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 | ||
d352ac68 | 59 | /* this also releases the path */ |
df24a2b9 | 60 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 61 | { |
ff175d57 JJ |
62 | if (!p) |
63 | return; | |
b3b4aa74 | 64 | btrfs_release_path(p); |
df24a2b9 | 65 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
66 | } |
67 | ||
d352ac68 CM |
68 | /* |
69 | * path release drops references on the extent buffers in the path | |
70 | * and it drops any locks held by this path | |
71 | * | |
72 | * It is safe to call this on paths that no locks or extent buffers held. | |
73 | */ | |
b3b4aa74 | 74 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
75 | { |
76 | int i; | |
a2135011 | 77 | |
234b63a0 | 78 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 79 | p->slots[i] = 0; |
eb60ceac | 80 | if (!p->nodes[i]) |
925baedd CM |
81 | continue; |
82 | if (p->locks[i]) { | |
bd681513 | 83 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
84 | p->locks[i] = 0; |
85 | } | |
5f39d397 | 86 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 87 | p->nodes[i] = NULL; |
eb60ceac CM |
88 | } |
89 | } | |
90 | ||
d352ac68 CM |
91 | /* |
92 | * safely gets a reference on the root node of a tree. A lock | |
93 | * is not taken, so a concurrent writer may put a different node | |
94 | * at the root of the tree. See btrfs_lock_root_node for the | |
95 | * looping required. | |
96 | * | |
97 | * The extent buffer returned by this has a reference taken, so | |
98 | * it won't disappear. It may stop being the root of the tree | |
99 | * at any time because there are no locks held. | |
100 | */ | |
925baedd CM |
101 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
102 | { | |
103 | struct extent_buffer *eb; | |
240f62c8 | 104 | |
3083ee2e JB |
105 | while (1) { |
106 | rcu_read_lock(); | |
107 | eb = rcu_dereference(root->node); | |
108 | ||
109 | /* | |
110 | * RCU really hurts here, we could free up the root node because | |
01327610 | 111 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
112 | * the inc_not_zero dance and if it doesn't work then |
113 | * synchronize_rcu and try again. | |
114 | */ | |
115 | if (atomic_inc_not_zero(&eb->refs)) { | |
116 | rcu_read_unlock(); | |
117 | break; | |
118 | } | |
119 | rcu_read_unlock(); | |
120 | synchronize_rcu(); | |
121 | } | |
925baedd CM |
122 | return eb; |
123 | } | |
124 | ||
d352ac68 CM |
125 | /* loop around taking references on and locking the root node of the |
126 | * tree until you end up with a lock on the root. A locked buffer | |
127 | * is returned, with a reference held. | |
128 | */ | |
925baedd CM |
129 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
130 | { | |
131 | struct extent_buffer *eb; | |
132 | ||
d397712b | 133 | while (1) { |
925baedd CM |
134 | eb = btrfs_root_node(root); |
135 | btrfs_tree_lock(eb); | |
240f62c8 | 136 | if (eb == root->node) |
925baedd | 137 | break; |
925baedd CM |
138 | btrfs_tree_unlock(eb); |
139 | free_extent_buffer(eb); | |
140 | } | |
141 | return eb; | |
142 | } | |
143 | ||
bd681513 CM |
144 | /* loop around taking references on and locking the root node of the |
145 | * tree until you end up with a lock on the root. A locked buffer | |
146 | * is returned, with a reference held. | |
147 | */ | |
84f7d8e6 | 148 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
149 | { |
150 | struct extent_buffer *eb; | |
151 | ||
152 | while (1) { | |
153 | eb = btrfs_root_node(root); | |
154 | btrfs_tree_read_lock(eb); | |
155 | if (eb == root->node) | |
156 | break; | |
157 | btrfs_tree_read_unlock(eb); | |
158 | free_extent_buffer(eb); | |
159 | } | |
160 | return eb; | |
161 | } | |
162 | ||
d352ac68 CM |
163 | /* cowonly root (everything not a reference counted cow subvolume), just get |
164 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
165 | * get properly updated on disk. | |
166 | */ | |
0b86a832 CM |
167 | static void add_root_to_dirty_list(struct btrfs_root *root) |
168 | { | |
0b246afa JM |
169 | struct btrfs_fs_info *fs_info = root->fs_info; |
170 | ||
e7070be1 JB |
171 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
172 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
173 | return; | |
174 | ||
0b246afa | 175 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
176 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
177 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 178 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 179 | list_move_tail(&root->dirty_list, |
0b246afa | 180 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
181 | else |
182 | list_move(&root->dirty_list, | |
0b246afa | 183 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 184 | } |
0b246afa | 185 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
186 | } |
187 | ||
d352ac68 CM |
188 | /* |
189 | * used by snapshot creation to make a copy of a root for a tree with | |
190 | * a given objectid. The buffer with the new root node is returned in | |
191 | * cow_ret, and this func returns zero on success or a negative error code. | |
192 | */ | |
be20aa9d CM |
193 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
194 | struct btrfs_root *root, | |
195 | struct extent_buffer *buf, | |
196 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
197 | { | |
0b246afa | 198 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 199 | struct extent_buffer *cow; |
be20aa9d CM |
200 | int ret = 0; |
201 | int level; | |
5d4f98a2 | 202 | struct btrfs_disk_key disk_key; |
be20aa9d | 203 | |
27cdeb70 | 204 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 205 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
206 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
207 | trans->transid != root->last_trans); | |
be20aa9d CM |
208 | |
209 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
210 | if (level == 0) |
211 | btrfs_item_key(buf, &disk_key, 0); | |
212 | else | |
213 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 214 | |
4d75f8a9 DS |
215 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
216 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 217 | if (IS_ERR(cow)) |
be20aa9d CM |
218 | return PTR_ERR(cow); |
219 | ||
58e8012c | 220 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
221 | btrfs_set_header_bytenr(cow, cow->start); |
222 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
223 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
224 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
225 | BTRFS_HEADER_FLAG_RELOC); | |
226 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
227 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
228 | else | |
229 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 230 | |
de37aa51 | 231 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 232 | |
be20aa9d | 233 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 234 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 235 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 236 | else |
e339a6b0 | 237 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 238 | |
be20aa9d CM |
239 | if (ret) |
240 | return ret; | |
241 | ||
242 | btrfs_mark_buffer_dirty(cow); | |
243 | *cow_ret = cow; | |
244 | return 0; | |
245 | } | |
246 | ||
bd989ba3 JS |
247 | enum mod_log_op { |
248 | MOD_LOG_KEY_REPLACE, | |
249 | MOD_LOG_KEY_ADD, | |
250 | MOD_LOG_KEY_REMOVE, | |
251 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
252 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
253 | MOD_LOG_MOVE_KEYS, | |
254 | MOD_LOG_ROOT_REPLACE, | |
255 | }; | |
256 | ||
bd989ba3 JS |
257 | struct tree_mod_root { |
258 | u64 logical; | |
259 | u8 level; | |
260 | }; | |
261 | ||
262 | struct tree_mod_elem { | |
263 | struct rb_node node; | |
298cfd36 | 264 | u64 logical; |
097b8a7c | 265 | u64 seq; |
bd989ba3 JS |
266 | enum mod_log_op op; |
267 | ||
268 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
269 | int slot; | |
270 | ||
271 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
272 | u64 generation; | |
273 | ||
274 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
275 | struct btrfs_disk_key key; | |
276 | u64 blockptr; | |
277 | ||
278 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
279 | struct { |
280 | int dst_slot; | |
281 | int nr_items; | |
282 | } move; | |
bd989ba3 JS |
283 | |
284 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
285 | struct tree_mod_root old_root; | |
286 | }; | |
287 | ||
fc36ed7e | 288 | /* |
fcebe456 | 289 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 290 | */ |
fcebe456 | 291 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
292 | { |
293 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
294 | } | |
295 | ||
097b8a7c JS |
296 | /* |
297 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
298 | * passed does not already have a sequence number set. So when a caller expects | |
299 | * to record tree modifications, it should ensure to set elem->seq to zero | |
300 | * before calling btrfs_get_tree_mod_seq. | |
301 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
302 | * blocker was added. | |
303 | */ | |
304 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
305 | struct seq_list *elem) | |
bd989ba3 | 306 | { |
b1a09f1e | 307 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 308 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 309 | if (!elem->seq) { |
fcebe456 | 310 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
311 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
312 | } | |
bd989ba3 | 313 | spin_unlock(&fs_info->tree_mod_seq_lock); |
b1a09f1e | 314 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 315 | |
fcebe456 | 316 | return elem->seq; |
bd989ba3 JS |
317 | } |
318 | ||
319 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
320 | struct seq_list *elem) | |
321 | { | |
322 | struct rb_root *tm_root; | |
323 | struct rb_node *node; | |
324 | struct rb_node *next; | |
325 | struct seq_list *cur_elem; | |
326 | struct tree_mod_elem *tm; | |
327 | u64 min_seq = (u64)-1; | |
328 | u64 seq_putting = elem->seq; | |
329 | ||
330 | if (!seq_putting) | |
331 | return; | |
332 | ||
bd989ba3 JS |
333 | spin_lock(&fs_info->tree_mod_seq_lock); |
334 | list_del(&elem->list); | |
097b8a7c | 335 | elem->seq = 0; |
bd989ba3 JS |
336 | |
337 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 338 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
339 | if (seq_putting > cur_elem->seq) { |
340 | /* | |
341 | * blocker with lower sequence number exists, we | |
342 | * cannot remove anything from the log | |
343 | */ | |
097b8a7c JS |
344 | spin_unlock(&fs_info->tree_mod_seq_lock); |
345 | return; | |
bd989ba3 JS |
346 | } |
347 | min_seq = cur_elem->seq; | |
348 | } | |
349 | } | |
097b8a7c JS |
350 | spin_unlock(&fs_info->tree_mod_seq_lock); |
351 | ||
bd989ba3 JS |
352 | /* |
353 | * anything that's lower than the lowest existing (read: blocked) | |
354 | * sequence number can be removed from the tree. | |
355 | */ | |
b1a09f1e | 356 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
357 | tm_root = &fs_info->tree_mod_log; |
358 | for (node = rb_first(tm_root); node; node = next) { | |
359 | next = rb_next(node); | |
6b4df8b6 | 360 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 361 | if (tm->seq > min_seq) |
bd989ba3 JS |
362 | continue; |
363 | rb_erase(node, tm_root); | |
bd989ba3 JS |
364 | kfree(tm); |
365 | } | |
b1a09f1e | 366 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
367 | } |
368 | ||
369 | /* | |
370 | * key order of the log: | |
298cfd36 | 371 | * node/leaf start address -> sequence |
bd989ba3 | 372 | * |
298cfd36 CR |
373 | * The 'start address' is the logical address of the *new* root node |
374 | * for root replace operations, or the logical address of the affected | |
375 | * block for all other operations. | |
bd989ba3 JS |
376 | */ |
377 | static noinline int | |
378 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
379 | { | |
380 | struct rb_root *tm_root; | |
381 | struct rb_node **new; | |
382 | struct rb_node *parent = NULL; | |
383 | struct tree_mod_elem *cur; | |
c8cc6341 | 384 | |
73e82fe4 DS |
385 | lockdep_assert_held_write(&fs_info->tree_mod_log_lock); |
386 | ||
fcebe456 | 387 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 388 | |
bd989ba3 JS |
389 | tm_root = &fs_info->tree_mod_log; |
390 | new = &tm_root->rb_node; | |
391 | while (*new) { | |
6b4df8b6 | 392 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 393 | parent = *new; |
298cfd36 | 394 | if (cur->logical < tm->logical) |
bd989ba3 | 395 | new = &((*new)->rb_left); |
298cfd36 | 396 | else if (cur->logical > tm->logical) |
bd989ba3 | 397 | new = &((*new)->rb_right); |
097b8a7c | 398 | else if (cur->seq < tm->seq) |
bd989ba3 | 399 | new = &((*new)->rb_left); |
097b8a7c | 400 | else if (cur->seq > tm->seq) |
bd989ba3 | 401 | new = &((*new)->rb_right); |
5de865ee FDBM |
402 | else |
403 | return -EEXIST; | |
bd989ba3 JS |
404 | } |
405 | ||
406 | rb_link_node(&tm->node, parent, new); | |
407 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 408 | return 0; |
bd989ba3 JS |
409 | } |
410 | ||
097b8a7c JS |
411 | /* |
412 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
413 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
414 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 415 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 416 | */ |
e9b7fd4d JS |
417 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
418 | struct extent_buffer *eb) { | |
419 | smp_mb(); | |
420 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
421 | return 1; | |
097b8a7c JS |
422 | if (eb && btrfs_header_level(eb) == 0) |
423 | return 1; | |
5de865ee | 424 | |
b1a09f1e | 425 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 426 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 427 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
428 | return 1; |
429 | } | |
430 | ||
e9b7fd4d JS |
431 | return 0; |
432 | } | |
433 | ||
5de865ee FDBM |
434 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
435 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
436 | struct extent_buffer *eb) | |
437 | { | |
438 | smp_mb(); | |
439 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
440 | return 0; | |
441 | if (eb && btrfs_header_level(eb) == 0) | |
442 | return 0; | |
443 | ||
444 | return 1; | |
445 | } | |
446 | ||
447 | static struct tree_mod_elem * | |
448 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
449 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 450 | { |
097b8a7c | 451 | struct tree_mod_elem *tm; |
bd989ba3 | 452 | |
c8cc6341 JB |
453 | tm = kzalloc(sizeof(*tm), flags); |
454 | if (!tm) | |
5de865ee | 455 | return NULL; |
bd989ba3 | 456 | |
298cfd36 | 457 | tm->logical = eb->start; |
bd989ba3 JS |
458 | if (op != MOD_LOG_KEY_ADD) { |
459 | btrfs_node_key(eb, &tm->key, slot); | |
460 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
461 | } | |
462 | tm->op = op; | |
463 | tm->slot = slot; | |
464 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 465 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 466 | |
5de865ee | 467 | return tm; |
097b8a7c JS |
468 | } |
469 | ||
e09c2efe DS |
470 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
471 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 472 | { |
5de865ee FDBM |
473 | struct tree_mod_elem *tm; |
474 | int ret; | |
475 | ||
e09c2efe | 476 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
477 | return 0; |
478 | ||
479 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
480 | if (!tm) | |
481 | return -ENOMEM; | |
482 | ||
e09c2efe | 483 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 484 | kfree(tm); |
097b8a7c | 485 | return 0; |
5de865ee FDBM |
486 | } |
487 | ||
e09c2efe | 488 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 489 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
490 | if (ret) |
491 | kfree(tm); | |
097b8a7c | 492 | |
5de865ee | 493 | return ret; |
097b8a7c JS |
494 | } |
495 | ||
6074d45f DS |
496 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
497 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 498 | { |
5de865ee FDBM |
499 | struct tree_mod_elem *tm = NULL; |
500 | struct tree_mod_elem **tm_list = NULL; | |
501 | int ret = 0; | |
bd989ba3 | 502 | int i; |
5de865ee | 503 | int locked = 0; |
bd989ba3 | 504 | |
6074d45f | 505 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 506 | return 0; |
bd989ba3 | 507 | |
176ef8f5 | 508 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
509 | if (!tm_list) |
510 | return -ENOMEM; | |
511 | ||
176ef8f5 | 512 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
513 | if (!tm) { |
514 | ret = -ENOMEM; | |
515 | goto free_tms; | |
516 | } | |
517 | ||
298cfd36 | 518 | tm->logical = eb->start; |
5de865ee FDBM |
519 | tm->slot = src_slot; |
520 | tm->move.dst_slot = dst_slot; | |
521 | tm->move.nr_items = nr_items; | |
522 | tm->op = MOD_LOG_MOVE_KEYS; | |
523 | ||
524 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
525 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 526 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
527 | if (!tm_list[i]) { |
528 | ret = -ENOMEM; | |
529 | goto free_tms; | |
530 | } | |
531 | } | |
532 | ||
6074d45f | 533 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
534 | goto free_tms; |
535 | locked = 1; | |
536 | ||
01763a2e JS |
537 | /* |
538 | * When we override something during the move, we log these removals. | |
539 | * This can only happen when we move towards the beginning of the | |
540 | * buffer, i.e. dst_slot < src_slot. | |
541 | */ | |
bd989ba3 | 542 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 543 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
544 | if (ret) |
545 | goto free_tms; | |
bd989ba3 JS |
546 | } |
547 | ||
6074d45f | 548 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
549 | if (ret) |
550 | goto free_tms; | |
b1a09f1e | 551 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 552 | kfree(tm_list); |
f395694c | 553 | |
5de865ee FDBM |
554 | return 0; |
555 | free_tms: | |
556 | for (i = 0; i < nr_items; i++) { | |
557 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 558 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
559 | kfree(tm_list[i]); |
560 | } | |
561 | if (locked) | |
b1a09f1e | 562 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
563 | kfree(tm_list); |
564 | kfree(tm); | |
bd989ba3 | 565 | |
5de865ee | 566 | return ret; |
bd989ba3 JS |
567 | } |
568 | ||
5de865ee FDBM |
569 | static inline int |
570 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
571 | struct tree_mod_elem **tm_list, | |
572 | int nritems) | |
097b8a7c | 573 | { |
5de865ee | 574 | int i, j; |
097b8a7c JS |
575 | int ret; |
576 | ||
097b8a7c | 577 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
578 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
579 | if (ret) { | |
580 | for (j = nritems - 1; j > i; j--) | |
581 | rb_erase(&tm_list[j]->node, | |
582 | &fs_info->tree_mod_log); | |
583 | return ret; | |
584 | } | |
097b8a7c | 585 | } |
5de865ee FDBM |
586 | |
587 | return 0; | |
097b8a7c JS |
588 | } |
589 | ||
95b757c1 DS |
590 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
591 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 592 | { |
95b757c1 | 593 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
594 | struct tree_mod_elem *tm = NULL; |
595 | struct tree_mod_elem **tm_list = NULL; | |
596 | int nritems = 0; | |
597 | int ret = 0; | |
598 | int i; | |
bd989ba3 | 599 | |
5de865ee | 600 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
601 | return 0; |
602 | ||
5de865ee FDBM |
603 | if (log_removal && btrfs_header_level(old_root) > 0) { |
604 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 605 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 606 | GFP_NOFS); |
5de865ee FDBM |
607 | if (!tm_list) { |
608 | ret = -ENOMEM; | |
609 | goto free_tms; | |
610 | } | |
611 | for (i = 0; i < nritems; i++) { | |
612 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 613 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
614 | if (!tm_list[i]) { |
615 | ret = -ENOMEM; | |
616 | goto free_tms; | |
617 | } | |
618 | } | |
619 | } | |
d9abbf1c | 620 | |
bcc8e07f | 621 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
622 | if (!tm) { |
623 | ret = -ENOMEM; | |
624 | goto free_tms; | |
625 | } | |
bd989ba3 | 626 | |
298cfd36 | 627 | tm->logical = new_root->start; |
bd989ba3 JS |
628 | tm->old_root.logical = old_root->start; |
629 | tm->old_root.level = btrfs_header_level(old_root); | |
630 | tm->generation = btrfs_header_generation(old_root); | |
631 | tm->op = MOD_LOG_ROOT_REPLACE; | |
632 | ||
5de865ee FDBM |
633 | if (tree_mod_dont_log(fs_info, NULL)) |
634 | goto free_tms; | |
635 | ||
636 | if (tm_list) | |
637 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
638 | if (!ret) | |
639 | ret = __tree_mod_log_insert(fs_info, tm); | |
640 | ||
b1a09f1e | 641 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
642 | if (ret) |
643 | goto free_tms; | |
644 | kfree(tm_list); | |
645 | ||
646 | return ret; | |
647 | ||
648 | free_tms: | |
649 | if (tm_list) { | |
650 | for (i = 0; i < nritems; i++) | |
651 | kfree(tm_list[i]); | |
652 | kfree(tm_list); | |
653 | } | |
654 | kfree(tm); | |
655 | ||
656 | return ret; | |
bd989ba3 JS |
657 | } |
658 | ||
659 | static struct tree_mod_elem * | |
660 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
661 | int smallest) | |
662 | { | |
663 | struct rb_root *tm_root; | |
664 | struct rb_node *node; | |
665 | struct tree_mod_elem *cur = NULL; | |
666 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 667 | |
b1a09f1e | 668 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
669 | tm_root = &fs_info->tree_mod_log; |
670 | node = tm_root->rb_node; | |
671 | while (node) { | |
6b4df8b6 | 672 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 673 | if (cur->logical < start) { |
bd989ba3 | 674 | node = node->rb_left; |
298cfd36 | 675 | } else if (cur->logical > start) { |
bd989ba3 | 676 | node = node->rb_right; |
097b8a7c | 677 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
678 | node = node->rb_left; |
679 | } else if (!smallest) { | |
680 | /* we want the node with the highest seq */ | |
681 | if (found) | |
097b8a7c | 682 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
683 | found = cur; |
684 | node = node->rb_left; | |
097b8a7c | 685 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
686 | /* we want the node with the smallest seq */ |
687 | if (found) | |
097b8a7c | 688 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
689 | found = cur; |
690 | node = node->rb_right; | |
691 | } else { | |
692 | found = cur; | |
693 | break; | |
694 | } | |
695 | } | |
b1a09f1e | 696 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
697 | |
698 | return found; | |
699 | } | |
700 | ||
701 | /* | |
702 | * this returns the element from the log with the smallest time sequence | |
703 | * value that's in the log (the oldest log item). any element with a time | |
704 | * sequence lower than min_seq will be ignored. | |
705 | */ | |
706 | static struct tree_mod_elem * | |
707 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
708 | u64 min_seq) | |
709 | { | |
710 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
711 | } | |
712 | ||
713 | /* | |
714 | * this returns the element from the log with the largest time sequence | |
715 | * value that's in the log (the most recent log item). any element with | |
716 | * a time sequence lower than min_seq will be ignored. | |
717 | */ | |
718 | static struct tree_mod_elem * | |
719 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
720 | { | |
721 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
722 | } | |
723 | ||
ed874f0d | 724 | static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, |
bd989ba3 | 725 | struct extent_buffer *src, unsigned long dst_offset, |
90f8d62e | 726 | unsigned long src_offset, int nr_items) |
bd989ba3 | 727 | { |
ed874f0d | 728 | struct btrfs_fs_info *fs_info = dst->fs_info; |
5de865ee FDBM |
729 | int ret = 0; |
730 | struct tree_mod_elem **tm_list = NULL; | |
731 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 732 | int i; |
5de865ee | 733 | int locked = 0; |
bd989ba3 | 734 | |
5de865ee FDBM |
735 | if (!tree_mod_need_log(fs_info, NULL)) |
736 | return 0; | |
bd989ba3 | 737 | |
c8cc6341 | 738 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
739 | return 0; |
740 | ||
31e818fe | 741 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
742 | GFP_NOFS); |
743 | if (!tm_list) | |
744 | return -ENOMEM; | |
bd989ba3 | 745 | |
5de865ee FDBM |
746 | tm_list_add = tm_list; |
747 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 748 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
749 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
750 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
751 | if (!tm_list_rem[i]) { | |
752 | ret = -ENOMEM; | |
753 | goto free_tms; | |
754 | } | |
755 | ||
756 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
757 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
758 | if (!tm_list_add[i]) { | |
759 | ret = -ENOMEM; | |
760 | goto free_tms; | |
761 | } | |
762 | } | |
763 | ||
764 | if (tree_mod_dont_log(fs_info, NULL)) | |
765 | goto free_tms; | |
766 | locked = 1; | |
767 | ||
768 | for (i = 0; i < nr_items; i++) { | |
769 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
770 | if (ret) | |
771 | goto free_tms; | |
772 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
773 | if (ret) | |
774 | goto free_tms; | |
bd989ba3 | 775 | } |
5de865ee | 776 | |
b1a09f1e | 777 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
778 | kfree(tm_list); |
779 | ||
780 | return 0; | |
781 | ||
782 | free_tms: | |
783 | for (i = 0; i < nr_items * 2; i++) { | |
