]>
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; | |
b4e967be | 35 | const char *driver; |
af024ed2 JT |
36 | } btrfs_csums[] = { |
37 | [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, | |
3951e7f0 | 38 | [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, |
3831bf00 | 39 | [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, |
352ae07b DS |
40 | [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", |
41 | .driver = "blake2b-256" }, | |
af024ed2 JT |
42 | }; |
43 | ||
44 | int btrfs_super_csum_size(const struct btrfs_super_block *s) | |
45 | { | |
46 | u16 t = btrfs_super_csum_type(s); | |
47 | /* | |
48 | * csum type is validated at mount time | |
49 | */ | |
50 | return btrfs_csums[t].size; | |
51 | } | |
52 | ||
53 | const char *btrfs_super_csum_name(u16 csum_type) | |
54 | { | |
55 | /* csum type is validated at mount time */ | |
56 | return btrfs_csums[csum_type].name; | |
57 | } | |
58 | ||
b4e967be DS |
59 | /* |
60 | * Return driver name if defined, otherwise the name that's also a valid driver | |
61 | * name | |
62 | */ | |
63 | const char *btrfs_super_csum_driver(u16 csum_type) | |
64 | { | |
65 | /* csum type is validated at mount time */ | |
66 | return btrfs_csums[csum_type].driver ?: | |
67 | btrfs_csums[csum_type].name; | |
68 | } | |
69 | ||
f7cea56c DS |
70 | size_t __const btrfs_get_num_csums(void) |
71 | { | |
72 | return ARRAY_SIZE(btrfs_csums); | |
73 | } | |
74 | ||
df24a2b9 | 75 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 76 | { |
e2c89907 | 77 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
78 | } |
79 | ||
d352ac68 | 80 | /* this also releases the path */ |
df24a2b9 | 81 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 82 | { |
ff175d57 JJ |
83 | if (!p) |
84 | return; | |
b3b4aa74 | 85 | btrfs_release_path(p); |
df24a2b9 | 86 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
87 | } |
88 | ||
d352ac68 CM |
89 | /* |
90 | * path release drops references on the extent buffers in the path | |
91 | * and it drops any locks held by this path | |
92 | * | |
93 | * It is safe to call this on paths that no locks or extent buffers held. | |
94 | */ | |
b3b4aa74 | 95 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
96 | { |
97 | int i; | |
a2135011 | 98 | |
234b63a0 | 99 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 100 | p->slots[i] = 0; |
eb60ceac | 101 | if (!p->nodes[i]) |
925baedd CM |
102 | continue; |
103 | if (p->locks[i]) { | |
bd681513 | 104 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
105 | p->locks[i] = 0; |
106 | } | |
5f39d397 | 107 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 108 | p->nodes[i] = NULL; |
eb60ceac CM |
109 | } |
110 | } | |
111 | ||
d352ac68 CM |
112 | /* |
113 | * safely gets a reference on the root node of a tree. A lock | |
114 | * is not taken, so a concurrent writer may put a different node | |
115 | * at the root of the tree. See btrfs_lock_root_node for the | |
116 | * looping required. | |
117 | * | |
118 | * The extent buffer returned by this has a reference taken, so | |
119 | * it won't disappear. It may stop being the root of the tree | |
120 | * at any time because there are no locks held. | |
121 | */ | |
925baedd CM |
122 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
123 | { | |
124 | struct extent_buffer *eb; | |
240f62c8 | 125 | |
3083ee2e JB |
126 | while (1) { |
127 | rcu_read_lock(); | |
128 | eb = rcu_dereference(root->node); | |
129 | ||
130 | /* | |
131 | * RCU really hurts here, we could free up the root node because | |
01327610 | 132 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
133 | * the inc_not_zero dance and if it doesn't work then |
134 | * synchronize_rcu and try again. | |
135 | */ | |
136 | if (atomic_inc_not_zero(&eb->refs)) { | |
137 | rcu_read_unlock(); | |
138 | break; | |
139 | } | |
140 | rcu_read_unlock(); | |
141 | synchronize_rcu(); | |
142 | } | |
925baedd CM |
143 | return eb; |
144 | } | |
145 | ||
d352ac68 CM |
146 | /* loop around taking references on and locking the root node of the |
147 | * tree until you end up with a lock on the root. A locked buffer | |
148 | * is returned, with a reference held. | |
149 | */ | |
925baedd CM |
150 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
151 | { | |
152 | struct extent_buffer *eb; | |
153 | ||
d397712b | 154 | while (1) { |
925baedd CM |
155 | eb = btrfs_root_node(root); |
156 | btrfs_tree_lock(eb); | |
240f62c8 | 157 | if (eb == root->node) |
925baedd | 158 | break; |
925baedd CM |
159 | btrfs_tree_unlock(eb); |
160 | free_extent_buffer(eb); | |
161 | } | |
162 | return eb; | |
163 | } | |
164 | ||
bd681513 CM |
