc6341 | 409 | |
73e82fe4 DS |
410 | lockdep_assert_held_write(&fs_info->tree_mod_log_lock); |
411 | ||
fcebe456 | 412 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 413 | |
bd989ba3 JS |
414 | tm_root = &fs_info->tree_mod_log; |
415 | new = &tm_root->rb_node; | |
416 | while (*new) { | |
6b4df8b6 | 417 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 418 | parent = *new; |
298cfd36 | 419 | if (cur->logical < tm->logical) |
bd989ba3 | 420 | new = &((*new)->rb_left); |
298cfd36 | 421 | else if (cur->logical > tm->logical) |
bd989ba3 | 422 | new = &((*new)->rb_right); |
097b8a7c | 423 | else if (cur->seq < tm->seq) |
bd989ba3 | 424 | new = &((*new)->rb_left); |
097b8a7c | 425 | else if (cur->seq > tm->seq) |
bd989ba3 | 426 | new = &((*new)->rb_right); |
5de865ee FDBM |
427 | else |
428 | return -EEXIST; | |
bd989ba3 JS |
429 | } |
430 | ||
431 | rb_link_node(&tm->node, parent, new); | |
432 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 433 | return 0; |
bd989ba3 JS |
434 | } |
435 | ||
097b8a7c JS |
436 | /* |
437 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
438 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
439 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 440 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 441 | */ |
e9b7fd4d JS |
442 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
443 | struct extent_buffer *eb) { | |
444 | smp_mb(); | |
445 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
446 | return 1; | |
097b8a7c JS |
447 | if (eb && btrfs_header_level(eb) == 0) |
448 | return 1; | |
5de865ee | 449 | |
b1a09f1e | 450 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 451 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 452 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
453 | return 1; |
454 | } | |
455 | ||
e9b7fd4d JS |
456 | return 0; |
457 | } | |
458 | ||
5de865ee FDBM |
459 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
460 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
461 | struct extent_buffer *eb) | |
462 | { | |
463 | smp_mb(); | |
464 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
465 | return 0; | |
466 | if (eb && btrfs_header_level(eb) == 0) | |
467 | return 0; | |
468 | ||
469 | return 1; | |
470 | } | |
471 | ||
472 | static struct tree_mod_elem * | |
473 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
474 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 475 | { |
097b8a7c | 476 | struct tree_mod_elem *tm; |
bd989ba3 | 477 | |
c8cc6341 JB |
478 | tm = kzalloc(sizeof(*tm), flags); |
479 | if (!tm) | |
5de865ee | 480 | return NULL; |
bd989ba3 | 481 | |
298cfd36 | 482 | tm->logical = eb->start; |
bd989ba3 JS |
483 | if (op != MOD_LOG_KEY_ADD) { |
484 | btrfs_node_key(eb, &tm->key, slot); | |
485 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
486 | } | |
487 | tm->op = op; | |
488 | tm->slot = slot; | |
489 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 490 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 491 | |
5de865ee | 492 | return tm; |
097b8a7c JS |
493 | } |
494 | ||
e09c2efe DS |
495 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
496 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 497 | { |
5de865ee FDBM |
498 | struct tree_mod_elem *tm; |
499 | int ret; | |
500 | ||
e09c2efe | 501 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
502 | return 0; |
503 | ||
504 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
505 | if (!tm) | |
506 | return -ENOMEM; | |
507 | ||
e09c2efe | 508 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 509 | kfree(tm); |
097b8a7c | 510 | return 0; |
5de865ee FDBM |
511 | } |
512 | ||
e09c2efe | 513 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 514 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
515 | if (ret) |
516 | kfree(tm); | |
097b8a7c | 517 | |
5de865ee | 518 | return ret; |
097b8a7c JS |
519 | } |
520 | ||
6074d45f DS |
521 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
522 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 523 | { |
5de865ee FDBM |
524 | struct tree_mod_elem *tm = NULL; |
525 | struct tree_mod_elem **tm_list = NULL; | |
526 | int ret = 0; | |
bd989ba3 | 527 | int i; |
5de865ee | 528 | int locked = 0; |
bd989ba3 | 529 | |
6074d45f | 530 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 531 | return 0; |
bd989ba3 | 532 | |
176ef8f5 | 533 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
534 | if (!tm_list) |
535 | return -ENOMEM; | |
536 | ||
176ef8f5 | 537 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
538 | if (!tm) { |
539 | ret = -ENOMEM; | |
540 | goto free_tms; | |
541 | } | |
542 | ||
298cfd36 | 543 | tm->logical = eb->start; |
5de865ee FDBM |
544 | tm->slot = src_slot; |
545 | tm->move.dst_slot = dst_slot; | |
546 | tm->move.nr_items = nr_items; | |
547 | tm->op = MOD_LOG_MOVE_KEYS; | |
548 | ||
549 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
550 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 551 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
552 | if (!tm_list[i]) { |
553 | ret = -ENOMEM; | |
554 | goto free_tms; | |
555 | } | |
556 | } | |
557 | ||
6074d45f | 558 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
559 | goto free_tms; |
560 | locked = 1; | |
561 | ||
01763a2e JS |
562 | /* |
563 | * When we override something during the move, we log these removals. | |
564 | * This can only happen when we move towards the beginning of the | |
565 | * buffer, i.e. dst_slot < src_slot. | |
566 | */ | |
bd989ba3 | 567 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 568 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
569 | if (ret) |
570 | goto free_tms; | |
bd989ba3 JS |
571 | } |
572 | ||
6074d45f | 573 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
574 | if (ret) |
575 | goto free_tms; | |
b1a09f1e | 576 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 577 | kfree(tm_list); |
f395694c | 578 | |
5de865ee FDBM |
579 | return 0; |
580 | free_tms: | |
581 | for (i = 0; i < nr_items; i++) { | |
582 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 583 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
584 | kfree(tm_list[i]); |
585 | } | |
586 | if (locked) | |
b1a09f1e | 587 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
588 | kfree(tm_list); |
589 | kfree(tm); | |
bd989ba3 | 590 | |
5de865ee | 591 | return ret; |
bd989ba3 JS |
592 | } |
593 | ||
5de865ee FDBM |
594 | static inline int |
595 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
596 | struct tree_mod_elem **tm_list, | |
597 | int nritems) | |
097b8a7c | 598 | { |
5de865ee | 599 | int i, j; |
097b8a7c JS |
600 | int ret; |
601 | ||
097b8a7c | 602 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
603 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
604 | if (ret) { | |
605 | for (j = nritems - 1; j > i; j--) | |
606 | rb_erase(&tm_list[j]->node, | |
607 | &fs_info->tree_mod_log); | |
608 | return ret; | |
609 | } | |
097b8a7c | 610 | } |
5de865ee FDBM |
611 | |
612 | return 0; | |
097b8a7c JS |
613 | } |
614 | ||
95b757c1 DS |
615 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
616 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 617 | { |
95b757c1 | 618 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
619 | struct tree_mod_elem *tm = NULL; |
620 | struct tree_mod_elem **tm_list = NULL; | |
621 | int nritems = 0; | |
622 | int ret = 0; | |
623 | int i; | |
bd989ba3 | 624 | |
5de865ee | 625 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
626 | return 0; |
627 | ||
5de865ee FDBM |
628 | if (log_removal && btrfs_header_level(old_root) > 0) { |
629 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 630 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 631 | GFP_NOFS); |
5de865ee FDBM |
632 | if (!tm_list) { |
633 | ret = -ENOMEM; | |
634 | goto free_tms; | |
635 | } | |
636 | for (i = 0; i < nritems; i++) { | |
637 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 638 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
639 | if (!tm_list[i]) { |
640 | ret = -ENOMEM; | |
641 | goto free_tms; | |
642 | } | |
643 | } | |
644 | } | |
d9abbf1c | 645 | |
bcc8e07f | 646 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
647 | if (!tm) { |
648 | ret = -ENOMEM; | |
649 | goto free_tms; | |
650 | } | |
bd989ba3 | 651 | |
298cfd36 | 652 | tm->logical = new_root->start; |
bd989ba3 JS |
653 | tm->old_root.logical = old_root->start; |
654 | tm->old_root.level = btrfs_header_level(old_root); | |
655 | tm->generation = btrfs_header_generation(old_root); | |
656 | tm->op = MOD_LOG_ROOT_REPLACE; | |
657 | ||
5de865ee FDBM |
658 | if (tree_mod_dont_log(fs_info, NULL)) |
659 | goto free_tms; | |
660 | ||
661 | if (tm_list) | |
662 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
663 | if (!ret) | |
664 | ret = __tree_mod_log_insert(fs_info, tm); | |
665 | ||
b1a09f1e | 666 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
667 | if (ret) |
668 | goto free_tms; | |
669 | kfree(tm_list); | |
670 | ||
671 | return ret; | |
672 | ||
673 | free_tms: | |
674 | if (tm_list) { | |
675 | for (i = 0; i < nritems; i++) | |
676 | kfree(tm_list[i]); | |
677 | kfree(tm_list); | |
678 | } | |
679 | kfree(tm); | |
680 | ||
681 | return ret; | |
bd989ba3 JS |
682 | } |
683 | ||
684 | static struct tree_mod_elem * | |
685 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
686 | int smallest) | |
687 | { | |
688 | struct rb_root *tm_root; | |
689 | struct rb_node *node; | |
690 | struct tree_mod_elem *cur = NULL; | |
691 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 692 | |
b1a09f1e | 693 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
694 | tm_root = &fs_info->tree_mod_log; |
695 | node = tm_root->rb_node; | |
696 | while (node) { | |
6b4df8b6 | 697 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 698 | if (cur->logical < start) { |
bd989ba3 | 699 | node = node->rb_left; |
298cfd36 | 700 | } else if (cur->logical > start) { |
bd989ba3 | 701 | node = node->rb_right; |
097b8a7c | 702 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
703 | node = node->rb_left; |
704 | } else if (!smallest) { | |
705 | /* we want the node with the highest seq */ | |
706 | if (found) | |
097b8a7c | 707 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
708 | found = cur; |
709 | node = node->rb_left; | |
097b8a7c | 710 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
711 | /* we want the node with the smallest seq */ |
712 | if (found) | |
097b8a7c | 713 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
714 | found = cur; |
715 | node = node->rb_right; | |
716 | } else { | |
717 | found = cur; | |
718 | break; | |
719 | } | |
720 | } | |
b1a09f1e | 721 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
722 | |
723 | return found; | |
724 | } | |
725 | ||
726 | /* | |
727 | * this returns the element from the log with the smallest time sequence | |
728 | * value that's in the log (the oldest log item). any element with a time | |
729 | * sequence lower than min_seq will be ignored. | |
730 | */ | |
731 | static struct tree_mod_elem * | |
732 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
733 | u64 min_seq) | |
734 | { | |
735 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
736 | } | |
737 | ||
738 | /* | |
739 | * this returns the element from the log with the largest time sequence | |
740 | * value that's in the log (the most recent log item). any element with | |
741 | * a time sequence lower than min_seq will be ignored. | |
742 | */ | |
743 | static struct tree_mod_elem * | |
744 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
745 | { | |
746 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
747 | } | |
748 | ||
ed874f0d | 749 | static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, |
bd989ba3 | 750 | struct extent_buffer *src, unsigned long dst_offset, |
90f8d62e | 751 | unsigned long src_offset, int nr_items) |
bd989ba3 | 752 | { |
ed874f0d | 753 | struct btrfs_fs_info *fs_info = dst->fs_info; |
5de865ee FDBM |
754 | int ret = 0; |
755 | struct tree_mod_elem **tm_list = NULL; | |
756 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 757 | int i; |
5de865ee | 758 | int locked = 0; |
bd989ba3 | 759 | |
5de865ee FDBM |
760 | if (!tree_mod_need_log(fs_info, NULL)) |
761 | return 0; | |
bd989ba3 | 762 | |
c8cc6341 | 763 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
764 | return 0; |
765 | ||
31e818fe | 766 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
767 | GFP_NOFS); |
768 | if (!tm_list) | |
769 | return -ENOMEM; | |
bd989ba3 | 770 | |
5de865ee FDBM |
771 | tm_list_add = tm_list; |
772 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 773 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
774 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
775 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
776 | if (!tm_list_rem[i]) { | |
777 | ret = -ENOMEM; | |
778 | goto free_tms; | |
779 | } | |
780 | ||
781 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
782 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
783 | if (!tm_list_add[i]) { | |
784 | ret = -ENOMEM; | |
785 | goto free_tms; | |
786 | } | |
787 | } | |
788 | ||
789 | if (tree_mod_dont_log(fs_info, NULL)) | |
790 | goto free_tms; | |
791 | locked = 1; | |
792 | ||
793 | for (i = 0; i < nr_items; i++) { | |
794 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
795 | if (ret) | |
796 | goto free_tms; | |
797 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
798 | if (ret) | |
799 | goto free_tms; | |
bd989ba3 | 800 | } |
5de865ee | 801 | |
b1a09f1e | 802 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
803 | kfree(tm_list); |
804 | ||
805 | return 0; | |
806 | ||
807 | free_tms: | |
808 | for (i = 0; i < nr_items * 2; i++) { | |
809 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
810 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
811 | kfree(tm_list[i]); | |
812 | } | |
813 | if (locked) | |
b1a09f1e | 814 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
815 | kfree(tm_list); |
816 | ||
817 | return ret; | |
bd989ba3 JS |
818 | } |
819 | ||
db7279a2 | 820 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 821 | { |
5de865ee FDBM |
822 | struct tree_mod_elem **tm_list = NULL; |
823 | int nritems = 0; | |
824 | int i; | |
825 | int ret = 0; | |
826 | ||
827 | if (btrfs_header_level(eb) == 0) | |
828 | return 0; | |
829 | ||
db7279a2 | 830 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
831 | return 0; |
832 | ||
833 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 834 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
835 | if (!tm_list) |
836 | return -ENOMEM; | |
837 | ||
838 | for (i = 0; i < nritems; i++) { | |
839 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
840 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
841 | if (!tm_list[i]) { | |
842 | ret = -ENOMEM; | |
843 | goto free_tms; | |
844 | } | |
845 | } | |
846 | ||
db7279a2 | 847 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
848 | goto free_tms; |
849 | ||
db7279a2 | 850 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 851 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
852 | if (ret) |
853 | goto free_tms; | |
854 | kfree(tm_list); | |
855 | ||
856 | return 0; | |
857 | ||
858 | free_tms: | |
859 | for (i = 0; i < nritems; i++) | |
860 | kfree(tm_list[i]); | |
861 | kfree(tm_list); | |
862 | ||
863 | return ret; | |
bd989ba3 JS |
864 | } |
865 | ||
5d4f98a2 YZ |
866 | /* |
867 | * check if the tree block can be shared by multiple trees | |
868 | */ | |
869 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
870 | struct extent_buffer *buf) | |
871 | { | |
872 | /* | |
01327610 | 873 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
874 | * are never shared. If a block was allocated after the last |
875 | * snapshot and the block was not allocated by tree relocation, | |
876 | * we know the block is not shared. | |
877 | */ | |
27cdeb70 | 878 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
879 | buf != root->node && buf != root->commit_root && |
880 | (btrfs_header_generation(buf) <= | |
881 | btrfs_root_last_snapshot(&root->root_item) || | |
882 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
883 | return 1; | |
a79865c6 | 884 | |
5d4f98a2 YZ |
885 | return 0; |
886 | } | |
887 | ||
888 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
889 | struct btrfs_root *root, | |
890 | struct extent_buffer *buf, | |
f0486c68 YZ |
891 | struct extent_buffer *cow, |
892 | int *last_ref) | |
5d4f98a2 | 893 | { |
0b246afa | 894 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
895 | u64 refs; |
896 | u64 owner; | |
897 | u64 flags; | |
898 | u64 new_flags = 0; | |
899 | int ret; | |
900 | ||
901 | /* | |
902 | * Backrefs update rules: | |
903 | * | |
904 | * Always use full backrefs for extent pointers in tree block | |
905 | * allocated by tree relocation. | |
906 | * | |
907 | * If a shared tree block is no longer referenced by its owner | |
908 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
909 | * use full backrefs for extent pointers in tree block. | |
910 | * | |
911 | * If a tree block is been relocating | |
912 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
913 | * use full backrefs for extent pointers in tree block. | |
914 | * The reason for this is some operations (such as drop tree) | |
915 | * are only allowed for blocks use full backrefs. | |
916 | */ | |
917 | ||
918 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 919 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
920 | btrfs_header_level(buf), 1, |
921 | &refs, &flags); | |
be1a5564 MF |
922 | if (ret) |
923 | return ret; | |
e5df9573 MF |
924 | if (refs == 0) { |
925 | ret = -EROFS; | |
0b246afa | 926 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
927 | return ret; |
928 | } | |
5d4f98a2 YZ |
929 | } else { |
930 | refs = 1; | |
931 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
932 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
933 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
934 | else | |
935 | flags = 0; | |
936 | } | |
937 | ||
938 | owner = btrfs_header_owner(buf); | |
939 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
940 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
941 | ||
942 | if (refs > 1) { | |
943 | if ((owner == root->root_key.objectid || | |
944 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
945 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 946 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
947 | if (ret) |
948 | return ret; | |
5d4f98a2 YZ |
949 | |
950 | if (root->root_key.objectid == | |
951 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 952 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
953 | if (ret) |
954 | return ret; | |
e339a6b0 | 955 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
956 | if (ret) |
957 | return ret; | |
5d4f98a2 YZ |
958 | } |
959 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
960 | } else { | |
961 | ||
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; | |
5d4f98a2 YZ |
969 | } |
970 | if (new_flags != 0) { | |
b1c79e09 JB |
971 | int level = btrfs_header_level(buf); |
972 | ||
f5c8daa5 | 973 | ret = btrfs_set_disk_extent_flags(trans, |
5d4f98a2 YZ |
974 | buf->start, |
975 | buf->len, | |
b1c79e09 | 976 | new_flags, level, 0); |
be1a5564 MF |
977 | if (ret) |
978 | return ret; | |
5d4f98a2 YZ |
979 | } |
980 | } else { | |
981 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
982 | if (root->root_key.objectid == | |
983 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 984 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 985 | else |
e339a6b0 | 986 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
987 | if (ret) |
988 | return ret; | |
e339a6b0 | 989 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
990 | if (ret) |
991 | return ret; | |
5d4f98a2 | 992 | } |
6a884d7d | 993 | btrfs_clean_tree_block(buf); |
f0486c68 | 994 | *last_ref = 1; |
5d4f98a2 YZ |
995 | } |
996 | return 0; | |
997 | } | |
998 | ||
a6279470 FM |
999 | static struct extent_buffer *alloc_tree_block_no_bg_flush( |
1000 | struct btrfs_trans_handle *trans, | |
1001 | struct btrfs_root *root, | |
1002 | u64 parent_start, | |
1003 | const struct btrfs_disk_key *disk_key, | |
1004 | int level, | |
1005 | u64 hint, | |
1006 | u64 empty_size) | |
1007 | { | |
1008 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1009 | struct extent_buffer *ret; | |
1010 | ||
1011 | /* | |
1012 | * If we are COWing a node/leaf from the extent, chunk, device or free | |
1013 | * space trees, make sure that we do not finish block group creation of | |
1014 | * pending block groups. We do this to avoid a deadlock. | |
1015 | * COWing can result in allocation of a new chunk, and flushing pending | |
1016 | * block groups (btrfs_create_pending_block_groups()) can be triggered | |
1017 | * when finishing allocation of a new chunk. Creation of a pending block | |
1018 | * group modifies the extent, chunk, device and free space trees, | |
1019 | * therefore we could deadlock with ourselves since we are holding a | |
1020 | * lock on an extent buffer that btrfs_create_pending_block_groups() may | |
1021 | * try to COW later. | |
1022 | * For similar reasons, we also need to delay flushing pending block | |
1023 | * groups when splitting a leaf or node, from one of those trees, since | |
1024 | * we are holding a write lock on it and its parent or when inserting a | |
1025 | * new root node for one of those trees. | |
1026 | */ | |
1027 | if (root == fs_info->extent_root || | |
1028 | root == fs_info->chunk_root || | |
1029 | root == fs_info->dev_root || | |
1030 | root == fs_info->free_space_root) | |
1031 | trans->can_flush_pending_bgs = false; | |
1032 | ||
1033 | ret = btrfs_alloc_tree_block(trans, root, parent_start, | |
1034 | root->root_key.objectid, disk_key, level, | |
1035 | hint, empty_size); | |
1036 | trans->can_flush_pending_bgs = true; | |
1037 | ||
1038 | return ret; | |
1039 | } | |
1040 | ||
d352ac68 | 1041 | /* |
d397712b CM |
1042 | * does the dirty work in cow of a single block. The parent block (if |
1043 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1044 | * dirty and returned locked. If you modify the block it needs to be marked | |
1045 | * dirty again. | |
d352ac68 CM |
1046 | * |
1047 | * search_start -- an allocation hint for the new block | |
1048 | * | |
d397712b CM |
1049 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1050 | * bytes the allocator should try to find free next to the block it returns. | |
1051 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1052 | */ |
d397712b | 1053 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1054 | struct btrfs_root *root, |
1055 | struct extent_buffer *buf, | |
1056 | struct extent_buffer *parent, int parent_slot, | |
1057 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1058 | u64 search_start, u64 empty_size) |
02217ed2 | 1059 | { |
0b246afa | 1060 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1061 | struct btrfs_disk_key disk_key; |
5f39d397 | 1062 | struct extent_buffer *cow; |
be1a5564 | 1063 | int level, ret; |
f0486c68 | 1064 | int last_ref = 0; |
925baedd | 1065 | int unlock_orig = 0; |
0f5053eb | 1066 | u64 parent_start = 0; |
7bb86316 | 1067 | |
925baedd CM |
1068 | if (*cow_ret == buf) |
1069 | unlock_orig = 1; | |
1070 | ||
b9447ef8 | 1071 | btrfs_assert_tree_locked(buf); |
925baedd | 1072 | |
27cdeb70 | 1073 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1074 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1075 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1076 | trans->transid != root->last_trans); | |
5f39d397 | 1077 | |
7bb86316 | 1078 | level = btrfs_header_level(buf); |
31840ae1 | 1079 | |
5d4f98a2 YZ |
1080 | if (level == 0) |
1081 | btrfs_item_key(buf, &disk_key, 0); | |
1082 | else | |
1083 | btrfs_node_key(buf, &disk_key, 0); | |
1084 | ||
0f5053eb GR |
1085 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1086 | parent_start = parent->start; | |
5d4f98a2 | 1087 | |
a6279470 FM |
1088 | cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, |
1089 | level, search_start, empty_size); | |
54aa1f4d CM |
1090 | if (IS_ERR(cow)) |
1091 | return PTR_ERR(cow); | |
6702ed49 | 1092 | |
b4ce94de CM |
1093 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1094 | ||
58e8012c | 1095 | copy_extent_buffer_full(cow, buf); |