784 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
785 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
786 | kfree(tm_list[i]); | |
787 | } | |
788 | if (locked) | |
b1a09f1e | 789 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
790 | kfree(tm_list); |
791 | ||
792 | return ret; | |
bd989ba3 JS |
793 | } |
794 | ||
db7279a2 | 795 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 796 | { |
5de865ee FDBM |
797 | struct tree_mod_elem **tm_list = NULL; |
798 | int nritems = 0; | |
799 | int i; | |
800 | int ret = 0; | |
801 | ||
802 | if (btrfs_header_level(eb) == 0) | |
803 | return 0; | |
804 | ||
db7279a2 | 805 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
806 | return 0; |
807 | ||
808 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 809 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
810 | if (!tm_list) |
811 | return -ENOMEM; | |
812 | ||
813 | for (i = 0; i < nritems; i++) { | |
814 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
815 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
816 | if (!tm_list[i]) { | |
817 | ret = -ENOMEM; | |
818 | goto free_tms; | |
819 | } | |
820 | } | |
821 | ||
db7279a2 | 822 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
823 | goto free_tms; |
824 | ||
db7279a2 | 825 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 826 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
827 | if (ret) |
828 | goto free_tms; | |
829 | kfree(tm_list); | |
830 | ||
831 | return 0; | |
832 | ||
833 | free_tms: | |
834 | for (i = 0; i < nritems; i++) | |
835 | kfree(tm_list[i]); | |
836 | kfree(tm_list); | |
837 | ||
838 | return ret; | |
bd989ba3 JS |
839 | } |
840 | ||
5d4f98a2 YZ |
841 | /* |
842 | * check if the tree block can be shared by multiple trees | |
843 | */ | |
844 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
845 | struct extent_buffer *buf) | |
846 | { | |
847 | /* | |
01327610 | 848 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
849 | * are never shared. If a block was allocated after the last |
850 | * snapshot and the block was not allocated by tree relocation, | |
851 | * we know the block is not shared. | |
852 | */ | |
27cdeb70 | 853 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
854 | buf != root->node && buf != root->commit_root && |
855 | (btrfs_header_generation(buf) <= | |
856 | btrfs_root_last_snapshot(&root->root_item) || | |
857 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
858 | return 1; | |
a79865c6 | 859 | |
5d4f98a2 YZ |
860 | return 0; |
861 | } | |
862 | ||
863 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
864 | struct btrfs_root *root, | |
865 | struct extent_buffer *buf, | |
f0486c68 YZ |
866 | struct extent_buffer *cow, |
867 | int *last_ref) | |
5d4f98a2 | 868 | { |
0b246afa | 869 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
870 | u64 refs; |
871 | u64 owner; | |
872 | u64 flags; | |
873 | u64 new_flags = 0; | |
874 | int ret; | |
875 | ||
876 | /* | |
877 | * Backrefs update rules: | |
878 | * | |
879 | * Always use full backrefs for extent pointers in tree block | |
880 | * allocated by tree relocation. | |
881 | * | |
882 | * If a shared tree block is no longer referenced by its owner | |
883 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
884 | * use full backrefs for extent pointers in tree block. | |
885 | * | |
886 | * If a tree block is been relocating | |
887 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
888 | * use full backrefs for extent pointers in tree block. | |
889 | * The reason for this is some operations (such as drop tree) | |
890 | * are only allowed for blocks use full backrefs. | |
891 | */ | |
892 | ||
893 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 894 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
895 | btrfs_header_level(buf), 1, |
896 | &refs, &flags); | |
be1a5564 MF |
897 | if (ret) |
898 | return ret; | |
e5df9573 MF |
899 | if (refs == 0) { |
900 | ret = -EROFS; | |
0b246afa | 901 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
902 | return ret; |
903 | } | |
5d4f98a2 YZ |
904 | } else { |
905 | refs = 1; | |
906 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
907 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
908 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
909 | else | |
910 | flags = 0; | |
911 | } | |
912 | ||
913 | owner = btrfs_header_owner(buf); | |
914 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
915 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
916 | ||
917 | if (refs > 1) { | |
918 | if ((owner == root->root_key.objectid || | |
919 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
920 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 921 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
922 | if (ret) |
923 | return ret; | |
5d4f98a2 YZ |
924 | |
925 | if (root->root_key.objectid == | |
926 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 927 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
928 | if (ret) |
929 | return ret; | |
e339a6b0 | 930 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
931 | if (ret) |
932 | return ret; | |
5d4f98a2 YZ |
933 | } |
934 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
935 | } else { | |
936 | ||
937 | if (root->root_key.objectid == | |
938 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 939 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 940 | else |
e339a6b0 | 941 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
942 | if (ret) |
943 | return ret; | |
5d4f98a2 YZ |
944 | } |
945 | if (new_flags != 0) { | |
b1c79e09 JB |
946 | int level = btrfs_header_level(buf); |
947 | ||
f5c8daa5 | 948 | ret = btrfs_set_disk_extent_flags(trans, |
5d4f98a2 YZ |
949 | buf->start, |
950 | buf->len, | |
b1c79e09 | 951 | new_flags, level, 0); |
be1a5564 MF |
952 | if (ret) |
953 | return ret; | |
5d4f98a2 YZ |
954 | } |
955 | } else { | |
956 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
957 | if (root->root_key.objectid == | |
958 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 959 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 960 | else |
e339a6b0 | 961 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
962 | if (ret) |
963 | return ret; | |
e339a6b0 | 964 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
965 | if (ret) |
966 | return ret; | |
5d4f98a2 | 967 | } |
6a884d7d | 968 | btrfs_clean_tree_block(buf); |
f0486c68 | 969 | *last_ref = 1; |
5d4f98a2 YZ |
970 | } |
971 | return 0; | |
972 | } | |
973 | ||
a6279470 FM |
974 | static struct extent_buffer *alloc_tree_block_no_bg_flush( |
975 | struct btrfs_trans_handle *trans, | |
976 | struct btrfs_root *root, | |
977 | u64 parent_start, | |
978 | const struct btrfs_disk_key *disk_key, | |
979 | int level, | |
980 | u64 hint, | |
981 | u64 empty_size) | |
982 | { | |
983 | struct btrfs_fs_info *fs_info = root->fs_info; | |
984 | struct extent_buffer *ret; | |
985 | ||
986 | /* | |
987 | * If we are COWing a node/leaf from the extent, chunk, device or free | |
988 | * space trees, make sure that we do not finish block group creation of | |
989 | * pending block groups. We do this to avoid a deadlock. | |
990 | * COWing can result in allocation of a new chunk, and flushing pending | |
991 | * block groups (btrfs_create_pending_block_groups()) can be triggered | |
992 | * when finishing allocation of a new chunk. Creation of a pending block | |
993 | * group modifies the extent, chunk, device and free space trees, | |
994 | * therefore we could deadlock with ourselves since we are holding a | |
995 | * lock on an extent buffer that btrfs_create_pending_block_groups() may | |
996 | * try to COW later. | |
997 | * For similar reasons, we also need to delay flushing pending block | |
998 | * groups when splitting a leaf or node, from one of those trees, since | |
999 | * we are holding a write lock on it and its parent or when inserting a | |
1000 | * new root node for one of those trees. | |
1001 | */ | |
1002 | if (root == fs_info->extent_root || | |
1003 | root == fs_info->chunk_root || | |
1004 | root == fs_info->dev_root || | |
1005 | root == fs_info->free_space_root) | |
1006 | trans->can_flush_pending_bgs = false; | |
1007 | ||
1008 | ret = btrfs_alloc_tree_block(trans, root, parent_start, | |
1009 | root->root_key.objectid, disk_key, level, | |
1010 | hint, empty_size); | |
1011 | trans->can_flush_pending_bgs = true; | |
1012 | ||
1013 | return ret; | |
1014 | } | |
1015 | ||
d352ac68 | 1016 | /* |
d397712b CM |
1017 | * does the dirty work in cow of a single block. The parent block (if |
1018 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1019 | * dirty and returned locked. If you modify the block it needs to be marked | |
1020 | * dirty again. | |
d352ac68 CM |
1021 | * |
1022 | * search_start -- an allocation hint for the new block | |
1023 | * | |
d397712b CM |
1024 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1025 | * bytes the allocator should try to find free next to the block it returns. | |
1026 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1027 | */ |
d397712b | 1028 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1029 | struct btrfs_root *root, |
1030 | struct extent_buffer *buf, | |
1031 | struct extent_buffer *parent, int parent_slot, | |
1032 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1033 | u64 search_start, u64 empty_size) |
02217ed2 | 1034 | { |
0b246afa | 1035 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1036 | struct btrfs_disk_key disk_key; |
5f39d397 | 1037 | struct extent_buffer *cow; |
be1a5564 | 1038 | int level, ret; |
f0486c68 | 1039 | int last_ref = 0; |
925baedd | 1040 | int unlock_orig = 0; |
0f5053eb | 1041 | u64 parent_start = 0; |
7bb86316 | 1042 | |
925baedd CM |
1043 | if (*cow_ret == buf) |
1044 | unlock_orig = 1; | |
1045 | ||
b9447ef8 | 1046 | btrfs_assert_tree_locked(buf); |
925baedd | 1047 | |
27cdeb70 | 1048 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1049 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1050 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1051 | trans->transid != root->last_trans); | |
5f39d397 | 1052 | |
7bb86316 | 1053 | level = btrfs_header_level(buf); |
31840ae1 | 1054 | |
5d4f98a2 YZ |
1055 | if (level == 0) |
1056 | btrfs_item_key(buf, &disk_key, 0); | |
1057 | else | |
1058 | btrfs_node_key(buf, &disk_key, 0); | |
1059 | ||
0f5053eb GR |
1060 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1061 | parent_start = parent->start; | |
5d4f98a2 | 1062 | |
a6279470 FM |
1063 | cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, |
1064 | level, search_start, empty_size); | |
54aa1f4d CM |
1065 | if (IS_ERR(cow)) |
1066 | return PTR_ERR(cow); | |
6702ed49 | 1067 | |
b4ce94de CM |
1068 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1069 | ||
58e8012c | 1070 | copy_extent_buffer_full(cow, buf); |
db94535d | 1071 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1072 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1073 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1074 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1075 | BTRFS_HEADER_FLAG_RELOC); | |
1076 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1077 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1078 | else | |
1079 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1080 | |
de37aa51 | 1081 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 1082 | |
be1a5564 | 1083 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1084 | if (ret) { |
66642832 | 1085 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1086 | return ret; |
1087 | } | |
1a40e23b | 1088 | |
27cdeb70 | 1089 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1090 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1091 | if (ret) { |
66642832 | 1092 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1093 | return ret; |
93314e3b | 1094 | } |
83d4cfd4 | 1095 | } |
3fd0a558 | 1096 | |
02217ed2 | 1097 | if (buf == root->node) { |
925baedd | 1098 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1099 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1100 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1101 | parent_start = buf->start; | |
925baedd | 1102 | |
5f39d397 | 1103 | extent_buffer_get(cow); |
d9d19a01 DS |
1104 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1105 | BUG_ON(ret < 0); | |
240f62c8 | 1106 | rcu_assign_pointer(root->node, cow); |
925baedd | 1107 | |
f0486c68 | 1108 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1109 | last_ref); |
5f39d397 | 1110 | free_extent_buffer(buf); |
0b86a832 | 1111 | add_root_to_dirty_list(root); |
02217ed2 | 1112 | } else { |
5d4f98a2 | 1113 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1114 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1115 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1116 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1117 | cow->start); |
74493f7a CM |
1118 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1119 | trans->transid); | |
d6025579 | 1120 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1121 | if (last_ref) { |
db7279a2 | 1122 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1123 | if (ret) { |
66642832 | 1124 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1125 | return ret; |
1126 | } | |
1127 | } | |
f0486c68 | 1128 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1129 | last_ref); |
02217ed2 | 1130 | } |
925baedd CM |
1131 | if (unlock_orig) |
1132 | btrfs_tree_unlock(buf); | |
3083ee2e | 1133 | free_extent_buffer_stale(buf); |
ccd467d6 | 1134 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1135 | *cow_ret = cow; |
02217ed2 CM |
1136 | return 0; |
1137 | } | |
1138 | ||
5d9e75c4 JS |
1139 | /* |
1140 | * returns the logical address of the oldest predecessor of the given root. | |
1141 | * entries older than time_seq are ignored. | |
1142 | */ | |
bcd24dab DS |
1143 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1144 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1145 | { |
1146 | struct tree_mod_elem *tm; | |
1147 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1148 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1149 | int looped = 0; |
1150 | ||
1151 | if (!time_seq) | |
35a3621b | 1152 | return NULL; |
5d9e75c4 JS |
1153 | |
1154 | /* | |
298cfd36 CR |
1155 | * the very last operation that's logged for a root is the |
1156 | * replacement operation (if it is replaced at all). this has | |
1157 | * the logical address of the *new* root, making it the very | |
1158 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1159 | */ |
1160 | while (1) { | |
bcd24dab | 1161 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1162 | time_seq); |
1163 | if (!looped && !tm) | |
35a3621b | 1164 | return NULL; |
5d9e75c4 | 1165 | /* |
28da9fb4 JS |
1166 | * if there are no tree operation for the oldest root, we simply |
1167 | * return it. this should only happen if that (old) root is at | |
1168 | * level 0. | |
5d9e75c4 | 1169 | */ |
28da9fb4 JS |
1170 | if (!tm) |
1171 | break; | |
5d9e75c4 | 1172 | |
28da9fb4 JS |
1173 | /* |
1174 | * if there's an operation that's not a root replacement, we | |
1175 | * found the oldest version of our root. normally, we'll find a | |
1176 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1177 | */ | |
5d9e75c4 JS |
1178 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1179 | break; | |
1180 | ||
1181 | found = tm; | |
1182 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1183 | looped = 1; |
1184 | } | |
1185 | ||
a95236d9 JS |
1186 | /* if there's no old root to return, return what we found instead */ |
1187 | if (!found) | |
1188 | found = tm; | |
1189 | ||
5d9e75c4 JS |
1190 | return found; |
1191 | } | |
1192 | ||
1193 | /* | |
1194 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1195 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1196 | * time_seq). |
1197 | */ | |
1198 | static void | |
f1ca7e98 JB |
1199 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1200 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1201 | { |
1202 | u32 n; | |
1203 | struct rb_node *next; | |
1204 | struct tree_mod_elem *tm = first_tm; | |
1205 | unsigned long o_dst; | |
1206 | unsigned long o_src; | |
1207 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1208 | ||
1209 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1210 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1211 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1212 | /* |
1213 | * all the operations are recorded with the operator used for | |
1214 | * the modification. as we're going backwards, we do the | |
1215 | * opposite of each operation here. | |
1216 | */ | |
1217 | switch (tm->op) { | |
1218 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1219 | BUG_ON(tm->slot < n); | |
1c697d4a | 1220 | /* Fallthrough */ |
95c80bb1 | 1221 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1222 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1223 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1224 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1225 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1226 | tm->generation); | |
4c3e6969 | 1227 | n++; |
5d9e75c4 JS |
1228 | break; |
1229 | case MOD_LOG_KEY_REPLACE: | |
1230 | BUG_ON(tm->slot >= n); | |
1231 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1232 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1233 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1234 | tm->generation); | |
1235 | break; | |
1236 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1237 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1238 | n--; |
1239 | break; | |
1240 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1241 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1242 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1243 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1244 | tm->move.nr_items * p_size); |
1245 | break; | |
1246 | case MOD_LOG_ROOT_REPLACE: | |
1247 | /* | |
1248 | * this operation is special. for roots, this must be | |
1249 | * handled explicitly before rewinding. | |
1250 | * for non-roots, this operation may exist if the node | |
1251 | * was a root: root A -> child B; then A gets empty and | |
1252 | * B is promoted to the new root. in the mod log, we'll | |
1253 | * have a root-replace operation for B, a tree block | |
1254 | * that is no root. we simply ignore that operation. | |
1255 | */ | |
1256 | break; | |
1257 | } | |
1258 | next = rb_next(&tm->node); | |
1259 | if (!next) | |
1260 | break; | |
6b4df8b6 | 1261 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1262 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1263 | break; |
1264 | } | |
b1a09f1e | 1265 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1266 | btrfs_set_header_nritems(eb, n); |
1267 | } | |
1268 | ||
47fb091f | 1269 | /* |
01327610 | 1270 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1271 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1272 | * returned buffer is always read-locked. If the returned buffer is not the | |
1273 | * input buffer, the lock on the input buffer is released and the input buffer | |
1274 | * is freed (its refcount is decremented). | |
1275 | */ | |
5d9e75c4 | 1276 | static struct extent_buffer * |
9ec72677 JB |
1277 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1278 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1279 | { |
1280 | struct extent_buffer *eb_rewin; | |
1281 | struct tree_mod_elem *tm; | |
1282 | ||
1283 | if (!time_seq) | |
1284 | return eb; | |
1285 | ||
1286 | if (btrfs_header_level(eb) == 0) | |
1287 | return eb; | |
1288 | ||
1289 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1290 | if (!tm) | |
1291 | return eb; | |
1292 | ||
9ec72677 | 1293 | btrfs_set_path_blocking(path); |
300aa896 | 1294 | btrfs_set_lock_blocking_read(eb); |
9ec72677 | 1295 | |
5d9e75c4 JS |
1296 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1297 | BUG_ON(tm->slot != 0); | |
da17066c | 1298 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1299 | if (!eb_rewin) { |
9ec72677 | 1300 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1301 | free_extent_buffer(eb); |
1302 | return NULL; | |
1303 | } | |
5d9e75c4 JS |
1304 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1305 | btrfs_set_header_backref_rev(eb_rewin, | |
1306 | btrfs_header_backref_rev(eb)); | |
1307 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1308 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1309 | } else { |
1310 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1311 | if (!eb_rewin) { |
9ec72677 | 1312 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1313 | free_extent_buffer(eb); |
1314 | return NULL; | |
1315 | } | |
5d9e75c4 JS |
1316 | } |
1317 | ||
9ec72677 | 1318 | btrfs_tree_read_unlock_blocking(eb); |
5d9e75c4 JS |
1319 | free_extent_buffer(eb); |
1320 | ||
47fb091f | 1321 | btrfs_tree_read_lock(eb_rewin); |
f1ca7e98 | 1322 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1323 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1324 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1325 | |
1326 | return eb_rewin; | |
1327 | } | |
1328 | ||
8ba97a15 JS |
1329 | /* |
1330 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1331 | * value. If there are no changes, the current root->root_node is returned. If | |
1332 | * anything changed in between, there's a fresh buffer allocated on which the | |
1333 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1334 | * Returns NULL on error (with no locks held). | |
1335 | */ | |
5d9e75c4 JS |
1336 | static inline struct extent_buffer * |
1337 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1338 | { | |
0b246afa | 1339 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1340 | struct tree_mod_elem *tm; |
30b0463a JS |
1341 | struct extent_buffer *eb = NULL; |
1342 | struct extent_buffer *eb_root; | |
efad8a85 | 1343 | u64 eb_root_owner = 0; |
7bfdcf7f | 1344 | struct extent_buffer *old; |
a95236d9 | 1345 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1346 | u64 old_generation = 0; |
a95236d9 | 1347 | u64 logical; |
581c1760 | 1348 | int level; |
5d9e75c4 | 1349 | |
30b0463a | 1350 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1351 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1352 | if (!tm) |
30b0463a | 1353 | return eb_root; |
5d9e75c4 | 1354 | |
a95236d9 JS |
1355 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1356 | old_root = &tm->old_root; | |
1357 | old_generation = tm->generation; | |
1358 | logical = old_root->logical; | |
581c1760 | 1359 | level = old_root->level; |
a95236d9 | 1360 | } else { |
30b0463a | 1361 | logical = eb_root->start; |
581c1760 | 1362 | level = btrfs_header_level(eb_root); |
a95236d9 | 1363 | } |
5d9e75c4 | 1364 | |
0b246afa | 1365 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1366 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1367 | btrfs_tree_read_unlock(eb_root); |
1368 | free_extent_buffer(eb_root); | |
581c1760 | 1369 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1370 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1371 | if (!IS_ERR(old)) | |
1372 | free_extent_buffer(old); | |
0b246afa JM |
1373 | btrfs_warn(fs_info, |
1374 | "failed to read tree block %llu from get_old_root", | |
1375 | logical); | |
834328a8 | 1376 | } else { |
7bfdcf7f LB |
1377 | eb = btrfs_clone_extent_buffer(old); |
1378 | free_extent_buffer(old); | |
834328a8 JS |
1379 | } |
1380 | } else if (old_root) { | |
efad8a85 | 1381 | eb_root_owner = btrfs_header_owner(eb_root); |
30b0463a JS |
1382 | btrfs_tree_read_unlock(eb_root); |
1383 | free_extent_buffer(eb_root); | |
0b246afa | 1384 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1385 | } else { |
300aa896 | 1386 | btrfs_set_lock_blocking_read(eb_root); |
30b0463a | 1387 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1388 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1389 | free_extent_buffer(eb_root); |
834328a8 JS |
1390 | } |
1391 | ||
8ba97a15 JS |
1392 | if (!eb) |
1393 | return NULL; | |
1394 | btrfs_tree_read_lock(eb); | |
a95236d9 | 1395 | if (old_root) { |
5d9e75c4 JS |
1396 | btrfs_set_header_bytenr(eb, eb->start); |
1397 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
efad8a85 | 1398 | btrfs_set_header_owner(eb, eb_root_owner); |
a95236d9 JS |
1399 | btrfs_set_header_level(eb, old_root->level); |
1400 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1401 | } |
28da9fb4 | 1402 | if (tm) |
0b246afa | 1403 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1404 | else |
1405 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1406 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1407 | |
1408 | return eb; | |
1409 | } | |
1410 | ||
5b6602e7 JS |
1411 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1412 | { | |
1413 | struct tree_mod_elem *tm; | |
1414 | int level; | |
30b0463a | 1415 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1416 | |
bcd24dab | 1417 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1418 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1419 | level = tm->old_root.level; | |
1420 | } else { | |
30b0463a | 1421 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1422 | } |
30b0463a | 1423 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1424 | |
1425 | return level; | |
1426 | } | |
1427 | ||
5d4f98a2 YZ |
1428 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1429 | struct btrfs_root *root, | |
1430 | struct extent_buffer *buf) | |
1431 | { | |
f5ee5c9a | 1432 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1433 | return 0; |
fccb84c9 | 1434 | |
d1980131 DS |
1435 | /* Ensure we can see the FORCE_COW bit */ |
1436 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1437 | |
1438 | /* | |
1439 | * We do not need to cow a block if | |