165 | /* loop around taking references on and locking the root node of the |
166 | * tree until you end up with a lock on the root. A locked buffer | |
167 | * is returned, with a reference held. | |
168 | */ | |
84f7d8e6 | 169 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
170 | { |
171 | struct extent_buffer *eb; | |
172 | ||
173 | while (1) { | |
174 | eb = btrfs_root_node(root); | |
175 | btrfs_tree_read_lock(eb); | |
176 | if (eb == root->node) | |
177 | break; | |
178 | btrfs_tree_read_unlock(eb); | |
179 | free_extent_buffer(eb); | |
180 | } | |
181 | return eb; | |
182 | } | |
183 | ||
d352ac68 CM |
184 | /* cowonly root (everything not a reference counted cow subvolume), just get |
185 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
186 | * get properly updated on disk. | |
187 | */ | |
0b86a832 CM |
188 | static void add_root_to_dirty_list(struct btrfs_root *root) |
189 | { | |
0b246afa JM |
190 | struct btrfs_fs_info *fs_info = root->fs_info; |
191 | ||
e7070be1 JB |
192 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
193 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
194 | return; | |
195 | ||
0b246afa | 196 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
197 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
198 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 199 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 200 | list_move_tail(&root->dirty_list, |
0b246afa | 201 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
202 | else |
203 | list_move(&root->dirty_list, | |
0b246afa | 204 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 205 | } |
0b246afa | 206 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
207 | } |
208 | ||
d352ac68 CM |
209 | /* |
210 | * used by snapshot creation to make a copy of a root for a tree with | |
211 | * a given objectid. The buffer with the new root node is returned in | |
212 | * cow_ret, and this func returns zero on success or a negative error code. | |
213 | */ | |
be20aa9d CM |
214 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
215 | struct btrfs_root *root, | |
216 | struct extent_buffer *buf, | |
217 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
218 | { | |
0b246afa | 219 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 220 | struct extent_buffer *cow; |
be20aa9d CM |
221 | int ret = 0; |
222 | int level; | |
5d4f98a2 | 223 | struct btrfs_disk_key disk_key; |
be20aa9d | 224 | |
27cdeb70 | 225 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 226 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
227 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
228 | trans->transid != root->last_trans); | |
be20aa9d CM |
229 | |
230 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
231 | if (level == 0) |
232 | btrfs_item_key(buf, &disk_key, 0); | |
233 | else | |
234 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 235 | |
4d75f8a9 DS |
236 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
237 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 238 | if (IS_ERR(cow)) |
be20aa9d CM |
239 | return PTR_ERR(cow); |
240 | ||
58e8012c | 241 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
242 | btrfs_set_header_bytenr(cow, cow->start); |
243 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
244 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
245 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
246 | BTRFS_HEADER_FLAG_RELOC); | |
247 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
248 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
249 | else | |
250 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 251 | |
de37aa51 | 252 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 253 | |
be20aa9d | 254 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 255 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 256 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 257 | else |
e339a6b0 | 258 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 259 | |
be20aa9d CM |
260 | if (ret) |
261 | return ret; | |
262 | ||
263 | btrfs_mark_buffer_dirty(cow); | |
264 | *cow_ret = cow; | |
265 | return 0; | |
266 | } | |
267 | ||
bd989ba3 JS |
268 | enum mod_log_op { |
269 | MOD_LOG_KEY_REPLACE, | |
270 | MOD_LOG_KEY_ADD, | |
271 | MOD_LOG_KEY_REMOVE, | |
272 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
273 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
274 | MOD_LOG_MOVE_KEYS, | |
275 | MOD_LOG_ROOT_REPLACE, | |
276 | }; | |
277 | ||
bd989ba3 JS |
278 | struct tree_mod_root { |
279 | u64 logical; | |
280 | u8 level; | |
281 | }; | |
282 | ||
283 | struct tree_mod_elem { | |
284 | struct rb_node node; | |
298cfd36 | 285 | u64 logical; |