db94535d | 1096 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1097 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1098 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1099 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1100 | BTRFS_HEADER_FLAG_RELOC); | |
1101 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1102 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1103 | else | |
1104 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1105 | |
de37aa51 | 1106 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 1107 | |
be1a5564 | 1108 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1109 | if (ret) { |
66642832 | 1110 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1111 | return ret; |
1112 | } | |
1a40e23b | 1113 | |
27cdeb70 | 1114 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1115 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1116 | if (ret) { |
66642832 | 1117 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1118 | return ret; |
93314e3b | 1119 | } |
83d4cfd4 | 1120 | } |
3fd0a558 | 1121 | |
02217ed2 | 1122 | if (buf == root->node) { |
925baedd | 1123 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1124 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1125 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1126 | parent_start = buf->start; | |
925baedd | 1127 | |
5f39d397 | 1128 | extent_buffer_get(cow); |
d9d19a01 DS |
1129 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1130 | BUG_ON(ret < 0); | |
240f62c8 | 1131 | rcu_assign_pointer(root->node, cow); |
925baedd | 1132 | |
f0486c68 | 1133 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1134 | last_ref); |
5f39d397 | 1135 | free_extent_buffer(buf); |
0b86a832 | 1136 | add_root_to_dirty_list(root); |
02217ed2 | 1137 | } else { |
5d4f98a2 | 1138 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1139 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1140 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1141 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1142 | cow->start); |
74493f7a CM |
1143 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1144 | trans->transid); | |
d6025579 | 1145 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1146 | if (last_ref) { |
db7279a2 | 1147 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1148 | if (ret) { |
66642832 | 1149 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1150 | return ret; |
1151 | } | |
1152 | } | |
f0486c68 | 1153 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1154 | last_ref); |
02217ed2 | 1155 | } |
925baedd CM |
1156 | if (unlock_orig) |
1157 | btrfs_tree_unlock(buf); | |
3083ee2e | 1158 | free_extent_buffer_stale(buf); |
ccd467d6 | 1159 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1160 | *cow_ret = cow; |
02217ed2 CM |
1161 | return 0; |
1162 | } | |
1163 | ||
5d9e75c4 JS |
1164 | /* |
1165 | * returns the logical address of the oldest predecessor of the given root. | |
1166 | * entries older than time_seq are ignored. | |
1167 | */ | |
bcd24dab DS |
1168 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1169 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1170 | { |
1171 | struct tree_mod_elem *tm; | |
1172 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1173 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1174 | int looped = 0; |
1175 | ||
1176 | if (!time_seq) | |
35a3621b | 1177 | return NULL; |
5d9e75c4 JS |
1178 | |
1179 | /* | |
298cfd36 CR |
1180 | * the very last operation that's logged for a root is the |
1181 | * replacement operation (if it is replaced at all). this has | |
1182 | * the logical address of the *new* root, making it the very | |
1183 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1184 | */ |
1185 | while (1) { | |
bcd24dab | 1186 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1187 | time_seq); |
1188 | if (!looped && !tm) | |
35a3621b | 1189 | return NULL; |
5d9e75c4 | 1190 | /* |
28da9fb4 JS |
1191 | * if there are no tree operation for the oldest root, we simply |
1192 | * return it. this should only happen if that (old) root is at | |
1193 | * level 0. | |
5d9e75c4 | 1194 | */ |
28da9fb4 JS |
1195 | if (!tm) |
1196 | break; | |
5d9e75c4 | 1197 | |
28da9fb4 JS |
1198 | /* |
1199 | * if there's an operation that's not a root replacement, we | |
1200 | * found the oldest version of our root. normally, we'll find a | |
1201 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1202 | */ | |
5d9e75c4 JS |
1203 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1204 | break; | |
1205 | ||
1206 | found = tm; | |
1207 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1208 | looped = 1; |
1209 | } | |
1210 | ||
a95236d9 JS |
1211 | /* if there's no old root to return, return what we found instead */ |
1212 | if (!found) | |
1213 | found = tm; | |
1214 | ||
5d9e75c4 JS |
1215 | return found; |
1216 | } | |
1217 | ||
1218 | /* | |
1219 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1220 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1221 | * time_seq). |
1222 | */ | |
1223 | static void | |
f1ca7e98 JB |
1224 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1225 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1226 | { |
1227 | u32 n; | |
1228 | struct rb_node *next; | |
1229 | struct tree_mod_elem *tm = first_tm; | |
1230 | unsigned long o_dst; | |
1231 | unsigned long o_src; | |
1232 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1233 | ||
1234 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1235 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1236 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1237 | /* |
1238 | * all the operations are recorded with the operator used for | |
1239 | * the modification. as we're going backwards, we do the | |
1240 | * opposite of each operation here. | |
1241 | */ | |
1242 | switch (tm->op) { | |
1243 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1244 | BUG_ON(tm->slot < n); | |
1c697d4a | 1245 | /* Fallthrough */ |
95c80bb1 | 1246 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1247 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1248 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1249 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1250 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1251 | tm->generation); | |
4c3e6969 | 1252 | n++; |
5d9e75c4 JS |
1253 | break; |
1254 | case MOD_LOG_KEY_REPLACE: | |
1255 | BUG_ON(tm->slot >= n); | |
1256 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1257 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1258 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1259 | tm->generation); | |
1260 | break; | |
1261 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1262 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1263 | n--; |
1264 | break; | |
1265 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1266 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1267 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1268 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1269 | tm->move.nr_items * p_size); |
1270 | break; | |
1271 | case MOD_LOG_ROOT_REPLACE: | |
1272 | /* | |
1273 | * this operation is special. for roots, this must be | |
1274 | * handled explicitly before rewinding. | |
1275 | * for non-roots, this operation may exist if the node | |
1276 | * was a root: root A -> child B; then A gets empty and | |
1277 | * B is promoted to the new root. in the mod log, we'll | |
1278 | * have a root-replace operation for B, a tree block | |
1279 | * that is no root. we simply ignore that operation. | |
1280 | */ | |
1281 | break; | |
1282 | } | |
1283 | next = rb_next(&tm->node); | |
1284 | if (!next) | |
1285 | break; | |
6b4df8b6 | 1286 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1287 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1288 | break; |
1289 | } | |
b1a09f1e | 1290 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1291 | btrfs_set_header_nritems(eb, n); |
1292 | } | |
1293 | ||
47fb091f | 1294 | /* |
01327610 | 1295 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1296 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1297 | * returned buffer is always read-locked. If the returned buffer is not the | |
1298 | * input buffer, the lock on the input buffer is released and the input buffer | |
1299 | * is freed (its refcount is decremented). | |
1300 | */ | |
5d9e75c4 | 1301 | static struct extent_buffer * |
9ec72677 JB |
1302 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1303 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1304 | { |
1305 | struct extent_buffer *eb_rewin; | |
1306 | struct tree_mod_elem *tm; | |
1307 | ||
1308 | if (!time_seq) | |
1309 | return eb; | |
1310 | ||
1311 | if (btrfs_header_level(eb) == 0) | |
1312 | return eb; | |
1313 | ||
1314 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1315 | if (!tm) | |
1316 | return eb; | |
1317 | ||
9ec72677 | 1318 | btrfs_set_path_blocking(path); |
300aa896 | 1319 | btrfs_set_lock_blocking_read(eb); |
9ec72677 | 1320 | |
5d9e75c4 JS |
1321 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1322 | BUG_ON(tm->slot != 0); | |
da17066c | 1323 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1324 | if (!eb_rewin) { |
9ec72677 | 1325 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1326 | free_extent_buffer(eb); |
1327 | return NULL; | |
1328 | } | |
5d9e75c4 JS |
1329 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1330 | btrfs_set_header_backref_rev(eb_rewin, | |
1331 | btrfs_header_backref_rev(eb)); | |
1332 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1333 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1334 | } else { |
1335 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1336 | if (!eb_rewin) { |
9ec72677 | 1337 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1338 | free_extent_buffer(eb); |
1339 | return NULL; | |
1340 | } | |
5d9e75c4 JS |
1341 | } |
1342 | ||
9ec72677 | 1343 | btrfs_tree_read_unlock_blocking(eb); |
5d9e75c4 JS |
1344 | free_extent_buffer(eb); |
1345 | ||
47fb091f | 1346 | btrfs_tree_read_lock(eb_rewin); |
f1ca7e98 | 1347 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1348 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1349 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1350 | |
1351 | return eb_rewin; | |
1352 | } | |
1353 | ||
8ba97a15 JS |
1354 | /* |
1355 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1356 | * value. If there are no changes, the current root->root_node is returned. If | |
1357 | * anything changed in between, there's a fresh buffer allocated on which the | |
1358 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1359 | * Returns NULL on error (with no locks held). | |
1360 | */ | |
5d9e75c4 JS |
1361 | static inline struct extent_buffer * |
1362 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1363 | { | |
0b246afa | 1364 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1365 | struct tree_mod_elem *tm; |
30b0463a JS |
1366 | struct extent_buffer *eb = NULL; |
1367 | struct extent_buffer *eb_root; | |
efad8a85 | 1368 | u64 eb_root_owner = 0; |
7bfdcf7f | 1369 | struct extent_buffer *old; |
a95236d9 | 1370 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1371 | u64 old_generation = 0; |
a95236d9 | 1372 | u64 logical; |
581c1760 | 1373 | int level; |
5d9e75c4 | 1374 | |
30b0463a | 1375 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1376 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1377 | if (!tm) |
30b0463a | 1378 | return eb_root; |
5d9e75c4 | 1379 | |
a95236d9 JS |
1380 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1381 | old_root = &tm->old_root; | |
1382 | old_generation = tm->generation; | |
1383 | logical = old_root->logical; | |
581c1760 | 1384 | level = old_root->level; |
a95236d9 | 1385 | } else { |
30b0463a | 1386 | logical = eb_root->start; |
581c1760 | 1387 | level = btrfs_header_level(eb_root); |
a95236d9 | 1388 | } |
5d9e75c4 | 1389 | |
0b246afa | 1390 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1391 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1392 | btrfs_tree_read_unlock(eb_root); |
1393 | free_extent_buffer(eb_root); | |
581c1760 | 1394 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1395 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1396 | if (!IS_ERR(old)) | |
1397 | free_extent_buffer(old); | |
0b246afa JM |
1398 | btrfs_warn(fs_info, |
1399 | "failed to read tree block %llu from get_old_root", | |
1400 | logical); | |
834328a8 | 1401 | } else { |
7bfdcf7f LB |
1402 | eb = btrfs_clone_extent_buffer(old); |
1403 | free_extent_buffer(old); | |
834328a8 JS |
1404 | } |
1405 | } else if (old_root) { | |
efad8a85 | 1406 | eb_root_owner = btrfs_header_owner(eb_root); |
30b0463a JS |
1407 | btrfs_tree_read_unlock(eb_root); |
1408 | free_extent_buffer(eb_root); | |
0b246afa | 1409 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1410 | } else { |
300aa896 | 1411 | btrfs_set_lock_blocking_read(eb_root); |
30b0463a | 1412 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1413 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1414 | free_extent_buffer(eb_root); |
834328a8 JS |
1415 | } |
1416 | ||
8ba97a15 JS |
1417 | if (!eb) |
1418 | return NULL; | |
1419 | btrfs_tree_read_lock(eb); | |
a95236d9 | 1420 | if (old_root) { |
5d9e75c4 JS |
1421 | btrfs_set_header_bytenr(eb, eb->start); |
1422 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
efad8a85 | 1423 | btrfs_set_header_owner(eb, eb_root_owner); |
a95236d9 JS |
1424 | btrfs_set_header_level(eb, old_root->level); |
1425 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1426 | } |
28da9fb4 | 1427 | if (tm) |
0b246afa | 1428 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1429 | else |
1430 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1431 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1432 | |
1433 | return eb; | |
1434 | } | |
1435 | ||
5b6602e7 JS |
1436 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1437 | { | |
1438 | struct tree_mod_elem *tm; | |
1439 | int level; | |
30b0463a | 1440 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1441 | |
bcd24dab | 1442 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1443 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1444 | level = tm->old_root.level; | |
1445 | } else { | |
30b0463a | 1446 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1447 | } |
30b0463a | 1448 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1449 | |
1450 | return level; | |
1451 | } | |
1452 | ||
5d4f98a2 YZ |
1453 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1454 | struct btrfs_root *root, | |
1455 | struct extent_buffer *buf) | |
1456 | { | |
f5ee5c9a | 1457 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1458 | return 0; |
fccb84c9 | 1459 | |
d1980131 DS |
1460 | /* Ensure we can see the FORCE_COW bit */ |
1461 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1462 | |
1463 | /* | |
1464 | * We do not need to cow a block if | |
1465 | * 1) this block is not created or changed in this transaction; | |
1466 | * 2) this block does not belong to TREE_RELOC tree; | |
1467 | * 3) the root is not forced COW. | |
1468 | * | |
1469 | * What is forced COW: | |
01327610 | 1470 | * when we create snapshot during committing the transaction, |
52042d8e | 1471 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
1472 | * block to ensure the metadata consistency. |
1473 | */ | |
5d4f98a2 YZ |
1474 | if (btrfs_header_generation(buf) == trans->transid && |
1475 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1476 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1477 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1478 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1479 | return 0; |
1480 | return 1; | |
1481 | } | |
1482 | ||
d352ac68 CM |
1483 | /* |
1484 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1485 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1486 | * once per transaction, as long as it hasn't been written yet |
1487 | */ | |
d397712b | 1488 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1489 | struct btrfs_root *root, struct extent_buffer *buf, |
1490 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1491 | struct extent_buffer **cow_ret) |
6702ed49 | 1492 | { |
0b246afa | 1493 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1494 | u64 search_start; |
f510cfec | 1495 | int ret; |
dc17ff8f | 1496 | |
83354f07 JB |
1497 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
1498 | btrfs_err(fs_info, | |
1499 | "COW'ing blocks on a fs root that's being dropped"); | |
1500 | ||
0b246afa | 1501 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1502 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1503 | trans->transid, |
0b246afa | 1504 | fs_info->running_transaction->transid); |
31b1a2bd | 1505 | |
0b246afa | 1506 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1507 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1508 | trans->transid, fs_info->generation); |
dc17ff8f | 1509 | |
5d4f98a2 | 1510 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1511 | trans->dirty = true; |
6702ed49 CM |
1512 | *cow_ret = buf; |
1513 | return 0; | |
1514 | } | |
c487685d | 1515 | |
ee22184b | 1516 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1517 | |
1518 | if (parent) | |
8bead258 DS |
1519 | btrfs_set_lock_blocking_write(parent); |
1520 | btrfs_set_lock_blocking_write(buf); | |
b4ce94de | 1521 | |
f616f5cd QW |
1522 | /* |
1523 | * Before CoWing this block for later modification, check if it's | |
1524 | * the subtree root and do the delayed subtree trace if needed. | |
1525 | * | |
1526 | * Also We don't care about the error, as it's handled internally. | |
1527 | */ | |
1528 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 1529 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1530 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1531 | |
1532 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1533 | ||
f510cfec | 1534 | return ret; |
6702ed49 CM |
1535 | } |
1536 | ||
d352ac68 CM |
1537 | /* |
1538 | * helper function for defrag to decide if two blocks pointed to by a | |
1539 | * node are actually close by | |
1540 | */ | |
6b80053d | 1541 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1542 | { |
6b80053d | 1543 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1544 | return 1; |
6b80053d | 1545 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1546 | return 1; |
1547 | return 0; | |
1548 | } | |
1549 | ||
081e9573 CM |
1550 | /* |
1551 | * compare two keys in a memcmp fashion | |
1552 | */ | |
310712b2 OS |
1553 | static int comp_keys(const struct btrfs_disk_key *disk, |
1554 | const struct btrfs_key *k2) | |
081e9573 CM |
1555 | { |
1556 | struct btrfs_key k1; | |
1557 | ||
1558 | btrfs_disk_key_to_cpu(&k1, disk); | |
1559 | ||
20736aba | 1560 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1561 | } |
1562 | ||
f3465ca4 JB |
1563 | /* |
1564 | * same as comp_keys only with two btrfs_key's | |
1565 | */ | |
310712b2 | 1566 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1567 | { |
1568 | if (k1->objectid > k2->objectid) | |
1569 | return 1; | |
1570 | if (k1->objectid < k2->objectid) | |
1571 | return -1; | |
1572 | if (k1->type > k2->type) | |
1573 | return 1; | |
1574 | if (k1->type < k2->type) | |
1575 | return -1; | |
1576 | if (k1->offset > k2->offset) | |
1577 | return 1; | |
1578 | if (k1->offset < k2->offset) | |
1579 | return -1; | |
1580 | return 0; | |
1581 | } | |
081e9573 | 1582 | |
d352ac68 CM |
1583 | /* |
1584 | * this is used by the defrag code to go through all the | |
1585 | * leaves pointed to by a node and reallocate them so that | |
1586 | * disk order is close to key order | |
1587 | */ | |
6702ed49 | 1588 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1589 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1590 | int start_slot, u64 *last_ret, |
a6b6e75e | 1591 | struct btrfs_key *progress) |
6702ed49 | 1592 | { |
0b246afa | 1593 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1594 | struct extent_buffer *cur; |
6702ed49 | 1595 | u64 blocknr; |
ca7a79ad | 1596 | u64 gen; |
e9d0b13b CM |
1597 | u64 search_start = *last_ret; |
1598 | u64 last_block = 0; | |
6702ed49 CM |
1599 | u64 other; |
1600 | u32 parent_nritems; | |
6702ed49 CM |
1601 | int end_slot; |
1602 | int i; | |
1603 | int err = 0; | |
f2183bde | 1604 | int parent_level; |
6b80053d CM |
1605 | int uptodate; |
1606 | u32 blocksize; | |
081e9573 CM |
1607 | int progress_passed = 0; |
1608 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1609 | |
5708b959 | 1610 | parent_level = btrfs_header_level(parent); |
5708b959 | 1611 | |
0b246afa JM |
1612 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1613 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1614 | |
6b80053d | 1615 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1616 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1617 | end_slot = parent_nritems - 1; |
6702ed49 | 1618 | |
5dfe2be7 | 1619 | if (parent_nritems <= 1) |
6702ed49 CM |
1620 | return 0; |
1621 | ||
8bead258 | 1622 | btrfs_set_lock_blocking_write(parent); |
b4ce94de | 1623 | |
5dfe2be7 | 1624 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1625 | struct btrfs_key first_key; |
6702ed49 | 1626 | int close = 1; |
a6b6e75e | 1627 | |
081e9573 CM |
1628 | btrfs_node_key(parent, &disk_key, i); |
1629 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1630 | continue; | |
1631 | ||
1632 | progress_passed = 1; | |
6b80053d | 1633 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1634 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1635 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1636 | if (last_block == 0) |
1637 | last_block = blocknr; | |
5708b959 | 1638 | |
6702ed49 | 1639 | if (i > 0) { |
6b80053d CM |
1640 | other = btrfs_node_blockptr(parent, i - 1); |
1641 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1642 | } |
5dfe2be7 | 1643 | if (!close && i < end_slot) { |
6b80053d CM |
1644 | other = btrfs_node_blockptr(parent, i + 1); |
1645 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1646 | } |
e9d0b13b CM |
1647 | if (close) { |
1648 | last_block = blocknr; | |
6702ed49 | 1649 | continue; |
e9d0b13b | 1650 | } |
6702ed49 | 1651 | |
0b246afa | 1652 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1653 | if (cur) |
b9fab919 | 1654 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1655 | else |
1656 | uptodate = 0; | |
5708b959 | 1657 | if (!cur || !uptodate) { |
6b80053d | 1658 | if (!cur) { |
581c1760 QW |
1659 | cur = read_tree_block(fs_info, blocknr, gen, |
1660 | parent_level - 1, | |
1661 | &first_key); | |
64c043de LB |
1662 | if (IS_ERR(cur)) { |
1663 | return PTR_ERR(cur); | |
1664 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1665 | free_extent_buffer(cur); |
97d9a8a4 | 1666 | return -EIO; |
416bc658 | 1667 | } |
6b80053d | 1668 | } else if (!uptodate) { |
581c1760 QW |
1669 | err = btrfs_read_buffer(cur, gen, |
1670 | parent_level - 1,&first_key); | |
018642a1 TI |
1671 | if (err) { |
1672 | free_extent_buffer(cur); | |
1673 | return err; | |
1674 | } | |
f2183bde | 1675 | } |
6702ed49 | 1676 | } |
e9d0b13b | 1677 | if (search_start == 0) |
6b80053d | 1678 | search_start = last_block; |
e9d0b13b | 1679 | |
e7a84565 | 1680 | btrfs_tree_lock(cur); |
8bead258 | 1681 | btrfs_set_lock_blocking_write(cur); |
6b80053d | 1682 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1683 | &cur, search_start, |
6b80053d | 1684 | min(16 * blocksize, |
9fa8cfe7 | 1685 | (end_slot - i) * blocksize)); |
252c38f0 | 1686 | if (err) { |
e7a84565 | 1687 | btrfs_tree_unlock(cur); |
6b80053d | 1688 | free_extent_buffer(cur); |
6702ed49 | 1689 | break; |
252c38f0 | 1690 | } |
e7a84565 CM |
1691 | search_start = cur->start; |
1692 | last_block = cur->start; | |
f2183bde | 1693 | *last_ret = search_start; |
e7a84565 CM |
1694 | btrfs_tree_unlock(cur); |
1695 | free_extent_buffer(cur); | |
6702ed49 CM |
1696 | } |
1697 | return err; | |
1698 | } | |
1699 | ||
74123bd7 | 1700 | /* |
5f39d397 CM |
1701 | * search for key in the extent_buffer. The items start at offset p, |
1702 | * and they are item_size apart. There are 'max' items in p. | |
1703 | * | |
74123bd7 CM |
1704 | * the slot in the array is returned via slot, and it points to |
1705 | * the place where you would insert key if it is not found in | |
1706 | * the array. | |
1707 | * | |
1708 | * slot may point to max if the key is bigger than all of the keys | |
1709 | */ | |
e02119d5 | 1710 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1711 | unsigned long p, int item_size, |