1440 | * 1) this block is not created or changed in this transaction; | |
1441 | * 2) this block does not belong to TREE_RELOC tree; | |
1442 | * 3) the root is not forced COW. | |
1443 | * | |
1444 | * What is forced COW: | |
01327610 | 1445 | * when we create snapshot during committing the transaction, |
52042d8e | 1446 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
1447 | * block to ensure the metadata consistency. |
1448 | */ | |
5d4f98a2 YZ |
1449 | if (btrfs_header_generation(buf) == trans->transid && |
1450 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1451 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1452 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1453 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1454 | return 0; |
1455 | return 1; | |
1456 | } | |
1457 | ||
d352ac68 CM |
1458 | /* |
1459 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1460 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1461 | * once per transaction, as long as it hasn't been written yet |
1462 | */ | |
d397712b | 1463 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1464 | struct btrfs_root *root, struct extent_buffer *buf, |
1465 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1466 | struct extent_buffer **cow_ret) |
6702ed49 | 1467 | { |
0b246afa | 1468 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1469 | u64 search_start; |
f510cfec | 1470 | int ret; |
dc17ff8f | 1471 | |
83354f07 JB |
1472 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
1473 | btrfs_err(fs_info, | |
1474 | "COW'ing blocks on a fs root that's being dropped"); | |
1475 | ||
0b246afa | 1476 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1477 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1478 | trans->transid, |
0b246afa | 1479 | fs_info->running_transaction->transid); |
31b1a2bd | 1480 | |
0b246afa | 1481 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1482 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1483 | trans->transid, fs_info->generation); |
dc17ff8f | 1484 | |
5d4f98a2 | 1485 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1486 | trans->dirty = true; |
6702ed49 CM |
1487 | *cow_ret = buf; |
1488 | return 0; | |
1489 | } | |
c487685d | 1490 | |
ee22184b | 1491 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1492 | |
1493 | if (parent) | |
8bead258 DS |
1494 | btrfs_set_lock_blocking_write(parent); |
1495 | btrfs_set_lock_blocking_write(buf); | |
b4ce94de | 1496 | |
f616f5cd QW |
1497 | /* |
1498 | * Before CoWing this block for later modification, check if it's | |
1499 | * the subtree root and do the delayed subtree trace if needed. | |
1500 | * | |
1501 | * Also We don't care about the error, as it's handled internally. | |
1502 | */ | |
1503 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 1504 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1505 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1506 | |
1507 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1508 | ||
f510cfec | 1509 | return ret; |
6702ed49 CM |
1510 | } |
1511 | ||
d352ac68 CM |
1512 | /* |
1513 | * helper function for defrag to decide if two blocks pointed to by a | |
1514 | * node are actually close by | |
1515 | */ | |
6b80053d | 1516 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1517 | { |
6b80053d | 1518 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1519 | return 1; |
6b80053d | 1520 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1521 | return 1; |
1522 | return 0; | |
1523 | } | |
1524 | ||
081e9573 CM |
1525 | /* |
1526 | * compare two keys in a memcmp fashion | |
1527 | */ | |
310712b2 OS |
1528 | static int comp_keys(const struct btrfs_disk_key *disk, |
1529 | const struct btrfs_key *k2) | |
081e9573 CM |
1530 | { |
1531 | struct btrfs_key k1; | |
1532 | ||
1533 | btrfs_disk_key_to_cpu(&k1, disk); | |
1534 | ||
20736aba | 1535 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1536 | } |
1537 | ||
f3465ca4 JB |
1538 | /* |
1539 | * same as comp_keys only with two btrfs_key's | |
1540 | */ | |
e1f60a65 | 1541 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1542 | { |
1543 | if (k1->objectid > k2->objectid) | |
1544 | return 1; | |
1545 | if (k1->objectid < k2->objectid) | |
1546 | return -1; | |
1547 | if (k1->type > k2->type) | |
1548 | return 1; | |
1549 | if (k1->type < k2->type) | |
1550 | return -1; | |
1551 | if (k1->offset > k2->offset) | |
1552 | return 1; | |
1553 | if (k1->offset < k2->offset) | |
1554 | return -1; | |
1555 | return 0; | |
1556 | } | |
081e9573 | 1557 | |
d352ac68 CM |
1558 | /* |
1559 | * this is used by the defrag code to go through all the | |
1560 | * leaves pointed to by a node and reallocate them so that | |
1561 | * disk order is close to key order | |
1562 | */ | |
6702ed49 | 1563 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1564 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1565 | int start_slot, u64 *last_ret, |
a6b6e75e | 1566 | struct btrfs_key *progress) |
6702ed49 | 1567 | { |
0b246afa | 1568 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1569 | struct extent_buffer *cur; |
6702ed49 | 1570 | u64 blocknr; |
ca7a79ad | 1571 | u64 gen; |
e9d0b13b CM |
1572 | u64 search_start = *last_ret; |
1573 | u64 last_block = 0; | |
6702ed49 CM |
1574 | u64 other; |
1575 | u32 parent_nritems; | |
6702ed49 CM |
1576 | int end_slot; |
1577 | int i; | |
1578 | int err = 0; | |
f2183bde | 1579 | int parent_level; |
6b80053d CM |
1580 | int uptodate; |
1581 | u32 blocksize; | |
081e9573 CM |
1582 | int progress_passed = 0; |
1583 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1584 | |
5708b959 | 1585 | parent_level = btrfs_header_level(parent); |
5708b959 | 1586 | |
0b246afa JM |
1587 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1588 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1589 | |
6b80053d | 1590 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1591 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1592 | end_slot = parent_nritems - 1; |
6702ed49 | 1593 | |
5dfe2be7 | 1594 | if (parent_nritems <= 1) |
6702ed49 CM |
1595 | return 0; |
1596 | ||
8bead258 | 1597 | btrfs_set_lock_blocking_write(parent); |
b4ce94de | 1598 | |
5dfe2be7 | 1599 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1600 | struct btrfs_key first_key; |
6702ed49 | 1601 | int close = 1; |
a6b6e75e | 1602 | |
081e9573 CM |
1603 | btrfs_node_key(parent, &disk_key, i); |
1604 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1605 | continue; | |
1606 | ||
1607 | progress_passed = 1; | |
6b80053d | 1608 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1609 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1610 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1611 | if (last_block == 0) |
1612 | last_block = blocknr; | |
5708b959 | 1613 | |
6702ed49 | 1614 | if (i > 0) { |
6b80053d CM |
1615 | other = btrfs_node_blockptr(parent, i - 1); |
1616 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1617 | } |
5dfe2be7 | 1618 | if (!close && i < end_slot) { |
6b80053d CM |
1619 | other = btrfs_node_blockptr(parent, i + 1); |
1620 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1621 | } |
e9d0b13b CM |
1622 | if (close) { |
1623 | last_block = blocknr; | |
6702ed49 | 1624 | continue; |
e9d0b13b | 1625 | } |
6702ed49 | 1626 | |
0b246afa | 1627 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1628 | if (cur) |
b9fab919 | 1629 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1630 | else |
1631 | uptodate = 0; | |
5708b959 | 1632 | if (!cur || !uptodate) { |
6b80053d | 1633 | if (!cur) { |
581c1760 QW |
1634 | cur = read_tree_block(fs_info, blocknr, gen, |
1635 | parent_level - 1, | |
1636 | &first_key); | |
64c043de LB |
1637 | if (IS_ERR(cur)) { |
1638 | return PTR_ERR(cur); | |
1639 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1640 | free_extent_buffer(cur); |
97d9a8a4 | 1641 | return -EIO; |
416bc658 | 1642 | } |
6b80053d | 1643 | } else if (!uptodate) { |
581c1760 QW |
1644 | err = btrfs_read_buffer(cur, gen, |
1645 | parent_level - 1,&first_key); | |
018642a1 TI |
1646 | if (err) { |
1647 | free_extent_buffer(cur); | |
1648 | return err; | |
1649 | } | |
f2183bde | 1650 | } |
6702ed49 | 1651 | } |
e9d0b13b | 1652 | if (search_start == 0) |
6b80053d | 1653 | search_start = last_block; |
e9d0b13b | 1654 | |
e7a84565 | 1655 | btrfs_tree_lock(cur); |
8bead258 | 1656 | btrfs_set_lock_blocking_write(cur); |
6b80053d | 1657 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1658 | &cur, search_start, |
6b80053d | 1659 | min(16 * blocksize, |
9fa8cfe7 | 1660 | (end_slot - i) * blocksize)); |
252c38f0 | 1661 | if (err) { |
e7a84565 | 1662 | btrfs_tree_unlock(cur); |
6b80053d | 1663 | free_extent_buffer(cur); |
6702ed49 | 1664 | break; |
252c38f0 | 1665 | } |
e7a84565 CM |
1666 | search_start = cur->start; |
1667 | last_block = cur->start; | |
f2183bde | 1668 | *last_ret = search_start; |
e7a84565 CM |
1669 | btrfs_tree_unlock(cur); |
1670 | free_extent_buffer(cur); | |
6702ed49 CM |
1671 | } |
1672 | return err; | |
1673 | } | |
1674 | ||
74123bd7 | 1675 | /* |
5f39d397 CM |
1676 | * search for key in the extent_buffer. The items start at offset p, |
1677 | * and they are item_size apart. There are 'max' items in p. | |
1678 | * | |
74123bd7 CM |
1679 | * the slot in the array is returned via slot, and it points to |
1680 | * the place where you would insert key if it is not found in | |
1681 | * the array. | |
1682 | * | |
1683 | * slot may point to max if the key is bigger than all of the keys | |
1684 | */ | |
e02119d5 | 1685 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1686 | unsigned long p, int item_size, |
1687 | const struct btrfs_key *key, | |
e02119d5 | 1688 | int max, int *slot) |
be0e5c09 CM |
1689 | { |
1690 | int low = 0; | |
1691 | int high = max; | |
1692 | int mid; | |
1693 | int ret; | |
479965d6 | 1694 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1695 | struct btrfs_disk_key unaligned; |
1696 | unsigned long offset; | |
5f39d397 CM |
1697 | char *kaddr = NULL; |
1698 | unsigned long map_start = 0; | |
1699 | unsigned long map_len = 0; | |
479965d6 | 1700 | int err; |
be0e5c09 | 1701 | |
5e24e9af LB |
1702 | if (low > high) { |
1703 | btrfs_err(eb->fs_info, | |
1704 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1705 | __func__, low, high, eb->start, | |
1706 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1707 | return -EINVAL; | |
1708 | } | |
1709 | ||
d397712b | 1710 | while (low < high) { |
be0e5c09 | 1711 | mid = (low + high) / 2; |
5f39d397 CM |
1712 | offset = p + mid * item_size; |
1713 | ||
a6591715 | 1714 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1715 | (offset + sizeof(struct btrfs_disk_key)) > |
1716 | map_start + map_len) { | |
934d375b CM |
1717 | |
1718 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1719 | sizeof(struct btrfs_disk_key), |
a6591715 | 1720 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1721 | |
1722 | if (!err) { | |
1723 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1724 | map_start); | |
415b35a5 | 1725 | } else if (err == 1) { |
479965d6 CM |
1726 | read_extent_buffer(eb, &unaligned, |
1727 | offset, sizeof(unaligned)); | |
1728 | tmp = &unaligned; | |
415b35a5 LB |
1729 | } else { |
1730 | return err; | |
479965d6 | 1731 | } |
5f39d397 | 1732 | |
5f39d397 CM |
1733 | } else { |
1734 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1735 | map_start); | |
1736 | } | |
be0e5c09 CM |
1737 | ret = comp_keys(tmp, key); |
1738 | ||
1739 | if (ret < 0) | |
1740 | low = mid + 1; | |
1741 | else if (ret > 0) | |
1742 | high = mid; | |
1743 | else { | |
1744 | *slot = mid; | |
1745 | return 0; | |
1746 | } | |
1747 | } | |
1748 | *slot = low; | |
1749 | return 1; | |
1750 | } | |
1751 | ||
97571fd0 CM |
1752 | /* |
1753 | * simple bin_search frontend that does the right thing for | |
1754 | * leaves vs nodes | |
1755 | */ | |
a74b35ec NB |
1756 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1757 | int level, int *slot) | |
be0e5c09 | 1758 | { |
f775738f | 1759 | if (level == 0) |
5f39d397 CM |
1760 | return generic_bin_search(eb, |
1761 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1762 | sizeof(struct btrfs_item), |
5f39d397 | 1763 | key, btrfs_header_nritems(eb), |
7518a238 | 1764 | slot); |
f775738f | 1765 | else |
5f39d397 CM |
1766 | return generic_bin_search(eb, |
1767 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1768 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1769 | key, btrfs_header_nritems(eb), |
7518a238 | 1770 | slot); |
be0e5c09 CM |
1771 | } |
1772 | ||
f0486c68 YZ |
1773 | static void root_add_used(struct btrfs_root *root, u32 size) |
1774 | { | |
1775 | spin_lock(&root->accounting_lock); | |
1776 | btrfs_set_root_used(&root->root_item, | |
1777 | btrfs_root_used(&root->root_item) + size); | |
1778 | spin_unlock(&root->accounting_lock); | |
1779 | } | |
1780 | ||
1781 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1782 | { | |
1783 | spin_lock(&root->accounting_lock); | |
1784 | btrfs_set_root_used(&root->root_item, | |
1785 | btrfs_root_used(&root->root_item) - size); | |
1786 | spin_unlock(&root->accounting_lock); | |
1787 | } | |
1788 | ||
d352ac68 CM |
1789 | /* given a node and slot number, this reads the blocks it points to. The |
1790 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1791 | */ |
4b231ae4 DS |
1792 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
1793 | int slot) | |
bb803951 | 1794 | { |
ca7a79ad | 1795 | int level = btrfs_header_level(parent); |
416bc658 | 1796 | struct extent_buffer *eb; |
581c1760 | 1797 | struct btrfs_key first_key; |
416bc658 | 1798 | |
fb770ae4 LB |
1799 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1800 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1801 | |
1802 | BUG_ON(level == 0); | |
1803 | ||
581c1760 | 1804 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 1805 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1806 | btrfs_node_ptr_generation(parent, slot), |
1807 | level - 1, &first_key); | |
fb770ae4 LB |
1808 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1809 | free_extent_buffer(eb); | |
1810 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1811 | } |
1812 | ||
1813 | return eb; | |
bb803951 CM |
1814 | } |
1815 | ||
d352ac68 CM |
1816 | /* |
1817 | * node level balancing, used to make sure nodes are in proper order for | |
1818 | * item deletion. We balance from the top down, so we have to make sure | |
1819 | * that a deletion won't leave an node completely empty later on. | |
1820 | */ | |
e02119d5 | 1821 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1822 | struct btrfs_root *root, |
1823 | struct btrfs_path *path, int level) | |
bb803951 | 1824 | { |
0b246afa | 1825 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1826 | struct extent_buffer *right = NULL; |
1827 | struct extent_buffer *mid; | |
1828 | struct extent_buffer *left = NULL; | |
1829 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1830 | int ret = 0; |
1831 | int wret; | |
1832 | int pslot; | |
bb803951 | 1833 | int orig_slot = path->slots[level]; |
79f95c82 | 1834 | u64 orig_ptr; |
bb803951 | 1835 | |
98e6b1eb | 1836 | ASSERT(level > 0); |
bb803951 | 1837 | |
5f39d397 | 1838 | mid = path->nodes[level]; |
b4ce94de | 1839 | |
bd681513 CM |
1840 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1841 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1842 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1843 | ||
1d4f8a0c | 1844 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1845 | |
a05a9bb1 | 1846 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1847 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1848 | pslot = path->slots[level + 1]; |
1849 | } | |
bb803951 | 1850 | |
40689478 CM |
1851 | /* |
1852 | * deal with the case where there is only one pointer in the root | |
1853 | * by promoting the node below to a root | |
1854 | */ | |
5f39d397 CM |
1855 | if (!parent) { |
1856 | struct extent_buffer *child; | |
bb803951 | 1857 | |
5f39d397 | 1858 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1859 | return 0; |
1860 | ||
1861 | /* promote the child to a root */ | |
4b231ae4 | 1862 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
1863 | if (IS_ERR(child)) { |
1864 | ret = PTR_ERR(child); | |
0b246afa | 1865 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1866 | goto enospc; |
1867 | } | |
1868 | ||
925baedd | 1869 | btrfs_tree_lock(child); |
8bead258 | 1870 | btrfs_set_lock_blocking_write(child); |
9fa8cfe7 | 1871 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1872 | if (ret) { |
1873 | btrfs_tree_unlock(child); | |
1874 | free_extent_buffer(child); | |
1875 | goto enospc; | |
1876 | } | |
2f375ab9 | 1877 | |
d9d19a01 DS |
1878 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1879 | BUG_ON(ret < 0); | |
240f62c8 | 1880 | rcu_assign_pointer(root->node, child); |
925baedd | 1881 | |
0b86a832 | 1882 | add_root_to_dirty_list(root); |
925baedd | 1883 | btrfs_tree_unlock(child); |
b4ce94de | 1884 | |
925baedd | 1885 | path->locks[level] = 0; |
bb803951 | 1886 | path->nodes[level] = NULL; |
6a884d7d | 1887 | btrfs_clean_tree_block(mid); |
925baedd | 1888 | btrfs_tree_unlock(mid); |
bb803951 | 1889 | /* once for the path */ |
5f39d397 | 1890 | free_extent_buffer(mid); |
f0486c68 YZ |
1891 | |
1892 | root_sub_used(root, mid->len); | |
5581a51a | 1893 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1894 | /* once for the root ptr */ |
3083ee2e | 1895 | free_extent_buffer_stale(mid); |
f0486c68 | 1896 | return 0; |
bb803951 | 1897 | } |
5f39d397 | 1898 | if (btrfs_header_nritems(mid) > |
0b246afa | 1899 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1900 | return 0; |
1901 | ||
4b231ae4 | 1902 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1903 | if (IS_ERR(left)) |
1904 | left = NULL; | |
1905 | ||
5f39d397 | 1906 | if (left) { |
925baedd | 1907 | btrfs_tree_lock(left); |
8bead258 | 1908 | btrfs_set_lock_blocking_write(left); |
5f39d397 | 1909 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1910 | parent, pslot - 1, &left); |
54aa1f4d CM |
1911 | if (wret) { |
1912 | ret = wret; | |
1913 | goto enospc; | |
1914 | } | |
2cc58cf2 | 1915 | } |
fb770ae4 | 1916 | |
4b231ae4 | 1917 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1918 | if (IS_ERR(right)) |
1919 | right = NULL; | |
1920 | ||
5f39d397 | 1921 | if (right) { |
925baedd | 1922 | btrfs_tree_lock(right); |
8bead258 | 1923 | btrfs_set_lock_blocking_write(right); |
5f39d397 | 1924 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1925 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1926 | if (wret) { |
1927 | ret = wret; | |
1928 | goto enospc; | |
1929 | } | |
1930 | } | |
1931 | ||
1932 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1933 | if (left) { |
1934 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 1935 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
1936 | if (wret < 0) |
1937 | ret = wret; | |
bb803951 | 1938 | } |
79f95c82 CM |
1939 | |
1940 | /* | |
1941 | * then try to empty the right most buffer into the middle | |
1942 | */ | |
5f39d397 | 1943 | if (right) { |
d30a668f | 1944 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1945 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1946 | ret = wret; |
5f39d397 | 1947 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 1948 | btrfs_clean_tree_block(right); |
925baedd | 1949 | btrfs_tree_unlock(right); |
afe5fea7 | 1950 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1951 | root_sub_used(root, right->len); |
5581a51a | 1952 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1953 | free_extent_buffer_stale(right); |
f0486c68 | 1954 | right = NULL; |
bb803951 | 1955 | } else { |
5f39d397 CM |
1956 | struct btrfs_disk_key right_key; |
1957 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1958 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1959 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1960 | BUG_ON(ret < 0); | |
5f39d397 CM |
1961 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1962 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1963 | } |
1964 | } | |
5f39d397 | 1965 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1966 | /* |
1967 | * we're not allowed to leave a node with one item in the | |
1968 | * tree during a delete. A deletion from lower in the tree | |
1969 | * could try to delete the only pointer in this node. | |
1970 | * So, pull some keys from the left. | |
1971 | * There has to be a left pointer at this point because | |
1972 | * otherwise we would have pulled some pointers from the | |
1973 | * right | |
1974 | */ | |
305a26af MF |
1975 | if (!left) { |
1976 | ret = -EROFS; | |
0b246afa | 1977 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1978 | goto enospc; |
1979 | } | |
55d32ed8 | 1980 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1981 | if (wret < 0) { |
79f95c82 | 1982 | ret = wret; |
54aa1f4d CM |
1983 | goto enospc; |
1984 | } | |
bce4eae9 | 1985 | if (wret == 1) { |
d30a668f | 1986 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1987 | if (wret < 0) |
1988 | ret = wret; | |
1989 | } | |
79f95c82 CM |
1990 | BUG_ON(wret == 1); |
1991 | } | |
5f39d397 | 1992 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 1993 | btrfs_clean_tree_block(mid); |
925baedd | 1994 | btrfs_tree_unlock(mid); |
afe5fea7 | 1995 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1996 | root_sub_used(root, mid->len); |
5581a51a | 1997 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1998 | free_extent_buffer_stale(mid); |
f0486c68 | 1999 | mid = NULL; |
79f95c82 CM |
2000 | } else { |
2001 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2002 | struct btrfs_disk_key mid_key; |
2003 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
2004 | ret = tree_mod_log_insert_key(parent, pslot, |
2005 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2006 | BUG_ON(ret < 0); | |
5f39d397 CM |
2007 | btrfs_set_node_key(parent, &mid_key, pslot); |
2008 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2009 | } |
bb803951 | 2010 | |
79f95c82 | 2011 | /* update the path */ |
5f39d397 CM |
2012 | if (left) { |
2013 | if (btrfs_header_nritems(left) > orig_slot) { | |
2014 | extent_buffer_get(left); | |
925baedd | 2015 | /* left was locked after cow */ |
5f39d397 | 2016 | path->nodes[level] = left; |
bb803951 CM |
2017 | path->slots[level + 1] -= 1; |
2018 | path->slots[level] = orig_slot; | |
925baedd CM |
2019 | if (mid) { |
2020 | btrfs_tree_unlock(mid); | |
5f39d397 | 2021 | free_extent_buffer(mid); |
925baedd | 2022 | } |
bb803951 | 2023 | } else { |
5f39d397 | 2024 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2025 | path->slots[level] = orig_slot; |
2026 | } | |
2027 | } | |
79f95c82 | 2028 | /* double check we haven't messed things up */ |
e20d96d6 | 2029 | if (orig_ptr != |
5f39d397 | 2030 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2031 | BUG(); |
54aa1f4d | 2032 | enospc: |
925baedd CM |
2033 | if (right) { |
2034 | btrfs_tree_unlock(right); | |
5f39d397 | 2035 | free_extent_buffer(right); |
925baedd CM |
2036 | } |
2037 | if (left) { | |
2038 | if (path->nodes[level] != left) | |
2039 | btrfs_tree_unlock(left); | |
5f39d397 | 2040 | free_extent_buffer(left); |
925baedd | 2041 | } |
bb803951 CM |
2042 | return ret; |
2043 | } | |
2044 | ||
d352ac68 CM |
2045 | /* Node balancing for insertion. Here we only split or push nodes around |
2046 | * when they are completely full. This is also done top down, so we | |
2047 | * have to be pessimistic. | |
2048 | */ | |
d397712b | 2049 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2050 | struct btrfs_root *root, |
2051 | struct btrfs_path *path, int level) | |
e66f709b | 2052 | { |
0b246afa | 2053 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2054 | struct extent_buffer *right = NULL; |
2055 | struct extent_buffer *mid; | |
2056 | struct extent_buffer *left = NULL; | |
2057 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2058 | int ret = 0; |
2059 | int wret; | |
2060 | int pslot; | |
2061 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2062 | |
2063 | if (level == 0) | |
2064 | return 1; | |
2065 | ||
5f39d397 | 2066 | mid = path->nodes[level]; |
7bb86316 | 2067 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2068 | |
a05a9bb1 | 2069 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2070 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2071 | pslot = path->slots[level + 1]; |
2072 | } | |
e66f709b | 2073 | |
5f39d397 | 2074 | if (!parent) |
e66f709b | 2075 | return 1; |
e66f709b | 2076 | |
4b231ae4 | 2077 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
2078 | if (IS_ERR(left)) |
2079 | left = NULL; | |
e66f709b CM |
2080 | |
2081 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2082 | if (left) { |
e66f709b | 2083 | u32 left_nr; |
925baedd CM |
2084 | |
2085 | btrfs_tree_lock(left); | |
8bead258 | 2086 | btrfs_set_lock_blocking_write(left); |
b4ce94de | 2087 | |
5f39d397 | 2088 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2089 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2090 | wret = 1; |
2091 | } else { | |
5f39d397 | 2092 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2093 | pslot - 1, &left); |
54aa1f4d CM |
2094 | if (ret) |
2095 | wret = 1; | |
2096 | else { | |
d30a668f | 2097 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 2098 | } |
33ade1f8 | 2099 | } |
e66f709b CM |
2100 | if (wret < 0) |
2101 | ret = wret; | |
2102 | if (wret == 0) { | |
5f39d397 | 2103 | struct btrfs_disk_key disk_key; |
e66f709b | 2104 | orig_slot += left_nr; |
5f39d397 | 2105 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2106 | ret = tree_mod_log_insert_key(parent, pslot, |
2107 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2108 | BUG_ON(ret < 0); | |
5f39d397 CM |
2109 | btrfs_set_node_key(parent, &disk_key, pslot); |
2110 | btrfs_mark_buffer_dirty(parent); | |
2111 | if (btrfs_header_nritems(left) > orig_slot) { | |
2112 | path->nodes[level] = left; | |
e66f709b CM |
2113 | path->slots[level + 1] -= 1; |
2114 | path->slots[level] = orig_slot; | |
925baedd | 2115 | btrfs_tree_unlock(mid); |
5f39d397 | 2116 | free_extent_buffer(mid); |
e66f709b CM |
2117 | } else { |
2118 | orig_slot -= | |
5f39d397 | 2119 | btrfs_header_nritems(left); |
e66f709b | 2120 | path->slots[level] = orig_slot; |
925baedd | 2121 | btrfs_tree_unlock(left); |
5f39d397 | 2122 | free_extent_buffer(left); |
e66f709b | 2123 | } |
e66f709b CM |
2124 | return 0; |
2125 | } | |
925baedd | 2126 | btrfs_tree_unlock(left); |
5f39d397 | 2127 | free_extent_buffer(left); |
e66f709b | 2128 | } |
4b231ae4 | 2129 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
2130 | if (IS_ERR(right)) |
2131 | right = NULL; | |
e66f709b CM |
2132 | |
2133 | /* | |
2134 | * then try to empty the right most buffer into the middle | |
2135 | */ | |
5f39d397 | 2136 | if (right) { |
33ade1f8 | 2137 | u32 right_nr; |
b4ce94de | 2138 | |
925baedd | 2139 | btrfs_tree_lock(right); |
8bead258 | 2140 | btrfs_set_lock_blocking_write(right); |
b4ce94de | 2141 | |
5f39d397 | 2142 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2143 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2144 | wret = 1; |
2145 | } else { | |
5f39d397 CM |
2146 | ret = btrfs_cow_block(trans, root, right, |
2147 | parent, pslot + 1, | |
9fa8cfe7 | 2148 | &right); |
54aa1f4d CM |
2149 | if (ret) |
2150 | wret = 1; | |
2151 | else { | |
55d32ed8 | 2152 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 2153 | } |
33ade1f8 | 2154 | } |
e66f709b CM |
2155 | if (wret < 0) |
2156 | ret = wret; | |
2157 | if (wret == 0) { | |
5f39d397 CM |
2158 | struct btrfs_disk_key disk_key; |
2159 | ||
2160 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2161 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2162 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2163 | BUG_ON(ret < 0); | |
5f39d397 CM |
2164 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2165 | btrfs_mark_buffer_dirty(parent); | |
2166 | ||
2167 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2168 | path->nodes[level] = right; | |
e66f709b CM |
2169 | path->slots[level + 1] += 1; |
2170 | path->slots[level] = orig_slot - | |
5f39d397 | 2171 | btrfs_header_nritems(mid); |
925baedd | 2172 | btrfs_tree_unlock(mid); |
5f39d397 | 2173 | free_extent_buffer(mid); |
e66f709b | 2174 | } else { |
925baedd | 2175 | btrfs_tree_unlock(right); |
5f39d397 | 2176 | free_extent_buffer(right); |
e66f709b | 2177 | } |
e66f709b CM |
2178 | return 0; |
2179 | } | |
925baedd | 2180 | btrfs_tree_unlock(right); |
5f39d397 | 2181 | free_extent_buffer(right); |
e66f709b | 2182 | } |
e66f709b CM |
2183 | return 1; |
2184 | } | |
2185 | ||
3c69faec | 2186 | /* |
d352ac68 CM |
2187 | * readahead one full node of leaves, finding things that are close |
2188 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2189 | */ |
2ff7e61e | 2190 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2191 | struct btrfs_path *path, |
2192 | int level, int slot, u64 objectid) | |
3c69faec | 2193 | { |
5f39d397 | 2194 | struct extent_buffer *node; |
01f46658 | 2195 | struct btrfs_disk_key disk_key; |
3c69faec | 2196 | u32 nritems; |
3c69faec | 2197 | u64 search; |
a7175319 | 2198 | u64 target; |
6b80053d | 2199 | u64 nread = 0; |
5f39d397 | 2200 | struct extent_buffer *eb; |
6b80053d CM |
2201 | u32 nr; |
2202 | u32 blocksize; | |
2203 | u32 nscan = 0; | |
db94535d | 2204 | |
a6b6e75e | 2205 | if (level != 1) |
6702ed49 CM |
2206 | return; |
2207 | ||
2208 | if (!path->nodes[level]) | |
3c69faec CM |
2209 | return; |
2210 | ||
5f39d397 | 2211 | node = path->nodes[level]; |
925baedd | 2212 | |
3c69faec | 2213 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2214 | blocksize = fs_info->nodesize; |
2215 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2216 | if (eb) { |
2217 | free_extent_buffer(eb); | |
3c69faec CM |
2218 | return; |
2219 | } | |
2220 | ||
a7175319 | 2221 | target = search; |
6b80053d | 2222 | |
5f39d397 | 2223 | nritems = btrfs_header_nritems(node); |
6b80053d | 2224 | nr = slot; |
25b8b936 | 2225 | |
d397712b | 2226 | while (1) { |
e4058b54 | 2227 | if (path->reada == READA_BACK) { |
6b80053d CM |
2228 | if (nr == 0) |
2229 | break; | |
2230 | nr--; | |
e4058b54 | 2231 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2232 | nr++; |
2233 | if (nr >= nritems) | |
2234 | break; | |
3c69faec | 2235 | } |
e4058b54 | 2236 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2237 | btrfs_node_key(node, &disk_key, nr); |
2238 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2239 | break; | |
2240 | } | |
6b80053d | 2241 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2242 | if ((search <= target && target - search <= 65536) || |
2243 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2244 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2245 | nread += blocksize; |
2246 | } | |
2247 | nscan++; | |
a7175319 | 2248 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2249 | break; |
3c69faec CM |
2250 | } |
2251 | } | |
925baedd | 2252 | |
2ff7e61e | 2253 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2254 | struct btrfs_path *path, int level) |
b4ce94de CM |
2255 | { |
2256 | int slot; | |
2257 | int nritems; | |
2258 | struct extent_buffer *parent; | |
2259 | struct extent_buffer *eb; | |
2260 | u64 gen; | |
2261 | u64 block1 = 0; | |
2262 | u64 block2 = 0; | |
b4ce94de | 2263 | |
8c594ea8 | 2264 | parent = path->nodes[level + 1]; |
b4ce94de | 2265 | if (!parent) |
0b08851f | 2266 | return; |
b4ce94de CM |
2267 | |
2268 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2269 | slot = path->slots[level + 1]; |
b4ce94de CM |
2270 | |
2271 | if (slot > 0) { | |
2272 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2273 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2274 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2275 | /* |
2276 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2277 | * don't want to return eagain here. That will loop | |
2278 | * forever | |
2279 | */ | |
2280 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2281 | block1 = 0; |
2282 | free_extent_buffer(eb); | |
2283 | } | |
8c594ea8 | 2284 | if (slot + 1 < nritems) { |
b4ce94de CM |
2285 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2286 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2287 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2288 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2289 | block2 = 0; |
2290 | free_extent_buffer(eb); | |
2291 | } | |
8c594ea8 | 2292 | |
0b08851f | 2293 | if (block1) |
2ff7e61e | 2294 | readahead_tree_block(fs_info, block1); |
0b08851f | 2295 | if (block2) |
2ff7e61e | 2296 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2297 | } |
2298 | ||
2299 | ||
d352ac68 | 2300 | /* |
d397712b CM |
2301 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2302 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2303 | * operations on the tree might require changing key pointers higher up in the | |
2304 | * tree. | |
d352ac68 | 2305 | * |
d397712b CM |
2306 | * callers might also have set path->keep_locks, which tells this code to keep |
2307 | * the lock if the path points to the last slot in the block. This is part of | |
2308 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2309 | * |
d397712b CM |
2310 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2311 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2312 | */ |
e02119d5 | 2313 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2314 | int lowest_unlock, int min_write_lock_level, |
2315 | int *write_lock_level) | |
925baedd CM |
2316 | { |
2317 | int i; | |
2318 | int skip_level = level; | |
051e1b9f | 2319 | int no_skips = 0; |
925baedd CM |
2320 | struct extent_buffer *t; |
2321 | ||
2322 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2323 | if (!path->nodes[i]) | |
2324 | break; | |
2325 | if (!path->locks[i]) | |
2326 | break; | |
051e1b9f | 2327 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2328 | skip_level = i + 1; |
2329 | continue; | |
2330 | } | |
051e1b9f | 2331 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2332 | u32 nritems; |
2333 | t = path->nodes[i]; | |
2334 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2335 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2336 | skip_level = i + 1; |
2337 | continue; | |
2338 | } | |
2339 | } | |
051e1b9f CM |
2340 | if (skip_level < i && i >= lowest_unlock) |
2341 | no_skips = 1; | |
2342 | ||
925baedd | 2343 | t = path->nodes[i]; |
d80bb3f9 | 2344 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2345 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2346 | path->locks[i] = 0; |
f7c79f30 CM |
2347 | if (write_lock_level && |
2348 | i > min_write_lock_level && | |
2349 | i <= *write_lock_level) { | |
2350 | *write_lock_level = i - 1; | |
2351 | } | |
925baedd CM |
2352 | } |
2353 | } | |
2354 | } | |
2355 | ||
c8c42864 CM |
2356 | /* |
2357 | * helper function for btrfs_search_slot. The goal is to find a block | |
2358 | * in cache without setting the path to blocking. If we find the block | |
2359 | * we return zero and the path is unchanged. | |
2360 | * | |
2361 | * If we can't find the block, we set the path blocking and do some | |
2362 | * reada. -EAGAIN is returned and the search must be repeated. | |
2363 | */ | |
2364 | static int | |
d07b8528 LB |
2365 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2366 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2367 | const struct btrfs_key *key) |
c8c42864 | 2368 | { |
0b246afa | 2369 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2370 | u64 blocknr; |
2371 | u64 gen; | |
c8c42864 CM |
2372 | struct extent_buffer *b = *eb_ret; |
2373 | struct extent_buffer *tmp; | |
581c1760 | 2374 | struct btrfs_key first_key; |
76a05b35 | 2375 | int ret; |
581c1760 | 2376 | int parent_level; |
c8c42864 CM |
2377 | |
2378 | blocknr = btrfs_node_blockptr(b, slot); | |
2379 | gen = btrfs_node_ptr_generation(b, slot); | |
581c1760 QW |
2380 | parent_level = btrfs_header_level(b); |
2381 | btrfs_node_key_to_cpu(b, &first_key, slot); | |
c8c42864 | 2382 | |
0b246afa | 2383 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2384 | if (tmp) { |
b9fab919 | 2385 | /* first we do an atomic uptodate check */ |
bdf7c00e | 2386 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
2387 | /* |
2388 | * Do extra check for first_key, eb can be stale due to | |
2389 | * being cached, read from scrub, or have multiple | |
2390 | * parents (shared tree blocks). | |
2391 | */ | |
e064d5e9 | 2392 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
2393 | parent_level - 1, &first_key, gen)) { |
2394 | free_extent_buffer(tmp); | |
2395 | return -EUCLEAN; | |
2396 | } | |
bdf7c00e JB |
2397 | *eb_ret = tmp; |
2398 | return 0; | |
2399 | } | |
2400 | ||
2401 | /* the pages were up to date, but we failed | |
2402 | * the generation number check. Do a full | |
2403 | * read for the generation number that is correct. | |
2404 | * We must do this without dropping locks so | |
2405 | * we can trust our generation number | |
2406 | */ | |
2407 | btrfs_set_path_blocking(p); | |
2408 | ||
2409 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2410 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2411 | if (!ret) { |
2412 | *eb_ret = tmp; | |
2413 | return 0; | |
cb44921a | 2414 | } |
bdf7c00e JB |
2415 | free_extent_buffer(tmp); |
2416 | btrfs_release_path(p); | |
2417 | return -EIO; | |
c8c42864 CM |
2418 | } |
2419 | ||
2420 | /* | |
2421 | * reduce lock contention at high levels | |
2422 | * of the btree by dropping locks before | |
76a05b35 CM |
2423 | * we read. Don't release the lock on the current |
2424 | * level because we need to walk this node to figure | |
2425 | * out which blocks to read. | |
c8c42864 | 2426 | */ |
8c594ea8 CM |
2427 | btrfs_unlock_up_safe(p, level + 1); |
2428 | btrfs_set_path_blocking(p); | |
2429 | ||
e4058b54 | 2430 | if (p->reada != READA_NONE) |
2ff7e61e | 2431 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2432 | |
76a05b35 | 2433 | ret = -EAGAIN; |
02a3307a | 2434 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2435 | &first_key); |
64c043de | 2436 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2437 | /* |
2438 | * If the read above didn't mark this buffer up to date, | |
2439 | * it will never end up being up to date. Set ret to EIO now | |
2440 | * and give up so that our caller doesn't loop forever | |
2441 | * on our EAGAINs. | |
2442 | */ | |
e6a1d6fd | 2443 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2444 | ret = -EIO; |
c8c42864 | 2445 | free_extent_buffer(tmp); |
c871b0f2 LB |
2446 | } else { |
2447 | ret = PTR_ERR(tmp); | |
76a05b35 | 2448 | } |
02a3307a LB |
2449 | |
2450 | btrfs_release_path(p); | |
76a05b35 | 2451 | return ret; |
c8c42864 CM |
2452 | } |
2453 | ||
2454 | /* | |
2455 | * helper function for btrfs_search_slot. This does all of the checks | |
2456 | * for node-level blocks and does any balancing required based on | |
2457 | * the ins_len. | |
2458 | * | |
2459 | * If no extra work was required, zero is returned. If we had to | |
2460 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2461 | * start over | |
2462 | */ | |
2463 | static int | |
2464 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2465 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2466 | struct extent_buffer *b, int level, int ins_len, |
2467 | int *write_lock_level) | |
c8c42864 | 2468 | { |
0b246afa | 2469 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2470 | int ret; |
0b246afa | 2471 | |
c8c42864 | 2472 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2473 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2474 | int sret; |
2475 | ||
bd681513 CM |
2476 | if (*write_lock_level < level + 1) { |
2477 | *write_lock_level = level + 1; | |
2478 | btrfs_release_path(p); | |
2479 | goto again; | |
2480 | } | |
2481 | ||
c8c42864 | 2482 | btrfs_set_path_blocking(p); |
2ff7e61e | 2483 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2484 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2485 | |
2486 | BUG_ON(sret > 0); | |
2487 | if (sret) { | |
2488 | ret = sret; | |
2489 | goto done; | |
2490 | } | |
2491 | b = p->nodes[level]; | |
2492 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2493 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2494 | int sret; |
2495 | ||
bd681513 CM |
2496 | if (*write_lock_level < level + 1) { |
2497 | *write_lock_level = level + 1; | |
2498 | btrfs_release_path(p); | |
2499 | goto again; | |
2500 | } | |
2501 | ||
c8c42864 | 2502 | btrfs_set_path_blocking(p); |
2ff7e61e | 2503 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2504 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2505 | |
2506 | if (sret) { | |
2507 | ret = sret; | |
2508 | goto done; | |
2509 | } | |
2510 | b = p->nodes[level]; | |
2511 | if (!b) { | |
b3b4aa74 | 2512 | btrfs_release_path(p); |
c8c42864 CM |
2513 | goto again; |
2514 | } | |
2515 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2516 | } | |
2517 | return 0; | |
2518 | ||
2519 | again: | |
2520 | ret = -EAGAIN; | |
2521 | done: | |
2522 | return ret; | |
2523 | } | |
2524 | ||
310712b2 | 2525 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2526 | int level, int *prev_cmp, int *slot) |
2527 | { | |
2528 | if (*prev_cmp != 0) { | |
a74b35ec | 2529 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2530 | return *prev_cmp; |
2531 | } | |
2532 | ||
d7396f07 FDBM |
2533 | *slot = 0; |
2534 | ||
2535 | return 0; | |
2536 | } | |
2537 | ||
381cf658 | 2538 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2539 | u64 iobjectid, u64 ioff, u8 key_type, |
2540 | struct btrfs_key *found_key) | |
2541 | { | |
2542 | int ret; | |
2543 | struct btrfs_key key; | |
2544 | struct extent_buffer *eb; | |
381cf658 DS |
2545 | |
2546 | ASSERT(path); | |
1d4c08e0 | 2547 | ASSERT(found_key); |
e33d5c3d KN |
2548 | |
2549 | key.type = key_type; | |
2550 | key.objectid = iobjectid; | |
2551 | key.offset = ioff; | |
2552 | ||
2553 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2554 | if (ret < 0) |
e33d5c3d KN |
2555 | return ret; |
2556 | ||
2557 | eb = path->nodes[0]; | |
2558 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2559 | ret = btrfs_next_leaf(fs_root, path); | |
2560 | if (ret) | |
2561 | return ret; | |
2562 | eb = path->nodes[0]; | |
2563 | } | |
2564 | ||
2565 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2566 | if (found_key->type != key.type || | |
2567 | found_key->objectid != key.objectid) | |
2568 | return 1; | |
2569 | ||
2570 | return 0; | |
2571 | } | |
2572 | ||
1fc28d8e LB |
2573 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2574 | struct btrfs_path *p, | |
2575 | int write_lock_level) | |
2576 | { | |
2577 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2578 | struct extent_buffer *b; | |
2579 | int root_lock; | |
2580 | int level = 0; | |
2581 | ||
2582 | /* We try very hard to do read locks on the root */ | |
2583 | root_lock = BTRFS_READ_LOCK; | |
2584 | ||
2585 | if (p->search_commit_root) { | |
be6821f8 FM |
2586 | /* |
2587 | * The commit roots are read only so we always do read locks, | |
2588 | * and we always must hold the commit_root_sem when doing | |
2589 | * searches on them, the only exception is send where we don't | |
2590 | * want to block transaction commits for a long time, so | |
2591 | * we need to clone the commit root in order to avoid races | |
2592 | * with transaction commits that create a snapshot of one of | |
2593 | * the roots used by a send operation. | |
2594 | */ | |
2595 | if (p->need_commit_sem) { | |
1fc28d8e | 2596 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 2597 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 2598 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
2599 | if (!b) |
2600 | return ERR_PTR(-ENOMEM); | |
2601 | ||
2602 | } else { | |
2603 | b = root->commit_root; | |
2604 | extent_buffer_get(b); | |
2605 | } | |
2606 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
2607 | /* |
2608 | * Ensure that all callers have set skip_locking when | |
2609 | * p->search_commit_root = 1. | |
2610 | */ | |
2611 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2612 | |
2613 | goto out; | |
2614 | } | |
2615 | ||
2616 | if (p->skip_locking) { | |
2617 | b = btrfs_root_node(root); | |
2618 | level = btrfs_header_level(b); | |
2619 | goto out; | |
2620 | } | |
2621 | ||
2622 | /* | |
662c653b LB |
2623 | * If the level is set to maximum, we can skip trying to get the read |
2624 | * lock. | |
1fc28d8e | 2625 | */ |
662c653b LB |
2626 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2627 | /* | |
2628 | * We don't know the level of the root node until we actually | |
2629 | * have it read locked | |
2630 | */ | |
2631 | b = btrfs_read_lock_root_node(root); | |
2632 | level = btrfs_header_level(b); | |
2633 | if (level > write_lock_level) | |
2634 | goto out; | |
2635 | ||
2636 | /* Whoops, must trade for write lock */ | |
2637 | btrfs_tree_read_unlock(b); | |
2638 | free_extent_buffer(b); | |
2639 | } | |
1fc28d8e | 2640 | |
1fc28d8e LB |
2641 | b = btrfs_lock_root_node(root); |
2642 | root_lock = BTRFS_WRITE_LOCK; | |
2643 | ||
2644 | /* The level might have changed, check again */ | |
2645 | level = btrfs_header_level(b); | |
2646 | ||
2647 | out: | |
2648 | p->nodes[level] = b; | |
2649 | if (!p->skip_locking) | |
2650 | p->locks[level] = root_lock; | |
2651 | /* | |
2652 | * Callers are responsible for dropping b's references. | |
2653 | */ | |
2654 | return b; | |
2655 | } | |
2656 | ||
2657 | ||
74123bd7 | 2658 | /* |
4271ecea NB |
2659 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2660 | * modifications to preserve tree invariants. | |
74123bd7 | 2661 | * |
4271ecea NB |
2662 | * @trans: Handle of transaction, used when modifying the tree |
2663 | * @p: Holds all btree nodes along the search path | |
2664 | * @root: The root node of the tree | |
2665 | * @key: The key we are looking for | |
2666 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2667 | * deletions it's -1. 0 for plain searches | |
2668 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2669 | * when modifying the tree. | |
97571fd0 | 2670 | * |
4271ecea NB |
2671 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2672 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2673 | * | |
2674 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2675 | * of the path (level 0) | |
2676 | * | |
2677 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2678 | * points to the slot where it should be inserted | |
2679 | * | |
2680 | * If an error is encountered while searching the tree a negative error number | |
2681 | * is returned | |
74123bd7 | 2682 | */ |
310712b2 OS |
2683 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2684 | const struct btrfs_key *key, struct btrfs_path *p, | |
2685 | int ins_len, int cow) | |
be0e5c09 | 2686 | { |
5f39d397 | 2687 | struct extent_buffer *b; |
be0e5c09 CM |
2688 | int slot; |
2689 | int ret; | |
33c66f43 | 2690 | int err; |
be0e5c09 | 2691 | int level; |
925baedd | 2692 | int lowest_unlock = 1; |
bd681513 CM |
2693 | /* everything at write_lock_level or lower must be write locked */ |
2694 | int write_lock_level = 0; | |
9f3a7427 | 2695 | u8 lowest_level = 0; |
f7c79f30 | 2696 | int min_write_lock_level; |
d7396f07 | 2697 | int prev_cmp; |
9f3a7427 | 2698 | |
6702ed49 | 2699 | lowest_level = p->lowest_level; |
323ac95b | 2700 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2701 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2702 | BUG_ON(!cow && ins_len); |
25179201 | 2703 | |
bd681513 | 2704 | if (ins_len < 0) { |
925baedd | 2705 | lowest_unlock = 2; |
65b51a00 | 2706 | |
bd681513 CM |
2707 | /* when we are removing items, we might have to go up to level |
2708 | * two as we update tree pointers Make sure we keep write | |
2709 | * for those levels as well | |
2710 | */ | |
2711 | write_lock_level = 2; | |
2712 | } else if (ins_len > 0) { | |
2713 | /* | |
2714 | * for inserting items, make sure we have a write lock on | |
2715 | * level 1 so we can update keys | |
2716 | */ | |
2717 | write_lock_level = 1; | |
2718 | } | |
2719 | ||
2720 | if (!cow) | |
2721 | write_lock_level = -1; | |
2722 | ||
09a2a8f9 | 2723 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2724 | write_lock_level = BTRFS_MAX_LEVEL; |
2725 | ||
f7c79f30 CM |
2726 | min_write_lock_level = write_lock_level; |
2727 | ||
bb803951 | 2728 | again: |
d7396f07 | 2729 | prev_cmp = -1; |
1fc28d8e | 2730 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2731 | if (IS_ERR(b)) { |
2732 | ret = PTR_ERR(b); | |
2733 | goto done; | |
2734 | } | |
925baedd | 2735 | |
eb60ceac | 2736 | while (b) { |
f624d976 QW |
2737 | int dec = 0; |
2738 | ||
5f39d397 | 2739 | level = btrfs_header_level(b); |
65b51a00 | 2740 | |
02217ed2 | 2741 | if (cow) { |
9ea2c7c9 NB |
2742 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2743 | ||
c8c42864 CM |
2744 | /* |
2745 | * if we don't really need to cow this block | |
2746 | * then we don't want to set the path blocking, | |
2747 | * so we test it here | |
2748 | */ | |
64c12921 JM |
2749 | if (!should_cow_block(trans, root, b)) { |
2750 | trans->dirty = true; | |
65b51a00 | 2751 | goto cow_done; |
64c12921 | 2752 | } |
5d4f98a2 | 2753 | |
bd681513 CM |
2754 | /* |
2755 | * must have write locks on this node and the | |
2756 | * parent | |
2757 | */ | |
5124e00e JB |
2758 | if (level > write_lock_level || |
2759 | (level + 1 > write_lock_level && | |
2760 | level + 1 < BTRFS_MAX_LEVEL && | |
2761 | p->nodes[level + 1])) { | |
bd681513 CM |
2762 | write_lock_level = level + 1; |
2763 | btrfs_release_path(p); | |
2764 | goto again; | |
2765 | } | |
2766 | ||
160f4089 | 2767 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2768 | if (last_level) |
2769 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2770 | &b); | |
2771 | else | |
2772 | err = btrfs_cow_block(trans, root, b, | |
2773 | p->nodes[level + 1], | |
2774 | p->slots[level + 1], &b); | |
33c66f43 | 2775 | if (err) { |
33c66f43 | 2776 | ret = err; |
65b51a00 | 2777 | goto done; |
54aa1f4d | 2778 | } |
02217ed2 | 2779 | } |
65b51a00 | 2780 | cow_done: |
eb60ceac | 2781 | p->nodes[level] = b; |
52398340 LB |
2782 | /* |
2783 | * Leave path with blocking locks to avoid massive | |
2784 | * lock context switch, this is made on purpose. | |
2785 | */ | |
b4ce94de CM |
2786 | |
2787 | /* | |
2788 | * we have a lock on b and as long as we aren't changing | |
2789 | * the tree, there is no way to for the items in b to change. | |
2790 | * It is safe to drop the lock on our parent before we | |
2791 | * go through the expensive btree search on b. | |
2792 | * | |
eb653de1 FDBM |
2793 | * If we're inserting or deleting (ins_len != 0), then we might |
2794 | * be changing slot zero, which may require changing the parent. | |
2795 | * So, we can't drop the lock until after we know which slot | |
2796 | * we're operating on. | |
b4ce94de | 2797 | */ |
eb653de1 FDBM |
2798 | if (!ins_len && !p->keep_locks) { |
2799 | int u = level + 1; | |
2800 | ||
2801 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2802 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2803 | p->locks[u] = 0; | |
2804 | } | |
2805 | } | |
b4ce94de | 2806 | |
d7396f07 | 2807 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2808 | if (ret < 0) |
2809 | goto done; | |
b4ce94de | 2810 | |
f624d976 | 2811 | if (level == 0) { |
be0e5c09 | 2812 | p->slots[level] = slot; |
87b29b20 | 2813 | if (ins_len > 0 && |
e902baac | 2814 | btrfs_leaf_free_space(b) < ins_len) { |
bd681513 CM |
2815 | if (write_lock_level < 1) { |