1712 | const struct btrfs_key *key, | |
e02119d5 | 1713 | int max, int *slot) |
be0e5c09 CM |
1714 | { |
1715 | int low = 0; | |
1716 | int high = max; | |
1717 | int mid; | |
1718 | int ret; | |
479965d6 | 1719 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1720 | struct btrfs_disk_key unaligned; |
1721 | unsigned long offset; | |
5f39d397 CM |
1722 | char *kaddr = NULL; |
1723 | unsigned long map_start = 0; | |
1724 | unsigned long map_len = 0; | |
479965d6 | 1725 | int err; |
be0e5c09 | 1726 | |
5e24e9af LB |
1727 | if (low > high) { |
1728 | btrfs_err(eb->fs_info, | |
1729 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1730 | __func__, low, high, eb->start, | |
1731 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1732 | return -EINVAL; | |
1733 | } | |
1734 | ||
d397712b | 1735 | while (low < high) { |
be0e5c09 | 1736 | mid = (low + high) / 2; |
5f39d397 CM |
1737 | offset = p + mid * item_size; |
1738 | ||
a6591715 | 1739 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1740 | (offset + sizeof(struct btrfs_disk_key)) > |
1741 | map_start + map_len) { | |
934d375b CM |
1742 | |
1743 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1744 | sizeof(struct btrfs_disk_key), |
a6591715 | 1745 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1746 | |
1747 | if (!err) { | |
1748 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1749 | map_start); | |
415b35a5 | 1750 | } else if (err == 1) { |
479965d6 CM |
1751 | read_extent_buffer(eb, &unaligned, |
1752 | offset, sizeof(unaligned)); | |
1753 | tmp = &unaligned; | |
415b35a5 LB |
1754 | } else { |
1755 | return err; | |
479965d6 | 1756 | } |
5f39d397 | 1757 | |
5f39d397 CM |
1758 | } else { |
1759 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1760 | map_start); | |
1761 | } | |
be0e5c09 CM |
1762 | ret = comp_keys(tmp, key); |
1763 | ||
1764 | if (ret < 0) | |
1765 | low = mid + 1; | |
1766 | else if (ret > 0) | |
1767 | high = mid; | |
1768 | else { | |
1769 | *slot = mid; | |
1770 | return 0; | |
1771 | } | |
1772 | } | |
1773 | *slot = low; | |
1774 | return 1; | |
1775 | } | |
1776 | ||
97571fd0 CM |
1777 | /* |
1778 | * simple bin_search frontend that does the right thing for | |
1779 | * leaves vs nodes | |
1780 | */ | |
a74b35ec NB |
1781 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1782 | int level, int *slot) | |
be0e5c09 | 1783 | { |
f775738f | 1784 | if (level == 0) |
5f39d397 CM |
1785 | return generic_bin_search(eb, |
1786 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1787 | sizeof(struct btrfs_item), |
5f39d397 | 1788 | key, btrfs_header_nritems(eb), |
7518a238 | 1789 | slot); |
f775738f | 1790 | else |
5f39d397 CM |
1791 | return generic_bin_search(eb, |
1792 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1793 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1794 | key, btrfs_header_nritems(eb), |
7518a238 | 1795 | slot); |
be0e5c09 CM |
1796 | } |
1797 | ||
f0486c68 YZ |
1798 | static void root_add_used(struct btrfs_root *root, u32 size) |
1799 | { | |
1800 | spin_lock(&root->accounting_lock); | |
1801 | btrfs_set_root_used(&root->root_item, | |
1802 | btrfs_root_used(&root->root_item) + size); | |
1803 | spin_unlock(&root->accounting_lock); | |
1804 | } | |
1805 | ||
1806 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1807 | { | |
1808 | spin_lock(&root->accounting_lock); | |
1809 | btrfs_set_root_used(&root->root_item, | |
1810 | btrfs_root_used(&root->root_item) - size); | |
1811 | spin_unlock(&root->accounting_lock); | |
1812 | } | |
1813 | ||
d352ac68 CM |
1814 | /* given a node and slot number, this reads the blocks it points to. The |
1815 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1816 | */ |
4b231ae4 DS |
1817 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
1818 | int slot) | |
bb803951 | 1819 | { |
ca7a79ad | 1820 | int level = btrfs_header_level(parent); |
416bc658 | 1821 | struct extent_buffer *eb; |
581c1760 | 1822 | struct btrfs_key first_key; |
416bc658 | 1823 | |
fb770ae4 LB |
1824 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1825 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1826 | |
1827 | BUG_ON(level == 0); | |
1828 | ||
581c1760 | 1829 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 1830 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1831 | btrfs_node_ptr_generation(parent, slot), |
1832 | level - 1, &first_key); | |
fb770ae4 LB |
1833 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1834 | free_extent_buffer(eb); | |
1835 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1836 | } |
1837 | ||
1838 | return eb; | |
bb803951 CM |
1839 | } |
1840 | ||
d352ac68 CM |
1841 | /* |
1842 | * node level balancing, used to make sure nodes are in proper order for | |
1843 | * item deletion. We balance from the top down, so we have to make sure | |
1844 | * that a deletion won't leave an node completely empty later on. | |
1845 | */ | |
e02119d5 | 1846 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1847 | struct btrfs_root *root, |
1848 | struct btrfs_path *path, int level) | |
bb803951 | 1849 | { |
0b246afa | 1850 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1851 | struct extent_buffer *right = NULL; |
1852 | struct extent_buffer *mid; | |
1853 | struct extent_buffer *left = NULL; | |
1854 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1855 | int ret = 0; |
1856 | int wret; | |
1857 | int pslot; | |
bb803951 | 1858 | int orig_slot = path->slots[level]; |
79f95c82 | 1859 | u64 orig_ptr; |
bb803951 | 1860 | |
98e6b1eb | 1861 | ASSERT(level > 0); |
bb803951 | 1862 | |
5f39d397 | 1863 | mid = path->nodes[level]; |
b4ce94de | 1864 | |
bd681513 CM |
1865 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1866 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1867 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1868 | ||
1d4f8a0c | 1869 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1870 | |
a05a9bb1 | 1871 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1872 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1873 | pslot = path->slots[level + 1]; |
1874 | } | |
bb803951 | 1875 | |
40689478 CM |
1876 | /* |
1877 | * deal with the case where there is only one pointer in the root | |
1878 | * by promoting the node below to a root | |
1879 | */ | |
5f39d397 CM |
1880 | if (!parent) { |
1881 | struct extent_buffer *child; | |
bb803951 | 1882 | |
5f39d397 | 1883 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1884 | return 0; |
1885 | ||
1886 | /* promote the child to a root */ | |
4b231ae4 | 1887 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
1888 | if (IS_ERR(child)) { |
1889 | ret = PTR_ERR(child); | |
0b246afa | 1890 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1891 | goto enospc; |
1892 | } | |
1893 | ||
925baedd | 1894 | btrfs_tree_lock(child); |
8bead258 | 1895 | btrfs_set_lock_blocking_write(child); |
9fa8cfe7 | 1896 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1897 | if (ret) { |
1898 | btrfs_tree_unlock(child); | |
1899 | free_extent_buffer(child); | |
1900 | goto enospc; | |
1901 | } | |
2f375ab9 | 1902 | |
d9d19a01 DS |
1903 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1904 | BUG_ON(ret < 0); | |
240f62c8 | 1905 | rcu_assign_pointer(root->node, child); |
925baedd | 1906 | |
0b86a832 | 1907 | add_root_to_dirty_list(root); |
925baedd | 1908 | btrfs_tree_unlock(child); |
b4ce94de | 1909 | |
925baedd | 1910 | path->locks[level] = 0; |
bb803951 | 1911 | path->nodes[level] = NULL; |
6a884d7d | 1912 | btrfs_clean_tree_block(mid); |
925baedd | 1913 | btrfs_tree_unlock(mid); |
bb803951 | 1914 | /* once for the path */ |
5f39d397 | 1915 | free_extent_buffer(mid); |
f0486c68 YZ |
1916 | |
1917 | root_sub_used(root, mid->len); | |
5581a51a | 1918 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1919 | /* once for the root ptr */ |
3083ee2e | 1920 | free_extent_buffer_stale(mid); |
f0486c68 | 1921 | return 0; |
bb803951 | 1922 | } |
5f39d397 | 1923 | if (btrfs_header_nritems(mid) > |
0b246afa | 1924 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1925 | return 0; |
1926 | ||
4b231ae4 | 1927 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1928 | if (IS_ERR(left)) |
1929 | left = NULL; | |
1930 | ||
5f39d397 | 1931 | if (left) { |
925baedd | 1932 | btrfs_tree_lock(left); |
8bead258 | 1933 | btrfs_set_lock_blocking_write(left); |
5f39d397 | 1934 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1935 | parent, pslot - 1, &left); |
54aa1f4d CM |
1936 | if (wret) { |
1937 | ret = wret; | |
1938 | goto enospc; | |
1939 | } | |
2cc58cf2 | 1940 | } |
fb770ae4 | 1941 | |
4b231ae4 | 1942 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1943 | if (IS_ERR(right)) |
1944 | right = NULL; | |
1945 | ||
5f39d397 | 1946 | if (right) { |
925baedd | 1947 | btrfs_tree_lock(right); |
8bead258 | 1948 | btrfs_set_lock_blocking_write(right); |
5f39d397 | 1949 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1950 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1951 | if (wret) { |
1952 | ret = wret; | |
1953 | goto enospc; | |
1954 | } | |
1955 | } | |
1956 | ||
1957 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1958 | if (left) { |
1959 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 1960 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
1961 | if (wret < 0) |
1962 | ret = wret; | |
bb803951 | 1963 | } |
79f95c82 CM |
1964 | |
1965 | /* | |
1966 | * then try to empty the right most buffer into the middle | |
1967 | */ | |
5f39d397 | 1968 | if (right) { |
d30a668f | 1969 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1970 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1971 | ret = wret; |
5f39d397 | 1972 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 1973 | btrfs_clean_tree_block(right); |
925baedd | 1974 | btrfs_tree_unlock(right); |
afe5fea7 | 1975 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1976 | root_sub_used(root, right->len); |
5581a51a | 1977 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1978 | free_extent_buffer_stale(right); |
f0486c68 | 1979 | right = NULL; |
bb803951 | 1980 | } else { |
5f39d397 CM |
1981 | struct btrfs_disk_key right_key; |
1982 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1983 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1984 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1985 | BUG_ON(ret < 0); | |
5f39d397 CM |
1986 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1987 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1988 | } |
1989 | } | |
5f39d397 | 1990 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1991 | /* |
1992 | * we're not allowed to leave a node with one item in the | |
1993 | * tree during a delete. A deletion from lower in the tree | |
1994 | * could try to delete the only pointer in this node. | |
1995 | * So, pull some keys from the left. | |
1996 | * There has to be a left pointer at this point because | |
1997 | * otherwise we would have pulled some pointers from the | |
1998 | * right | |
1999 | */ | |
305a26af MF |
2000 | if (!left) { |
2001 | ret = -EROFS; | |
0b246afa | 2002 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
2003 | goto enospc; |
2004 | } | |
55d32ed8 | 2005 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 2006 | if (wret < 0) { |
79f95c82 | 2007 | ret = wret; |
54aa1f4d CM |
2008 | goto enospc; |
2009 | } | |
bce4eae9 | 2010 | if (wret == 1) { |
d30a668f | 2011 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
2012 | if (wret < 0) |
2013 | ret = wret; | |
2014 | } | |
79f95c82 CM |
2015 | BUG_ON(wret == 1); |
2016 | } | |
5f39d397 | 2017 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 2018 | btrfs_clean_tree_block(mid); |
925baedd | 2019 | btrfs_tree_unlock(mid); |
afe5fea7 | 2020 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2021 | root_sub_used(root, mid->len); |
5581a51a | 2022 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2023 | free_extent_buffer_stale(mid); |
f0486c68 | 2024 | mid = NULL; |
79f95c82 CM |
2025 | } else { |
2026 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2027 | struct btrfs_disk_key mid_key; |
2028 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
2029 | ret = tree_mod_log_insert_key(parent, pslot, |
2030 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2031 | BUG_ON(ret < 0); | |
5f39d397 CM |
2032 | btrfs_set_node_key(parent, &mid_key, pslot); |
2033 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2034 | } |
bb803951 | 2035 | |
79f95c82 | 2036 | /* update the path */ |
5f39d397 CM |
2037 | if (left) { |
2038 | if (btrfs_header_nritems(left) > orig_slot) { | |
2039 | extent_buffer_get(left); | |
925baedd | 2040 | /* left was locked after cow */ |
5f39d397 | 2041 | path->nodes[level] = left; |
bb803951 CM |
2042 | path->slots[level + 1] -= 1; |
2043 | path->slots[level] = orig_slot; | |
925baedd CM |
2044 | if (mid) { |
2045 | btrfs_tree_unlock(mid); | |
5f39d397 | 2046 | free_extent_buffer(mid); |
925baedd | 2047 | } |
bb803951 | 2048 | } else { |
5f39d397 | 2049 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2050 | path->slots[level] = orig_slot; |
2051 | } | |
2052 | } | |
79f95c82 | 2053 | /* double check we haven't messed things up */ |
e20d96d6 | 2054 | if (orig_ptr != |
5f39d397 | 2055 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2056 | BUG(); |
54aa1f4d | 2057 | enospc: |
925baedd CM |
2058 | if (right) { |
2059 | btrfs_tree_unlock(right); | |
5f39d397 | 2060 | free_extent_buffer(right); |
925baedd CM |
2061 | } |
2062 | if (left) { | |
2063 | if (path->nodes[level] != left) | |
2064 | btrfs_tree_unlock(left); | |
5f39d397 | 2065 | free_extent_buffer(left); |
925baedd | 2066 | } |
bb803951 CM |
2067 | return ret; |
2068 | } | |
2069 | ||
d352ac68 CM |
2070 | /* Node balancing for insertion. Here we only split or push nodes around |
2071 | * when they are completely full. This is also done top down, so we | |
2072 | * have to be pessimistic. | |
2073 | */ | |
d397712b | 2074 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2075 | struct btrfs_root *root, |
2076 | struct btrfs_path *path, int level) | |
e66f709b | 2077 | { |
0b246afa | 2078 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2079 | struct extent_buffer *right = NULL; |
2080 | struct extent_buffer *mid; | |
2081 | struct extent_buffer *left = NULL; | |
2082 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2083 | int ret = 0; |
2084 | int wret; | |
2085 | int pslot; | |
2086 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2087 | |
2088 | if (level == 0) | |
2089 | return 1; | |
2090 | ||
5f39d397 | 2091 | mid = path->nodes[level]; |
7bb86316 | 2092 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2093 | |
a05a9bb1 | 2094 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2095 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2096 | pslot = path->slots[level + 1]; |
2097 | } | |
e66f709b | 2098 | |
5f39d397 | 2099 | if (!parent) |
e66f709b | 2100 | return 1; |
e66f709b | 2101 | |
4b231ae4 | 2102 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
2103 | if (IS_ERR(left)) |
2104 | left = NULL; | |
e66f709b CM |
2105 | |
2106 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2107 | if (left) { |
e66f709b | 2108 | u32 left_nr; |
925baedd CM |
2109 | |
2110 | btrfs_tree_lock(left); | |
8bead258 | 2111 | btrfs_set_lock_blocking_write(left); |
b4ce94de | 2112 | |
5f39d397 | 2113 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2114 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2115 | wret = 1; |
2116 | } else { | |
5f39d397 | 2117 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2118 | pslot - 1, &left); |
54aa1f4d CM |
2119 | if (ret) |
2120 | wret = 1; | |
2121 | else { | |
d30a668f | 2122 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 2123 | } |
33ade1f8 | 2124 | } |
e66f709b CM |
2125 | if (wret < 0) |
2126 | ret = wret; | |
2127 | if (wret == 0) { | |
5f39d397 | 2128 | struct btrfs_disk_key disk_key; |
e66f709b | 2129 | orig_slot += left_nr; |
5f39d397 | 2130 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2131 | ret = tree_mod_log_insert_key(parent, pslot, |
2132 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2133 | BUG_ON(ret < 0); | |
5f39d397 CM |
2134 | btrfs_set_node_key(parent, &disk_key, pslot); |
2135 | btrfs_mark_buffer_dirty(parent); | |
2136 | if (btrfs_header_nritems(left) > orig_slot) { | |
2137 | path->nodes[level] = left; | |
e66f709b CM |
2138 | path->slots[level + 1] -= 1; |
2139 | path->slots[level] = orig_slot; | |
925baedd | 2140 | btrfs_tree_unlock(mid); |
5f39d397 | 2141 | free_extent_buffer(mid); |
e66f709b CM |
2142 | } else { |
2143 | orig_slot -= | |
5f39d397 | 2144 | btrfs_header_nritems(left); |
e66f709b | 2145 | path->slots[level] = orig_slot; |
925baedd | 2146 | btrfs_tree_unlock(left); |
5f39d397 | 2147 | free_extent_buffer(left); |
e66f709b | 2148 | } |
e66f709b CM |
2149 | return 0; |
2150 | } | |
925baedd | 2151 | btrfs_tree_unlock(left); |
5f39d397 | 2152 | free_extent_buffer(left); |
e66f709b | 2153 | } |
4b231ae4 | 2154 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
2155 | if (IS_ERR(right)) |
2156 | right = NULL; | |
e66f709b CM |
2157 | |
2158 | /* | |
2159 | * then try to empty the right most buffer into the middle | |
2160 | */ | |
5f39d397 | 2161 | if (right) { |
33ade1f8 | 2162 | u32 right_nr; |
b4ce94de | 2163 | |
925baedd | 2164 | btrfs_tree_lock(right); |
8bead258 | 2165 | btrfs_set_lock_blocking_write(right); |
b4ce94de | 2166 | |
5f39d397 | 2167 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2168 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2169 | wret = 1; |
2170 | } else { | |
5f39d397 CM |
2171 | ret = btrfs_cow_block(trans, root, right, |
2172 | parent, pslot + 1, | |
9fa8cfe7 | 2173 | &right); |
54aa1f4d CM |
2174 | if (ret) |
2175 | wret = 1; | |
2176 | else { | |
55d32ed8 | 2177 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 2178 | } |
33ade1f8 | 2179 | } |
e66f709b CM |
2180 | if (wret < 0) |
2181 | ret = wret; | |
2182 | if (wret == 0) { | |
5f39d397 CM |
2183 | struct btrfs_disk_key disk_key; |
2184 | ||
2185 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2186 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2187 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2188 | BUG_ON(ret < 0); | |
5f39d397 CM |
2189 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2190 | btrfs_mark_buffer_dirty(parent); | |
2191 | ||
2192 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2193 | path->nodes[level] = right; | |
e66f709b CM |
2194 | path->slots[level + 1] += 1; |
2195 | path->slots[level] = orig_slot - | |
5f39d397 | 2196 | btrfs_header_nritems(mid); |
925baedd | 2197 | btrfs_tree_unlock(mid); |
5f39d397 | 2198 | free_extent_buffer(mid); |
e66f709b | 2199 | } else { |
925baedd | 2200 | btrfs_tree_unlock(right); |
5f39d397 | 2201 | free_extent_buffer(right); |
e66f709b | 2202 | } |
e66f709b CM |
2203 | return 0; |
2204 | } | |
925baedd | 2205 | btrfs_tree_unlock(right); |
5f39d397 | 2206 | free_extent_buffer(right); |
e66f709b | 2207 | } |
e66f709b CM |
2208 | return 1; |
2209 | } | |
2210 | ||
3c69faec | 2211 | /* |
d352ac68 CM |
2212 | * readahead one full node of leaves, finding things that are close |
2213 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2214 | */ |
2ff7e61e | 2215 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2216 | struct btrfs_path *path, |
2217 | int level, int slot, u64 objectid) | |
3c69faec | 2218 | { |
5f39d397 | 2219 | struct extent_buffer *node; |
01f46658 | 2220 | struct btrfs_disk_key disk_key; |
3c69faec | 2221 | u32 nritems; |
3c69faec | 2222 | u64 search; |
a7175319 | 2223 | u64 target; |
6b80053d | 2224 | u64 nread = 0; |
5f39d397 | 2225 | struct extent_buffer *eb; |
6b80053d CM |
2226 | u32 nr; |
2227 | u32 blocksize; | |
2228 | u32 nscan = 0; | |
db94535d | 2229 | |
a6b6e75e | 2230 | if (level != 1) |
6702ed49 CM |
2231 | return; |
2232 | ||
2233 | if (!path->nodes[level]) | |
3c69faec CM |
2234 | return; |
2235 | ||
5f39d397 | 2236 | node = path->nodes[level]; |
925baedd | 2237 | |
3c69faec | 2238 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2239 | blocksize = fs_info->nodesize; |
2240 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2241 | if (eb) { |
2242 | free_extent_buffer(eb); | |
3c69faec CM |
2243 | return; |
2244 | } | |
2245 | ||
a7175319 | 2246 | target = search; |
6b80053d | 2247 | |
5f39d397 | 2248 | nritems = btrfs_header_nritems(node); |
6b80053d | 2249 | nr = slot; |
25b8b936 | 2250 | |
d397712b | 2251 | while (1) { |
e4058b54 | 2252 | if (path->reada == READA_BACK) { |
6b80053d CM |
2253 | if (nr == 0) |
2254 | break; | |
2255 | nr--; | |
e4058b54 | 2256 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2257 | nr++; |
2258 | if (nr >= nritems) | |
2259 | break; | |
3c69faec | 2260 | } |
e4058b54 | 2261 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2262 | btrfs_node_key(node, &disk_key, nr); |
2263 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2264 | break; | |
2265 | } | |
6b80053d | 2266 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2267 | if ((search <= target && target - search <= 65536) || |
2268 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2269 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2270 | nread += blocksize; |
2271 | } | |
2272 | nscan++; | |
a7175319 | 2273 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2274 | break; |
3c69faec CM |
2275 | } |
2276 | } | |
925baedd | 2277 | |
2ff7e61e | 2278 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2279 | struct btrfs_path *path, int level) |
b4ce94de CM |
2280 | { |
2281 | int slot; | |
2282 | int nritems; | |
2283 | struct extent_buffer *parent; | |
2284 | struct extent_buffer *eb; | |
2285 | u64 gen; | |
2286 | u64 block1 = 0; | |
2287 | u64 block2 = 0; | |
b4ce94de | 2288 | |
8c594ea8 | 2289 | parent = path->nodes[level + 1]; |
b4ce94de | 2290 | if (!parent) |
0b08851f | 2291 | return; |
b4ce94de CM |
2292 | |
2293 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2294 | slot = path->slots[level + 1]; |
b4ce94de CM |
2295 | |
2296 | if (slot > 0) { | |
2297 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2298 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2299 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2300 | /* |
2301 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2302 | * don't want to return eagain here. That will loop | |
2303 | * forever | |
2304 | */ | |
2305 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2306 | block1 = 0; |
2307 | free_extent_buffer(eb); | |
2308 | } | |
8c594ea8 | 2309 | if (slot + 1 < nritems) { |
b4ce94de CM |
2310 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2311 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2312 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2313 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2314 | block2 = 0; |
2315 | free_extent_buffer(eb); | |
2316 | } | |
8c594ea8 | 2317 | |
0b08851f | 2318 | if (block1) |
2ff7e61e | 2319 | readahead_tree_block(fs_info, block1); |
0b08851f | 2320 | if (block2) |
2ff7e61e | 2321 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2322 | } |
2323 | ||
2324 | ||
d352ac68 | 2325 | /* |
d397712b CM |
2326 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2327 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2328 | * operations on the tree might require changing key pointers higher up in the | |
2329 | * tree. | |
d352ac68 | 2330 | * |
d397712b CM |
2331 | * callers might also have set path->keep_locks, which tells this code to keep |
2332 | * the lock if the path points to the last slot in the block. This is part of | |
2333 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2334 | * |
d397712b CM |
2335 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2336 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2337 | */ |
e02119d5 | 2338 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2339 | int lowest_unlock, int min_write_lock_level, |
2340 | int *write_lock_level) | |
925baedd CM |
2341 | { |
2342 | int i; | |
2343 | int skip_level = level; | |
051e1b9f | 2344 | int no_skips = 0; |
925baedd CM |
2345 | struct extent_buffer *t; |
2346 | ||
2347 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2348 | if (!path->nodes[i]) | |
2349 | break; | |
2350 | if (!path->locks[i]) | |
2351 | break; | |
051e1b9f | 2352 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2353 | skip_level = i + 1; |
2354 | continue; | |
2355 | } | |
051e1b9f | 2356 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2357 | u32 nritems; |
2358 | t = path->nodes[i]; | |
2359 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2360 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2361 | skip_level = i + 1; |
2362 | continue; | |
2363 | } | |
2364 | } | |
051e1b9f CM |
2365 | if (skip_level < i && i >= lowest_unlock) |
2366 | no_skips = 1; | |
2367 | ||
925baedd | 2368 | t = path->nodes[i]; |
d80bb3f9 | 2369 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2370 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2371 | path->locks[i] = 0; |
f7c79f30 CM |
2372 | if (write_lock_level && |
2373 | i > min_write_lock_level && | |
2374 | i <= *write_lock_level) { | |
2375 | *write_lock_level = i - 1; | |
2376 | } | |
925baedd CM |
2377 | } |
2378 | } | |
2379 | } | |
2380 | ||
b4ce94de CM |
2381 | /* |
2382 | * This releases any locks held in the path starting at level and | |
2383 | * going all the way up to the root. | |
2384 | * | |
2385 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2386 | * corner cases, such as COW of the block at slot zero in the node. This | |
2387 | * ignores those rules, and it should only be called when there are no | |
2388 | * more updates to be done higher up in the tree. | |
2389 | */ | |
2390 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2391 | { | |
2392 | int i; | |
2393 | ||
09a2a8f9 | 2394 | if (path->keep_locks) |
b4ce94de CM |
2395 | return; |
2396 | ||
2397 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2398 | if (!path->nodes[i]) | |
12f4dacc | 2399 | continue; |
b4ce94de | 2400 | if (!path->locks[i]) |
12f4dacc | 2401 | continue; |
bd681513 | 2402 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2403 | path->locks[i] = 0; |
2404 | } | |
2405 | } | |
2406 | ||
c8c42864 CM |
2407 | /* |
2408 | * helper function for btrfs_search_slot. The goal is to find a block | |
2409 | * in cache without setting the path to blocking. If we find the block | |
2410 | * we return zero and the path is unchanged. | |
2411 | * | |
2412 | * If we can't find the block, we set the path blocking and do some | |
2413 | * reada. -EAGAIN is returned and the search must be repeated. | |
2414 | */ | |
2415 | static int | |
d07b8528 LB |
2416 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2417 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2418 | const struct btrfs_key *key) |
c8c42864 | 2419 | { |
0b246afa | 2420 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2421 | u64 blocknr; |
2422 | u64 gen; | |
c8c42864 CM |
2423 | struct extent_buffer *b = *eb_ret; |
2424 | struct extent_buffer *tmp; | |
581c1760 | 2425 | struct btrfs_key first_key; |
76a05b35 | 2426 | int ret; |
581c1760 | 2427 | int parent_level; |
c8c42864 CM |
2428 | |
2429 | blocknr = btrfs_node_blockptr(b, slot); | |
2430 | gen = btrfs_node_ptr_generation(b, slot); | |
581c1760 QW |
2431 | parent_level = btrfs_header_level(b); |
2432 | btrfs_node_key_to_cpu(b, &first_key, slot); | |
c8c42864 | 2433 | |
0b246afa | 2434 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2435 | if (tmp) { |
b9fab919 | 2436 | /* first we do an atomic uptodate check */ |
bdf7c00e | 2437 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
2438 | /* |
2439 | * Do extra check for first_key, eb can be stale due to | |
2440 | * being cached, read from scrub, or have multiple | |
2441 | * parents (shared tree blocks). | |
2442 | */ | |
e064d5e9 | 2443 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
2444 | parent_level - 1, &first_key, gen)) { |
2445 | free_extent_buffer(tmp); | |
2446 | return -EUCLEAN; | |
2447 | } | |
bdf7c00e JB |
2448 | *eb_ret = tmp; |
2449 | return 0; | |
2450 | } | |
2451 | ||
2452 | /* the pages were up to date, but we failed | |
2453 | * the generation number check. Do a full | |
2454 | * read for the generation number that is correct. | |
2455 | * We must do this without dropping locks so | |
2456 | * we can trust our generation number | |
2457 | */ | |
2458 | btrfs_set_path_blocking(p); | |
2459 | ||
2460 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2461 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2462 | if (!ret) { |
2463 | *eb_ret = tmp; | |
2464 | return 0; | |
cb44921a | 2465 | } |
bdf7c00e JB |
2466 | free_extent_buffer(tmp); |
2467 | btrfs_release_path(p); | |
2468 | return -EIO; | |
c8c42864 CM |
2469 | } |
2470 | ||
2471 | /* | |
2472 | * reduce lock contention at high levels | |
2473 | * of the btree by dropping locks before | |
76a05b35 CM |
2474 | * we read. Don't release the lock on the current |
2475 | * level because we need to walk this node to figure | |
2476 | * out which blocks to read. | |
c8c42864 | 2477 | */ |
8c594ea8 CM |
2478 | btrfs_unlock_up_safe(p, level + 1); |
2479 | btrfs_set_path_blocking(p); | |
2480 | ||
e4058b54 | 2481 | if (p->reada != READA_NONE) |
2ff7e61e | 2482 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2483 | |
76a05b35 | 2484 | ret = -EAGAIN; |
02a3307a | 2485 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2486 | &first_key); |
64c043de | 2487 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2488 | /* |
2489 | * If the read above didn't mark this buffer up to date, | |
2490 | * it will never end up being up to date. Set ret to EIO now | |
2491 | * and give up so that our caller doesn't loop forever | |
2492 | * on our EAGAINs. | |
2493 | */ | |
e6a1d6fd | 2494 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2495 | ret = -EIO; |
c8c42864 | 2496 | free_extent_buffer(tmp); |
c871b0f2 LB |
2497 | } else { |
2498 | ret = PTR_ERR(tmp); | |
76a05b35 | 2499 | } |
02a3307a LB |
2500 | |
2501 | btrfs_release_path(p); | |
76a05b35 | 2502 | return ret; |
c8c42864 CM |
2503 | } |
2504 | ||
2505 | /* | |
2506 | * helper function for btrfs_search_slot. This does all of the checks | |
2507 | * for node-level blocks and does any balancing required based on | |
2508 | * the ins_len. | |
2509 | * | |
2510 | * If no extra work was required, zero is returned. If we had to | |
2511 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2512 | * start over | |
2513 | */ | |
2514 | static int | |
2515 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2516 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2517 | struct extent_buffer *b, int level, int ins_len, |
2518 | int *write_lock_level) | |
c8c42864 | 2519 | { |
0b246afa | 2520 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2521 | int ret; |
0b246afa | 2522 | |
c8c42864 | 2523 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2524 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2525 | int sret; |
2526 | ||
bd681513 CM |
2527 | if (*write_lock_level < level + 1) { |
2528 | *write_lock_level = level + 1; | |
2529 | btrfs_release_path(p); | |
2530 | goto again; | |
2531 | } | |
2532 | ||
c8c42864 | 2533 | btrfs_set_path_blocking(p); |
2ff7e61e | 2534 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2535 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2536 | |
2537 | BUG_ON(sret > 0); | |
2538 | if (sret) { | |
2539 | ret = sret; | |
2540 | goto done; | |
2541 | } | |
2542 | b = p->nodes[level]; | |
2543 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2544 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2545 | int sret; |
2546 | ||
bd681513 CM |
2547 | if (*write_lock_level < level + 1) { |
2548 | *write_lock_level = level + 1; | |
2549 | btrfs_release_path(p); | |
2550 | goto again; | |
2551 | } | |
2552 | ||
c8c42864 | 2553 | btrfs_set_path_blocking(p); |
2ff7e61e | 2554 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2555 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2556 | |
2557 | if (sret) { | |
2558 | ret = sret; | |
2559 | goto done; | |
2560 | } | |
2561 | b = p->nodes[level]; | |
2562 | if (!b) { | |
b3b4aa74 | 2563 | btrfs_release_path(p); |
c8c42864 CM |
2564 | goto again; |
2565 | } | |
2566 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2567 | } | |
2568 | return 0; | |
2569 | ||
2570 | again: | |
2571 | ret = -EAGAIN; | |
2572 | done: | |
2573 | return ret; | |
2574 | } | |
2575 | ||
310712b2 | 2576 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2577 | int level, int *prev_cmp, int *slot) |
2578 | { | |
2579 | if (*prev_cmp != 0) { | |
a74b35ec | 2580 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2581 | return *prev_cmp; |
2582 | } | |
2583 | ||
d7396f07 FDBM |
2584 | *slot = 0; |
2585 | ||
2586 | return 0; | |
2587 | } | |
2588 | ||
381cf658 | 2589 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2590 | u64 iobjectid, u64 ioff, u8 key_type, |
2591 | struct btrfs_key *found_key) | |
2592 | { | |
2593 | int ret; | |
2594 | struct btrfs_key key; | |
2595 | struct extent_buffer *eb; | |
381cf658 DS |
2596 | |
2597 | ASSERT(path); | |
1d4c08e0 | 2598 | ASSERT(found_key); |
e33d5c3d KN |
2599 | |
2600 | key.type = key_type; | |
2601 | key.objectid = iobjectid; | |
2602 | key.offset = ioff; | |
2603 | ||
2604 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2605 | if (ret < 0) |
e33d5c3d KN |
2606 | return ret; |
2607 | ||
2608 | eb = path->nodes[0]; | |
2609 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2610 | ret = btrfs_next_leaf(fs_root, path); | |
2611 | if (ret) | |
2612 | return ret; | |
2613 | eb = path->nodes[0]; | |
2614 | } | |
2615 | ||
2616 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2617 | if (found_key->type != key.type || | |
2618 | found_key->objectid != key.objectid) | |
2619 | return 1; | |
2620 | ||
2621 | return 0; | |
2622 | } | |
2623 | ||
1fc28d8e LB |
2624 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2625 | struct btrfs_path *p, | |
2626 | int write_lock_level) | |
2627 | { | |
2628 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2629 | struct extent_buffer *b; | |
2630 | int root_lock; | |
2631 | int level = 0; | |
2632 | ||
2633 | /* We try very hard to do read locks on the root */ | |
2634 | root_lock = BTRFS_READ_LOCK; | |
2635 | ||
2636 | if (p->search_commit_root) { | |
be6821f8 FM |
2637 | /* |
2638 | * The commit roots are read only so we always do read locks, | |
2639 | * and we always must hold the commit_root_sem when doing | |
2640 | * searches on them, the only exception is send where we don't | |
2641 | * want to block transaction commits for a long time, so | |
2642 | * we need to clone the commit root in order to avoid races | |
2643 | * with transaction commits that create a snapshot of one of | |
2644 | * the roots used by a send operation. | |
2645 | */ | |
2646 | if (p->need_commit_sem) { | |
1fc28d8e | 2647 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 2648 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 2649 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
2650 | if (!b) |
2651 | return ERR_PTR(-ENOMEM); | |
2652 | ||
2653 | } else { | |
2654 | b = root->commit_root; | |
2655 | extent_buffer_get(b); | |
2656 | } | |
2657 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
2658 | /* |
2659 | * Ensure that all callers have set skip_locking when | |
2660 | * p->search_commit_root = 1. | |
2661 | */ | |
2662 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2663 | |
2664 | goto out; | |
2665 | } | |
2666 | ||
2667 | if (p->skip_locking) { | |
2668 | b = btrfs_root_node(root); | |
2669 | level = btrfs_header_level(b); | |
2670 | goto out; | |
2671 | } | |
2672 | ||
2673 | /* | |
662c653b LB |
2674 | * If the level is set to maximum, we can skip trying to get the read |
2675 | * lock. | |
1fc28d8e | 2676 | */ |
662c653b LB |
2677 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2678 | /* | |
2679 | * We don't know the level of the root node until we actually | |
2680 | * have it read locked | |
2681 | */ | |
2682 | b = btrfs_read_lock_root_node(root); | |
2683 | level = btrfs_header_level(b); | |
2684 | if (level > write_lock_level) | |
2685 | goto out; | |
2686 | ||
2687 | /* Whoops, must trade for write lock */ | |
2688 | btrfs_tree_read_unlock(b); | |
2689 | free_extent_buffer(b); | |
2690 | } | |
1fc28d8e | 2691 | |
1fc28d8e LB |
2692 | b = btrfs_lock_root_node(root); |
2693 | root_lock = BTRFS_WRITE_LOCK; | |
2694 | ||
2695 | /* The level might have changed, check again */ | |
2696 | level = btrfs_header_level(b); | |
2697 | ||
2698 | out: | |
2699 | p->nodes[level] = b; | |
2700 | if (!p->skip_locking) | |
2701 | p->locks[level] = root_lock; | |
2702 | /* | |
2703 | * Callers are responsible for dropping b's references. | |
2704 | */ | |
2705 | return b; | |
2706 | } | |
2707 | ||
2708 | ||
74123bd7 | 2709 | /* |
4271ecea NB |
2710 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2711 | * modifications to preserve tree invariants. | |
74123bd7 | 2712 | * |
4271ecea NB |
2713 | * @trans: Handle of transaction, used when modifying the tree |
2714 | * @p: Holds all btree nodes along the search path | |
2715 | * @root: The root node of the tree | |
2716 | * @key: The key we are looking for | |
2717 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2718 | * deletions it's -1. 0 for plain searches | |
2719 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2720 | * when modifying the tree. | |
97571fd0 | 2721 | * |
4271ecea NB |
2722 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2723 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2724 | * | |
2725 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2726 | * of the path (level 0) | |
2727 | * | |
2728 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2729 | * points to the slot where it should be inserted | |
2730 | * | |
2731 | * If an error is encountered while searching the tree a negative error number | |
2732 | * is returned | |
74123bd7 | 2733 | */ |
310712b2 OS |
2734 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2735 | const struct btrfs_key *key, struct btrfs_path *p, | |
2736 | int ins_len, int cow) | |
be0e5c09 | 2737 | { |
5f39d397 | 2738 | struct extent_buffer *b; |
be0e5c09 CM |
2739 | int slot; |
2740 | int ret; | |
33c66f43 | 2741 | int err; |
be0e5c09 | 2742 | int level; |
925baedd | 2743 | int lowest_unlock = 1; |
bd681513 CM |
2744 | /* everything at write_lock_level or lower must be write locked */ |
2745 | int write_lock_level = 0; | |
9f3a7427 | 2746 | u8 lowest_level = 0; |
f7c79f30 | 2747 | int min_write_lock_level; |
d7396f07 | 2748 | int prev_cmp; |
9f3a7427 | 2749 | |
6702ed49 | 2750 | lowest_level = p->lowest_level; |
323ac95b | 2751 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2752 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2753 | BUG_ON(!cow && ins_len); |
25179201 | 2754 | |
bd681513 | 2755 | if (ins_len < 0) { |
925baedd | 2756 | lowest_unlock = 2; |
65b51a00 | 2757 | |
bd681513 CM |
2758 | /* when we are removing items, we might have to go up to level |
2759 | * two as we update tree pointers Make sure we keep write | |
2760 | * for those levels as well | |
2761 | */ | |
2762 | write_lock_level = 2; | |
2763 | } else if (ins_len > 0) { | |
2764 | /* | |
2765 | * for inserting items, make sure we have a write lock on | |
2766 | * level 1 so we can update keys | |
2767 | */ | |
2768 | write_lock_level = 1; | |
2769 | } | |
2770 | ||
2771 | if (!cow) | |
2772 | write_lock_level = -1; | |
2773 | ||
09a2a8f9 | 2774 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2775 | write_lock_level = BTRFS_MAX_LEVEL; |
2776 | ||
f7c79f30 CM |
2777 | min_write_lock_level = write_lock_level; |
2778 | ||
bb803951 | 2779 | again: |
d7396f07 | 2780 | prev_cmp = -1; |
1fc28d8e | 2781 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2782 | if (IS_ERR(b)) { |
2783 | ret = PTR_ERR(b); | |
2784 | goto done; | |
2785 | } | |
925baedd | 2786 | |
eb60ceac | 2787 | while (b) { |
5f39d397 | 2788 | level = btrfs_header_level(b); |
65b51a00 CM |
2789 | |
2790 | /* | |
2791 | * setup the path here so we can release it under lock | |
2792 | * contention with the cow code | |
2793 | */ | |
02217ed2 | 2794 | if (cow) { |
9ea2c7c9 NB |
2795 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2796 | ||
c8c42864 CM |
2797 | /* |
2798 | * if we don't really need to cow this block | |
2799 | * then we don't want to set the path blocking, | |
2800 | * so we test it here | |
2801 | */ | |
64c12921 JM |
2802 | if (!should_cow_block(trans, root, b)) { |
2803 | trans->dirty = true; | |
65b51a00 | 2804 | goto cow_done; |
64c12921 | 2805 | } |
5d4f98a2 | 2806 | |
bd681513 CM |
2807 | /* |
2808 | * must have write locks on this node and the | |
2809 | * parent | |
2810 | */ | |
5124e00e JB |
2811 | if (level > write_lock_level || |
2812 | (level + 1 > write_lock_level && | |
2813 | level + 1 < BTRFS_MAX_LEVEL && | |
2814 | p->nodes[level + 1])) { | |
bd681513 CM |
2815 | write_lock_level = level + 1; |
2816 | btrfs_release_path(p); | |
2817 | goto again; | |
2818 | } | |
2819 | ||
160f4089 | 2820 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2821 | if (last_level) |
2822 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2823 | &b); | |
2824 | else | |
2825 | err = btrfs_cow_block(trans, root, b, | |
2826 | p->nodes[level + 1], | |
2827 | p->slots[level + 1], &b); | |
33c66f43 | 2828 | if (err) { |
33c66f43 | 2829 | ret = err; |
65b51a00 | 2830 | goto done; |
54aa1f4d | 2831 | } |
02217ed2 | 2832 | } |
65b51a00 | 2833 | cow_done: |
eb60ceac | 2834 | p->nodes[level] = b; |
52398340 LB |
2835 | /* |
2836 | * Leave path with blocking locks to avoid massive | |
2837 | * lock context switch, this is made on purpose. | |
2838 | */ | |
b4ce94de CM |
2839 | |
2840 | /* | |
2841 | * we have a lock on b and as long as we aren't changing | |
2842 | * the tree, there is no way to for the items in b to change. | |
2843 | * It is safe to drop the lock on our parent before we | |
2844 | * go through the expensive btree search on b. | |
2845 | * | |
eb653de1 FDBM |
2846 | * If we're inserting or deleting (ins_len != 0), then we might |
2847 | * be changing slot zero, which may require changing the parent. | |
2848 | * So, we can't drop the lock until after we know which slot | |
2849 | * we're operating on. | |
b4ce94de | 2850 | */ |
eb653de1 FDBM |
2851 | if (!ins_len && !p->keep_locks) { |
2852 | int u = level + 1; | |
2853 | ||
2854 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2855 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2856 | p->locks[u] = 0; | |
2857 | } | |
2858 | } | |
b4ce94de | 2859 | |
d7396f07 | 2860 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2861 | if (ret < 0) |
2862 | goto done; | |
b4ce94de | 2863 | |
5f39d397 | 2864 | if (level != 0) { |
33c66f43 YZ |
2865 | int dec = 0; |
2866 | if (ret && slot > 0) { | |
2867 | dec = 1; | |
be0e5c09 | 2868 | slot -= 1; |
33c66f43 | 2869 | } |
be0e5c09 | 2870 | p->slots[level] = slot; |
33c66f43 | 2871 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2872 | ins_len, &write_lock_level); |
33c66f43 | 2873 | if (err == -EAGAIN) |
c8c42864 | 2874 | goto again; |
33c66f43 YZ |
2875 | if (err) { |
2876 | ret = err; | |
c8c42864 | 2877 | goto done; |
33c66f43 | 2878 | } |
c8c42864 CM |
2879 | b = p->nodes[level]; |
2880 | slot = p->slots[level]; | |
b4ce94de | 2881 | |
bd681513 CM |
2882 | /* |
2883 | * slot 0 is special, if we change the key | |
2884 | * we have to update the parent pointer | |
2885 | * which means we must have a write lock | |
2886 | * on the parent | |
2887 | */ | |
eb653de1 | 2888 | if (slot == 0 && ins_len && |
bd681513 CM |
2889 | write_lock_level < level + 1) { |
2890 | write_lock_level = level + 1; | |
2891 | btrfs_release_path(p); | |
2892 | goto again; | |
2893 | } | |
2894 | ||
f7c79f30 CM |
2895 | unlock_up(p, level, lowest_unlock, |
2896 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2897 | |
925baedd | 2898 | if (level == lowest_level) { |
33c66f43 YZ |
2899 | if (dec) |
2900 | p->slots[level]++; | |
5b21f2ed | 2901 | goto done; |
925baedd | 2902 | } |
ca7a79ad | 2903 | |
d07b8528 | 2904 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2905 | slot, key); |
33c66f43 | 2906 | if (err == -EAGAIN) |
c8c42864 | 2907 | goto again; |
33c66f43 YZ |
2908 | if (err) { |
2909 | ret = err; | |
76a05b35 | 2910 | goto done; |
33c66f43 | 2911 | } |
76a05b35 | 2912 | |
b4ce94de | 2913 | if (!p->skip_locking) { |
bd681513 CM |
2914 | level = btrfs_header_level(b); |
2915 | if (level <= write_lock_level) { | |
2916 | err = btrfs_try_tree_write_lock(b); | |
2917 | if (!err) { | |
2918 | btrfs_set_path_blocking(p); | |
2919 | btrfs_tree_lock(b); | |
bd681513 CM |
2920 | } |
2921 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2922 | } else { | |
f82c458a | 2923 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2924 | if (!err) { |
2925 | btrfs_set_path_blocking(p); | |
2926 | btrfs_tree_read_lock(b); | |
bd681513 CM |
2927 | } |
2928 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2929 | } |
bd681513 | 2930 | p->nodes[level] = b; |
b4ce94de | 2931 | } |
be0e5c09 CM |
2932 | } else { |
2933 | p->slots[level] = slot; | |
87b29b20 | 2934 | if (ins_len > 0 && |
e902baac | 2935 | btrfs_leaf_free_space(b) < ins_len) { |
bd681513 CM |
2936 | if (write_lock_level < 1) { |
2937 | write_lock_level = 1; | |
2938 | btrfs_release_path(p); | |
2939 | goto again; | |
2940 | } | |
2941 | ||
b4ce94de | 2942 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2943 | err = split_leaf(trans, root, key, |
2944 | p, ins_len, ret == 0); | |
b4ce94de | 2945 | |
33c66f43 YZ |
2946 | BUG_ON(err > 0); |
2947 | if (err) { | |
2948 | ret = err; | |
65b51a00 CM |
2949 | goto done; |
2950 | } | |
5c680ed6 | 2951 | } |
459931ec | 2952 | if (!p->search_for_split) |
f7c79f30 | 2953 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2954 | min_write_lock_level, NULL); |
65b51a00 | 2955 | goto done; |
be0e5c09 CM |
2956 | } |
2957 | } | |
65b51a00 CM |
2958 | ret = 1; |
2959 | done: | |
b4ce94de CM |
2960 | /* |
2961 | * we don't really know what they plan on doing with the path | |
2962 | * from here on, so for now just mark it as blocking | |
2963 | */ | |
b9473439 CM |
2964 | if (!p->leave_spinning) |
2965 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2966 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2967 | btrfs_release_path(p); |
65b51a00 | 2968 | return ret; |
be0e5c09 CM |
2969 | } |
2970 | ||
5d9e75c4 JS |
2971 | /* |
2972 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2973 | * current state of the tree together with the operations recorded in the tree | |
2974 | * modification log to search for the key in a previous version of this tree, as | |
2975 | * denoted by the time_seq parameter. | |
2976 | * | |
2977 | * Naturally, there is no support for insert, delete or cow operations. | |
2978 | * | |
2979 | * The resulting path and return value will be set up as if we called | |
2980 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2981 | */ | |
310712b2 | 2982 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2983 | struct btrfs_path *p, u64 time_seq) |
2984 | { | |
0b246afa | 2985 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2986 | struct extent_buffer *b; |
2987 | int slot; | |
2988 | int ret; | |
2989 | int err; | |
2990 | int level; | |
2991 | int lowest_unlock = 1; | |
2992 | u8 lowest_level = 0; | |
d4b4087c | 2993 | int prev_cmp = -1; |
5d9e75c4 JS |
2994 | |
2995 | lowest_level = p->lowest_level; | |
2996 | WARN_ON(p->nodes[0] != NULL); | |
2997 | ||
2998 | if (p->search_commit_root) { | |
2999 | BUG_ON(time_seq); | |
3000 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3001 | } | |
3002 | ||
3003 | again: | |
5d9e75c4 | 3004 | b = get_old_root(root, time_seq); |
315bed43 NB |
3005 | if (!b) { |
3006 | ret = -EIO; | |
3007 | goto done; | |
3008 | } | |
5d9e75c4 | 3009 | level = btrfs_header_level(b); |
5d9e75c4 JS |
3010 | p->locks[level] = BTRFS_READ_LOCK; |
3011 | ||
3012 | while (b) { | |
3013 | level = btrfs_header_level(b); | |
3014 | p->nodes[level] = b; | |
5d9e75c4 JS |
3015 | |
3016 | /* | |
3017 | * we have a lock on b and as long as we aren't changing | |
3018 | * the tree, there is no way to for the items in b to change. | |
3019 | * It is safe to drop the lock on our parent before we | |
3020 | * go through the expensive btree search on b. | |
3021 | */ | |
3022 | btrfs_unlock_up_safe(p, level + 1); | |
3023 | ||
d4b4087c | 3024 | /* |
01327610 | 3025 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
3026 | * time. |
3027 | */ | |
3028 | prev_cmp = -1; | |
d7396f07 | 3029 | ret = key_search(b, key, level, &prev_cmp, &slot); |
cbca7d59 FM |
3030 | if (ret < 0) |
3031 | goto done; | |
5d9e75c4 JS |
3032 | |
3033 | if (level != 0) { | |
3034 | int dec = 0; | |
3035 | if (ret && slot > 0) { | |