2816 | write_lock_level = 1; | |
2817 | btrfs_release_path(p); | |
2818 | goto again; | |
2819 | } | |
2820 | ||
b4ce94de | 2821 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2822 | err = split_leaf(trans, root, key, |
2823 | p, ins_len, ret == 0); | |
b4ce94de | 2824 | |
33c66f43 YZ |
2825 | BUG_ON(err > 0); |
2826 | if (err) { | |
2827 | ret = err; | |
65b51a00 CM |
2828 | goto done; |
2829 | } | |
5c680ed6 | 2830 | } |
459931ec | 2831 | if (!p->search_for_split) |
f7c79f30 | 2832 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2833 | min_write_lock_level, NULL); |
65b51a00 | 2834 | goto done; |
be0e5c09 | 2835 | } |
f624d976 QW |
2836 | if (ret && slot > 0) { |
2837 | dec = 1; | |
2838 | slot--; | |
2839 | } | |
2840 | p->slots[level] = slot; | |
2841 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
2842 | &write_lock_level); | |
2843 | if (err == -EAGAIN) | |
2844 | goto again; | |
2845 | if (err) { | |
2846 | ret = err; | |
2847 | goto done; | |
2848 | } | |
2849 | b = p->nodes[level]; | |
2850 | slot = p->slots[level]; | |
2851 | ||
2852 | /* | |
2853 | * Slot 0 is special, if we change the key we have to update | |
2854 | * the parent pointer which means we must have a write lock on | |
2855 | * the parent | |
2856 | */ | |
2857 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
2858 | write_lock_level = level + 1; | |
2859 | btrfs_release_path(p); | |
2860 | goto again; | |
2861 | } | |
2862 | ||
2863 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
2864 | &write_lock_level); | |
2865 | ||
2866 | if (level == lowest_level) { | |
2867 | if (dec) | |
2868 | p->slots[level]++; | |
2869 | goto done; | |
2870 | } | |
2871 | ||
2872 | err = read_block_for_search(root, p, &b, level, slot, key); | |
2873 | if (err == -EAGAIN) | |
2874 | goto again; | |
2875 | if (err) { | |
2876 | ret = err; | |
2877 | goto done; | |
2878 | } | |
2879 | ||
2880 | if (!p->skip_locking) { | |
2881 | level = btrfs_header_level(b); | |
2882 | if (level <= write_lock_level) { | |
2883 | if (!btrfs_try_tree_write_lock(b)) { | |
2884 | btrfs_set_path_blocking(p); | |
2885 | btrfs_tree_lock(b); | |
2886 | } | |
2887 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2888 | } else { | |
2889 | if (!btrfs_tree_read_lock_atomic(b)) { | |
2890 | btrfs_set_path_blocking(p); | |
2891 | btrfs_tree_read_lock(b); | |
2892 | } | |
2893 | p->locks[level] = BTRFS_READ_LOCK; | |
2894 | } | |
2895 | p->nodes[level] = b; | |
2896 | } | |
be0e5c09 | 2897 | } |
65b51a00 CM |
2898 | ret = 1; |
2899 | done: | |
b4ce94de CM |
2900 | /* |
2901 | * we don't really know what they plan on doing with the path | |
2902 | * from here on, so for now just mark it as blocking | |
2903 | */ | |
b9473439 CM |
2904 | if (!p->leave_spinning) |
2905 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2906 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2907 | btrfs_release_path(p); |
65b51a00 | 2908 | return ret; |
be0e5c09 CM |
2909 | } |
2910 | ||
5d9e75c4 JS |
2911 | /* |
2912 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2913 | * current state of the tree together with the operations recorded in the tree | |
2914 | * modification log to search for the key in a previous version of this tree, as | |
2915 | * denoted by the time_seq parameter. | |
2916 | * | |
2917 | * Naturally, there is no support for insert, delete or cow operations. | |
2918 | * | |
2919 | * The resulting path and return value will be set up as if we called | |
2920 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2921 | */ | |
310712b2 | 2922 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2923 | struct btrfs_path *p, u64 time_seq) |
2924 | { | |
0b246afa | 2925 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2926 | struct extent_buffer *b; |
2927 | int slot; | |
2928 | int ret; | |
2929 | int err; | |
2930 | int level; | |
2931 | int lowest_unlock = 1; | |
2932 | u8 lowest_level = 0; | |
d4b4087c | 2933 | int prev_cmp = -1; |
5d9e75c4 JS |
2934 | |
2935 | lowest_level = p->lowest_level; | |
2936 | WARN_ON(p->nodes[0] != NULL); | |
2937 | ||
2938 | if (p->search_commit_root) { | |
2939 | BUG_ON(time_seq); | |
2940 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2941 | } | |
2942 | ||
2943 | again: | |
5d9e75c4 | 2944 | b = get_old_root(root, time_seq); |
315bed43 NB |
2945 | if (!b) { |
2946 | ret = -EIO; | |
2947 | goto done; | |
2948 | } | |
5d9e75c4 | 2949 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2950 | p->locks[level] = BTRFS_READ_LOCK; |
2951 | ||
2952 | while (b) { | |
abe9339d QW |
2953 | int dec = 0; |
2954 | ||
5d9e75c4 JS |
2955 | level = btrfs_header_level(b); |
2956 | p->nodes[level] = b; | |
5d9e75c4 JS |
2957 | |
2958 | /* | |
2959 | * we have a lock on b and as long as we aren't changing | |
2960 | * the tree, there is no way to for the items in b to change. | |
2961 | * It is safe to drop the lock on our parent before we | |
2962 | * go through the expensive btree search on b. | |
2963 | */ | |
2964 | btrfs_unlock_up_safe(p, level + 1); | |
2965 | ||
d4b4087c | 2966 | /* |
01327610 | 2967 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2968 | * time. |
2969 | */ | |
2970 | prev_cmp = -1; | |
d7396f07 | 2971 | ret = key_search(b, key, level, &prev_cmp, &slot); |
cbca7d59 FM |
2972 | if (ret < 0) |
2973 | goto done; | |
5d9e75c4 | 2974 | |
abe9339d | 2975 | if (level == 0) { |
5d9e75c4 JS |
2976 | p->slots[level] = slot; |
2977 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
2978 | goto done; |
2979 | } | |
5d9e75c4 | 2980 | |
abe9339d QW |
2981 | if (ret && slot > 0) { |
2982 | dec = 1; | |
2983 | slot--; | |
2984 | } | |
2985 | p->slots[level] = slot; | |
2986 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 2987 | |
abe9339d QW |
2988 | if (level == lowest_level) { |
2989 | if (dec) | |
2990 | p->slots[level]++; | |
2991 | goto done; | |
2992 | } | |
5d9e75c4 | 2993 | |
abe9339d QW |
2994 | err = read_block_for_search(root, p, &b, level, slot, key); |
2995 | if (err == -EAGAIN) | |
2996 | goto again; | |
2997 | if (err) { | |
2998 | ret = err; | |
5d9e75c4 JS |
2999 | goto done; |
3000 | } | |
abe9339d QW |
3001 | |
3002 | level = btrfs_header_level(b); | |
3003 | if (!btrfs_tree_read_lock_atomic(b)) { | |
3004 | btrfs_set_path_blocking(p); | |
3005 | btrfs_tree_read_lock(b); | |
3006 | } | |
3007 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); | |
3008 | if (!b) { | |
3009 | ret = -ENOMEM; | |
3010 | goto done; | |
3011 | } | |
3012 | p->locks[level] = BTRFS_READ_LOCK; | |
3013 | p->nodes[level] = b; | |
5d9e75c4 JS |
3014 | } |
3015 | ret = 1; | |
3016 | done: | |
3017 | if (!p->leave_spinning) | |
3018 | btrfs_set_path_blocking(p); | |
3019 | if (ret < 0) | |
3020 | btrfs_release_path(p); | |
3021 | ||
3022 | return ret; | |
3023 | } | |
3024 | ||
2f38b3e1 AJ |
3025 | /* |
3026 | * helper to use instead of search slot if no exact match is needed but | |
3027 | * instead the next or previous item should be returned. | |
3028 | * When find_higher is true, the next higher item is returned, the next lower | |
3029 | * otherwise. | |
3030 | * When return_any and find_higher are both true, and no higher item is found, | |
3031 | * return the next lower instead. | |
3032 | * When return_any is true and find_higher is false, and no lower item is found, | |
3033 | * return the next higher instead. | |
3034 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3035 | * < 0 on error | |
3036 | */ | |
3037 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3038 | const struct btrfs_key *key, |
3039 | struct btrfs_path *p, int find_higher, | |
3040 | int return_any) | |
2f38b3e1 AJ |
3041 | { |
3042 | int ret; | |
3043 | struct extent_buffer *leaf; | |
3044 | ||
3045 | again: | |
3046 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3047 | if (ret <= 0) | |
3048 | return ret; | |
3049 | /* | |
3050 | * a return value of 1 means the path is at the position where the | |
3051 | * item should be inserted. Normally this is the next bigger item, | |
3052 | * but in case the previous item is the last in a leaf, path points | |
3053 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3054 | * item. | |
3055 | */ | |
3056 | leaf = p->nodes[0]; | |
3057 | ||
3058 | if (find_higher) { | |
3059 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3060 | ret = btrfs_next_leaf(root, p); | |
3061 | if (ret <= 0) | |
3062 | return ret; | |
3063 | if (!return_any) | |
3064 | return 1; | |
3065 | /* | |
3066 | * no higher item found, return the next | |
3067 | * lower instead | |
3068 | */ | |
3069 | return_any = 0; | |
3070 | find_higher = 0; | |
3071 | btrfs_release_path(p); | |
3072 | goto again; | |
3073 | } | |
3074 | } else { | |
e6793769 AJ |
3075 | if (p->slots[0] == 0) { |
3076 | ret = btrfs_prev_leaf(root, p); | |
3077 | if (ret < 0) | |
3078 | return ret; | |
3079 | if (!ret) { | |
23c6bf6a FDBM |
3080 | leaf = p->nodes[0]; |
3081 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3082 | p->slots[0]--; | |
e6793769 | 3083 | return 0; |
2f38b3e1 | 3084 | } |
e6793769 AJ |
3085 | if (!return_any) |
3086 | return 1; | |
3087 | /* | |
3088 | * no lower item found, return the next | |
3089 | * higher instead | |
3090 | */ | |
3091 | return_any = 0; | |
3092 | find_higher = 1; | |
3093 | btrfs_release_path(p); | |
3094 | goto again; | |
3095 | } else { | |
2f38b3e1 AJ |
3096 | --p->slots[0]; |
3097 | } | |
3098 | } | |
3099 | return 0; | |
3100 | } | |
3101 | ||
74123bd7 CM |
3102 | /* |
3103 | * adjust the pointers going up the tree, starting at level | |
3104 | * making sure the right key of each node is points to 'key'. | |
3105 | * This is used after shifting pointers to the left, so it stops | |
3106 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3107 | * higher levels | |
aa5d6bed | 3108 | * |
74123bd7 | 3109 | */ |
b167fa91 | 3110 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3111 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3112 | { |
3113 | int i; | |
5f39d397 | 3114 | struct extent_buffer *t; |
0e82bcfe | 3115 | int ret; |
5f39d397 | 3116 | |
234b63a0 | 3117 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3118 | int tslot = path->slots[i]; |
0e82bcfe | 3119 | |
eb60ceac | 3120 | if (!path->nodes[i]) |
be0e5c09 | 3121 | break; |
5f39d397 | 3122 | t = path->nodes[i]; |
0e82bcfe DS |
3123 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3124 | GFP_ATOMIC); | |
3125 | BUG_ON(ret < 0); | |
5f39d397 | 3126 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3127 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3128 | if (tslot != 0) |
3129 | break; | |
3130 | } | |
3131 | } | |
3132 | ||
31840ae1 ZY |
3133 | /* |
3134 | * update item key. | |
3135 | * | |
3136 | * This function isn't completely safe. It's the caller's responsibility | |
3137 | * that the new key won't break the order | |
3138 | */ | |
b7a0365e DD |
3139 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3140 | struct btrfs_path *path, | |
310712b2 | 3141 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3142 | { |
3143 | struct btrfs_disk_key disk_key; | |
3144 | struct extent_buffer *eb; | |
3145 | int slot; | |
3146 | ||
3147 | eb = path->nodes[0]; | |
3148 | slot = path->slots[0]; | |
3149 | if (slot > 0) { | |
3150 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
3151 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
3152 | btrfs_crit(fs_info, | |
3153 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3154 | slot, btrfs_disk_key_objectid(&disk_key), | |
3155 | btrfs_disk_key_type(&disk_key), | |
3156 | btrfs_disk_key_offset(&disk_key), | |
3157 | new_key->objectid, new_key->type, | |
3158 | new_key->offset); | |
3159 | btrfs_print_leaf(eb); | |
3160 | BUG(); | |
3161 | } | |
31840ae1 ZY |
3162 | } |
3163 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3164 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
3165 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
3166 | btrfs_crit(fs_info, | |
3167 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3168 | slot, btrfs_disk_key_objectid(&disk_key), | |
3169 | btrfs_disk_key_type(&disk_key), | |
3170 | btrfs_disk_key_offset(&disk_key), | |
3171 | new_key->objectid, new_key->type, | |
3172 | new_key->offset); | |
3173 | btrfs_print_leaf(eb); | |
3174 | BUG(); | |
3175 | } | |
31840ae1 ZY |
3176 | } |
3177 | ||
3178 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3179 | btrfs_set_item_key(eb, &disk_key, slot); | |
3180 | btrfs_mark_buffer_dirty(eb); | |
3181 | if (slot == 0) | |
b167fa91 | 3182 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3183 | } |
3184 | ||
74123bd7 CM |
3185 | /* |
3186 | * try to push data from one node into the next node left in the | |
79f95c82 | 3187 | * tree. |
aa5d6bed CM |
3188 | * |
3189 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3190 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3191 | */ |
98ed5174 | 3192 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 3193 | struct extent_buffer *dst, |
971a1f66 | 3194 | struct extent_buffer *src, int empty) |
be0e5c09 | 3195 | { |
d30a668f | 3196 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 3197 | int push_items = 0; |
bb803951 CM |
3198 | int src_nritems; |
3199 | int dst_nritems; | |
aa5d6bed | 3200 | int ret = 0; |
be0e5c09 | 3201 | |
5f39d397 CM |
3202 | src_nritems = btrfs_header_nritems(src); |
3203 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3204 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3205 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3206 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3207 | |
bce4eae9 | 3208 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3209 | return 1; |
3210 | ||
d397712b | 3211 | if (push_items <= 0) |
be0e5c09 CM |
3212 | return 1; |
3213 | ||
bce4eae9 | 3214 | if (empty) { |
971a1f66 | 3215 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3216 | if (push_items < src_nritems) { |
3217 | /* leave at least 8 pointers in the node if | |
3218 | * we aren't going to empty it | |
3219 | */ | |
3220 | if (src_nritems - push_items < 8) { | |
3221 | if (push_items <= 8) | |
3222 | return 1; | |
3223 | push_items -= 8; | |
3224 | } | |
3225 | } | |
3226 | } else | |
3227 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3228 | |
ed874f0d | 3229 | ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 3230 | if (ret) { |
66642832 | 3231 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3232 | return ret; |
3233 | } | |
5f39d397 CM |
3234 | copy_extent_buffer(dst, src, |
3235 | btrfs_node_key_ptr_offset(dst_nritems), | |
3236 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3237 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3238 | |
bb803951 | 3239 | if (push_items < src_nritems) { |
57911b8b | 3240 | /* |
bf1d3425 DS |
3241 | * Don't call tree_mod_log_insert_move here, key removal was |
3242 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3243 | */ |
5f39d397 CM |
3244 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3245 | btrfs_node_key_ptr_offset(push_items), | |
3246 | (src_nritems - push_items) * | |
3247 | sizeof(struct btrfs_key_ptr)); | |
3248 | } | |
3249 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3250 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3251 | btrfs_mark_buffer_dirty(src); | |
3252 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3253 | |
79f95c82 CM |
3254 | return ret; |
3255 | } | |
3256 | ||
3257 | /* | |
3258 | * try to push data from one node into the next node right in the | |
3259 | * tree. | |
3260 | * | |
3261 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3262 | * error, and > 0 if there was no room in the right hand block. | |
3263 | * | |
3264 | * this will only push up to 1/2 the contents of the left node over | |
3265 | */ | |
5f39d397 | 3266 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3267 | struct extent_buffer *dst, |
3268 | struct extent_buffer *src) | |
79f95c82 | 3269 | { |
55d32ed8 | 3270 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
3271 | int push_items = 0; |
3272 | int max_push; | |
3273 | int src_nritems; | |
3274 | int dst_nritems; | |
3275 | int ret = 0; | |
79f95c82 | 3276 | |
7bb86316 CM |
3277 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3278 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3279 | ||
5f39d397 CM |
3280 | src_nritems = btrfs_header_nritems(src); |
3281 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3282 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3283 | if (push_items <= 0) |
79f95c82 | 3284 | return 1; |
bce4eae9 | 3285 | |
d397712b | 3286 | if (src_nritems < 4) |
bce4eae9 | 3287 | return 1; |
79f95c82 CM |
3288 | |
3289 | max_push = src_nritems / 2 + 1; | |
3290 | /* don't try to empty the node */ | |
d397712b | 3291 | if (max_push >= src_nritems) |
79f95c82 | 3292 | return 1; |
252c38f0 | 3293 | |
79f95c82 CM |
3294 | if (max_push < push_items) |
3295 | push_items = max_push; | |
3296 | ||
bf1d3425 DS |
3297 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3298 | BUG_ON(ret < 0); | |
5f39d397 CM |
3299 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3300 | btrfs_node_key_ptr_offset(0), | |
3301 | (dst_nritems) * | |
3302 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3303 | |
ed874f0d DS |
3304 | ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
3305 | push_items); | |
5de865ee | 3306 | if (ret) { |
66642832 | 3307 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3308 | return ret; |
3309 | } | |
5f39d397 CM |
3310 | copy_extent_buffer(dst, src, |
3311 | btrfs_node_key_ptr_offset(0), | |
3312 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3313 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3314 | |
5f39d397 CM |
3315 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3316 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3317 | |
5f39d397 CM |
3318 | btrfs_mark_buffer_dirty(src); |
3319 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3320 | |
aa5d6bed | 3321 | return ret; |
be0e5c09 CM |
3322 | } |
3323 | ||
97571fd0 CM |
3324 | /* |
3325 | * helper function to insert a new root level in the tree. | |
3326 | * A new node is allocated, and a single item is inserted to | |
3327 | * point to the existing root | |
aa5d6bed CM |
3328 | * |
3329 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3330 | */ |
d397712b | 3331 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3332 | struct btrfs_root *root, |
fdd99c72 | 3333 | struct btrfs_path *path, int level) |
5c680ed6 | 3334 | { |
0b246afa | 3335 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3336 | u64 lower_gen; |
5f39d397 CM |
3337 | struct extent_buffer *lower; |
3338 | struct extent_buffer *c; | |
925baedd | 3339 | struct extent_buffer *old; |
5f39d397 | 3340 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3341 | int ret; |
5c680ed6 CM |
3342 | |
3343 | BUG_ON(path->nodes[level]); | |
3344 | BUG_ON(path->nodes[level-1] != root->node); | |
3345 | ||
7bb86316 CM |
3346 | lower = path->nodes[level-1]; |
3347 | if (level == 1) | |
3348 | btrfs_item_key(lower, &lower_key, 0); | |
3349 | else | |
3350 | btrfs_node_key(lower, &lower_key, 0); | |
3351 | ||
a6279470 FM |
3352 | c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, |
3353 | root->node->start, 0); | |
5f39d397 CM |
3354 | if (IS_ERR(c)) |
3355 | return PTR_ERR(c); | |
925baedd | 3356 | |
0b246afa | 3357 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3358 | |
5f39d397 | 3359 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3360 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3361 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3362 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3363 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3364 | |
3365 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3366 | |
5f39d397 | 3367 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3368 | |
925baedd | 3369 | old = root->node; |
d9d19a01 DS |
3370 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3371 | BUG_ON(ret < 0); | |
240f62c8 | 3372 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3373 | |
3374 | /* the super has an extra ref to root->node */ | |
3375 | free_extent_buffer(old); | |
3376 | ||
0b86a832 | 3377 | add_root_to_dirty_list(root); |
5f39d397 CM |
3378 | extent_buffer_get(c); |
3379 | path->nodes[level] = c; | |
95449a16 | 3380 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3381 | path->slots[level] = 0; |
3382 | return 0; | |
3383 | } | |
3384 | ||
74123bd7 CM |
3385 | /* |
3386 | * worker function to insert a single pointer in a node. | |
3387 | * the node should have enough room for the pointer already | |
97571fd0 | 3388 | * |
74123bd7 CM |
3389 | * slot and level indicate where you want the key to go, and |
3390 | * blocknr is the block the key points to. | |
3391 | */ | |
143bede5 | 3392 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 3393 | struct btrfs_path *path, |
143bede5 | 3394 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3395 | int slot, int level) |
74123bd7 | 3396 | { |
5f39d397 | 3397 | struct extent_buffer *lower; |
74123bd7 | 3398 | int nritems; |
f3ea38da | 3399 | int ret; |
5c680ed6 CM |
3400 | |
3401 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3402 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3403 | lower = path->nodes[level]; |
3404 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3405 | BUG_ON(slot > nritems); |
6ad3cf6d | 3406 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 3407 | if (slot != nritems) { |
bf1d3425 DS |
3408 | if (level) { |
3409 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3410 | nritems - slot); |
bf1d3425 DS |
3411 | BUG_ON(ret < 0); |
3412 | } | |
5f39d397 CM |
3413 | memmove_extent_buffer(lower, |
3414 | btrfs_node_key_ptr_offset(slot + 1), | |
3415 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3416 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3417 | } |
c3e06965 | 3418 | if (level) { |
e09c2efe DS |
3419 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3420 | GFP_NOFS); | |
f3ea38da JS |
3421 | BUG_ON(ret < 0); |
3422 | } | |
5f39d397 | 3423 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3424 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3425 | WARN_ON(trans->transid == 0); |
3426 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3427 | btrfs_set_header_nritems(lower, nritems + 1); |
3428 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3429 | } |
3430 | ||
97571fd0 CM |
3431 | /* |
3432 | * split the node at the specified level in path in two. | |
3433 | * The path is corrected to point to the appropriate node after the split | |
3434 | * | |
3435 | * Before splitting this tries to make some room in the node by pushing | |
3436 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3437 | * |
3438 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3439 | */ |
e02119d5 CM |
3440 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3441 | struct btrfs_root *root, | |
3442 | struct btrfs_path *path, int level) | |
be0e5c09 | 3443 | { |
0b246afa | 3444 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3445 | struct extent_buffer *c; |
3446 | struct extent_buffer *split; | |
3447 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3448 | int mid; |
5c680ed6 | 3449 | int ret; |
7518a238 | 3450 | u32 c_nritems; |
eb60ceac | 3451 | |
5f39d397 | 3452 | c = path->nodes[level]; |
7bb86316 | 3453 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3454 | if (c == root->node) { |
d9abbf1c | 3455 | /* |
90f8d62e JS |
3456 | * trying to split the root, lets make a new one |
3457 | * | |
fdd99c72 | 3458 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3459 | * insert_new_root, because that root buffer will be kept as a |
3460 | * normal node. We are going to log removal of half of the | |
3461 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3462 | * tree lock on the buffer, which is why we cannot race with | |
3463 | * other tree_mod_log users. | |
d9abbf1c | 3464 | */ |
fdd99c72 | 3465 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3466 | if (ret) |
3467 | return ret; | |
b3612421 | 3468 | } else { |
e66f709b | 3469 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3470 | c = path->nodes[level]; |
3471 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3472 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3473 | return 0; |
54aa1f4d CM |
3474 | if (ret < 0) |
3475 | return ret; | |
be0e5c09 | 3476 | } |
e66f709b | 3477 | |
5f39d397 | 3478 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3479 | mid = (c_nritems + 1) / 2; |
3480 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3481 | |
a6279470 FM |
3482 | split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, |
3483 | c->start, 0); | |
5f39d397 CM |
3484 | if (IS_ERR(split)) |
3485 | return PTR_ERR(split); | |
3486 | ||
0b246afa | 3487 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3488 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3489 | |
ed874f0d | 3490 | ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 3491 | if (ret) { |
66642832 | 3492 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3493 | return ret; |
3494 | } | |
5f39d397 CM |
3495 | copy_extent_buffer(split, c, |
3496 | btrfs_node_key_ptr_offset(0), | |
3497 | btrfs_node_key_ptr_offset(mid), | |
3498 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3499 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3500 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3501 | ret = 0; |
3502 | ||
5f39d397 CM |
3503 | btrfs_mark_buffer_dirty(c); |
3504 | btrfs_mark_buffer_dirty(split); | |
3505 | ||
6ad3cf6d | 3506 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 3507 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3508 | |
5de08d7d | 3509 | if (path->slots[level] >= mid) { |
5c680ed6 | 3510 | path->slots[level] -= mid; |
925baedd | 3511 | btrfs_tree_unlock(c); |