3036 | dec = 1; | |
3037 | slot -= 1; | |
3038 | } | |
3039 | p->slots[level] = slot; | |
3040 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3041 | ||
3042 | if (level == lowest_level) { | |
3043 | if (dec) | |
3044 | p->slots[level]++; | |
3045 | goto done; | |
3046 | } | |
3047 | ||
d07b8528 | 3048 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 3049 | slot, key); |
5d9e75c4 JS |
3050 | if (err == -EAGAIN) |
3051 | goto again; | |
3052 | if (err) { | |
3053 | ret = err; | |
3054 | goto done; | |
3055 | } | |
3056 | ||
3057 | level = btrfs_header_level(b); | |
f82c458a | 3058 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3059 | if (!err) { |
3060 | btrfs_set_path_blocking(p); | |
3061 | btrfs_tree_read_lock(b); | |
5d9e75c4 | 3062 | } |
0b246afa | 3063 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3064 | if (!b) { |
3065 | ret = -ENOMEM; | |
3066 | goto done; | |
3067 | } | |
5d9e75c4 JS |
3068 | p->locks[level] = BTRFS_READ_LOCK; |
3069 | p->nodes[level] = b; | |
5d9e75c4 JS |
3070 | } else { |
3071 | p->slots[level] = slot; | |
3072 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3073 | goto done; | |
3074 | } | |
3075 | } | |
3076 | ret = 1; | |
3077 | done: | |
3078 | if (!p->leave_spinning) | |
3079 | btrfs_set_path_blocking(p); | |
3080 | if (ret < 0) | |
3081 | btrfs_release_path(p); | |
3082 | ||
3083 | return ret; | |
3084 | } | |
3085 | ||
2f38b3e1 AJ |
3086 | /* |
3087 | * helper to use instead of search slot if no exact match is needed but | |
3088 | * instead the next or previous item should be returned. | |
3089 | * When find_higher is true, the next higher item is returned, the next lower | |
3090 | * otherwise. | |
3091 | * When return_any and find_higher are both true, and no higher item is found, | |
3092 | * return the next lower instead. | |
3093 | * When return_any is true and find_higher is false, and no lower item is found, | |
3094 | * return the next higher instead. | |
3095 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3096 | * < 0 on error | |
3097 | */ | |
3098 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3099 | const struct btrfs_key *key, |
3100 | struct btrfs_path *p, int find_higher, | |
3101 | int return_any) | |
2f38b3e1 AJ |
3102 | { |
3103 | int ret; | |
3104 | struct extent_buffer *leaf; | |
3105 | ||
3106 | again: | |
3107 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3108 | if (ret <= 0) | |
3109 | return ret; | |
3110 | /* | |
3111 | * a return value of 1 means the path is at the position where the | |
3112 | * item should be inserted. Normally this is the next bigger item, | |
3113 | * but in case the previous item is the last in a leaf, path points | |
3114 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3115 | * item. | |
3116 | */ | |
3117 | leaf = p->nodes[0]; | |
3118 | ||
3119 | if (find_higher) { | |
3120 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3121 | ret = btrfs_next_leaf(root, p); | |
3122 | if (ret <= 0) | |
3123 | return ret; | |
3124 | if (!return_any) | |
3125 | return 1; | |
3126 | /* | |
3127 | * no higher item found, return the next | |
3128 | * lower instead | |
3129 | */ | |
3130 | return_any = 0; | |
3131 | find_higher = 0; | |
3132 | btrfs_release_path(p); | |
3133 | goto again; | |
3134 | } | |
3135 | } else { | |
e6793769 AJ |
3136 | if (p->slots[0] == 0) { |
3137 | ret = btrfs_prev_leaf(root, p); | |
3138 | if (ret < 0) | |
3139 | return ret; | |
3140 | if (!ret) { | |
23c6bf6a FDBM |
3141 | leaf = p->nodes[0]; |
3142 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3143 | p->slots[0]--; | |
e6793769 | 3144 | return 0; |
2f38b3e1 | 3145 | } |
e6793769 AJ |
3146 | if (!return_any) |
3147 | return 1; | |
3148 | /* | |
3149 | * no lower item found, return the next | |
3150 | * higher instead | |
3151 | */ | |
3152 | return_any = 0; | |
3153 | find_higher = 1; | |
3154 | btrfs_release_path(p); | |
3155 | goto again; | |
3156 | } else { | |
2f38b3e1 AJ |
3157 | --p->slots[0]; |
3158 | } | |
3159 | } | |
3160 | return 0; | |
3161 | } | |
3162 | ||
74123bd7 CM |
3163 | /* |
3164 | * adjust the pointers going up the tree, starting at level | |
3165 | * making sure the right key of each node is points to 'key'. | |
3166 | * This is used after shifting pointers to the left, so it stops | |
3167 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3168 | * higher levels | |
aa5d6bed | 3169 | * |
74123bd7 | 3170 | */ |
b167fa91 | 3171 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3172 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3173 | { |
3174 | int i; | |
5f39d397 | 3175 | struct extent_buffer *t; |
0e82bcfe | 3176 | int ret; |
5f39d397 | 3177 | |
234b63a0 | 3178 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3179 | int tslot = path->slots[i]; |
0e82bcfe | 3180 | |
eb60ceac | 3181 | if (!path->nodes[i]) |
be0e5c09 | 3182 | break; |
5f39d397 | 3183 | t = path->nodes[i]; |
0e82bcfe DS |
3184 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3185 | GFP_ATOMIC); | |
3186 | BUG_ON(ret < 0); | |
5f39d397 | 3187 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3188 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3189 | if (tslot != 0) |
3190 | break; | |
3191 | } | |
3192 | } | |
3193 | ||
31840ae1 ZY |
3194 | /* |
3195 | * update item key. | |
3196 | * | |
3197 | * This function isn't completely safe. It's the caller's responsibility | |
3198 | * that the new key won't break the order | |
3199 | */ | |
b7a0365e DD |
3200 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3201 | struct btrfs_path *path, | |
310712b2 | 3202 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3203 | { |
3204 | struct btrfs_disk_key disk_key; | |
3205 | struct extent_buffer *eb; | |
3206 | int slot; | |
3207 | ||
3208 | eb = path->nodes[0]; | |
3209 | slot = path->slots[0]; | |
3210 | if (slot > 0) { | |
3211 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
3212 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
3213 | btrfs_crit(fs_info, | |
3214 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3215 | slot, btrfs_disk_key_objectid(&disk_key), | |
3216 | btrfs_disk_key_type(&disk_key), | |
3217 | btrfs_disk_key_offset(&disk_key), | |
3218 | new_key->objectid, new_key->type, | |
3219 | new_key->offset); | |
3220 | btrfs_print_leaf(eb); | |
3221 | BUG(); | |
3222 | } | |
31840ae1 ZY |
3223 | } |
3224 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3225 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
3226 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
3227 | btrfs_crit(fs_info, | |
3228 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3229 | slot, btrfs_disk_key_objectid(&disk_key), | |
3230 | btrfs_disk_key_type(&disk_key), | |
3231 | btrfs_disk_key_offset(&disk_key), | |
3232 | new_key->objectid, new_key->type, | |
3233 | new_key->offset); | |
3234 | btrfs_print_leaf(eb); | |
3235 | BUG(); | |
3236 | } | |
31840ae1 ZY |
3237 | } |
3238 | ||
3239 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3240 | btrfs_set_item_key(eb, &disk_key, slot); | |
3241 | btrfs_mark_buffer_dirty(eb); | |
3242 | if (slot == 0) | |
b167fa91 | 3243 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3244 | } |
3245 | ||
74123bd7 CM |
3246 | /* |
3247 | * try to push data from one node into the next node left in the | |
79f95c82 | 3248 | * tree. |
aa5d6bed CM |
3249 | * |
3250 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3251 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3252 | */ |
98ed5174 | 3253 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 3254 | struct extent_buffer *dst, |
971a1f66 | 3255 | struct extent_buffer *src, int empty) |
be0e5c09 | 3256 | { |
d30a668f | 3257 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 3258 | int push_items = 0; |
bb803951 CM |
3259 | int src_nritems; |
3260 | int dst_nritems; | |
aa5d6bed | 3261 | int ret = 0; |
be0e5c09 | 3262 | |
5f39d397 CM |
3263 | src_nritems = btrfs_header_nritems(src); |
3264 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3265 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3266 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3267 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3268 | |
bce4eae9 | 3269 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3270 | return 1; |
3271 | ||
d397712b | 3272 | if (push_items <= 0) |
be0e5c09 CM |
3273 | return 1; |
3274 | ||
bce4eae9 | 3275 | if (empty) { |
971a1f66 | 3276 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3277 | if (push_items < src_nritems) { |
3278 | /* leave at least 8 pointers in the node if | |
3279 | * we aren't going to empty it | |
3280 | */ | |
3281 | if (src_nritems - push_items < 8) { | |
3282 | if (push_items <= 8) | |
3283 | return 1; | |
3284 | push_items -= 8; | |
3285 | } | |
3286 | } | |
3287 | } else | |
3288 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3289 | |
ed874f0d | 3290 | ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 3291 | if (ret) { |
66642832 | 3292 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3293 | return ret; |
3294 | } | |
5f39d397 CM |
3295 | copy_extent_buffer(dst, src, |
3296 | btrfs_node_key_ptr_offset(dst_nritems), | |
3297 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3298 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3299 | |
bb803951 | 3300 | if (push_items < src_nritems) { |
57911b8b | 3301 | /* |
bf1d3425 DS |
3302 | * Don't call tree_mod_log_insert_move here, key removal was |
3303 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3304 | */ |
5f39d397 CM |
3305 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3306 | btrfs_node_key_ptr_offset(push_items), | |
3307 | (src_nritems - push_items) * | |
3308 | sizeof(struct btrfs_key_ptr)); | |
3309 | } | |
3310 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3311 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3312 | btrfs_mark_buffer_dirty(src); | |
3313 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3314 | |
79f95c82 CM |
3315 | return ret; |
3316 | } | |
3317 | ||
3318 | /* | |
3319 | * try to push data from one node into the next node right in the | |
3320 | * tree. | |
3321 | * | |
3322 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3323 | * error, and > 0 if there was no room in the right hand block. | |
3324 | * | |
3325 | * this will only push up to 1/2 the contents of the left node over | |
3326 | */ | |
5f39d397 | 3327 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3328 | struct extent_buffer *dst, |
3329 | struct extent_buffer *src) | |
79f95c82 | 3330 | { |
55d32ed8 | 3331 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
3332 | int push_items = 0; |
3333 | int max_push; | |
3334 | int src_nritems; | |
3335 | int dst_nritems; | |
3336 | int ret = 0; | |
79f95c82 | 3337 | |
7bb86316 CM |
3338 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3339 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3340 | ||
5f39d397 CM |
3341 | src_nritems = btrfs_header_nritems(src); |
3342 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3343 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3344 | if (push_items <= 0) |
79f95c82 | 3345 | return 1; |
bce4eae9 | 3346 | |
d397712b | 3347 | if (src_nritems < 4) |
bce4eae9 | 3348 | return 1; |
79f95c82 CM |
3349 | |
3350 | max_push = src_nritems / 2 + 1; | |
3351 | /* don't try to empty the node */ | |
d397712b | 3352 | if (max_push >= src_nritems) |
79f95c82 | 3353 | return 1; |
252c38f0 | 3354 | |
79f95c82 CM |
3355 | if (max_push < push_items) |
3356 | push_items = max_push; | |
3357 | ||
bf1d3425 DS |
3358 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3359 | BUG_ON(ret < 0); | |
5f39d397 CM |
3360 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3361 | btrfs_node_key_ptr_offset(0), | |
3362 | (dst_nritems) * | |
3363 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3364 | |
ed874f0d DS |
3365 | ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
3366 | push_items); | |
5de865ee | 3367 | if (ret) { |
66642832 | 3368 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3369 | return ret; |
3370 | } | |
5f39d397 CM |
3371 | copy_extent_buffer(dst, src, |
3372 | btrfs_node_key_ptr_offset(0), | |
3373 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3374 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3375 | |
5f39d397 CM |
3376 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3377 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3378 | |
5f39d397 CM |
3379 | btrfs_mark_buffer_dirty(src); |
3380 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3381 | |
aa5d6bed | 3382 | return ret; |
be0e5c09 CM |
3383 | } |
3384 | ||
97571fd0 CM |
3385 | /* |
3386 | * helper function to insert a new root level in the tree. | |
3387 | * A new node is allocated, and a single item is inserted to | |
3388 | * point to the existing root | |
aa5d6bed CM |
3389 | * |
3390 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3391 | */ |
d397712b | 3392 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3393 | struct btrfs_root *root, |
fdd99c72 | 3394 | struct btrfs_path *path, int level) |
5c680ed6 | 3395 | { |
0b246afa | 3396 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3397 | u64 lower_gen; |
5f39d397 CM |
3398 | struct extent_buffer *lower; |
3399 | struct extent_buffer *c; | |
925baedd | 3400 | struct extent_buffer *old; |
5f39d397 | 3401 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3402 | int ret; |
5c680ed6 CM |
3403 | |
3404 | BUG_ON(path->nodes[level]); | |
3405 | BUG_ON(path->nodes[level-1] != root->node); | |
3406 | ||
7bb86316 CM |
3407 | lower = path->nodes[level-1]; |
3408 | if (level == 1) | |
3409 | btrfs_item_key(lower, &lower_key, 0); | |
3410 | else | |
3411 | btrfs_node_key(lower, &lower_key, 0); | |
3412 | ||
a6279470 FM |
3413 | c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, |
3414 | root->node->start, 0); | |
5f39d397 CM |
3415 | if (IS_ERR(c)) |
3416 | return PTR_ERR(c); | |
925baedd | 3417 | |
0b246afa | 3418 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3419 | |
5f39d397 | 3420 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3421 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3422 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3423 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3424 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3425 | |
3426 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3427 | |
5f39d397 | 3428 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3429 | |
925baedd | 3430 | old = root->node; |
d9d19a01 DS |
3431 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3432 | BUG_ON(ret < 0); | |
240f62c8 | 3433 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3434 | |
3435 | /* the super has an extra ref to root->node */ | |
3436 | free_extent_buffer(old); | |
3437 | ||
0b86a832 | 3438 | add_root_to_dirty_list(root); |
5f39d397 CM |
3439 | extent_buffer_get(c); |
3440 | path->nodes[level] = c; | |
95449a16 | 3441 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3442 | path->slots[level] = 0; |
3443 | return 0; | |
3444 | } | |
3445 | ||
74123bd7 CM |
3446 | /* |
3447 | * worker function to insert a single pointer in a node. | |
3448 | * the node should have enough room for the pointer already | |
97571fd0 | 3449 | * |
74123bd7 CM |
3450 | * slot and level indicate where you want the key to go, and |
3451 | * blocknr is the block the key points to. | |
3452 | */ | |
143bede5 | 3453 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 3454 | struct btrfs_path *path, |
143bede5 | 3455 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3456 | int slot, int level) |
74123bd7 | 3457 | { |
5f39d397 | 3458 | struct extent_buffer *lower; |
74123bd7 | 3459 | int nritems; |
f3ea38da | 3460 | int ret; |
5c680ed6 CM |
3461 | |
3462 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3463 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3464 | lower = path->nodes[level]; |
3465 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3466 | BUG_ON(slot > nritems); |
6ad3cf6d | 3467 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 3468 | if (slot != nritems) { |
bf1d3425 DS |
3469 | if (level) { |
3470 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3471 | nritems - slot); |
bf1d3425 DS |
3472 | BUG_ON(ret < 0); |
3473 | } | |
5f39d397 CM |
3474 | memmove_extent_buffer(lower, |
3475 | btrfs_node_key_ptr_offset(slot + 1), | |
3476 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3477 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3478 | } |
c3e06965 | 3479 | if (level) { |
e09c2efe DS |
3480 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3481 | GFP_NOFS); | |
f3ea38da JS |
3482 | BUG_ON(ret < 0); |
3483 | } | |
5f39d397 | 3484 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3485 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3486 | WARN_ON(trans->transid == 0); |
3487 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3488 | btrfs_set_header_nritems(lower, nritems + 1); |
3489 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3490 | } |
3491 | ||
97571fd0 CM |
3492 | /* |
3493 | * split the node at the specified level in path in two. | |
3494 | * The path is corrected to point to the appropriate node after the split | |
3495 | * | |
3496 | * Before splitting this tries to make some room in the node by pushing | |
3497 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3498 | * |
3499 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3500 | */ |
e02119d5 CM |
3501 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3502 | struct btrfs_root *root, | |
3503 | struct btrfs_path *path, int level) | |
be0e5c09 | 3504 | { |
0b246afa | 3505 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3506 | struct extent_buffer *c; |
3507 | struct extent_buffer *split; | |
3508 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3509 | int mid; |
5c680ed6 | 3510 | int ret; |
7518a238 | 3511 | u32 c_nritems; |
eb60ceac | 3512 | |
5f39d397 | 3513 | c = path->nodes[level]; |
7bb86316 | 3514 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3515 | if (c == root->node) { |
d9abbf1c | 3516 | /* |
90f8d62e JS |
3517 | * trying to split the root, lets make a new one |
3518 | * | |
fdd99c72 | 3519 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3520 | * insert_new_root, because that root buffer will be kept as a |
3521 | * normal node. We are going to log removal of half of the | |
3522 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3523 | * tree lock on the buffer, which is why we cannot race with | |
3524 | * other tree_mod_log users. | |
d9abbf1c | 3525 | */ |
fdd99c72 | 3526 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3527 | if (ret) |
3528 | return ret; | |
b3612421 | 3529 | } else { |
e66f709b | 3530 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3531 | c = path->nodes[level]; |
3532 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3533 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3534 | return 0; |
54aa1f4d CM |
3535 | if (ret < 0) |
3536 | return ret; | |
be0e5c09 | 3537 | } |
e66f709b | 3538 | |
5f39d397 | 3539 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3540 | mid = (c_nritems + 1) / 2; |
3541 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3542 | |
a6279470 FM |
3543 | split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, |
3544 | c->start, 0); | |
5f39d397 CM |
3545 | if (IS_ERR(split)) |
3546 | return PTR_ERR(split); | |
3547 | ||
0b246afa | 3548 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3549 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3550 | |
ed874f0d | 3551 | ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 3552 | if (ret) { |
66642832 | 3553 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3554 | return ret; |
3555 | } | |
5f39d397 CM |
3556 | copy_extent_buffer(split, c, |
3557 | btrfs_node_key_ptr_offset(0), | |
3558 | btrfs_node_key_ptr_offset(mid), | |
3559 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3560 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3561 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3562 | ret = 0; |
3563 | ||
5f39d397 CM |
3564 | btrfs_mark_buffer_dirty(c); |
3565 | btrfs_mark_buffer_dirty(split); | |
3566 | ||
6ad3cf6d | 3567 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 3568 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3569 | |
5de08d7d | 3570 | if (path->slots[level] >= mid) { |
5c680ed6 | 3571 | path->slots[level] -= mid; |
925baedd | 3572 | btrfs_tree_unlock(c); |
5f39d397 CM |
3573 | free_extent_buffer(c); |
3574 | path->nodes[level] = split; | |
5c680ed6 CM |
3575 | path->slots[level + 1] += 1; |
3576 | } else { | |
925baedd | 3577 | btrfs_tree_unlock(split); |
5f39d397 | 3578 | free_extent_buffer(split); |
be0e5c09 | 3579 | } |
aa5d6bed | 3580 | return ret; |
be0e5c09 CM |
3581 | } |
3582 | ||
74123bd7 CM |
3583 | /* |
3584 | * how many bytes are required to store the items in a leaf. start | |
3585 | * and nr indicate which items in the leaf to check. This totals up the | |
3586 | * space used both by the item structs and the item data | |
3587 | */ | |
5f39d397 | 3588 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3589 | { |
41be1f3b JB |
3590 | struct btrfs_item *start_item; |
3591 | struct btrfs_item *end_item; | |
3592 | struct btrfs_map_token token; | |
be0e5c09 | 3593 | int data_len; |
5f39d397 | 3594 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3595 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3596 | |
3597 | if (!nr) | |
3598 | return 0; | |
c82f823c | 3599 | btrfs_init_map_token(&token, l); |
dd3cc16b RK |
3600 | start_item = btrfs_item_nr(start); |
3601 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3602 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3603 | btrfs_token_item_size(l, start_item, &token); | |
3604 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3605 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3606 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3607 | return data_len; |
3608 | } | |
3609 | ||
d4dbff95 CM |
3610 | /* |
3611 | * The space between the end of the leaf items and | |
3612 | * the start of the leaf data. IOW, how much room | |
3613 | * the leaf has left for both items and data | |
3614 | */ | |
e902baac | 3615 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 3616 | { |
e902baac | 3617 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
3618 | int nritems = btrfs_header_nritems(leaf); |
3619 | int ret; | |
0b246afa JM |
3620 | |
3621 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3622 | if (ret < 0) { |
0b246afa JM |
3623 | btrfs_crit(fs_info, |
3624 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3625 | ret, | |
3626 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3627 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3628 | } |
3629 | return ret; | |
d4dbff95 CM |
3630 | } |
3631 | ||
99d8f83c CM |
3632 | /* |
3633 | * min slot controls the lowest index we're willing to push to the | |
3634 | * right. We'll push up to and including min_slot, but no lower | |
3635 | */ | |
f72f0010 | 3636 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
3637 | int data_size, int empty, |
3638 | struct extent_buffer *right, | |
99d8f83c CM |
3639 | int free_space, u32 left_nritems, |
3640 | u32 min_slot) | |
00ec4c51 | 3641 | { |
f72f0010 | 3642 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 3643 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3644 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3645 | struct btrfs_map_token token; |
5f39d397 | 3646 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3647 | int slot; |
34a38218 | 3648 | u32 i; |
00ec4c51 CM |
3649 | int push_space = 0; |
3650 | int push_items = 0; | |
0783fcfc | 3651 | struct btrfs_item *item; |
34a38218 | 3652 | u32 nr; |
7518a238 | 3653 | u32 right_nritems; |
5f39d397 | 3654 | u32 data_end; |
db94535d | 3655 | u32 this_item_size; |
00ec4c51 | 3656 | |
34a38218 CM |
3657 | if (empty) |
3658 | nr = 0; | |
3659 | else | |
99d8f83c | 3660 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3661 | |
31840ae1 | 3662 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3663 | push_space += data_size; |
31840ae1 | 3664 | |
44871b1b | 3665 | slot = path->slots[1]; |
34a38218 CM |
3666 | i = left_nritems - 1; |
3667 | while (i >= nr) { | |
dd3cc16b | 3668 | item = btrfs_item_nr(i); |
db94535d | 3669 | |
31840ae1 ZY |
3670 | if (!empty && push_items > 0) { |
3671 | if (path->slots[0] > i) | |
3672 | break; | |
3673 | if (path->slots[0] == i) { | |
e902baac DS |
3674 | int space = btrfs_leaf_free_space(left); |
3675 | ||
31840ae1 ZY |
3676 | if (space + push_space * 2 > free_space) |
3677 | break; | |
3678 | } | |
3679 | } | |
3680 | ||
00ec4c51 | 3681 | if (path->slots[0] == i) |
87b29b20 | 3682 | push_space += data_size; |
db94535d | 3683 | |
db94535d CM |
3684 | this_item_size = btrfs_item_size(left, item); |
3685 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3686 | break; |
31840ae1 | 3687 | |
00ec4c51 | 3688 | push_items++; |
db94535d | 3689 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3690 | if (i == 0) |
3691 | break; | |
3692 | i--; | |
db94535d | 3693 | } |
5f39d397 | 3694 | |
925baedd CM |
3695 | if (push_items == 0) |
3696 | goto out_unlock; | |
5f39d397 | 3697 | |
6c1500f2 | 3698 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3699 | |
00ec4c51 | 3700 | /* push left to right */ |
5f39d397 | 3701 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3702 | |
5f39d397 | 3703 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
8f881e8c | 3704 | push_space -= leaf_data_end(left); |
5f39d397 | 3705 | |
00ec4c51 | 3706 | /* make room in the right data area */ |
8f881e8c | 3707 | data_end = leaf_data_end(right); |
5f39d397 | 3708 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3709 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3710 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3711 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3712 | |
00ec4c51 | 3713 | /* copy from the left data area */ |
3d9ec8c4 | 3714 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3715 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 3716 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 3717 | push_space); |
5f39d397 CM |
3718 | |
3719 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3720 | btrfs_item_nr_offset(0), | |
3721 | right_nritems * sizeof(struct btrfs_item)); | |
3722 | ||
00ec4c51 | 3723 | /* copy the items from left to right */ |
5f39d397 CM |
3724 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3725 | btrfs_item_nr_offset(left_nritems - push_items), | |
3726 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3727 | |
3728 | /* update the item pointers */ | |
c82f823c | 3729 | btrfs_init_map_token(&token, right); |
7518a238 | 3730 | right_nritems += push_items; |
5f39d397 | 3731 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3732 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3733 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3734 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3735 | push_space -= btrfs_token_item_size(right, item, &token); |
3736 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3737 | } |
3738 | ||
7518a238 | 3739 | left_nritems -= push_items; |
5f39d397 | 3740 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3741 | |
34a38218 CM |
3742 | if (left_nritems) |
3743 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3744 | else |
6a884d7d | 3745 | btrfs_clean_tree_block(left); |
f0486c68 | 3746 | |
5f39d397 | 3747 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3748 | |
5f39d397 CM |
3749 | btrfs_item_key(right, &disk_key, 0); |
3750 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3751 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3752 | |
00ec4c51 | 3753 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3754 | if (path->slots[0] >= left_nritems) { |
3755 | path->slots[0] -= left_nritems; | |
925baedd | 3756 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 3757 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 3758 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3759 | free_extent_buffer(path->nodes[0]); |
3760 | path->nodes[0] = right; | |
00ec4c51 CM |
3761 | path->slots[1] += 1; |
3762 | } else { | |
925baedd | 3763 | btrfs_tree_unlock(right); |
5f39d397 | 3764 | free_extent_buffer(right); |
00ec4c51 CM |
3765 | } |
3766 | return 0; | |
925baedd CM |
3767 | |
3768 | out_unlock: | |
3769 | btrfs_tree_unlock(right); | |
3770 | free_extent_buffer(right); | |
3771 | return 1; | |
00ec4c51 | 3772 | } |
925baedd | 3773 | |
44871b1b CM |
3774 | /* |
3775 | * push some data in the path leaf to the right, trying to free up at | |
3776 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3777 | * | |
3778 | * returns 1 if the push failed because the other node didn't have enough | |
3779 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3780 | * |
3781 | * this will push starting from min_slot to the end of the leaf. It won't | |
3782 | * push any slot lower than min_slot | |
44871b1b CM |
3783 | */ |
3784 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3785 | *root, struct btrfs_path *path, |
3786 | int min_data_size, int data_size, | |
3787 | int empty, u32 min_slot) | |
44871b1b CM |
3788 | { |
3789 | struct extent_buffer *left = path->nodes[0]; | |
3790 | struct extent_buffer *right; | |
3791 | struct extent_buffer *upper; | |
3792 | int slot; | |
3793 | int free_space; | |
3794 | u32 left_nritems; | |
3795 | int ret; | |
3796 | ||
3797 | if (!path->nodes[1]) | |
3798 | return 1; | |
3799 | ||
3800 | slot = path->slots[1]; | |
3801 | upper = path->nodes[1]; | |
3802 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3803 | return 1; | |
3804 | ||
3805 | btrfs_assert_tree_locked(path->nodes[1]); | |
3806 | ||
4b231ae4 | 3807 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3808 | /* |
3809 | * slot + 1 is not valid or we fail to read the right node, | |
3810 | * no big deal, just return. | |
3811 | */ | |
3812 | if (IS_ERR(right)) | |
91ca338d TI |
3813 | return 1; |
3814 | ||
44871b1b | 3815 | btrfs_tree_lock(right); |
8bead258 | 3816 | btrfs_set_lock_blocking_write(right); |
44871b1b | 3817 | |
e902baac | 3818 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3819 | if (free_space < data_size) |
3820 | goto out_unlock; | |
3821 | ||
3822 | /* cow and double check */ | |
3823 | ret = btrfs_cow_block(trans, root, right, upper, | |
3824 | slot + 1, &right); | |
3825 | if (ret) | |
3826 | goto out_unlock; | |
3827 | ||
e902baac | 3828 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3829 | if (free_space < data_size) |
3830 | goto out_unlock; | |
3831 | ||
3832 | left_nritems = btrfs_header_nritems(left); | |
3833 | if (left_nritems == 0) | |
3834 | goto out_unlock; | |
3835 | ||
2ef1fed2 FDBM |
3836 | if (path->slots[0] == left_nritems && !empty) { |
3837 | /* Key greater than all keys in the leaf, right neighbor has | |
3838 | * enough room for it and we're not emptying our leaf to delete | |
3839 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3840 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3841 | btrfs_tree_unlock(left); |
3842 | free_extent_buffer(left); | |
3843 | path->nodes[0] = right; | |
3844 | path->slots[0] = 0; | |
3845 | path->slots[1]++; | |
3846 | return 0; | |
3847 | } | |
3848 | ||
f72f0010 | 3849 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3850 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3851 | out_unlock: |
3852 | btrfs_tree_unlock(right); | |
3853 | free_extent_buffer(right); | |
3854 | return 1; | |
3855 | } | |
3856 | ||
74123bd7 CM |
3857 | /* |
3858 | * push some data in the path leaf to the left, trying to free up at | |
3859 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3860 | * |
3861 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3862 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3863 | * items | |
74123bd7 | 3864 | */ |
8087c193 | 3865 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3866 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3867 | int free_space, u32 right_nritems, |
3868 | u32 max_slot) | |
be0e5c09 | 3869 | { |
8087c193 | 3870 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3871 | struct btrfs_disk_key disk_key; |
3872 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3873 | int i; |
be0e5c09 CM |
3874 | int push_space = 0; |
3875 | int push_items = 0; | |
0783fcfc | 3876 | struct btrfs_item *item; |
7518a238 | 3877 | u32 old_left_nritems; |
34a38218 | 3878 | u32 nr; |
aa5d6bed | 3879 | int ret = 0; |
db94535d CM |
3880 | u32 this_item_size; |
3881 | u32 old_left_item_size; | |
cfed81a0 CM |
3882 | struct btrfs_map_token token; |
3883 | ||
34a38218 | 3884 | if (empty) |
99d8f83c | 3885 | nr = min(right_nritems, max_slot); |
34a38218 | 3886 | else |
99d8f83c | 3887 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3888 | |
3889 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3890 | item = btrfs_item_nr(i); |
db94535d | 3891 | |
31840ae1 ZY |
3892 | if (!empty && push_items > 0) { |
3893 | if (path->slots[0] < i) | |
3894 | break; | |
3895 | if (path->slots[0] == i) { | |
e902baac DS |
3896 | int space = btrfs_leaf_free_space(right); |
3897 | ||
31840ae1 ZY |
3898 | if (space + push_space * 2 > free_space) |
3899 | break; | |
3900 | } | |
3901 | } | |
3902 | ||
be0e5c09 | 3903 | if (path->slots[0] == i) |
87b29b20 | 3904 | push_space += data_size; |
db94535d CM |
3905 | |
3906 | this_item_size = btrfs_item_size(right, item); | |
3907 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3908 | break; |
db94535d | 3909 | |
be0e5c09 | 3910 | push_items++; |
db94535d CM |
3911 | push_space += this_item_size + sizeof(*item); |
3912 | } | |
3913 | ||
be0e5c09 | 3914 | if (push_items == 0) { |
925baedd CM |
3915 | ret = 1; |
3916 | goto out; | |
be0e5c09 | 3917 | } |
fae7f21c | 3918 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3919 | |
be0e5c09 | 3920 | /* push data from right to left */ |
5f39d397 CM |
3921 | copy_extent_buffer(left, right, |
3922 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3923 | btrfs_item_nr_offset(0), | |
3924 | push_items * sizeof(struct btrfs_item)); | |
3925 | ||
0b246afa | 3926 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3927 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3928 | |
3d9ec8c4 | 3929 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3930 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3931 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3932 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3933 | push_space); |
5f39d397 | 3934 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3935 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3936 | |
c82f823c | 3937 | btrfs_init_map_token(&token, left); |
db94535d | 3938 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3939 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3940 | u32 ioff; |
db94535d | 3941 | |
dd3cc16b | 3942 | item = btrfs_item_nr(i); |
db94535d | 3943 | |
cfed81a0 CM |
3944 | ioff = btrfs_token_item_offset(left, item, &token); |
3945 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3946 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3947 | &token); |
be0e5c09 | 3948 | } |
5f39d397 | 3949 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3950 | |
3951 | /* fixup right node */ | |
31b1a2bd JL |
3952 | if (push_items > right_nritems) |
3953 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3954 | right_nritems); |
34a38218 CM |
3955 | |
3956 | if (push_items < right_nritems) { | |
3957 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
8f881e8c | 3958 | leaf_data_end(right); |
3d9ec8c4 | 3959 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3960 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3961 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3962 | leaf_data_end(right), push_space); |
34a38218 CM |
3963 | |
3964 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3965 | btrfs_item_nr_offset(push_items), |
3966 | (btrfs_header_nritems(right) - push_items) * | |
3967 | sizeof(struct btrfs_item)); | |
34a38218 | 3968 | } |
c82f823c DS |
3969 | |
3970 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3971 | right_nritems -= push_items; |
3972 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3973 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3974 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3975 | item = btrfs_item_nr(i); |
db94535d | 3976 | |
cfed81a0 CM |
3977 | push_space = push_space - btrfs_token_item_size(right, |
3978 | item, &token); | |
3979 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3980 | } |
eb60ceac | 3981 | |
5f39d397 | 3982 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3983 | if (right_nritems) |
3984 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3985 | else |
6a884d7d | 3986 | btrfs_clean_tree_block(right); |
098f59c2 | 3987 | |
5f39d397 | 3988 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3989 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3990 | |
3991 | /* then fixup the leaf pointer in the path */ | |
3992 | if (path->slots[0] < push_items) { | |
3993 | path->slots[0] += old_left_nritems; | |
925baedd | 3994 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3995 | free_extent_buffer(path->nodes[0]); |
3996 | path->nodes[0] = left; | |
be0e5c09 CM |
3997 | path->slots[1] -= 1; |
3998 | } else { | |
925baedd | 3999 | btrfs_tree_unlock(left); |
5f39d397 | 4000 | free_extent_buffer(left); |
be0e5c09 CM |
4001 | path->slots[0] -= push_items; |
4002 | } | |
eb60ceac | 4003 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 4004 | return ret; |
925baedd CM |
4005 | out: |
4006 | btrfs_tree_unlock(left); | |
4007 | free_extent_buffer(left); | |
4008 | return ret; | |
be0e5c09 CM |
4009 | } |
4010 | ||
44871b1b CM |
4011 | /* |
4012 | * push some data in the path leaf to the left, trying to free up at | |
4013 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
4014 | * |
4015 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
4016 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
4017 | * items | |
44871b1b CM |
4018 | */ |
4019 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
4020 | *root, struct btrfs_path *path, int min_data_size, |
4021 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
4022 | { |
4023 | struct extent_buffer *right = path->nodes[0]; | |
4024 | struct extent_buffer *left; | |
4025 | int slot; | |
4026 | int free_space; | |
4027 | u32 right_nritems; | |
4028 | int ret = 0; | |
4029 | ||
4030 | slot = path->slots[1]; | |
4031 | if (slot == 0) | |
4032 | return 1; | |
4033 | if (!path->nodes[1]) | |
4034 | return 1; | |
4035 | ||
4036 | right_nritems = btrfs_header_nritems(right); | |
4037 | if (right_nritems == 0) | |
4038 | return 1; | |
4039 | ||
4040 | btrfs_assert_tree_locked(path->nodes[1]); | |
4041 | ||
4b231ae4 | 4042 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
4043 | /* |
4044 | * slot - 1 is not valid or we fail to read the left node, | |
4045 | * no big deal, just return. | |
4046 | */ | |
4047 | if (IS_ERR(left)) | |
91ca338d TI |
4048 | return 1; |
4049 | ||
44871b1b | 4050 | btrfs_tree_lock(left); |
8bead258 | 4051 | btrfs_set_lock_blocking_write(left); |
44871b1b | 4052 | |
e902baac | 4053 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
4054 | if (free_space < data_size) { |
4055 | ret = 1; | |
4056 | goto out; | |
4057 | } | |
4058 | ||
4059 | /* cow and double check */ | |
4060 | ret = btrfs_cow_block(trans, root, left, | |
4061 | path->nodes[1], slot - 1, &left); | |
4062 | if (ret) { | |
4063 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4064 | if (ret == -ENOSPC) |
4065 | ret = 1; | |
44871b1b CM |
4066 | goto out; |
4067 | } | |
4068 | ||
e902baac | 4069 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
4070 | if (free_space < data_size) { |
4071 | ret = 1; | |
4072 | goto out; | |
4073 | } | |
4074 | ||
8087c193 | 4075 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
4076 | empty, left, free_space, right_nritems, |
4077 | max_slot); | |
44871b1b CM |
4078 | out: |
4079 | btrfs_tree_unlock(left); | |
4080 | free_extent_buffer(left); | |
4081 | return ret; | |
4082 | } | |
4083 | ||
4084 | /* | |
4085 | * split the path's leaf in two, making sure there is at least data_size | |
4086 | * available for the resulting leaf level of the path. | |
44871b1b | 4087 | */ |
143bede5 | 4088 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
4089 | struct btrfs_path *path, |
4090 | struct extent_buffer *l, | |
4091 | struct extent_buffer *right, | |
4092 | int slot, int mid, int nritems) | |
44871b1b | 4093 | { |
94f94ad9 | 4094 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
4095 | int data_copy_size; |
4096 | int rt_data_off; | |
4097 | int i; | |
44871b1b | 4098 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4099 | struct btrfs_map_token token; |
4100 | ||
44871b1b CM |
4101 | nritems = nritems - mid; |
4102 | btrfs_set_header_nritems(right, nritems); | |
8f881e8c | 4103 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); |
44871b1b CM |
4104 | |
4105 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4106 | btrfs_item_nr_offset(mid), | |
4107 | nritems * sizeof(struct btrfs_item)); | |
4108 | ||
4109 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4110 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4111 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 4112 | leaf_data_end(l), data_copy_size); |
44871b1b | 4113 | |
0b246afa | 4114 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b | 4115 | |
c82f823c | 4116 | btrfs_init_map_token(&token, right); |
44871b1b | 4117 | for (i = 0; i < nritems; i++) { |
dd3cc16b | 4118 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4119 | u32 ioff; |
4120 | ||
cfed81a0 CM |
4121 | ioff = btrfs_token_item_offset(right, item, &token); |
4122 | btrfs_set_token_item_offset(right, item, | |
4123 | ioff + rt_data_off, &token); | |
44871b1b CM |
4124 | } |
4125 | ||
44871b1b | 4126 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4127 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 4128 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
4129 | |
4130 | btrfs_mark_buffer_dirty(right); | |
4131 | btrfs_mark_buffer_dirty(l); | |
4132 | BUG_ON(path->slots[0] != slot); | |
4133 | ||
44871b1b CM |
4134 | if (mid <= slot) { |
4135 | btrfs_tree_unlock(path->nodes[0]); | |
4136 | free_extent_buffer(path->nodes[0]); | |
4137 | path->nodes[0] = right; | |
4138 | path->slots[0] -= mid; | |
4139 | path->slots[1] += 1; | |
4140 | } else { | |
4141 | btrfs_tree_unlock(right); | |
4142 | free_extent_buffer(right); | |
4143 | } | |
4144 | ||
4145 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4146 | } |
4147 | ||
99d8f83c CM |
4148 | /* |
4149 | * double splits happen when we need to insert a big item in the middle | |
4150 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4151 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4152 | * A B C | |
4153 | * | |
4154 | * We avoid this by trying to push the items on either side of our target | |
4155 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4156 | * completely. | |
4157 | */ | |
4158 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4159 | struct btrfs_root *root, | |
4160 | struct btrfs_path *path, | |
4161 | int data_size) | |
4162 | { | |
4163 | int ret; | |
4164 | int progress = 0; | |
4165 | int slot; | |
4166 | u32 nritems; | |
5a4267ca | 4167 | int space_needed = data_size; |
99d8f83c CM |
4168 | |
4169 | slot = path->slots[0]; | |
5a4267ca | 4170 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 4171 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
4172 | |
4173 | /* | |
4174 | * try to push all the items after our slot into the | |
4175 | * right leaf | |
4176 | */ | |
5a4267ca | 4177 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4178 | if (ret < 0) |
4179 | return ret; | |
4180 | ||
4181 | if (ret == 0) | |
4182 | progress++; | |
4183 | ||
4184 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4185 | /* | |
4186 | * our goal is to get our slot at the start or end of a leaf. If | |
4187 | * we've done so we're done | |
4188 | */ | |
4189 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4190 | return 0; | |
4191 | ||
e902baac | 4192 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4193 | return 0; |
4194 | ||
4195 | /* try to push all the items before our slot into the next leaf */ | |
4196 | slot = path->slots[0]; | |
263d3995 FM |
4197 | space_needed = data_size; |
4198 | if (slot > 0) | |
e902baac | 4199 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 4200 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4201 | if (ret < 0) |
4202 | return ret; | |
4203 | ||
4204 | if (ret == 0) | |
4205 | progress++; | |
4206 | ||
4207 | if (progress) | |
4208 | return 0; | |
4209 | return 1; | |
4210 | } | |
4211 | ||
74123bd7 CM |
4212 | /* |
4213 | * split the path's leaf in two, making sure there is at least data_size | |
4214 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4215 | * |
4216 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4217 | */ |
e02119d5 CM |
4218 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4219 | struct btrfs_root *root, | |
310712b2 | 4220 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4221 | struct btrfs_path *path, int data_size, |
4222 | int extend) | |
be0e5c09 | 4223 | { |
5d4f98a2 | 4224 | struct btrfs_disk_key disk_key; |
5f39d397 | 4225 | struct extent_buffer *l; |
7518a238 | 4226 | u32 nritems; |
eb60ceac CM |
4227 | int mid; |
4228 | int slot; | |
5f39d397 | 4229 | struct extent_buffer *right; |
b7a0365e | 4230 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4231 | int ret = 0; |
aa5d6bed | 4232 | int wret; |
5d4f98a2 | 4233 | int split; |
cc0c5538 | 4234 | int num_doubles = 0; |
99d8f83c | 4235 | int tried_avoid_double = 0; |
aa5d6bed | 4236 | |
a5719521 YZ |
4237 | l = path->nodes[0]; |
4238 | slot = path->slots[0]; | |
4239 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4240 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4241 | return -EOVERFLOW; |
4242 | ||
40689478 | 4243 | /* first try to make some room by pushing left and right */ |
33157e05 | 4244 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4245 | int space_needed = data_size; |
4246 | ||
4247 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 4248 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4249 | |
4250 | wret = push_leaf_right(trans, root, path, space_needed, | |
4251 | space_needed, 0, 0); | |
d397712b | 4252 | if (wret < 0) |
eaee50e8 | 4253 | return wret; |
3685f791 | 4254 | if (wret) { |
263d3995 FM |
4255 | space_needed = data_size; |
4256 | if (slot > 0) | |
e902baac | 4257 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4258 | wret = push_leaf_left(trans, root, path, space_needed, |
4259 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4260 | if (wret < 0) |
4261 | return wret; | |
4262 | } | |
4263 | l = path->nodes[0]; | |
aa5d6bed | 4264 | |
3685f791 | 4265 | /* did the pushes work? */ |
e902baac | 4266 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 4267 | return 0; |
3326d1b0 | 4268 | } |
aa5d6bed | 4269 | |
5c680ed6 | 4270 | if (!path->nodes[1]) { |
fdd99c72 | 4271 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4272 | if (ret) |
4273 | return ret; | |
4274 | } | |
cc0c5538 | 4275 | again: |
5d4f98a2 | 4276 | split = 1; |
cc0c5538 | 4277 | l = path->nodes[0]; |
eb60ceac | 4278 | slot = path->slots[0]; |
5f39d397 | 4279 | nritems = btrfs_header_nritems(l); |
d397712b | 4280 | mid = (nritems + 1) / 2; |
54aa1f4d | 4281 | |
5d4f98a2 YZ |
4282 | if (mid <= slot) { |
4283 | if (nritems == 1 || | |
4284 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4285 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4286 | if (slot >= nritems) { |
4287 | split = 0; | |
4288 | } else { | |
4289 | mid = slot; | |
4290 | if (mid != nritems && | |
4291 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4292 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4293 | if (data_size && !tried_avoid_double) |
4294 | goto push_for_double; | |
5d4f98a2 YZ |
4295 | split = 2; |
4296 | } | |
4297 | } | |
4298 | } | |
4299 | } else { | |
4300 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4301 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4302 | if (!extend && data_size && slot == 0) { |
4303 | split = 0; | |
4304 | } else if ((extend || !data_size) && slot == 0) { | |
4305 | mid = 1; | |
4306 | } else { | |
4307 | mid = slot; | |
4308 | if (mid != nritems && | |
4309 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4310 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4311 | if (data_size && !tried_avoid_double) |
4312 | goto push_for_double; | |
67871254 | 4313 | split = 2; |
5d4f98a2 YZ |
4314 | } |
4315 | } | |
4316 | } | |
4317 | } | |
4318 | ||
4319 | if (split == 0) | |