5f39d397 CM |
3512 | free_extent_buffer(c); |
3513 | path->nodes[level] = split; | |
5c680ed6 CM |
3514 | path->slots[level + 1] += 1; |
3515 | } else { | |
925baedd | 3516 | btrfs_tree_unlock(split); |
5f39d397 | 3517 | free_extent_buffer(split); |
be0e5c09 | 3518 | } |
aa5d6bed | 3519 | return ret; |
be0e5c09 CM |
3520 | } |
3521 | ||
74123bd7 CM |
3522 | /* |
3523 | * how many bytes are required to store the items in a leaf. start | |
3524 | * and nr indicate which items in the leaf to check. This totals up the | |
3525 | * space used both by the item structs and the item data | |
3526 | */ | |
5f39d397 | 3527 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3528 | { |
41be1f3b JB |
3529 | struct btrfs_item *start_item; |
3530 | struct btrfs_item *end_item; | |
3531 | struct btrfs_map_token token; | |
be0e5c09 | 3532 | int data_len; |
5f39d397 | 3533 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3534 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3535 | |
3536 | if (!nr) | |
3537 | return 0; | |
c82f823c | 3538 | btrfs_init_map_token(&token, l); |
dd3cc16b RK |
3539 | start_item = btrfs_item_nr(start); |
3540 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3541 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3542 | btrfs_token_item_size(l, start_item, &token); | |
3543 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3544 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3545 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3546 | return data_len; |
3547 | } | |
3548 | ||
d4dbff95 CM |
3549 | /* |
3550 | * The space between the end of the leaf items and | |
3551 | * the start of the leaf data. IOW, how much room | |
3552 | * the leaf has left for both items and data | |
3553 | */ | |
e902baac | 3554 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 3555 | { |
e902baac | 3556 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
3557 | int nritems = btrfs_header_nritems(leaf); |
3558 | int ret; | |
0b246afa JM |
3559 | |
3560 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3561 | if (ret < 0) { |
0b246afa JM |
3562 | btrfs_crit(fs_info, |
3563 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3564 | ret, | |
3565 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3566 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3567 | } |
3568 | return ret; | |
d4dbff95 CM |
3569 | } |
3570 | ||
99d8f83c CM |
3571 | /* |
3572 | * min slot controls the lowest index we're willing to push to the | |
3573 | * right. We'll push up to and including min_slot, but no lower | |
3574 | */ | |
f72f0010 | 3575 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
3576 | int data_size, int empty, |
3577 | struct extent_buffer *right, | |
99d8f83c CM |
3578 | int free_space, u32 left_nritems, |
3579 | u32 min_slot) | |
00ec4c51 | 3580 | { |
f72f0010 | 3581 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 3582 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3583 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3584 | struct btrfs_map_token token; |
5f39d397 | 3585 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3586 | int slot; |
34a38218 | 3587 | u32 i; |
00ec4c51 CM |
3588 | int push_space = 0; |
3589 | int push_items = 0; | |
0783fcfc | 3590 | struct btrfs_item *item; |
34a38218 | 3591 | u32 nr; |
7518a238 | 3592 | u32 right_nritems; |
5f39d397 | 3593 | u32 data_end; |
db94535d | 3594 | u32 this_item_size; |
00ec4c51 | 3595 | |
34a38218 CM |
3596 | if (empty) |
3597 | nr = 0; | |
3598 | else | |
99d8f83c | 3599 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3600 | |
31840ae1 | 3601 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3602 | push_space += data_size; |
31840ae1 | 3603 | |
44871b1b | 3604 | slot = path->slots[1]; |
34a38218 CM |
3605 | i = left_nritems - 1; |
3606 | while (i >= nr) { | |
dd3cc16b | 3607 | item = btrfs_item_nr(i); |
db94535d | 3608 | |
31840ae1 ZY |
3609 | if (!empty && push_items > 0) { |
3610 | if (path->slots[0] > i) | |
3611 | break; | |
3612 | if (path->slots[0] == i) { | |
e902baac DS |
3613 | int space = btrfs_leaf_free_space(left); |
3614 | ||
31840ae1 ZY |
3615 | if (space + push_space * 2 > free_space) |
3616 | break; | |
3617 | } | |
3618 | } | |
3619 | ||
00ec4c51 | 3620 | if (path->slots[0] == i) |
87b29b20 | 3621 | push_space += data_size; |
db94535d | 3622 | |
db94535d CM |
3623 | this_item_size = btrfs_item_size(left, item); |
3624 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3625 | break; |
31840ae1 | 3626 | |
00ec4c51 | 3627 | push_items++; |
db94535d | 3628 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3629 | if (i == 0) |
3630 | break; | |
3631 | i--; | |
db94535d | 3632 | } |
5f39d397 | 3633 | |
925baedd CM |
3634 | if (push_items == 0) |
3635 | goto out_unlock; | |
5f39d397 | 3636 | |
6c1500f2 | 3637 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3638 | |
00ec4c51 | 3639 | /* push left to right */ |
5f39d397 | 3640 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3641 | |
5f39d397 | 3642 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
8f881e8c | 3643 | push_space -= leaf_data_end(left); |
5f39d397 | 3644 | |
00ec4c51 | 3645 | /* make room in the right data area */ |
8f881e8c | 3646 | data_end = leaf_data_end(right); |
5f39d397 | 3647 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3648 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3649 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3650 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3651 | |
00ec4c51 | 3652 | /* copy from the left data area */ |
3d9ec8c4 | 3653 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3654 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 3655 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 3656 | push_space); |
5f39d397 CM |
3657 | |
3658 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3659 | btrfs_item_nr_offset(0), | |
3660 | right_nritems * sizeof(struct btrfs_item)); | |
3661 | ||
00ec4c51 | 3662 | /* copy the items from left to right */ |
5f39d397 CM |
3663 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3664 | btrfs_item_nr_offset(left_nritems - push_items), | |
3665 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3666 | |
3667 | /* update the item pointers */ | |
c82f823c | 3668 | btrfs_init_map_token(&token, right); |
7518a238 | 3669 | right_nritems += push_items; |
5f39d397 | 3670 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3671 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3672 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3673 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3674 | push_space -= btrfs_token_item_size(right, item, &token); |
3675 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3676 | } |
3677 | ||
7518a238 | 3678 | left_nritems -= push_items; |
5f39d397 | 3679 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3680 | |
34a38218 CM |
3681 | if (left_nritems) |
3682 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3683 | else |
6a884d7d | 3684 | btrfs_clean_tree_block(left); |
f0486c68 | 3685 | |
5f39d397 | 3686 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3687 | |
5f39d397 CM |
3688 | btrfs_item_key(right, &disk_key, 0); |
3689 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3690 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3691 | |
00ec4c51 | 3692 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3693 | if (path->slots[0] >= left_nritems) { |
3694 | path->slots[0] -= left_nritems; | |
925baedd | 3695 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 3696 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 3697 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3698 | free_extent_buffer(path->nodes[0]); |
3699 | path->nodes[0] = right; | |
00ec4c51 CM |
3700 | path->slots[1] += 1; |
3701 | } else { | |
925baedd | 3702 | btrfs_tree_unlock(right); |
5f39d397 | 3703 | free_extent_buffer(right); |
00ec4c51 CM |
3704 | } |
3705 | return 0; | |
925baedd CM |
3706 | |
3707 | out_unlock: | |
3708 | btrfs_tree_unlock(right); | |
3709 | free_extent_buffer(right); | |
3710 | return 1; | |
00ec4c51 | 3711 | } |
925baedd | 3712 | |
44871b1b CM |
3713 | /* |
3714 | * push some data in the path leaf to the right, trying to free up at | |
3715 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3716 | * | |
3717 | * returns 1 if the push failed because the other node didn't have enough | |
3718 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3719 | * |
3720 | * this will push starting from min_slot to the end of the leaf. It won't | |
3721 | * push any slot lower than min_slot | |
44871b1b CM |
3722 | */ |
3723 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3724 | *root, struct btrfs_path *path, |
3725 | int min_data_size, int data_size, | |
3726 | int empty, u32 min_slot) | |
44871b1b CM |
3727 | { |
3728 | struct extent_buffer *left = path->nodes[0]; | |
3729 | struct extent_buffer *right; | |
3730 | struct extent_buffer *upper; | |
3731 | int slot; | |
3732 | int free_space; | |
3733 | u32 left_nritems; | |
3734 | int ret; | |
3735 | ||
3736 | if (!path->nodes[1]) | |
3737 | return 1; | |
3738 | ||
3739 | slot = path->slots[1]; | |
3740 | upper = path->nodes[1]; | |
3741 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3742 | return 1; | |
3743 | ||
3744 | btrfs_assert_tree_locked(path->nodes[1]); | |
3745 | ||
4b231ae4 | 3746 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3747 | /* |
3748 | * slot + 1 is not valid or we fail to read the right node, | |
3749 | * no big deal, just return. | |
3750 | */ | |
3751 | if (IS_ERR(right)) | |
91ca338d TI |
3752 | return 1; |
3753 | ||
44871b1b | 3754 | btrfs_tree_lock(right); |
8bead258 | 3755 | btrfs_set_lock_blocking_write(right); |
44871b1b | 3756 | |
e902baac | 3757 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3758 | if (free_space < data_size) |
3759 | goto out_unlock; | |
3760 | ||
3761 | /* cow and double check */ | |
3762 | ret = btrfs_cow_block(trans, root, right, upper, | |
3763 | slot + 1, &right); | |
3764 | if (ret) | |
3765 | goto out_unlock; | |
3766 | ||
e902baac | 3767 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3768 | if (free_space < data_size) |
3769 | goto out_unlock; | |
3770 | ||
3771 | left_nritems = btrfs_header_nritems(left); | |
3772 | if (left_nritems == 0) | |
3773 | goto out_unlock; | |
3774 | ||
2ef1fed2 FDBM |
3775 | if (path->slots[0] == left_nritems && !empty) { |
3776 | /* Key greater than all keys in the leaf, right neighbor has | |
3777 | * enough room for it and we're not emptying our leaf to delete | |
3778 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3779 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3780 | btrfs_tree_unlock(left); |
3781 | free_extent_buffer(left); | |
3782 | path->nodes[0] = right; | |
3783 | path->slots[0] = 0; | |
3784 | path->slots[1]++; | |
3785 | return 0; | |
3786 | } | |
3787 | ||
f72f0010 | 3788 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3789 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3790 | out_unlock: |
3791 | btrfs_tree_unlock(right); | |
3792 | free_extent_buffer(right); | |
3793 | return 1; | |
3794 | } | |
3795 | ||
74123bd7 CM |
3796 | /* |
3797 | * push some data in the path leaf to the left, trying to free up at | |
3798 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3799 | * |
3800 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3801 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3802 | * items | |
74123bd7 | 3803 | */ |
8087c193 | 3804 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3805 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3806 | int free_space, u32 right_nritems, |
3807 | u32 max_slot) | |
be0e5c09 | 3808 | { |
8087c193 | 3809 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3810 | struct btrfs_disk_key disk_key; |
3811 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3812 | int i; |
be0e5c09 CM |
3813 | int push_space = 0; |
3814 | int push_items = 0; | |
0783fcfc | 3815 | struct btrfs_item *item; |
7518a238 | 3816 | u32 old_left_nritems; |
34a38218 | 3817 | u32 nr; |
aa5d6bed | 3818 | int ret = 0; |
db94535d CM |
3819 | u32 this_item_size; |
3820 | u32 old_left_item_size; | |
cfed81a0 CM |
3821 | struct btrfs_map_token token; |
3822 | ||
34a38218 | 3823 | if (empty) |
99d8f83c | 3824 | nr = min(right_nritems, max_slot); |
34a38218 | 3825 | else |
99d8f83c | 3826 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3827 | |
3828 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3829 | item = btrfs_item_nr(i); |
db94535d | 3830 | |
31840ae1 ZY |
3831 | if (!empty && push_items > 0) { |
3832 | if (path->slots[0] < i) | |
3833 | break; | |
3834 | if (path->slots[0] == i) { | |
e902baac DS |
3835 | int space = btrfs_leaf_free_space(right); |
3836 | ||
31840ae1 ZY |
3837 | if (space + push_space * 2 > free_space) |
3838 | break; | |
3839 | } | |
3840 | } | |
3841 | ||
be0e5c09 | 3842 | if (path->slots[0] == i) |
87b29b20 | 3843 | push_space += data_size; |
db94535d CM |
3844 | |
3845 | this_item_size = btrfs_item_size(right, item); | |
3846 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3847 | break; |
db94535d | 3848 | |
be0e5c09 | 3849 | push_items++; |
db94535d CM |
3850 | push_space += this_item_size + sizeof(*item); |
3851 | } | |
3852 | ||
be0e5c09 | 3853 | if (push_items == 0) { |
925baedd CM |
3854 | ret = 1; |
3855 | goto out; | |
be0e5c09 | 3856 | } |
fae7f21c | 3857 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3858 | |
be0e5c09 | 3859 | /* push data from right to left */ |
5f39d397 CM |
3860 | copy_extent_buffer(left, right, |
3861 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3862 | btrfs_item_nr_offset(0), | |
3863 | push_items * sizeof(struct btrfs_item)); | |
3864 | ||
0b246afa | 3865 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3866 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3867 | |
3d9ec8c4 | 3868 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3869 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3870 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3871 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3872 | push_space); |
5f39d397 | 3873 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3874 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3875 | |
c82f823c | 3876 | btrfs_init_map_token(&token, left); |
db94535d | 3877 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3878 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3879 | u32 ioff; |
db94535d | 3880 | |
dd3cc16b | 3881 | item = btrfs_item_nr(i); |
db94535d | 3882 | |
cfed81a0 CM |
3883 | ioff = btrfs_token_item_offset(left, item, &token); |
3884 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3885 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3886 | &token); |
be0e5c09 | 3887 | } |
5f39d397 | 3888 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3889 | |
3890 | /* fixup right node */ | |
31b1a2bd JL |
3891 | if (push_items > right_nritems) |
3892 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3893 | right_nritems); |
34a38218 CM |
3894 | |
3895 | if (push_items < right_nritems) { | |
3896 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
8f881e8c | 3897 | leaf_data_end(right); |
3d9ec8c4 | 3898 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3899 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3900 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3901 | leaf_data_end(right), push_space); |
34a38218 CM |
3902 | |
3903 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3904 | btrfs_item_nr_offset(push_items), |
3905 | (btrfs_header_nritems(right) - push_items) * | |
3906 | sizeof(struct btrfs_item)); | |
34a38218 | 3907 | } |
c82f823c DS |
3908 | |
3909 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3910 | right_nritems -= push_items; |
3911 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3912 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3913 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3914 | item = btrfs_item_nr(i); |
db94535d | 3915 | |
cfed81a0 CM |
3916 | push_space = push_space - btrfs_token_item_size(right, |
3917 | item, &token); | |
3918 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3919 | } |
eb60ceac | 3920 | |
5f39d397 | 3921 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3922 | if (right_nritems) |
3923 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3924 | else |
6a884d7d | 3925 | btrfs_clean_tree_block(right); |
098f59c2 | 3926 | |
5f39d397 | 3927 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3928 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3929 | |
3930 | /* then fixup the leaf pointer in the path */ | |
3931 | if (path->slots[0] < push_items) { | |
3932 | path->slots[0] += old_left_nritems; | |
925baedd | 3933 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3934 | free_extent_buffer(path->nodes[0]); |
3935 | path->nodes[0] = left; | |
be0e5c09 CM |
3936 | path->slots[1] -= 1; |
3937 | } else { | |
925baedd | 3938 | btrfs_tree_unlock(left); |
5f39d397 | 3939 | free_extent_buffer(left); |
be0e5c09 CM |
3940 | path->slots[0] -= push_items; |
3941 | } | |
eb60ceac | 3942 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3943 | return ret; |
925baedd CM |
3944 | out: |
3945 | btrfs_tree_unlock(left); | |
3946 | free_extent_buffer(left); | |
3947 | return ret; | |
be0e5c09 CM |
3948 | } |
3949 | ||
44871b1b CM |
3950 | /* |
3951 | * push some data in the path leaf to the left, trying to free up at | |
3952 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3953 | * |
3954 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3955 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3956 | * items | |
44871b1b CM |
3957 | */ |
3958 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3959 | *root, struct btrfs_path *path, int min_data_size, |
3960 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3961 | { |
3962 | struct extent_buffer *right = path->nodes[0]; | |
3963 | struct extent_buffer *left; | |
3964 | int slot; | |
3965 | int free_space; | |
3966 | u32 right_nritems; | |
3967 | int ret = 0; | |
3968 | ||
3969 | slot = path->slots[1]; | |
3970 | if (slot == 0) | |
3971 | return 1; | |
3972 | if (!path->nodes[1]) | |
3973 | return 1; | |
3974 | ||
3975 | right_nritems = btrfs_header_nritems(right); | |
3976 | if (right_nritems == 0) | |
3977 | return 1; | |
3978 | ||
3979 | btrfs_assert_tree_locked(path->nodes[1]); | |
3980 | ||
4b231ae4 | 3981 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
3982 | /* |
3983 | * slot - 1 is not valid or we fail to read the left node, | |
3984 | * no big deal, just return. | |
3985 | */ | |
3986 | if (IS_ERR(left)) | |
91ca338d TI |
3987 | return 1; |
3988 | ||
44871b1b | 3989 | btrfs_tree_lock(left); |
8bead258 | 3990 | btrfs_set_lock_blocking_write(left); |
44871b1b | 3991 | |
e902baac | 3992 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3993 | if (free_space < data_size) { |
3994 | ret = 1; | |
3995 | goto out; | |
3996 | } | |
3997 | ||
3998 | /* cow and double check */ | |
3999 | ret = btrfs_cow_block(trans, root, left, | |
4000 | path->nodes[1], slot - 1, &left); | |
4001 | if (ret) { | |
4002 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4003 | if (ret == -ENOSPC) |
4004 | ret = 1; | |
44871b1b CM |
4005 | goto out; |
4006 | } | |
4007 | ||
e902baac | 4008 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
4009 | if (free_space < data_size) { |
4010 | ret = 1; | |
4011 | goto out; | |
4012 | } | |
4013 | ||
8087c193 | 4014 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
4015 | empty, left, free_space, right_nritems, |
4016 | max_slot); | |
44871b1b CM |
4017 | out: |
4018 | btrfs_tree_unlock(left); | |
4019 | free_extent_buffer(left); | |
4020 | return ret; | |
4021 | } | |
4022 | ||
4023 | /* | |
4024 | * split the path's leaf in two, making sure there is at least data_size | |
4025 | * available for the resulting leaf level of the path. | |
44871b1b | 4026 | */ |
143bede5 | 4027 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
4028 | struct btrfs_path *path, |
4029 | struct extent_buffer *l, | |
4030 | struct extent_buffer *right, | |
4031 | int slot, int mid, int nritems) | |
44871b1b | 4032 | { |
94f94ad9 | 4033 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
4034 | int data_copy_size; |
4035 | int rt_data_off; | |
4036 | int i; | |
44871b1b | 4037 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4038 | struct btrfs_map_token token; |
4039 | ||
44871b1b CM |
4040 | nritems = nritems - mid; |
4041 | btrfs_set_header_nritems(right, nritems); | |
8f881e8c | 4042 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); |
44871b1b CM |
4043 | |
4044 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4045 | btrfs_item_nr_offset(mid), | |
4046 | nritems * sizeof(struct btrfs_item)); | |
4047 | ||
4048 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4049 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4050 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 4051 | leaf_data_end(l), data_copy_size); |
44871b1b | 4052 | |
0b246afa | 4053 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b | 4054 | |
c82f823c | 4055 | btrfs_init_map_token(&token, right); |
44871b1b | 4056 | for (i = 0; i < nritems; i++) { |
dd3cc16b | 4057 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4058 | u32 ioff; |
4059 | ||
cfed81a0 CM |
4060 | ioff = btrfs_token_item_offset(right, item, &token); |
4061 | btrfs_set_token_item_offset(right, item, | |
4062 | ioff + rt_data_off, &token); | |
44871b1b CM |
4063 | } |
4064 | ||
44871b1b | 4065 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4066 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 4067 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
4068 | |
4069 | btrfs_mark_buffer_dirty(right); | |
4070 | btrfs_mark_buffer_dirty(l); | |
4071 | BUG_ON(path->slots[0] != slot); | |
4072 | ||
44871b1b CM |
4073 | if (mid <= slot) { |
4074 | btrfs_tree_unlock(path->nodes[0]); | |
4075 | free_extent_buffer(path->nodes[0]); | |
4076 | path->nodes[0] = right; | |
4077 | path->slots[0] -= mid; | |
4078 | path->slots[1] += 1; | |
4079 | } else { | |
4080 | btrfs_tree_unlock(right); | |
4081 | free_extent_buffer(right); | |
4082 | } | |
4083 | ||
4084 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4085 | } |
4086 | ||
99d8f83c CM |
4087 | /* |
4088 | * double splits happen when we need to insert a big item in the middle | |
4089 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4090 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4091 | * A B C | |
4092 | * | |
4093 | * We avoid this by trying to push the items on either side of our target | |
4094 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4095 | * completely. | |
4096 | */ | |
4097 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4098 | struct btrfs_root *root, | |
4099 | struct btrfs_path *path, | |
4100 | int data_size) | |
4101 | { | |
4102 | int ret; | |
4103 | int progress = 0; | |
4104 | int slot; | |
4105 | u32 nritems; | |
5a4267ca | 4106 | int space_needed = data_size; |
99d8f83c CM |
4107 | |
4108 | slot = path->slots[0]; | |
5a4267ca | 4109 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 4110 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
4111 | |
4112 | /* | |
4113 | * try to push all the items after our slot into the | |
4114 | * right leaf | |
4115 | */ | |
5a4267ca | 4116 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4117 | if (ret < 0) |
4118 | return ret; | |
4119 | ||
4120 | if (ret == 0) | |
4121 | progress++; | |
4122 | ||
4123 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4124 | /* | |
4125 | * our goal is to get our slot at the start or end of a leaf. If | |
4126 | * we've done so we're done | |
4127 | */ | |
4128 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4129 | return 0; | |
4130 | ||
e902baac | 4131 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4132 | return 0; |
4133 | ||
4134 | /* try to push all the items before our slot into the next leaf */ | |
4135 | slot = path->slots[0]; | |
263d3995 FM |
4136 | space_needed = data_size; |
4137 | if (slot > 0) | |
e902baac | 4138 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 4139 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4140 | if (ret < 0) |
4141 | return ret; | |
4142 | ||
4143 | if (ret == 0) | |
4144 | progress++; | |
4145 | ||
4146 | if (progress) | |
4147 | return 0; | |
4148 | return 1; | |
4149 | } | |
4150 | ||
74123bd7 CM |
4151 | /* |
4152 | * split the path's leaf in two, making sure there is at least data_size | |
4153 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4154 | * |
4155 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4156 | */ |
e02119d5 CM |
4157 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4158 | struct btrfs_root *root, | |
310712b2 | 4159 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4160 | struct btrfs_path *path, int data_size, |
4161 | int extend) | |
be0e5c09 | 4162 | { |
5d4f98a2 | 4163 | struct btrfs_disk_key disk_key; |
5f39d397 | 4164 | struct extent_buffer *l; |
7518a238 | 4165 | u32 nritems; |
eb60ceac CM |
4166 | int mid; |
4167 | int slot; | |
5f39d397 | 4168 | struct extent_buffer *right; |
b7a0365e | 4169 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4170 | int ret = 0; |
aa5d6bed | 4171 | int wret; |
5d4f98a2 | 4172 | int split; |
cc0c5538 | 4173 | int num_doubles = 0; |
99d8f83c | 4174 | int tried_avoid_double = 0; |
aa5d6bed | 4175 | |
a5719521 YZ |
4176 | l = path->nodes[0]; |
4177 | slot = path->slots[0]; | |
4178 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4179 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4180 | return -EOVERFLOW; |
4181 | ||
40689478 | 4182 | /* first try to make some room by pushing left and right */ |