4320 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4321 | else | |
4322 | btrfs_item_key(l, &disk_key, mid); | |
4323 | ||
a6279470 FM |
4324 | right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, |
4325 | l->start, 0); | |
f0486c68 | 4326 | if (IS_ERR(right)) |
5f39d397 | 4327 | return PTR_ERR(right); |
f0486c68 | 4328 | |
0b246afa | 4329 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4330 | |
5d4f98a2 YZ |
4331 | if (split == 0) { |
4332 | if (mid <= slot) { | |
4333 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4334 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4335 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4336 | btrfs_tree_unlock(path->nodes[0]); |
4337 | free_extent_buffer(path->nodes[0]); | |
4338 | path->nodes[0] = right; | |
4339 | path->slots[0] = 0; | |
4340 | path->slots[1] += 1; | |
4341 | } else { | |
4342 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4343 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4344 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
4345 | btrfs_tree_unlock(path->nodes[0]); |
4346 | free_extent_buffer(path->nodes[0]); | |
4347 | path->nodes[0] = right; | |
4348 | path->slots[0] = 0; | |
143bede5 | 4349 | if (path->slots[1] == 0) |
b167fa91 | 4350 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4351 | } |
196e0249 LB |
4352 | /* |
4353 | * We create a new leaf 'right' for the required ins_len and | |
4354 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4355 | * the content of ins_len to 'right'. | |
4356 | */ | |
5d4f98a2 | 4357 | return ret; |
d4dbff95 | 4358 | } |
74123bd7 | 4359 | |
94f94ad9 | 4360 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 4361 | |
5d4f98a2 | 4362 | if (split == 2) { |
cc0c5538 CM |
4363 | BUG_ON(num_doubles != 0); |
4364 | num_doubles++; | |
4365 | goto again; | |
a429e513 | 4366 | } |
44871b1b | 4367 | |
143bede5 | 4368 | return 0; |
99d8f83c CM |
4369 | |
4370 | push_for_double: | |
4371 | push_for_double_split(trans, root, path, data_size); | |
4372 | tried_avoid_double = 1; | |
e902baac | 4373 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4374 | return 0; |
4375 | goto again; | |
be0e5c09 CM |
4376 | } |
4377 | ||
ad48fd75 YZ |
4378 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4379 | struct btrfs_root *root, | |
4380 | struct btrfs_path *path, int ins_len) | |
459931ec | 4381 | { |
ad48fd75 | 4382 | struct btrfs_key key; |
459931ec | 4383 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4384 | struct btrfs_file_extent_item *fi; |
4385 | u64 extent_len = 0; | |
4386 | u32 item_size; | |
4387 | int ret; | |
459931ec CM |
4388 | |
4389 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4390 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4391 | ||
4392 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4393 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4394 | ||
e902baac | 4395 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 4396 | return 0; |
459931ec CM |
4397 | |
4398 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4399 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4400 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4401 | struct btrfs_file_extent_item); | |
4402 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4403 | } | |
b3b4aa74 | 4404 | btrfs_release_path(path); |
459931ec | 4405 | |
459931ec | 4406 | path->keep_locks = 1; |
ad48fd75 YZ |
4407 | path->search_for_split = 1; |
4408 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4409 | path->search_for_split = 0; |
a8df6fe6 FM |
4410 | if (ret > 0) |
4411 | ret = -EAGAIN; | |
ad48fd75 YZ |
4412 | if (ret < 0) |
4413 | goto err; | |
459931ec | 4414 | |
ad48fd75 YZ |
4415 | ret = -EAGAIN; |
4416 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4417 | /* if our item isn't there, return now */ |
4418 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4419 | goto err; |
4420 | ||
109f6aef | 4421 | /* the leaf has changed, it now has room. return now */ |
e902baac | 4422 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
4423 | goto err; |
4424 | ||
ad48fd75 YZ |
4425 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4426 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4427 | struct btrfs_file_extent_item); | |
4428 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4429 | goto err; | |
459931ec CM |
4430 | } |
4431 | ||
b9473439 | 4432 | btrfs_set_path_blocking(path); |
ad48fd75 | 4433 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4434 | if (ret) |
4435 | goto err; | |
459931ec | 4436 | |
ad48fd75 | 4437 | path->keep_locks = 0; |
b9473439 | 4438 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4439 | return 0; |
4440 | err: | |
4441 | path->keep_locks = 0; | |
4442 | return ret; | |
4443 | } | |
4444 | ||
25263cd7 | 4445 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 4446 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4447 | unsigned long split_offset) |
4448 | { | |
4449 | struct extent_buffer *leaf; | |
4450 | struct btrfs_item *item; | |
4451 | struct btrfs_item *new_item; | |
4452 | int slot; | |
4453 | char *buf; | |
4454 | u32 nritems; | |
4455 | u32 item_size; | |
4456 | u32 orig_offset; | |
4457 | struct btrfs_disk_key disk_key; | |
4458 | ||
b9473439 | 4459 | leaf = path->nodes[0]; |
e902baac | 4460 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4461 | |
b4ce94de CM |
4462 | btrfs_set_path_blocking(path); |
4463 | ||
dd3cc16b | 4464 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4465 | orig_offset = btrfs_item_offset(leaf, item); |
4466 | item_size = btrfs_item_size(leaf, item); | |
4467 | ||
459931ec | 4468 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4469 | if (!buf) |
4470 | return -ENOMEM; | |
4471 | ||
459931ec CM |
4472 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4473 | path->slots[0]), item_size); | |
459931ec | 4474 | |
ad48fd75 | 4475 | slot = path->slots[0] + 1; |
459931ec | 4476 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4477 | if (slot != nritems) { |
4478 | /* shift the items */ | |
4479 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4480 | btrfs_item_nr_offset(slot), |
4481 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4482 | } |
4483 | ||
4484 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4485 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4486 | ||
dd3cc16b | 4487 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4488 | |
4489 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4490 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4491 | ||
4492 | btrfs_set_item_offset(leaf, item, | |
4493 | orig_offset + item_size - split_offset); | |
4494 | btrfs_set_item_size(leaf, item, split_offset); | |
4495 | ||
4496 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4497 | ||
4498 | /* write the data for the start of the original item */ | |
4499 | write_extent_buffer(leaf, buf, | |
4500 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4501 | split_offset); | |
4502 | ||
4503 | /* write the data for the new item */ | |
4504 | write_extent_buffer(leaf, buf + split_offset, | |
4505 | btrfs_item_ptr_offset(leaf, slot), | |
4506 | item_size - split_offset); | |
4507 | btrfs_mark_buffer_dirty(leaf); | |
4508 | ||
e902baac | 4509 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 4510 | kfree(buf); |
ad48fd75 YZ |
4511 | return 0; |
4512 | } | |
4513 | ||
4514 | /* | |
4515 | * This function splits a single item into two items, | |
4516 | * giving 'new_key' to the new item and splitting the | |
4517 | * old one at split_offset (from the start of the item). | |
4518 | * | |
4519 | * The path may be released by this operation. After | |
4520 | * the split, the path is pointing to the old item. The | |
4521 | * new item is going to be in the same node as the old one. | |
4522 | * | |
4523 | * Note, the item being split must be smaller enough to live alone on | |
4524 | * a tree block with room for one extra struct btrfs_item | |
4525 | * | |
4526 | * This allows us to split the item in place, keeping a lock on the | |
4527 | * leaf the entire time. | |
4528 | */ | |
4529 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4530 | struct btrfs_root *root, | |
4531 | struct btrfs_path *path, | |
310712b2 | 4532 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4533 | unsigned long split_offset) |
4534 | { | |
4535 | int ret; | |
4536 | ret = setup_leaf_for_split(trans, root, path, | |
4537 | sizeof(struct btrfs_item)); | |
4538 | if (ret) | |
4539 | return ret; | |
4540 | ||
25263cd7 | 4541 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
4542 | return ret; |
4543 | } | |
4544 | ||
ad48fd75 YZ |
4545 | /* |
4546 | * This function duplicate a item, giving 'new_key' to the new item. | |
4547 | * It guarantees both items live in the same tree leaf and the new item | |
4548 | * is contiguous with the original item. | |
4549 | * | |
4550 | * This allows us to split file extent in place, keeping a lock on the | |
4551 | * leaf the entire time. | |
4552 | */ | |
4553 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4554 | struct btrfs_root *root, | |
4555 | struct btrfs_path *path, | |
310712b2 | 4556 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4557 | { |
4558 | struct extent_buffer *leaf; | |
4559 | int ret; | |
4560 | u32 item_size; | |
4561 | ||
4562 | leaf = path->nodes[0]; | |
4563 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4564 | ret = setup_leaf_for_split(trans, root, path, | |
4565 | item_size + sizeof(struct btrfs_item)); | |
4566 | if (ret) | |
4567 | return ret; | |
4568 | ||
4569 | path->slots[0]++; | |
afe5fea7 | 4570 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4571 | item_size, item_size + |
4572 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4573 | leaf = path->nodes[0]; |
4574 | memcpy_extent_buffer(leaf, | |
4575 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4576 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4577 | item_size); | |
4578 | return 0; | |
4579 | } | |
4580 | ||
d352ac68 CM |
4581 | /* |
4582 | * make the item pointed to by the path smaller. new_size indicates | |
4583 | * how small to make it, and from_end tells us if we just chop bytes | |
4584 | * off the end of the item or if we shift the item to chop bytes off | |
4585 | * the front. | |
4586 | */ | |
78ac4f9e | 4587 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 4588 | { |
b18c6685 | 4589 | int slot; |
5f39d397 CM |
4590 | struct extent_buffer *leaf; |
4591 | struct btrfs_item *item; | |
b18c6685 CM |
4592 | u32 nritems; |
4593 | unsigned int data_end; | |
4594 | unsigned int old_data_start; | |
4595 | unsigned int old_size; | |
4596 | unsigned int size_diff; | |
4597 | int i; | |
cfed81a0 CM |
4598 | struct btrfs_map_token token; |
4599 | ||
5f39d397 | 4600 | leaf = path->nodes[0]; |
179e29e4 CM |
4601 | slot = path->slots[0]; |
4602 | ||
4603 | old_size = btrfs_item_size_nr(leaf, slot); | |
4604 | if (old_size == new_size) | |
143bede5 | 4605 | return; |
b18c6685 | 4606 | |
5f39d397 | 4607 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4608 | data_end = leaf_data_end(leaf); |
b18c6685 | 4609 | |
5f39d397 | 4610 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4611 | |
b18c6685 CM |
4612 | size_diff = old_size - new_size; |
4613 | ||
4614 | BUG_ON(slot < 0); | |
4615 | BUG_ON(slot >= nritems); | |
4616 | ||
4617 | /* | |
4618 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4619 | */ | |
4620 | /* first correct the data pointers */ | |
c82f823c | 4621 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 4622 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4623 | u32 ioff; |
dd3cc16b | 4624 | item = btrfs_item_nr(i); |
db94535d | 4625 | |
cfed81a0 CM |
4626 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4627 | btrfs_set_token_item_offset(leaf, item, | |
4628 | ioff + size_diff, &token); | |
b18c6685 | 4629 | } |
db94535d | 4630 | |
b18c6685 | 4631 | /* shift the data */ |
179e29e4 | 4632 | if (from_end) { |
3d9ec8c4 NB |
4633 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4634 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4635 | data_end, old_data_start + new_size - data_end); |
4636 | } else { | |
4637 | struct btrfs_disk_key disk_key; | |
4638 | u64 offset; | |
4639 | ||
4640 | btrfs_item_key(leaf, &disk_key, slot); | |
4641 | ||
4642 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4643 | unsigned long ptr; | |
4644 | struct btrfs_file_extent_item *fi; | |
4645 | ||
4646 | fi = btrfs_item_ptr(leaf, slot, | |
4647 | struct btrfs_file_extent_item); | |
4648 | fi = (struct btrfs_file_extent_item *)( | |
4649 | (unsigned long)fi - size_diff); | |
4650 | ||
4651 | if (btrfs_file_extent_type(leaf, fi) == | |
4652 | BTRFS_FILE_EXTENT_INLINE) { | |
4653 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4654 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4655 | (unsigned long)fi, |
7ec20afb | 4656 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4657 | } |
4658 | } | |
4659 | ||
3d9ec8c4 NB |
4660 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4661 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4662 | data_end, old_data_start - data_end); |
4663 | ||
4664 | offset = btrfs_disk_key_offset(&disk_key); | |
4665 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4666 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4667 | if (slot == 0) | |
b167fa91 | 4668 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4669 | } |
5f39d397 | 4670 | |
dd3cc16b | 4671 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4672 | btrfs_set_item_size(leaf, item, new_size); |
4673 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4674 | |
e902baac | 4675 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4676 | btrfs_print_leaf(leaf); |
b18c6685 | 4677 | BUG(); |
5f39d397 | 4678 | } |
b18c6685 CM |
4679 | } |
4680 | ||
d352ac68 | 4681 | /* |
8f69dbd2 | 4682 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4683 | */ |
c71dd880 | 4684 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 4685 | { |
6567e837 | 4686 | int slot; |
5f39d397 CM |
4687 | struct extent_buffer *leaf; |
4688 | struct btrfs_item *item; | |
6567e837 CM |
4689 | u32 nritems; |
4690 | unsigned int data_end; | |
4691 | unsigned int old_data; | |
4692 | unsigned int old_size; | |
4693 | int i; | |
cfed81a0 CM |
4694 | struct btrfs_map_token token; |
4695 | ||
5f39d397 | 4696 | leaf = path->nodes[0]; |
6567e837 | 4697 | |
5f39d397 | 4698 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4699 | data_end = leaf_data_end(leaf); |
6567e837 | 4700 | |
e902baac | 4701 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 4702 | btrfs_print_leaf(leaf); |
6567e837 | 4703 | BUG(); |
5f39d397 | 4704 | } |
6567e837 | 4705 | slot = path->slots[0]; |
5f39d397 | 4706 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4707 | |
4708 | BUG_ON(slot < 0); | |
3326d1b0 | 4709 | if (slot >= nritems) { |
a4f78750 | 4710 | btrfs_print_leaf(leaf); |
c71dd880 | 4711 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 4712 | slot, nritems); |
290342f6 | 4713 | BUG(); |
3326d1b0 | 4714 | } |
6567e837 CM |
4715 | |
4716 | /* | |
4717 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4718 | */ | |
4719 | /* first correct the data pointers */ | |
c82f823c | 4720 | btrfs_init_map_token(&token, leaf); |
6567e837 | 4721 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4722 | u32 ioff; |
dd3cc16b | 4723 | item = btrfs_item_nr(i); |
db94535d | 4724 | |
cfed81a0 CM |
4725 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4726 | btrfs_set_token_item_offset(leaf, item, | |
4727 | ioff - data_size, &token); | |
6567e837 | 4728 | } |
5f39d397 | 4729 | |
6567e837 | 4730 | /* shift the data */ |
3d9ec8c4 NB |
4731 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4732 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4733 | data_end, old_data - data_end); |
5f39d397 | 4734 | |
6567e837 | 4735 | data_end = old_data; |
5f39d397 | 4736 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4737 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4738 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4739 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4740 | |
e902baac | 4741 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4742 | btrfs_print_leaf(leaf); |
6567e837 | 4743 | BUG(); |
5f39d397 | 4744 | } |
6567e837 CM |
4745 | } |
4746 | ||
74123bd7 | 4747 | /* |
44871b1b CM |
4748 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4749 | * to save stack depth by doing the bulk of the work in a function | |
4750 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4751 | */ |
afe5fea7 | 4752 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4753 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4754 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4755 | { |
0b246afa | 4756 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4757 | struct btrfs_item *item; |
9c58309d | 4758 | int i; |
7518a238 | 4759 | u32 nritems; |
be0e5c09 | 4760 | unsigned int data_end; |
e2fa7227 | 4761 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4762 | struct extent_buffer *leaf; |
4763 | int slot; | |
cfed81a0 CM |
4764 | struct btrfs_map_token token; |
4765 | ||
24cdc847 FM |
4766 | if (path->slots[0] == 0) { |
4767 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4768 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4769 | } |
4770 | btrfs_unlock_up_safe(path, 1); | |
4771 | ||
5f39d397 | 4772 | leaf = path->nodes[0]; |
44871b1b | 4773 | slot = path->slots[0]; |
74123bd7 | 4774 | |
5f39d397 | 4775 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4776 | data_end = leaf_data_end(leaf); |
eb60ceac | 4777 | |
e902baac | 4778 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4779 | btrfs_print_leaf(leaf); |
0b246afa | 4780 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4781 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4782 | BUG(); |
d4dbff95 | 4783 | } |
5f39d397 | 4784 | |
c82f823c | 4785 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4786 | if (slot != nritems) { |
5f39d397 | 4787 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4788 | |
5f39d397 | 4789 | if (old_data < data_end) { |
a4f78750 | 4790 | btrfs_print_leaf(leaf); |
0b246afa | 4791 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4792 | slot, old_data, data_end); |
290342f6 | 4793 | BUG(); |
5f39d397 | 4794 | } |
be0e5c09 CM |
4795 | /* |
4796 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4797 | */ | |
4798 | /* first correct the data pointers */ | |
0783fcfc | 4799 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4800 | u32 ioff; |
db94535d | 4801 | |
62e85577 | 4802 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4803 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4804 | btrfs_set_token_item_offset(leaf, item, | |
4805 | ioff - total_data, &token); | |
0783fcfc | 4806 | } |
be0e5c09 | 4807 | /* shift the items */ |
9c58309d | 4808 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4809 | btrfs_item_nr_offset(slot), |
d6025579 | 4810 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4811 | |
4812 | /* shift the data */ | |
3d9ec8c4 NB |
4813 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4814 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4815 | data_end, old_data - data_end); |
be0e5c09 CM |
4816 | data_end = old_data; |
4817 | } | |
5f39d397 | 4818 | |
62e2749e | 4819 | /* setup the item for the new data */ |
9c58309d CM |
4820 | for (i = 0; i < nr; i++) { |
4821 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4822 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4823 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4824 | btrfs_set_token_item_offset(leaf, item, |
4825 | data_end - data_size[i], &token); | |
9c58309d | 4826 | data_end -= data_size[i]; |
cfed81a0 | 4827 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4828 | } |
44871b1b | 4829 | |
9c58309d | 4830 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4831 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4832 | |
e902baac | 4833 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4834 | btrfs_print_leaf(leaf); |
be0e5c09 | 4835 | BUG(); |
5f39d397 | 4836 | } |
44871b1b CM |
4837 | } |
4838 | ||
4839 | /* | |
4840 | * Given a key and some data, insert items into the tree. | |
4841 | * This does all the path init required, making room in the tree if needed. | |
4842 | */ | |
4843 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4844 | struct btrfs_root *root, | |
4845 | struct btrfs_path *path, | |
310712b2 | 4846 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4847 | int nr) |
4848 | { | |
44871b1b CM |
4849 | int ret = 0; |
4850 | int slot; | |
4851 | int i; | |
4852 | u32 total_size = 0; | |
4853 | u32 total_data = 0; | |
4854 | ||
4855 | for (i = 0; i < nr; i++) | |
4856 | total_data += data_size[i]; | |
4857 | ||
4858 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4859 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4860 | if (ret == 0) | |
4861 | return -EEXIST; | |
4862 | if (ret < 0) | |
143bede5 | 4863 | return ret; |
44871b1b | 4864 | |
44871b1b CM |
4865 | slot = path->slots[0]; |
4866 | BUG_ON(slot < 0); | |
4867 | ||
afe5fea7 | 4868 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4869 | total_data, total_size, nr); |
143bede5 | 4870 | return 0; |
62e2749e CM |
4871 | } |
4872 | ||
4873 | /* | |
4874 | * Given a key and some data, insert an item into the tree. | |
4875 | * This does all the path init required, making room in the tree if needed. | |
4876 | */ | |
310712b2 OS |
4877 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4878 | const struct btrfs_key *cpu_key, void *data, | |
4879 | u32 data_size) | |
62e2749e CM |
4880 | { |
4881 | int ret = 0; | |
2c90e5d6 | 4882 | struct btrfs_path *path; |
5f39d397 CM |
4883 | struct extent_buffer *leaf; |
4884 | unsigned long ptr; | |
62e2749e | 4885 | |
2c90e5d6 | 4886 | path = btrfs_alloc_path(); |
db5b493a TI |
4887 | if (!path) |
4888 | return -ENOMEM; | |
2c90e5d6 | 4889 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4890 | if (!ret) { |
5f39d397 CM |
4891 | leaf = path->nodes[0]; |
4892 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4893 | write_extent_buffer(leaf, data, ptr, data_size); | |
4894 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4895 | } |
2c90e5d6 | 4896 | btrfs_free_path(path); |
aa5d6bed | 4897 | return ret; |
be0e5c09 CM |
4898 | } |
4899 | ||
74123bd7 | 4900 | /* |
5de08d7d | 4901 | * delete the pointer from a given node. |
74123bd7 | 4902 | * |
d352ac68 CM |
4903 | * the tree should have been previously balanced so the deletion does not |
4904 | * empty a node. | |
74123bd7 | 4905 | */ |
afe5fea7 TI |
4906 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4907 | int level, int slot) | |
be0e5c09 | 4908 | { |
5f39d397 | 4909 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4910 | u32 nritems; |
f3ea38da | 4911 | int ret; |
be0e5c09 | 4912 | |
5f39d397 | 4913 | nritems = btrfs_header_nritems(parent); |
d397712b | 4914 | if (slot != nritems - 1) { |
bf1d3425 DS |
4915 | if (level) { |
4916 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4917 | nritems - slot - 1); |
bf1d3425 DS |
4918 | BUG_ON(ret < 0); |
4919 | } | |
5f39d397 CM |
4920 | memmove_extent_buffer(parent, |
4921 | btrfs_node_key_ptr_offset(slot), | |
4922 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4923 | sizeof(struct btrfs_key_ptr) * |
4924 | (nritems - slot - 1)); | |
57ba86c0 | 4925 | } else if (level) { |
e09c2efe DS |
4926 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4927 | GFP_NOFS); | |
57ba86c0 | 4928 | BUG_ON(ret < 0); |
bb803951 | 4929 | } |
f3ea38da | 4930 | |
7518a238 | 4931 | nritems--; |
5f39d397 | 4932 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4933 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4934 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4935 | /* just turn the root into a leaf and break */ |
5f39d397 | 4936 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4937 | } else if (slot == 0) { |
5f39d397 CM |
4938 | struct btrfs_disk_key disk_key; |
4939 | ||
4940 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4941 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4942 | } |
d6025579 | 4943 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4944 | } |
4945 | ||
323ac95b CM |
4946 | /* |
4947 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4948 | * path->nodes[1]. |
323ac95b CM |
4949 | * |