33157e05 | 4183 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4184 | int space_needed = data_size; |
4185 | ||
4186 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 4187 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4188 | |
4189 | wret = push_leaf_right(trans, root, path, space_needed, | |
4190 | space_needed, 0, 0); | |
d397712b | 4191 | if (wret < 0) |
eaee50e8 | 4192 | return wret; |
3685f791 | 4193 | if (wret) { |
263d3995 FM |
4194 | space_needed = data_size; |
4195 | if (slot > 0) | |
e902baac | 4196 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4197 | wret = push_leaf_left(trans, root, path, space_needed, |
4198 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4199 | if (wret < 0) |
4200 | return wret; | |
4201 | } | |
4202 | l = path->nodes[0]; | |
aa5d6bed | 4203 | |
3685f791 | 4204 | /* did the pushes work? */ |
e902baac | 4205 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 4206 | return 0; |
3326d1b0 | 4207 | } |
aa5d6bed | 4208 | |
5c680ed6 | 4209 | if (!path->nodes[1]) { |
fdd99c72 | 4210 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4211 | if (ret) |
4212 | return ret; | |
4213 | } | |
cc0c5538 | 4214 | again: |
5d4f98a2 | 4215 | split = 1; |
cc0c5538 | 4216 | l = path->nodes[0]; |
eb60ceac | 4217 | slot = path->slots[0]; |
5f39d397 | 4218 | nritems = btrfs_header_nritems(l); |
d397712b | 4219 | mid = (nritems + 1) / 2; |
54aa1f4d | 4220 | |
5d4f98a2 YZ |
4221 | if (mid <= slot) { |
4222 | if (nritems == 1 || | |
4223 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4224 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4225 | if (slot >= nritems) { |
4226 | split = 0; | |
4227 | } else { | |
4228 | mid = slot; | |
4229 | if (mid != nritems && | |
4230 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4231 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4232 | if (data_size && !tried_avoid_double) |
4233 | goto push_for_double; | |
5d4f98a2 YZ |
4234 | split = 2; |
4235 | } | |
4236 | } | |
4237 | } | |
4238 | } else { | |
4239 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4240 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4241 | if (!extend && data_size && slot == 0) { |
4242 | split = 0; | |
4243 | } else if ((extend || !data_size) && slot == 0) { | |
4244 | mid = 1; | |
4245 | } else { | |
4246 | mid = slot; | |
4247 | if (mid != nritems && | |
4248 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4249 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4250 | if (data_size && !tried_avoid_double) |
4251 | goto push_for_double; | |
67871254 | 4252 | split = 2; |
5d4f98a2 YZ |
4253 | } |
4254 | } | |
4255 | } | |
4256 | } | |
4257 | ||
4258 | if (split == 0) | |
4259 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4260 | else | |
4261 | btrfs_item_key(l, &disk_key, mid); | |
4262 | ||
a6279470 FM |
4263 | right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, |
4264 | l->start, 0); | |
f0486c68 | 4265 | if (IS_ERR(right)) |
5f39d397 | 4266 | return PTR_ERR(right); |
f0486c68 | 4267 | |
0b246afa | 4268 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4269 | |
5d4f98a2 YZ |
4270 | if (split == 0) { |
4271 | if (mid <= slot) { | |
4272 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4273 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4274 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4275 | btrfs_tree_unlock(path->nodes[0]); |
4276 | free_extent_buffer(path->nodes[0]); | |
4277 | path->nodes[0] = right; | |
4278 | path->slots[0] = 0; | |
4279 | path->slots[1] += 1; | |
4280 | } else { | |
4281 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4282 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4283 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
4284 | btrfs_tree_unlock(path->nodes[0]); |
4285 | free_extent_buffer(path->nodes[0]); | |
4286 | path->nodes[0] = right; | |
4287 | path->slots[0] = 0; | |
143bede5 | 4288 | if (path->slots[1] == 0) |
b167fa91 | 4289 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4290 | } |
196e0249 LB |
4291 | /* |
4292 | * We create a new leaf 'right' for the required ins_len and | |
4293 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4294 | * the content of ins_len to 'right'. | |
4295 | */ | |
5d4f98a2 | 4296 | return ret; |
d4dbff95 | 4297 | } |
74123bd7 | 4298 | |
94f94ad9 | 4299 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 4300 | |
5d4f98a2 | 4301 | if (split == 2) { |
cc0c5538 CM |
4302 | BUG_ON(num_doubles != 0); |
4303 | num_doubles++; | |
4304 | goto again; | |
a429e513 | 4305 | } |
44871b1b | 4306 | |
143bede5 | 4307 | return 0; |
99d8f83c CM |
4308 | |
4309 | push_for_double: | |
4310 | push_for_double_split(trans, root, path, data_size); | |
4311 | tried_avoid_double = 1; | |
e902baac | 4312 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4313 | return 0; |
4314 | goto again; | |
be0e5c09 CM |
4315 | } |
4316 | ||
ad48fd75 YZ |
4317 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4318 | struct btrfs_root *root, | |
4319 | struct btrfs_path *path, int ins_len) | |
459931ec | 4320 | { |
ad48fd75 | 4321 | struct btrfs_key key; |
459931ec | 4322 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4323 | struct btrfs_file_extent_item *fi; |
4324 | u64 extent_len = 0; | |
4325 | u32 item_size; | |
4326 | int ret; | |
459931ec CM |
4327 | |
4328 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4329 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4330 | ||
4331 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4332 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4333 | ||
e902baac | 4334 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 4335 | return 0; |
459931ec CM |
4336 | |
4337 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4338 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4339 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4340 | struct btrfs_file_extent_item); | |
4341 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4342 | } | |
b3b4aa74 | 4343 | btrfs_release_path(path); |
459931ec | 4344 | |
459931ec | 4345 | path->keep_locks = 1; |
ad48fd75 YZ |
4346 | path->search_for_split = 1; |
4347 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4348 | path->search_for_split = 0; |
a8df6fe6 FM |
4349 | if (ret > 0) |
4350 | ret = -EAGAIN; | |
ad48fd75 YZ |
4351 | if (ret < 0) |
4352 | goto err; | |
459931ec | 4353 | |
ad48fd75 YZ |
4354 | ret = -EAGAIN; |
4355 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4356 | /* if our item isn't there, return now */ |
4357 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4358 | goto err; |
4359 | ||
109f6aef | 4360 | /* the leaf has changed, it now has room. return now */ |
e902baac | 4361 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
4362 | goto err; |
4363 | ||
ad48fd75 YZ |
4364 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4365 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4366 | struct btrfs_file_extent_item); | |
4367 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4368 | goto err; | |
459931ec CM |
4369 | } |
4370 | ||
b9473439 | 4371 | btrfs_set_path_blocking(path); |
ad48fd75 | 4372 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4373 | if (ret) |
4374 | goto err; | |
459931ec | 4375 | |
ad48fd75 | 4376 | path->keep_locks = 0; |
b9473439 | 4377 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4378 | return 0; |
4379 | err: | |
4380 | path->keep_locks = 0; | |
4381 | return ret; | |
4382 | } | |
4383 | ||
25263cd7 | 4384 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 4385 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4386 | unsigned long split_offset) |
4387 | { | |
4388 | struct extent_buffer *leaf; | |
4389 | struct btrfs_item *item; | |
4390 | struct btrfs_item *new_item; | |
4391 | int slot; | |
4392 | char *buf; | |
4393 | u32 nritems; | |
4394 | u32 item_size; | |
4395 | u32 orig_offset; | |
4396 | struct btrfs_disk_key disk_key; | |
4397 | ||
b9473439 | 4398 | leaf = path->nodes[0]; |
e902baac | 4399 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4400 | |
b4ce94de CM |
4401 | btrfs_set_path_blocking(path); |
4402 | ||
dd3cc16b | 4403 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4404 | orig_offset = btrfs_item_offset(leaf, item); |
4405 | item_size = btrfs_item_size(leaf, item); | |
4406 | ||
459931ec | 4407 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4408 | if (!buf) |
4409 | return -ENOMEM; | |
4410 | ||
459931ec CM |
4411 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4412 | path->slots[0]), item_size); | |
459931ec | 4413 | |
ad48fd75 | 4414 | slot = path->slots[0] + 1; |
459931ec | 4415 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4416 | if (slot != nritems) { |
4417 | /* shift the items */ | |
4418 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4419 | btrfs_item_nr_offset(slot), |
4420 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4421 | } |
4422 | ||
4423 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4424 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4425 | ||
dd3cc16b | 4426 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4427 | |
4428 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4429 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4430 | ||
4431 | btrfs_set_item_offset(leaf, item, | |
4432 | orig_offset + item_size - split_offset); | |
4433 | btrfs_set_item_size(leaf, item, split_offset); | |
4434 | ||
4435 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4436 | ||
4437 | /* write the data for the start of the original item */ | |
4438 | write_extent_buffer(leaf, buf, | |
4439 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4440 | split_offset); | |
4441 | ||
4442 | /* write the data for the new item */ | |
4443 | write_extent_buffer(leaf, buf + split_offset, | |
4444 | btrfs_item_ptr_offset(leaf, slot), | |
4445 | item_size - split_offset); | |
4446 | btrfs_mark_buffer_dirty(leaf); | |
4447 | ||
e902baac | 4448 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 4449 | kfree(buf); |
ad48fd75 YZ |
4450 | return 0; |
4451 | } | |
4452 | ||
4453 | /* | |
4454 | * This function splits a single item into two items, | |
4455 | * giving 'new_key' to the new item and splitting the | |
4456 | * old one at split_offset (from the start of the item). | |
4457 | * | |
4458 | * The path may be released by this operation. After | |
4459 | * the split, the path is pointing to the old item. The | |
4460 | * new item is going to be in the same node as the old one. | |
4461 | * | |
4462 | * Note, the item being split must be smaller enough to live alone on | |
4463 | * a tree block with room for one extra struct btrfs_item | |
4464 | * | |
4465 | * This allows us to split the item in place, keeping a lock on the | |
4466 | * leaf the entire time. | |
4467 | */ | |
4468 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4469 | struct btrfs_root *root, | |
4470 | struct btrfs_path *path, | |
310712b2 | 4471 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4472 | unsigned long split_offset) |
4473 | { | |
4474 | int ret; | |
4475 | ret = setup_leaf_for_split(trans, root, path, | |
4476 | sizeof(struct btrfs_item)); | |
4477 | if (ret) | |
4478 | return ret; | |
4479 | ||
25263cd7 | 4480 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
4481 | return ret; |
4482 | } | |
4483 | ||
ad48fd75 YZ |
4484 | /* |
4485 | * This function duplicate a item, giving 'new_key' to the new item. | |
4486 | * It guarantees both items live in the same tree leaf and the new item | |
4487 | * is contiguous with the original item. | |
4488 | * | |
4489 | * This allows us to split file extent in place, keeping a lock on the | |
4490 | * leaf the entire time. | |
4491 | */ | |
4492 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4493 | struct btrfs_root *root, | |
4494 | struct btrfs_path *path, | |
310712b2 | 4495 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4496 | { |
4497 | struct extent_buffer *leaf; | |
4498 | int ret; | |
4499 | u32 item_size; | |
4500 | ||
4501 | leaf = path->nodes[0]; | |
4502 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4503 | ret = setup_leaf_for_split(trans, root, path, | |
4504 | item_size + sizeof(struct btrfs_item)); | |
4505 | if (ret) | |
4506 | return ret; | |
4507 | ||
4508 | path->slots[0]++; | |
afe5fea7 | 4509 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4510 | item_size, item_size + |
4511 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4512 | leaf = path->nodes[0]; |
4513 | memcpy_extent_buffer(leaf, | |
4514 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4515 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4516 | item_size); | |
4517 | return 0; | |
4518 | } | |
4519 | ||
d352ac68 CM |
4520 | /* |
4521 | * make the item pointed to by the path smaller. new_size indicates | |
4522 | * how small to make it, and from_end tells us if we just chop bytes | |
4523 | * off the end of the item or if we shift the item to chop bytes off | |
4524 | * the front. | |
4525 | */ | |
78ac4f9e | 4526 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 4527 | { |
b18c6685 | 4528 | int slot; |
5f39d397 CM |
4529 | struct extent_buffer *leaf; |
4530 | struct btrfs_item *item; | |
b18c6685 CM |
4531 | u32 nritems; |
4532 | unsigned int data_end; | |
4533 | unsigned int old_data_start; | |
4534 | unsigned int old_size; | |
4535 | unsigned int size_diff; | |
4536 | int i; | |
cfed81a0 CM |
4537 | struct btrfs_map_token token; |
4538 | ||
5f39d397 | 4539 | leaf = path->nodes[0]; |
179e29e4 CM |
4540 | slot = path->slots[0]; |
4541 | ||
4542 | old_size = btrfs_item_size_nr(leaf, slot); | |
4543 | if (old_size == new_size) | |
143bede5 | 4544 | return; |
b18c6685 | 4545 | |
5f39d397 | 4546 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4547 | data_end = leaf_data_end(leaf); |
b18c6685 | 4548 | |
5f39d397 | 4549 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4550 | |
b18c6685 CM |
4551 | size_diff = old_size - new_size; |
4552 | ||
4553 | BUG_ON(slot < 0); | |
4554 | BUG_ON(slot >= nritems); | |
4555 | ||
4556 | /* | |
4557 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4558 | */ | |
4559 | /* first correct the data pointers */ | |
c82f823c | 4560 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 4561 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4562 | u32 ioff; |
dd3cc16b | 4563 | item = btrfs_item_nr(i); |
db94535d | 4564 | |
cfed81a0 CM |
4565 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4566 | btrfs_set_token_item_offset(leaf, item, | |
4567 | ioff + size_diff, &token); | |
b18c6685 | 4568 | } |
db94535d | 4569 | |
b18c6685 | 4570 | /* shift the data */ |
179e29e4 | 4571 | if (from_end) { |
3d9ec8c4 NB |
4572 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4573 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4574 | data_end, old_data_start + new_size - data_end); |
4575 | } else { | |
4576 | struct btrfs_disk_key disk_key; | |
4577 | u64 offset; | |
4578 | ||
4579 | btrfs_item_key(leaf, &disk_key, slot); | |
4580 | ||
4581 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4582 | unsigned long ptr; | |
4583 | struct btrfs_file_extent_item *fi; | |
4584 | ||
4585 | fi = btrfs_item_ptr(leaf, slot, | |
4586 | struct btrfs_file_extent_item); | |
4587 | fi = (struct btrfs_file_extent_item *)( | |
4588 | (unsigned long)fi - size_diff); | |
4589 | ||
4590 | if (btrfs_file_extent_type(leaf, fi) == | |
4591 | BTRFS_FILE_EXTENT_INLINE) { | |
4592 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4593 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4594 | (unsigned long)fi, |
7ec20afb | 4595 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4596 | } |
4597 | } | |
4598 | ||
3d9ec8c4 NB |
4599 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4600 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4601 | data_end, old_data_start - data_end); |
4602 | ||
4603 | offset = btrfs_disk_key_offset(&disk_key); | |
4604 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4605 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4606 | if (slot == 0) | |
b167fa91 | 4607 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4608 | } |
5f39d397 | 4609 | |
dd3cc16b | 4610 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4611 | btrfs_set_item_size(leaf, item, new_size); |
4612 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4613 | |
e902baac | 4614 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4615 | btrfs_print_leaf(leaf); |
b18c6685 | 4616 | BUG(); |
5f39d397 | 4617 | } |
b18c6685 CM |
4618 | } |
4619 | ||
d352ac68 | 4620 | /* |
8f69dbd2 | 4621 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4622 | */ |
c71dd880 | 4623 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 4624 | { |
6567e837 | 4625 | int slot; |
5f39d397 CM |
4626 | struct extent_buffer *leaf; |
4627 | struct btrfs_item *item; | |
6567e837 CM |
4628 | u32 nritems; |
4629 | unsigned int data_end; | |
4630 | unsigned int old_data; | |
4631 | unsigned int old_size; | |
4632 | int i; | |
cfed81a0 CM |
4633 | struct btrfs_map_token token; |
4634 | ||
5f39d397 | 4635 | leaf = path->nodes[0]; |
6567e837 | 4636 | |
5f39d397 | 4637 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4638 | data_end = leaf_data_end(leaf); |
6567e837 | 4639 | |
e902baac | 4640 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 4641 | btrfs_print_leaf(leaf); |
6567e837 | 4642 | BUG(); |
5f39d397 | 4643 | } |
6567e837 | 4644 | slot = path->slots[0]; |
5f39d397 | 4645 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4646 | |
4647 | BUG_ON(slot < 0); | |
3326d1b0 | 4648 | if (slot >= nritems) { |
a4f78750 | 4649 | btrfs_print_leaf(leaf); |
c71dd880 | 4650 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 4651 | slot, nritems); |
290342f6 | 4652 | BUG(); |
3326d1b0 | 4653 | } |
6567e837 CM |
4654 | |
4655 | /* | |
4656 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4657 | */ | |
4658 | /* first correct the data pointers */ | |
c82f823c | 4659 | btrfs_init_map_token(&token, leaf); |
6567e837 | 4660 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4661 | u32 ioff; |
dd3cc16b | 4662 | item = btrfs_item_nr(i); |
db94535d | 4663 | |
cfed81a0 CM |
4664 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4665 | btrfs_set_token_item_offset(leaf, item, | |
4666 | ioff - data_size, &token); | |
6567e837 | 4667 | } |
5f39d397 | 4668 | |
6567e837 | 4669 | /* shift the data */ |
3d9ec8c4 NB |
4670 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4671 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4672 | data_end, old_data - data_end); |
5f39d397 | 4673 | |
6567e837 | 4674 | data_end = old_data; |
5f39d397 | 4675 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4676 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4677 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4678 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4679 | |
e902baac | 4680 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4681 | btrfs_print_leaf(leaf); |
6567e837 | 4682 | BUG(); |
5f39d397 | 4683 | } |
6567e837 CM |
4684 | } |
4685 | ||
74123bd7 | 4686 | /* |
44871b1b CM |
4687 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4688 | * to save stack depth by doing the bulk of the work in a function | |
4689 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4690 | */ |
afe5fea7 | 4691 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4692 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4693 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4694 | { |
0b246afa | 4695 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4696 | struct btrfs_item *item; |
9c58309d | 4697 | int i; |
7518a238 | 4698 | u32 nritems; |
be0e5c09 | 4699 | unsigned int data_end; |
e2fa7227 | 4700 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4701 | struct extent_buffer *leaf; |
4702 | int slot; | |
cfed81a0 CM |
4703 | struct btrfs_map_token token; |
4704 | ||
24cdc847 FM |
4705 | if (path->slots[0] == 0) { |
4706 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4707 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4708 | } |
4709 | btrfs_unlock_up_safe(path, 1); | |
4710 | ||
5f39d397 | 4711 | leaf = path->nodes[0]; |
44871b1b | 4712 | slot = path->slots[0]; |
74123bd7 | 4713 | |
5f39d397 | 4714 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4715 | data_end = leaf_data_end(leaf); |
eb60ceac | 4716 | |
e902baac | 4717 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4718 | btrfs_print_leaf(leaf); |
0b246afa | 4719 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4720 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4721 | BUG(); |
d4dbff95 | 4722 | } |
5f39d397 | 4723 | |
c82f823c | 4724 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4725 | if (slot != nritems) { |
5f39d397 | 4726 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4727 | |
5f39d397 | 4728 | if (old_data < data_end) { |
a4f78750 | 4729 | btrfs_print_leaf(leaf); |
0b246afa | 4730 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4731 | slot, old_data, data_end); |
290342f6 | 4732 | BUG(); |
5f39d397 | 4733 | } |
be0e5c09 CM |
4734 | /* |
4735 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4736 | */ | |
4737 | /* first correct the data pointers */ | |
0783fcfc | 4738 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4739 | u32 ioff; |
db94535d | 4740 | |
62e85577 | 4741 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4742 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4743 | btrfs_set_token_item_offset(leaf, item, | |
4744 | ioff - total_data, &token); | |
0783fcfc | 4745 | } |
be0e5c09 | 4746 | /* shift the items */ |
9c58309d | 4747 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4748 | btrfs_item_nr_offset(slot), |
d6025579 | 4749 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4750 | |
4751 | /* shift the data */ | |
3d9ec8c4 NB |
4752 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4753 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4754 | data_end, old_data - data_end); |
be0e5c09 CM |
4755 | data_end = old_data; |
4756 | } | |
5f39d397 | 4757 | |
62e2749e | 4758 | /* setup the item for the new data */ |
9c58309d CM |
4759 | for (i = 0; i < nr; i++) { |
4760 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4761 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4762 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4763 | btrfs_set_token_item_offset(leaf, item, |
4764 | data_end - data_size[i], &token); | |
9c58309d | 4765 | data_end -= data_size[i]; |
cfed81a0 | 4766 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4767 | } |
44871b1b | 4768 | |
9c58309d | 4769 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4770 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4771 | |
e902baac | 4772 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4773 | btrfs_print_leaf(leaf); |
be0e5c09 | 4774 | BUG(); |
5f39d397 | 4775 | } |
44871b1b CM |
4776 | } |
4777 | ||
4778 | /* | |
4779 | * Given a key and some data, insert items into the tree. | |
4780 | * This does all the path init required, making room in the tree if needed. | |
4781 | */ | |
4782 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4783 | struct btrfs_root *root, | |
4784 | struct btrfs_path *path, | |
310712b2 | 4785 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4786 | int nr) |
4787 | { | |
44871b1b CM |
4788 | int ret = 0; |
4789 | int slot; | |
4790 | int i; | |
4791 | u32 total_size = 0; | |
4792 | u32 total_data = 0; | |
4793 | ||
4794 | for (i = 0; i < nr; i++) | |
4795 | total_data += data_size[i]; | |
4796 | ||
4797 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4798 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4799 | if (ret == 0) | |
4800 | return -EEXIST; | |
4801 | if (ret < 0) | |
143bede5 | 4802 | return ret; |
44871b1b | 4803 | |
44871b1b CM |
4804 | slot = path->slots[0]; |
4805 | BUG_ON(slot < 0); | |
4806 | ||
afe5fea7 | 4807 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4808 | total_data, total_size, nr); |
143bede5 | 4809 | return 0; |
62e2749e CM |
4810 | } |
4811 | ||
4812 | /* | |
4813 | * Given a key and some data, insert an item into the tree. | |
4814 | * This does all the path init required, making room in the tree if needed. | |
4815 | */ | |
310712b2 OS |
4816 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4817 | const struct btrfs_key *cpu_key, void *data, | |
4818 | u32 data_size) | |
62e2749e CM |
4819 | { |
4820 | int ret = 0; | |
2c90e5d6 | 4821 | struct btrfs_path *path; |
5f39d397 CM |
4822 | struct extent_buffer *leaf; |
4823 | unsigned long ptr; | |
62e2749e | 4824 | |
2c90e5d6 | 4825 | path = btrfs_alloc_path(); |
db5b493a TI |
4826 | if (!path) |
4827 | return -ENOMEM; | |
2c90e5d6 | 4828 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4829 | if (!ret) { |
5f39d397 CM |
4830 | leaf = path->nodes[0]; |
4831 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4832 | write_extent_buffer(leaf, data, ptr, data_size); | |
4833 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4834 | } |
2c90e5d6 | 4835 | btrfs_free_path(path); |
aa5d6bed | 4836 | return ret; |
be0e5c09 CM |
4837 | } |
4838 | ||
74123bd7 | 4839 | /* |
5de08d7d | 4840 | * delete the pointer from a given node. |
74123bd7 | 4841 | * |
d352ac68 CM |
4842 | * the tree should have been previously balanced so the deletion does not |
4843 | * empty a node. | |
74123bd7 | 4844 | */ |
afe5fea7 TI |
4845 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4846 | int level, int slot) | |
be0e5c09 | 4847 | { |
5f39d397 | 4848 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4849 | u32 nritems; |
f3ea38da | 4850 | int ret; |
be0e5c09 | 4851 | |
5f39d397 | 4852 | nritems = btrfs_header_nritems(parent); |
d397712b | 4853 | if (slot != nritems - 1) { |
bf1d3425 DS |
4854 | if (level) { |
4855 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4856 | nritems - slot - 1); |