4950 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4951 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4952 | * | |
4953 | * The path must have already been setup for deleting the leaf, including | |
4954 | * all the proper balancing. path->nodes[1] must be locked. | |
4955 | */ | |
143bede5 JM |
4956 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4957 | struct btrfs_root *root, | |
4958 | struct btrfs_path *path, | |
4959 | struct extent_buffer *leaf) | |
323ac95b | 4960 | { |
5d4f98a2 | 4961 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4962 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4963 | |
4d081c41 CM |
4964 | /* |
4965 | * btrfs_free_extent is expensive, we want to make sure we | |
4966 | * aren't holding any locks when we call it | |
4967 | */ | |
4968 | btrfs_unlock_up_safe(path, 0); | |
4969 | ||
f0486c68 YZ |
4970 | root_sub_used(root, leaf->len); |
4971 | ||
3083ee2e | 4972 | extent_buffer_get(leaf); |
5581a51a | 4973 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4974 | free_extent_buffer_stale(leaf); |
323ac95b | 4975 | } |
74123bd7 CM |
4976 | /* |
4977 | * delete the item at the leaf level in path. If that empties | |
4978 | * the leaf, remove it from the tree | |
4979 | */ | |
85e21bac CM |
4980 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4981 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4982 | { |
0b246afa | 4983 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4984 | struct extent_buffer *leaf; |
4985 | struct btrfs_item *item; | |
ce0eac2a AM |
4986 | u32 last_off; |
4987 | u32 dsize = 0; | |
aa5d6bed CM |
4988 | int ret = 0; |
4989 | int wret; | |
85e21bac | 4990 | int i; |
7518a238 | 4991 | u32 nritems; |
be0e5c09 | 4992 | |
5f39d397 | 4993 | leaf = path->nodes[0]; |
85e21bac CM |
4994 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4995 | ||
4996 | for (i = 0; i < nr; i++) | |
4997 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4998 | ||
5f39d397 | 4999 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 5000 | |
85e21bac | 5001 | if (slot + nr != nritems) { |
8f881e8c | 5002 | int data_end = leaf_data_end(leaf); |
c82f823c | 5003 | struct btrfs_map_token token; |
5f39d397 | 5004 | |
3d9ec8c4 | 5005 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 5006 | data_end + dsize, |
3d9ec8c4 | 5007 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 5008 | last_off - data_end); |
5f39d397 | 5009 | |
c82f823c | 5010 | btrfs_init_map_token(&token, leaf); |
85e21bac | 5011 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 5012 | u32 ioff; |
db94535d | 5013 | |
dd3cc16b | 5014 | item = btrfs_item_nr(i); |
cfed81a0 CM |
5015 | ioff = btrfs_token_item_offset(leaf, item, &token); |
5016 | btrfs_set_token_item_offset(leaf, item, | |
5017 | ioff + dsize, &token); | |
0783fcfc | 5018 | } |
db94535d | 5019 | |
5f39d397 | 5020 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5021 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5022 | sizeof(struct btrfs_item) * |
85e21bac | 5023 | (nritems - slot - nr)); |
be0e5c09 | 5024 | } |
85e21bac CM |
5025 | btrfs_set_header_nritems(leaf, nritems - nr); |
5026 | nritems -= nr; | |
5f39d397 | 5027 | |
74123bd7 | 5028 | /* delete the leaf if we've emptied it */ |
7518a238 | 5029 | if (nritems == 0) { |
5f39d397 CM |
5030 | if (leaf == root->node) { |
5031 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5032 | } else { |
f0486c68 | 5033 | btrfs_set_path_blocking(path); |
6a884d7d | 5034 | btrfs_clean_tree_block(leaf); |
143bede5 | 5035 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5036 | } |
be0e5c09 | 5037 | } else { |
7518a238 | 5038 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5039 | if (slot == 0) { |
5f39d397 CM |
5040 | struct btrfs_disk_key disk_key; |
5041 | ||
5042 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 5043 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 5044 | } |
aa5d6bed | 5045 | |
74123bd7 | 5046 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5047 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5048 | /* push_leaf_left fixes the path. |
5049 | * make sure the path still points to our leaf | |
5050 | * for possible call to del_ptr below | |
5051 | */ | |
4920c9ac | 5052 | slot = path->slots[1]; |
5f39d397 CM |
5053 | extent_buffer_get(leaf); |
5054 | ||
b9473439 | 5055 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5056 | wret = push_leaf_left(trans, root, path, 1, 1, |
5057 | 1, (u32)-1); | |
54aa1f4d | 5058 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5059 | ret = wret; |
5f39d397 CM |
5060 | |
5061 | if (path->nodes[0] == leaf && | |
5062 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5063 | wret = push_leaf_right(trans, root, path, 1, |
5064 | 1, 1, 0); | |
54aa1f4d | 5065 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5066 | ret = wret; |
5067 | } | |
5f39d397 CM |
5068 | |
5069 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5070 | path->slots[1] = slot; |
143bede5 | 5071 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5072 | free_extent_buffer(leaf); |
143bede5 | 5073 | ret = 0; |
5de08d7d | 5074 | } else { |
925baedd CM |
5075 | /* if we're still in the path, make sure |
5076 | * we're dirty. Otherwise, one of the | |
5077 | * push_leaf functions must have already | |
5078 | * dirtied this buffer | |
5079 | */ | |
5080 | if (path->nodes[0] == leaf) | |
5081 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5082 | free_extent_buffer(leaf); |
be0e5c09 | 5083 | } |
d5719762 | 5084 | } else { |
5f39d397 | 5085 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5086 | } |
5087 | } | |
aa5d6bed | 5088 | return ret; |
be0e5c09 CM |
5089 | } |
5090 | ||
7bb86316 | 5091 | /* |
925baedd | 5092 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5093 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5094 | * returns < 0 on io errors. | |
d352ac68 CM |
5095 | * |
5096 | * This may release the path, and so you may lose any locks held at the | |
5097 | * time you call it. | |
7bb86316 | 5098 | */ |
16e7549f | 5099 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5100 | { |
925baedd CM |
5101 | struct btrfs_key key; |
5102 | struct btrfs_disk_key found_key; | |
5103 | int ret; | |
7bb86316 | 5104 | |
925baedd | 5105 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5106 | |
e8b0d724 | 5107 | if (key.offset > 0) { |
925baedd | 5108 | key.offset--; |
e8b0d724 | 5109 | } else if (key.type > 0) { |
925baedd | 5110 | key.type--; |
e8b0d724 FDBM |
5111 | key.offset = (u64)-1; |
5112 | } else if (key.objectid > 0) { | |
925baedd | 5113 | key.objectid--; |
e8b0d724 FDBM |
5114 | key.type = (u8)-1; |
5115 | key.offset = (u64)-1; | |
5116 | } else { | |
925baedd | 5117 | return 1; |
e8b0d724 | 5118 | } |
7bb86316 | 5119 | |
b3b4aa74 | 5120 | btrfs_release_path(path); |
925baedd CM |
5121 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5122 | if (ret < 0) | |
5123 | return ret; | |
5124 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5125 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5126 | /* |
5127 | * We might have had an item with the previous key in the tree right | |
5128 | * before we released our path. And after we released our path, that | |
5129 | * item might have been pushed to the first slot (0) of the leaf we | |
5130 | * were holding due to a tree balance. Alternatively, an item with the | |
5131 | * previous key can exist as the only element of a leaf (big fat item). | |
5132 | * Therefore account for these 2 cases, so that our callers (like | |
5133 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5134 | * the previous key we computed above. | |
5135 | */ | |
5136 | if (ret <= 0) | |
925baedd CM |
5137 | return 0; |
5138 | return 1; | |
7bb86316 CM |
5139 | } |
5140 | ||
3f157a2f CM |
5141 | /* |
5142 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5143 | * for nodes or leaves that are have a minimum transaction id. |
5144 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5145 | * |
5146 | * This does not cow, but it does stuff the starting key it finds back | |
5147 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5148 | * key and get a writable path. | |
5149 | * | |
3f157a2f CM |
5150 | * This honors path->lowest_level to prevent descent past a given level |
5151 | * of the tree. | |
5152 | * | |
d352ac68 CM |
5153 | * min_trans indicates the oldest transaction that you are interested |
5154 | * in walking through. Any nodes or leaves older than min_trans are | |
5155 | * skipped over (without reading them). | |
5156 | * | |
3f157a2f CM |
5157 | * returns zero if something useful was found, < 0 on error and 1 if there |
5158 | * was nothing in the tree that matched the search criteria. | |
5159 | */ | |
5160 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5161 | struct btrfs_path *path, |
3f157a2f CM |
5162 | u64 min_trans) |
5163 | { | |
5164 | struct extent_buffer *cur; | |
5165 | struct btrfs_key found_key; | |
5166 | int slot; | |
9652480b | 5167 | int sret; |
3f157a2f CM |
5168 | u32 nritems; |
5169 | int level; | |
5170 | int ret = 1; | |
f98de9b9 | 5171 | int keep_locks = path->keep_locks; |
3f157a2f | 5172 | |
f98de9b9 | 5173 | path->keep_locks = 1; |
3f157a2f | 5174 | again: |
bd681513 | 5175 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5176 | level = btrfs_header_level(cur); |
e02119d5 | 5177 | WARN_ON(path->nodes[level]); |
3f157a2f | 5178 | path->nodes[level] = cur; |
bd681513 | 5179 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5180 | |
5181 | if (btrfs_header_generation(cur) < min_trans) { | |
5182 | ret = 1; | |
5183 | goto out; | |
5184 | } | |
d397712b | 5185 | while (1) { |
3f157a2f CM |
5186 | nritems = btrfs_header_nritems(cur); |
5187 | level = btrfs_header_level(cur); | |
a74b35ec | 5188 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
cbca7d59 FM |
5189 | if (sret < 0) { |
5190 | ret = sret; | |
5191 | goto out; | |
5192 | } | |
3f157a2f | 5193 | |
323ac95b CM |
5194 | /* at the lowest level, we're done, setup the path and exit */ |
5195 | if (level == path->lowest_level) { | |
e02119d5 CM |
5196 | if (slot >= nritems) |
5197 | goto find_next_key; | |
3f157a2f CM |
5198 | ret = 0; |
5199 | path->slots[level] = slot; | |
5200 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5201 | goto out; | |
5202 | } | |
9652480b Y |
5203 | if (sret && slot > 0) |
5204 | slot--; | |
3f157a2f | 5205 | /* |
de78b51a ES |
5206 | * check this node pointer against the min_trans parameters. |
5207 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5208 | */ |
d397712b | 5209 | while (slot < nritems) { |
3f157a2f | 5210 | u64 gen; |
e02119d5 | 5211 | |
3f157a2f CM |
5212 | gen = btrfs_node_ptr_generation(cur, slot); |
5213 | if (gen < min_trans) { | |
5214 | slot++; | |
5215 | continue; | |
5216 | } | |
de78b51a | 5217 | break; |
3f157a2f | 5218 | } |
e02119d5 | 5219 | find_next_key: |
3f157a2f CM |
5220 | /* |
5221 | * we didn't find a candidate key in this node, walk forward | |
5222 | * and find another one | |
5223 | */ | |
5224 | if (slot >= nritems) { | |
e02119d5 | 5225 | path->slots[level] = slot; |
b4ce94de | 5226 | btrfs_set_path_blocking(path); |
e02119d5 | 5227 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5228 | min_trans); |
e02119d5 | 5229 | if (sret == 0) { |
b3b4aa74 | 5230 | btrfs_release_path(path); |
3f157a2f CM |
5231 | goto again; |
5232 | } else { | |
5233 | goto out; | |
5234 | } | |
5235 | } | |
5236 | /* save our key for returning back */ | |
5237 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5238 | path->slots[level] = slot; | |
5239 | if (level == path->lowest_level) { | |
5240 | ret = 0; | |
3f157a2f CM |
5241 | goto out; |
5242 | } | |
b4ce94de | 5243 | btrfs_set_path_blocking(path); |
4b231ae4 | 5244 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
5245 | if (IS_ERR(cur)) { |
5246 | ret = PTR_ERR(cur); | |
5247 | goto out; | |
5248 | } | |
3f157a2f | 5249 | |
bd681513 | 5250 | btrfs_tree_read_lock(cur); |
b4ce94de | 5251 | |
bd681513 | 5252 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5253 | path->nodes[level - 1] = cur; |
f7c79f30 | 5254 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5255 | } |
5256 | out: | |
f98de9b9 FM |
5257 | path->keep_locks = keep_locks; |
5258 | if (ret == 0) { | |
5259 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5260 | btrfs_set_path_blocking(path); | |
3f157a2f | 5261 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5262 | } |
3f157a2f CM |
5263 | return ret; |
5264 | } | |
5265 | ||
5266 | /* | |
5267 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5268 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5269 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5270 | * |
5271 | * 0 is returned if another key is found, < 0 if there are any errors | |
5272 | * and 1 is returned if there are no higher keys in the tree | |
5273 | * | |
5274 | * path->keep_locks should be set to 1 on the search made before | |
5275 | * calling this function. | |
5276 | */ | |
e7a84565 | 5277 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5278 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5279 | { |
e7a84565 CM |
5280 | int slot; |
5281 | struct extent_buffer *c; | |
5282 | ||
6a9fb468 | 5283 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 5284 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5285 | if (!path->nodes[level]) |
5286 | return 1; | |
5287 | ||
5288 | slot = path->slots[level] + 1; | |
5289 | c = path->nodes[level]; | |
3f157a2f | 5290 | next: |
e7a84565 | 5291 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5292 | int ret; |
5293 | int orig_lowest; | |
5294 | struct btrfs_key cur_key; | |
5295 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5296 | !path->nodes[level + 1]) | |
e7a84565 | 5297 | return 1; |
33c66f43 | 5298 | |
6a9fb468 | 5299 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
5300 | level++; |
5301 | continue; | |
5302 | } | |
5303 | ||
5304 | slot = btrfs_header_nritems(c) - 1; | |
5305 | if (level == 0) | |
5306 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5307 | else | |
5308 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5309 | ||
5310 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5311 | btrfs_release_path(path); |
33c66f43 YZ |
5312 | path->lowest_level = level; |
5313 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5314 | 0, 0); | |
5315 | path->lowest_level = orig_lowest; | |
5316 | if (ret < 0) | |
5317 | return ret; | |
5318 | ||
5319 | c = path->nodes[level]; | |
5320 | slot = path->slots[level]; | |
5321 | if (ret == 0) | |
5322 | slot++; | |
5323 | goto next; | |
e7a84565 | 5324 | } |
33c66f43 | 5325 | |
e7a84565 CM |
5326 | if (level == 0) |
5327 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5328 | else { |
3f157a2f CM |
5329 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5330 | ||
3f157a2f CM |
5331 | if (gen < min_trans) { |
5332 | slot++; | |
5333 | goto next; | |
5334 | } | |
e7a84565 | 5335 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5336 | } |
e7a84565 CM |
5337 | return 0; |
5338 | } | |
5339 | return 1; | |
5340 | } | |
5341 | ||
97571fd0 | 5342 | /* |
925baedd | 5343 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5344 | * returns 0 if it found something or 1 if there are no greater leaves. |
5345 | * returns < 0 on io errors. | |
97571fd0 | 5346 | */ |
234b63a0 | 5347 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5348 | { |
5349 | return btrfs_next_old_leaf(root, path, 0); | |
5350 | } | |
5351 | ||
5352 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5353 | u64 time_seq) | |
d97e63b6 CM |
5354 | { |
5355 | int slot; | |
8e73f275 | 5356 | int level; |
5f39d397 | 5357 | struct extent_buffer *c; |
8e73f275 | 5358 | struct extent_buffer *next; |
925baedd CM |
5359 | struct btrfs_key key; |
5360 | u32 nritems; | |
5361 | int ret; | |
8e73f275 | 5362 | int old_spinning = path->leave_spinning; |
bd681513 | 5363 | int next_rw_lock = 0; |
925baedd CM |
5364 | |
5365 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5366 | if (nritems == 0) |
925baedd | 5367 | return 1; |
925baedd | 5368 | |
8e73f275 CM |
5369 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5370 | again: | |
5371 | level = 1; | |
5372 | next = NULL; | |
bd681513 | 5373 | next_rw_lock = 0; |
b3b4aa74 | 5374 | btrfs_release_path(path); |
8e73f275 | 5375 | |
a2135011 | 5376 | path->keep_locks = 1; |
31533fb2 | 5377 | path->leave_spinning = 1; |
8e73f275 | 5378 | |
3d7806ec JS |
5379 | if (time_seq) |
5380 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5381 | else | |
5382 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5383 | path->keep_locks = 0; |
5384 | ||
5385 | if (ret < 0) | |
5386 | return ret; | |
5387 | ||
a2135011 | 5388 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5389 | /* |
5390 | * by releasing the path above we dropped all our locks. A balance | |
5391 | * could have added more items next to the key that used to be | |
5392 | * at the very end of the block. So, check again here and | |
5393 | * advance the path if there are now more items available. | |
5394 | */ | |
a2135011 | 5395 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5396 | if (ret == 0) |
5397 | path->slots[0]++; | |
8e73f275 | 5398 | ret = 0; |
925baedd CM |
5399 | goto done; |
5400 | } | |
0b43e04f LB |
5401 | /* |
5402 | * So the above check misses one case: | |
5403 | * - after releasing the path above, someone has removed the item that | |
5404 | * used to be at the very end of the block, and balance between leafs | |
5405 | * gets another one with bigger key.offset to replace it. | |
5406 | * | |
5407 | * This one should be returned as well, or we can get leaf corruption | |
5408 | * later(esp. in __btrfs_drop_extents()). | |
5409 | * | |
5410 | * And a bit more explanation about this check, | |
5411 | * with ret > 0, the key isn't found, the path points to the slot | |
5412 | * where it should be inserted, so the path->slots[0] item must be the | |
5413 | * bigger one. | |
5414 | */ | |
5415 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5416 | ret = 0; | |
5417 | goto done; | |
5418 | } | |
d97e63b6 | 5419 | |
d397712b | 5420 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5421 | if (!path->nodes[level]) { |
5422 | ret = 1; | |
5423 | goto done; | |
5424 | } | |
5f39d397 | 5425 | |
d97e63b6 CM |
5426 | slot = path->slots[level] + 1; |
5427 | c = path->nodes[level]; | |
5f39d397 | 5428 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5429 | level++; |
8e73f275 CM |
5430 | if (level == BTRFS_MAX_LEVEL) { |
5431 | ret = 1; | |
5432 | goto done; | |
5433 | } | |
d97e63b6 CM |
5434 | continue; |
5435 | } | |
5f39d397 | 5436 | |
925baedd | 5437 | if (next) { |
bd681513 | 5438 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5439 | free_extent_buffer(next); |
925baedd | 5440 | } |
5f39d397 | 5441 | |
8e73f275 | 5442 | next = c; |
bd681513 | 5443 | next_rw_lock = path->locks[level]; |
d07b8528 | 5444 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5445 | slot, &key); |
8e73f275 CM |
5446 | if (ret == -EAGAIN) |
5447 | goto again; | |
5f39d397 | 5448 | |
76a05b35 | 5449 | if (ret < 0) { |
b3b4aa74 | 5450 | btrfs_release_path(path); |
76a05b35 CM |
5451 | goto done; |
5452 | } | |
5453 | ||
5cd57b2c | 5454 | if (!path->skip_locking) { |
bd681513 | 5455 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5456 | if (!ret && time_seq) { |
5457 | /* | |
5458 | * If we don't get the lock, we may be racing | |
5459 | * with push_leaf_left, holding that lock while | |
5460 | * itself waiting for the leaf we've currently | |
5461 | * locked. To solve this situation, we give up | |
5462 | * on our lock and cycle. | |
5463 | */ | |
cf538830 | 5464 | free_extent_buffer(next); |
d42244a0 JS |
5465 | btrfs_release_path(path); |
5466 | cond_resched(); | |
5467 | goto again; | |
5468 | } | |
8e73f275 CM |
5469 | if (!ret) { |
5470 | btrfs_set_path_blocking(path); | |
bd681513 | 5471 | btrfs_tree_read_lock(next); |
8e73f275 | 5472 | } |
31533fb2 | 5473 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5474 | } |
d97e63b6 CM |
5475 | break; |
5476 | } | |
5477 | path->slots[level] = slot; | |
d397712b | 5478 | while (1) { |
d97e63b6 CM |
5479 | level--; |
5480 | c = path->nodes[level]; | |
925baedd | 5481 | if (path->locks[level]) |
bd681513 | 5482 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5483 | |
5f39d397 | 5484 | free_extent_buffer(c); |
d97e63b6 CM |
5485 | path->nodes[level] = next; |
5486 | path->slots[level] = 0; | |
a74a4b97 | 5487 | if (!path->skip_locking) |
bd681513 | 5488 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5489 | if (!level) |
5490 | break; | |
b4ce94de | 5491 | |
d07b8528 | 5492 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5493 | 0, &key); |
8e73f275 CM |
5494 | if (ret == -EAGAIN) |
5495 | goto again; | |
5496 | ||
76a05b35 | 5497 | if (ret < 0) { |
b3b4aa74 | 5498 | btrfs_release_path(path); |
76a05b35 CM |
5499 | goto done; |
5500 | } | |
5501 | ||
5cd57b2c | 5502 | if (!path->skip_locking) { |
bd681513 | 5503 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5504 | if (!ret) { |
5505 | btrfs_set_path_blocking(path); | |
bd681513 | 5506 | btrfs_tree_read_lock(next); |
bd681513 | 5507 | } |
31533fb2 | 5508 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5509 | } |
d97e63b6 | 5510 | } |
8e73f275 | 5511 | ret = 0; |
925baedd | 5512 | done: |
f7c79f30 | 5513 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5514 | path->leave_spinning = old_spinning; |
5515 | if (!old_spinning) | |
5516 | btrfs_set_path_blocking(path); | |
5517 | ||
5518 | return ret; | |
d97e63b6 | 5519 | } |
0b86a832 | 5520 | |
3f157a2f CM |
5521 | /* |
5522 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5523 | * searching until it gets past min_objectid or finds an item of 'type' | |
5524 | * | |
5525 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5526 | */ | |
0b86a832 CM |
5527 | int btrfs_previous_item(struct btrfs_root *root, |
5528 | struct btrfs_path *path, u64 min_objectid, | |
5529 | int type) | |
5530 | { | |
5531 | struct btrfs_key found_key; | |
5532 | struct extent_buffer *leaf; | |
e02119d5 | 5533 | u32 nritems; |
0b86a832 CM |
5534 | int ret; |
5535 | ||
d397712b | 5536 | while (1) { |
0b86a832 | 5537 | if (path->slots[0] == 0) { |
b4ce94de | 5538 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5539 | ret = btrfs_prev_leaf(root, path); |
5540 | if (ret != 0) | |
5541 | return ret; | |
5542 | } else { | |
5543 | path->slots[0]--; | |
5544 | } | |
5545 | leaf = path->nodes[0]; | |
e02119d5 CM |
5546 | nritems = btrfs_header_nritems(leaf); |
5547 | if (nritems == 0) | |
5548 | return 1; | |
5549 | if (path->slots[0] == nritems) | |
5550 | path->slots[0]--; | |
5551 | ||
0b86a832 | 5552 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5553 | if (found_key.objectid < min_objectid) |
5554 | break; | |
0a4eefbb YZ |
5555 | if (found_key.type == type) |
5556 | return 0; | |
e02119d5 CM |
5557 | if (found_key.objectid == min_objectid && |
5558 | found_key.type < type) | |
5559 | break; | |
0b86a832 CM |
5560 | } |
5561 | return 1; | |
5562 | } | |
ade2e0b3 WS |
5563 | |
5564 | /* | |
5565 | * search in extent tree to find a previous Metadata/Data extent item with | |
5566 | * min objecitd. | |
5567 | * | |
5568 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5569 | */ | |
5570 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5571 | struct btrfs_path *path, u64 min_objectid) | |
5572 | { | |
5573 | struct btrfs_key found_key; | |
5574 | struct extent_buffer *leaf; | |
5575 | u32 nritems; | |
5576 | int ret; | |
5577 | ||
5578 | while (1) { | |
5579 | if (path->slots[0] == 0) { | |
5580 | btrfs_set_path_blocking(path); | |
5581 | ret = btrfs_prev_leaf(root, path); | |
5582 | if (ret != 0) | |
5583 | return ret; | |
5584 | } else { | |
5585 | path->slots[0]--; | |
5586 | } | |
5587 | leaf = path->nodes[0]; | |
5588 | nritems = btrfs_header_nritems(leaf); | |
5589 | if (nritems == 0) | |
5590 | return 1; | |
5591 | if (path->slots[0] == nritems) | |
5592 | path->slots[0]--; | |
5593 | ||
5594 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5595 | if (found_key.objectid < min_objectid) | |
5596 | break; | |
5597 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5598 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5599 | return 0; | |
5600 | if (found_key.objectid == min_objectid && | |
5601 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5602 | break; | |
5603 | } | |
5604 | return 1; | |
5605 | } |