bf1d3425 DS |
4857 | BUG_ON(ret < 0); |
4858 | } | |
5f39d397 CM |
4859 | memmove_extent_buffer(parent, |
4860 | btrfs_node_key_ptr_offset(slot), | |
4861 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4862 | sizeof(struct btrfs_key_ptr) * |
4863 | (nritems - slot - 1)); | |
57ba86c0 | 4864 | } else if (level) { |
e09c2efe DS |
4865 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4866 | GFP_NOFS); | |
57ba86c0 | 4867 | BUG_ON(ret < 0); |
bb803951 | 4868 | } |
f3ea38da | 4869 | |
7518a238 | 4870 | nritems--; |
5f39d397 | 4871 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4872 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4873 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4874 | /* just turn the root into a leaf and break */ |
5f39d397 | 4875 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4876 | } else if (slot == 0) { |
5f39d397 CM |
4877 | struct btrfs_disk_key disk_key; |
4878 | ||
4879 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4880 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4881 | } |
d6025579 | 4882 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4883 | } |
4884 | ||
323ac95b CM |
4885 | /* |
4886 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4887 | * path->nodes[1]. |
323ac95b CM |
4888 | * |
4889 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4890 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4891 | * | |
4892 | * The path must have already been setup for deleting the leaf, including | |
4893 | * all the proper balancing. path->nodes[1] must be locked. | |
4894 | */ | |
143bede5 JM |
4895 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4896 | struct btrfs_root *root, | |
4897 | struct btrfs_path *path, | |
4898 | struct extent_buffer *leaf) | |
323ac95b | 4899 | { |
5d4f98a2 | 4900 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4901 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4902 | |
4d081c41 CM |
4903 | /* |
4904 | * btrfs_free_extent is expensive, we want to make sure we | |
4905 | * aren't holding any locks when we call it | |
4906 | */ | |
4907 | btrfs_unlock_up_safe(path, 0); | |
4908 | ||
f0486c68 YZ |
4909 | root_sub_used(root, leaf->len); |
4910 | ||
3083ee2e | 4911 | extent_buffer_get(leaf); |
5581a51a | 4912 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4913 | free_extent_buffer_stale(leaf); |
323ac95b | 4914 | } |
74123bd7 CM |
4915 | /* |
4916 | * delete the item at the leaf level in path. If that empties | |
4917 | * the leaf, remove it from the tree | |
4918 | */ | |
85e21bac CM |
4919 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4920 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4921 | { |
0b246afa | 4922 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4923 | struct extent_buffer *leaf; |
4924 | struct btrfs_item *item; | |
ce0eac2a AM |
4925 | u32 last_off; |
4926 | u32 dsize = 0; | |
aa5d6bed CM |
4927 | int ret = 0; |
4928 | int wret; | |
85e21bac | 4929 | int i; |
7518a238 | 4930 | u32 nritems; |
be0e5c09 | 4931 | |
5f39d397 | 4932 | leaf = path->nodes[0]; |
85e21bac CM |
4933 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4934 | ||
4935 | for (i = 0; i < nr; i++) | |
4936 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4937 | ||
5f39d397 | 4938 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4939 | |
85e21bac | 4940 | if (slot + nr != nritems) { |
8f881e8c | 4941 | int data_end = leaf_data_end(leaf); |
c82f823c | 4942 | struct btrfs_map_token token; |
5f39d397 | 4943 | |
3d9ec8c4 | 4944 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4945 | data_end + dsize, |
3d9ec8c4 | 4946 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4947 | last_off - data_end); |
5f39d397 | 4948 | |
c82f823c | 4949 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4950 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4951 | u32 ioff; |
db94535d | 4952 | |
dd3cc16b | 4953 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4954 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4955 | btrfs_set_token_item_offset(leaf, item, | |
4956 | ioff + dsize, &token); | |
0783fcfc | 4957 | } |
db94535d | 4958 | |
5f39d397 | 4959 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4960 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4961 | sizeof(struct btrfs_item) * |
85e21bac | 4962 | (nritems - slot - nr)); |
be0e5c09 | 4963 | } |
85e21bac CM |
4964 | btrfs_set_header_nritems(leaf, nritems - nr); |
4965 | nritems -= nr; | |
5f39d397 | 4966 | |
74123bd7 | 4967 | /* delete the leaf if we've emptied it */ |
7518a238 | 4968 | if (nritems == 0) { |
5f39d397 CM |
4969 | if (leaf == root->node) { |
4970 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4971 | } else { |
f0486c68 | 4972 | btrfs_set_path_blocking(path); |
6a884d7d | 4973 | btrfs_clean_tree_block(leaf); |
143bede5 | 4974 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4975 | } |
be0e5c09 | 4976 | } else { |
7518a238 | 4977 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4978 | if (slot == 0) { |
5f39d397 CM |
4979 | struct btrfs_disk_key disk_key; |
4980 | ||
4981 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4982 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 4983 | } |
aa5d6bed | 4984 | |
74123bd7 | 4985 | /* delete the leaf if it is mostly empty */ |
0b246afa | 4986 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
4987 | /* push_leaf_left fixes the path. |
4988 | * make sure the path still points to our leaf | |
4989 | * for possible call to del_ptr below | |
4990 | */ | |
4920c9ac | 4991 | slot = path->slots[1]; |
5f39d397 CM |
4992 | extent_buffer_get(leaf); |
4993 | ||
b9473439 | 4994 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4995 | wret = push_leaf_left(trans, root, path, 1, 1, |
4996 | 1, (u32)-1); | |
54aa1f4d | 4997 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4998 | ret = wret; |
5f39d397 CM |
4999 | |
5000 | if (path->nodes[0] == leaf && | |
5001 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5002 | wret = push_leaf_right(trans, root, path, 1, |
5003 | 1, 1, 0); | |
54aa1f4d | 5004 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5005 | ret = wret; |
5006 | } | |
5f39d397 CM |
5007 | |
5008 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5009 | path->slots[1] = slot; |
143bede5 | 5010 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5011 | free_extent_buffer(leaf); |
143bede5 | 5012 | ret = 0; |
5de08d7d | 5013 | } else { |
925baedd CM |
5014 | /* if we're still in the path, make sure |
5015 | * we're dirty. Otherwise, one of the | |
5016 | * push_leaf functions must have already | |
5017 | * dirtied this buffer | |
5018 | */ | |
5019 | if (path->nodes[0] == leaf) | |
5020 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5021 | free_extent_buffer(leaf); |
be0e5c09 | 5022 | } |
d5719762 | 5023 | } else { |
5f39d397 | 5024 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5025 | } |
5026 | } | |
aa5d6bed | 5027 | return ret; |
be0e5c09 CM |
5028 | } |
5029 | ||
7bb86316 | 5030 | /* |
925baedd | 5031 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5032 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5033 | * returns < 0 on io errors. | |
d352ac68 CM |
5034 | * |
5035 | * This may release the path, and so you may lose any locks held at the | |
5036 | * time you call it. | |
7bb86316 | 5037 | */ |
16e7549f | 5038 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5039 | { |
925baedd CM |
5040 | struct btrfs_key key; |
5041 | struct btrfs_disk_key found_key; | |
5042 | int ret; | |
7bb86316 | 5043 | |
925baedd | 5044 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5045 | |
e8b0d724 | 5046 | if (key.offset > 0) { |
925baedd | 5047 | key.offset--; |
e8b0d724 | 5048 | } else if (key.type > 0) { |
925baedd | 5049 | key.type--; |
e8b0d724 FDBM |
5050 | key.offset = (u64)-1; |
5051 | } else if (key.objectid > 0) { | |
925baedd | 5052 | key.objectid--; |
e8b0d724 FDBM |
5053 | key.type = (u8)-1; |
5054 | key.offset = (u64)-1; | |
5055 | } else { | |
925baedd | 5056 | return 1; |
e8b0d724 | 5057 | } |
7bb86316 | 5058 | |
b3b4aa74 | 5059 | btrfs_release_path(path); |
925baedd CM |
5060 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5061 | if (ret < 0) | |
5062 | return ret; | |
5063 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5064 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5065 | /* |
5066 | * We might have had an item with the previous key in the tree right | |
5067 | * before we released our path. And after we released our path, that | |
5068 | * item might have been pushed to the first slot (0) of the leaf we | |
5069 | * were holding due to a tree balance. Alternatively, an item with the | |
5070 | * previous key can exist as the only element of a leaf (big fat item). | |
5071 | * Therefore account for these 2 cases, so that our callers (like | |
5072 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5073 | * the previous key we computed above. | |
5074 | */ | |
5075 | if (ret <= 0) | |
925baedd CM |
5076 | return 0; |
5077 | return 1; | |
7bb86316 CM |
5078 | } |
5079 | ||
3f157a2f CM |
5080 | /* |
5081 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5082 | * for nodes or leaves that are have a minimum transaction id. |
5083 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5084 | * |
5085 | * This does not cow, but it does stuff the starting key it finds back | |
5086 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5087 | * key and get a writable path. | |
5088 | * | |
3f157a2f CM |
5089 | * This honors path->lowest_level to prevent descent past a given level |
5090 | * of the tree. | |
5091 | * | |
d352ac68 CM |
5092 | * min_trans indicates the oldest transaction that you are interested |
5093 | * in walking through. Any nodes or leaves older than min_trans are | |
5094 | * skipped over (without reading them). | |
5095 | * | |
3f157a2f CM |
5096 | * returns zero if something useful was found, < 0 on error and 1 if there |
5097 | * was nothing in the tree that matched the search criteria. | |
5098 | */ | |
5099 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5100 | struct btrfs_path *path, |
3f157a2f CM |
5101 | u64 min_trans) |
5102 | { | |
5103 | struct extent_buffer *cur; | |
5104 | struct btrfs_key found_key; | |
5105 | int slot; | |
9652480b | 5106 | int sret; |
3f157a2f CM |
5107 | u32 nritems; |
5108 | int level; | |
5109 | int ret = 1; | |
f98de9b9 | 5110 | int keep_locks = path->keep_locks; |
3f157a2f | 5111 | |
f98de9b9 | 5112 | path->keep_locks = 1; |
3f157a2f | 5113 | again: |
bd681513 | 5114 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5115 | level = btrfs_header_level(cur); |
e02119d5 | 5116 | WARN_ON(path->nodes[level]); |
3f157a2f | 5117 | path->nodes[level] = cur; |
bd681513 | 5118 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5119 | |
5120 | if (btrfs_header_generation(cur) < min_trans) { | |
5121 | ret = 1; | |
5122 | goto out; | |
5123 | } | |
d397712b | 5124 | while (1) { |
3f157a2f CM |
5125 | nritems = btrfs_header_nritems(cur); |
5126 | level = btrfs_header_level(cur); | |
a74b35ec | 5127 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
cbca7d59 FM |
5128 | if (sret < 0) { |
5129 | ret = sret; | |
5130 | goto out; | |
5131 | } | |
3f157a2f | 5132 | |
323ac95b CM |
5133 | /* at the lowest level, we're done, setup the path and exit */ |
5134 | if (level == path->lowest_level) { | |
e02119d5 CM |
5135 | if (slot >= nritems) |
5136 | goto find_next_key; | |
3f157a2f CM |
5137 | ret = 0; |
5138 | path->slots[level] = slot; | |
5139 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5140 | goto out; | |
5141 | } | |
9652480b Y |
5142 | if (sret && slot > 0) |
5143 | slot--; | |
3f157a2f | 5144 | /* |
de78b51a ES |
5145 | * check this node pointer against the min_trans parameters. |
5146 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5147 | */ |
d397712b | 5148 | while (slot < nritems) { |
3f157a2f | 5149 | u64 gen; |
e02119d5 | 5150 | |
3f157a2f CM |
5151 | gen = btrfs_node_ptr_generation(cur, slot); |
5152 | if (gen < min_trans) { | |
5153 | slot++; | |
5154 | continue; | |
5155 | } | |
de78b51a | 5156 | break; |
3f157a2f | 5157 | } |
e02119d5 | 5158 | find_next_key: |
3f157a2f CM |
5159 | /* |
5160 | * we didn't find a candidate key in this node, walk forward | |
5161 | * and find another one | |
5162 | */ | |
5163 | if (slot >= nritems) { | |
e02119d5 | 5164 | path->slots[level] = slot; |
b4ce94de | 5165 | btrfs_set_path_blocking(path); |
e02119d5 | 5166 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5167 | min_trans); |
e02119d5 | 5168 | if (sret == 0) { |
b3b4aa74 | 5169 | btrfs_release_path(path); |
3f157a2f CM |
5170 | goto again; |
5171 | } else { | |
5172 | goto out; | |
5173 | } | |
5174 | } | |
5175 | /* save our key for returning back */ | |
5176 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5177 | path->slots[level] = slot; | |
5178 | if (level == path->lowest_level) { | |
5179 | ret = 0; | |
3f157a2f CM |
5180 | goto out; |
5181 | } | |
b4ce94de | 5182 | btrfs_set_path_blocking(path); |
4b231ae4 | 5183 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
5184 | if (IS_ERR(cur)) { |
5185 | ret = PTR_ERR(cur); | |
5186 | goto out; | |
5187 | } | |
3f157a2f | 5188 | |
bd681513 | 5189 | btrfs_tree_read_lock(cur); |
b4ce94de | 5190 | |
bd681513 | 5191 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5192 | path->nodes[level - 1] = cur; |
f7c79f30 | 5193 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5194 | } |
5195 | out: | |
f98de9b9 FM |
5196 | path->keep_locks = keep_locks; |
5197 | if (ret == 0) { | |
5198 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5199 | btrfs_set_path_blocking(path); | |
3f157a2f | 5200 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5201 | } |
3f157a2f CM |
5202 | return ret; |
5203 | } | |
5204 | ||
5205 | /* | |
5206 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5207 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5208 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5209 | * |
5210 | * 0 is returned if another key is found, < 0 if there are any errors | |
5211 | * and 1 is returned if there are no higher keys in the tree | |
5212 | * | |
5213 | * path->keep_locks should be set to 1 on the search made before | |
5214 | * calling this function. | |
5215 | */ | |
e7a84565 | 5216 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5217 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5218 | { |
e7a84565 CM |
5219 | int slot; |
5220 | struct extent_buffer *c; | |
5221 | ||
6a9fb468 | 5222 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 5223 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5224 | if (!path->nodes[level]) |
5225 | return 1; | |
5226 | ||
5227 | slot = path->slots[level] + 1; | |
5228 | c = path->nodes[level]; | |
3f157a2f | 5229 | next: |
e7a84565 | 5230 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5231 | int ret; |
5232 | int orig_lowest; | |
5233 | struct btrfs_key cur_key; | |
5234 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5235 | !path->nodes[level + 1]) | |
e7a84565 | 5236 | return 1; |
33c66f43 | 5237 | |
6a9fb468 | 5238 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
5239 | level++; |
5240 | continue; | |
5241 | } | |
5242 | ||
5243 | slot = btrfs_header_nritems(c) - 1; | |
5244 | if (level == 0) | |
5245 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5246 | else | |
5247 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5248 | ||
5249 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5250 | btrfs_release_path(path); |
33c66f43 YZ |
5251 | path->lowest_level = level; |
5252 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5253 | 0, 0); | |
5254 | path->lowest_level = orig_lowest; | |
5255 | if (ret < 0) | |
5256 | return ret; | |
5257 | ||
5258 | c = path->nodes[level]; | |
5259 | slot = path->slots[level]; | |
5260 | if (ret == 0) | |
5261 | slot++; | |
5262 | goto next; | |
e7a84565 | 5263 | } |
33c66f43 | 5264 | |
e7a84565 CM |
5265 | if (level == 0) |
5266 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5267 | else { |
3f157a2f CM |
5268 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5269 | ||
3f157a2f CM |
5270 | if (gen < min_trans) { |
5271 | slot++; | |
5272 | goto next; | |
5273 | } | |
e7a84565 | 5274 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5275 | } |
e7a84565 CM |
5276 | return 0; |
5277 | } | |
5278 | return 1; | |
5279 | } | |
5280 | ||
97571fd0 | 5281 | /* |
925baedd | 5282 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5283 | * returns 0 if it found something or 1 if there are no greater leaves. |
5284 | * returns < 0 on io errors. | |
97571fd0 | 5285 | */ |
234b63a0 | 5286 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5287 | { |
5288 | return btrfs_next_old_leaf(root, path, 0); | |
5289 | } | |
5290 | ||
5291 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5292 | u64 time_seq) | |
d97e63b6 CM |
5293 | { |
5294 | int slot; | |
8e73f275 | 5295 | int level; |
5f39d397 | 5296 | struct extent_buffer *c; |
8e73f275 | 5297 | struct extent_buffer *next; |
925baedd CM |
5298 | struct btrfs_key key; |
5299 | u32 nritems; | |
5300 | int ret; | |
8e73f275 | 5301 | int old_spinning = path->leave_spinning; |
bd681513 | 5302 | int next_rw_lock = 0; |
925baedd CM |
5303 | |
5304 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5305 | if (nritems == 0) |
925baedd | 5306 | return 1; |
925baedd | 5307 | |
8e73f275 CM |
5308 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5309 | again: | |
5310 | level = 1; | |
5311 | next = NULL; | |
bd681513 | 5312 | next_rw_lock = 0; |
b3b4aa74 | 5313 | btrfs_release_path(path); |
8e73f275 | 5314 | |
a2135011 | 5315 | path->keep_locks = 1; |
31533fb2 | 5316 | path->leave_spinning = 1; |
8e73f275 | 5317 | |
3d7806ec JS |
5318 | if (time_seq) |
5319 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5320 | else | |
5321 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5322 | path->keep_locks = 0; |
5323 | ||
5324 | if (ret < 0) | |
5325 | return ret; | |
5326 | ||
a2135011 | 5327 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5328 | /* |
5329 | * by releasing the path above we dropped all our locks. A balance | |
5330 | * could have added more items next to the key that used to be | |
5331 | * at the very end of the block. So, check again here and | |
5332 | * advance the path if there are now more items available. | |
5333 | */ | |
a2135011 | 5334 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5335 | if (ret == 0) |
5336 | path->slots[0]++; | |
8e73f275 | 5337 | ret = 0; |
925baedd CM |
5338 | goto done; |
5339 | } | |
0b43e04f LB |
5340 | /* |
5341 | * So the above check misses one case: | |
5342 | * - after releasing the path above, someone has removed the item that | |
5343 | * used to be at the very end of the block, and balance between leafs | |
5344 | * gets another one with bigger key.offset to replace it. | |
5345 | * | |
5346 | * This one should be returned as well, or we can get leaf corruption | |
5347 | * later(esp. in __btrfs_drop_extents()). | |
5348 | * | |
5349 | * And a bit more explanation about this check, | |
5350 | * with ret > 0, the key isn't found, the path points to the slot | |
5351 | * where it should be inserted, so the path->slots[0] item must be the | |
5352 | * bigger one. | |
5353 | */ | |
5354 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5355 | ret = 0; | |
5356 | goto done; | |
5357 | } | |
d97e63b6 | 5358 | |
d397712b | 5359 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5360 | if (!path->nodes[level]) { |
5361 | ret = 1; | |
5362 | goto done; | |
5363 | } | |
5f39d397 | 5364 | |
d97e63b6 CM |
5365 | slot = path->slots[level] + 1; |
5366 | c = path->nodes[level]; | |
5f39d397 | 5367 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5368 | level++; |
8e73f275 CM |
5369 | if (level == BTRFS_MAX_LEVEL) { |
5370 | ret = 1; | |
5371 | goto done; | |
5372 | } | |
d97e63b6 CM |
5373 | continue; |
5374 | } | |
5f39d397 | 5375 | |
925baedd | 5376 | if (next) { |
bd681513 | 5377 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5378 | free_extent_buffer(next); |
925baedd | 5379 | } |
5f39d397 | 5380 | |
8e73f275 | 5381 | next = c; |
bd681513 | 5382 | next_rw_lock = path->locks[level]; |
d07b8528 | 5383 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5384 | slot, &key); |
8e73f275 CM |
5385 | if (ret == -EAGAIN) |
5386 | goto again; | |
5f39d397 | 5387 | |
76a05b35 | 5388 | if (ret < 0) { |
b3b4aa74 | 5389 | btrfs_release_path(path); |
76a05b35 CM |
5390 | goto done; |
5391 | } | |
5392 | ||
5cd57b2c | 5393 | if (!path->skip_locking) { |
bd681513 | 5394 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5395 | if (!ret && time_seq) { |
5396 | /* | |
5397 | * If we don't get the lock, we may be racing | |
5398 | * with push_leaf_left, holding that lock while | |
5399 | * itself waiting for the leaf we've currently | |
5400 | * locked. To solve this situation, we give up | |
5401 | * on our lock and cycle. | |
5402 | */ | |
cf538830 | 5403 | free_extent_buffer(next); |
d42244a0 JS |
5404 | btrfs_release_path(path); |
5405 | cond_resched(); | |
5406 | goto again; | |
5407 | } | |
8e73f275 CM |
5408 | if (!ret) { |
5409 | btrfs_set_path_blocking(path); | |
bd681513 | 5410 | btrfs_tree_read_lock(next); |
8e73f275 | 5411 | } |
31533fb2 | 5412 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5413 | } |
d97e63b6 CM |
5414 | break; |
5415 | } | |
5416 | path->slots[level] = slot; | |
d397712b | 5417 | while (1) { |
d97e63b6 CM |
5418 | level--; |
5419 | c = path->nodes[level]; | |
925baedd | 5420 | if (path->locks[level]) |
bd681513 | 5421 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5422 | |
5f39d397 | 5423 | free_extent_buffer(c); |
d97e63b6 CM |
5424 | path->nodes[level] = next; |
5425 | path->slots[level] = 0; | |
a74a4b97 | 5426 | if (!path->skip_locking) |
bd681513 | 5427 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5428 | if (!level) |
5429 | break; | |
b4ce94de | 5430 | |
d07b8528 | 5431 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5432 | 0, &key); |
8e73f275 CM |
5433 | if (ret == -EAGAIN) |
5434 | goto again; | |
5435 | ||
76a05b35 | 5436 | if (ret < 0) { |
b3b4aa74 | 5437 | btrfs_release_path(path); |
76a05b35 CM |
5438 | goto done; |
5439 | } | |
5440 | ||
5cd57b2c | 5441 | if (!path->skip_locking) { |
bd681513 | 5442 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5443 | if (!ret) { |
5444 | btrfs_set_path_blocking(path); | |
bd681513 | 5445 | btrfs_tree_read_lock(next); |
bd681513 | 5446 | } |
31533fb2 | 5447 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5448 | } |
d97e63b6 | 5449 | } |
8e73f275 | 5450 | ret = 0; |
925baedd | 5451 | done: |
f7c79f30 | 5452 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5453 | path->leave_spinning = old_spinning; |
5454 | if (!old_spinning) | |
5455 | btrfs_set_path_blocking(path); | |
5456 | ||
5457 | return ret; | |
d97e63b6 | 5458 | } |
0b86a832 | 5459 | |
3f157a2f CM |
5460 | /* |
5461 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5462 | * searching until it gets past min_objectid or finds an item of 'type' | |
5463 | * | |
5464 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5465 | */ | |
0b86a832 CM |
5466 | int btrfs_previous_item(struct btrfs_root *root, |
5467 | struct btrfs_path *path, u64 min_objectid, | |
5468 | int type) | |
5469 | { | |
5470 | struct btrfs_key found_key; | |
5471 | struct extent_buffer *leaf; | |
e02119d5 | 5472 | u32 nritems; |
0b86a832 CM |
5473 | int ret; |
5474 | ||
d397712b | 5475 | while (1) { |
0b86a832 | 5476 | if (path->slots[0] == 0) { |
b4ce94de | 5477 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5478 | ret = btrfs_prev_leaf(root, path); |
5479 | if (ret != 0) | |
5480 | return ret; | |
5481 | } else { | |
5482 | path->slots[0]--; | |
5483 | } | |
5484 | leaf = path->nodes[0]; | |
e02119d5 CM |
5485 | nritems = btrfs_header_nritems(leaf); |
5486 | if (nritems == 0) | |
5487 | return 1; | |
5488 | if (path->slots[0] == nritems) | |
5489 | path->slots[0]--; | |
5490 | ||
0b86a832 | 5491 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5492 | if (found_key.objectid < min_objectid) |
5493 | break; | |
0a4eefbb YZ |
5494 | if (found_key.type == type) |
5495 | return 0; | |
e02119d5 CM |
5496 | if (found_key.objectid == min_objectid && |
5497 | found_key.type < type) | |
5498 | break; | |
0b86a832 CM |
5499 | } |
5500 | return 1; | |
5501 | } | |
ade2e0b3 WS |
5502 | |
5503 | /* | |
5504 | * search in extent tree to find a previous Metadata/Data extent item with | |
5505 | * min objecitd. | |
5506 | * | |
5507 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5508 | */ | |
5509 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5510 | struct btrfs_path *path, u64 min_objectid) | |
5511 | { | |
5512 | struct btrfs_key found_key; | |
5513 | struct extent_buffer *leaf; | |
5514 | u32 nritems; | |
5515 | int ret; | |
5516 | ||
5517 | while (1) { | |
5518 | if (path->slots[0] == 0) { | |
5519 | btrfs_set_path_blocking(path); | |
5520 | ret = btrfs_prev_leaf(root, path); | |
5521 | if (ret != 0) | |
5522 | return ret; | |
5523 | } else { | |
5524 | path->slots[0]--; | |
5525 | } | |
5526 | leaf = path->nodes[0]; | |
5527 | nritems = btrfs_header_nritems(leaf); | |
5528 | if (nritems == 0) | |
5529 | return 1; | |
5530 | if (path->slots[0] == nritems) | |
5531 | path->slots[0]--; | |
5532 | ||
5533 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5534 | if (found_key.objectid < min_objectid) | |
5535 | break; | |
5536 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5537 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5538 | return 0; | |
5539 | if (found_key.objectid == min_objectid && | |
5540 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5541 | break; | |
5542 | } | |
5543 | return 1; | |
5544 | } |