097b8a7c | 286 | u64 seq; |
bd989ba3 JS |
287 | enum mod_log_op op; |
288 | ||
289 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
290 | int slot; | |
291 | ||
292 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
293 | u64 generation; | |
294 | ||
295 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
296 | struct btrfs_disk_key key; | |
297 | u64 blockptr; | |
298 | ||
299 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
300 | struct { |
301 | int dst_slot; | |
302 | int nr_items; | |
303 | } move; | |
bd989ba3 JS |
304 | |
305 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
306 | struct tree_mod_root old_root; | |
307 | }; | |
308 | ||
fc36ed7e | 309 | /* |
fcebe456 | 310 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 311 | */ |
fcebe456 | 312 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
313 | { |
314 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
315 | } | |
316 | ||
097b8a7c JS |
317 | /* |
318 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
319 | * passed does not already have a sequence number set. So when a caller expects | |
320 | * to record tree modifications, it should ensure to set elem->seq to zero | |
321 | * before calling btrfs_get_tree_mod_seq. | |
322 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
323 | * blocker was added. | |
324 | */ | |
325 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
326 | struct seq_list *elem) | |
bd989ba3 | 327 | { |
b1a09f1e | 328 | write_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 329 | if (!elem->seq) { |
fcebe456 | 330 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
331 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
332 | } | |
b1a09f1e | 333 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 334 | |
fcebe456 | 335 | return elem->seq; |
bd989ba3 JS |
336 | } |
337 | ||
338 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
339 | struct seq_list *elem) | |
340 | { | |
341 | struct rb_root *tm_root; | |
342 | struct rb_node *node; | |
343 | struct rb_node *next; | |
344 | struct seq_list *cur_elem; | |
345 | struct tree_mod_elem *tm; | |
346 | u64 min_seq = (u64)-1; | |
347 | u64 seq_putting = elem->seq; | |
348 | ||
349 | if (!seq_putting) | |
350 | return; | |
351 | ||
7227ff4d | 352 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 353 | list_del(&elem->list); |
097b8a7c | 354 | elem->seq = 0; |
bd989ba3 JS |
355 | |
356 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 357 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
358 | if (seq_putting > cur_elem->seq) { |
359 | /* | |
360 | * blocker with lower sequence number exists, we | |
361 | * cannot remove anything from the log | |
362 | */ | |
7227ff4d | 363 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 364 | return; |
bd989ba3 JS |
365 | } |
366 | min_seq = cur_elem->seq; | |
367 | } | |
368 | } | |
097b8a7c | 369 | |
bd989ba3 JS |
370 | /* |
371 | * anything that's lower than the lowest existing (read: blocked) | |
372 | * sequence number can be removed from the tree. | |
373 | */ | |
bd989ba3 JS |
374 | tm_root = &fs_info->tree_mod_log; |
375 | for (node = rb_first(tm_root); node; node = next) { | |
376 | next = rb_next(node); | |
6b4df8b6 | 377 | tm = rb_entry(node, struct tree_mod_elem, node); |
6609fee8 | 378 | if (tm->seq >= min_seq) |
bd989ba3 JS |
379 | continue; |
380 | rb_erase(node, tm_root); | |
bd989ba3 JS |
381 | kfree(tm); |
382 | } | |
b1a09f1e | 383 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
384 | } |
385 | ||
386 | /* | |
387 | * key order of the log: | |
298cfd36 | 388 | * node/leaf start address -> sequence |
bd989ba3 | 389 | * |
298cfd36 CR |
390 | * The 'start address' is the logical address of the *new* root node |
391 | * for root replace operations, or the logical address of the affected | |
392 | * block for all other operations. | |
bd989ba3 JS |
393 | */ |
394 | static noinline int | |
395 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
396 | { | |
397 | struct rb_root *tm_root; | |
398 | struct rb_node **new; | |
399 | struct rb_node *parent = NULL; | |
400 | struct tree_mod_elem *cur; | |
c8cc6341 | 401 | |
73e82fe4 DS |
402 | lockdep_assert_held_write(&fs_info->tree_mod_log_lock); |
403 | ||
fcebe456 | 404 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 405 | |
bd989ba3 JS |
406 | tm_root = &fs_info->tree_mod_log; |
407 | new = &tm_root->rb_node; | |
408 | while (*new) { | |
6b4df8b6 | 409 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 410 | parent = *new; |
298cfd36 | 411 | if (cur->logical < tm->logical) |
bd989ba3 | 412 | new = &((*new)->rb_left); |
298cfd36 | 413 | else if (cur->logical > tm->logical) |
bd989ba3 | 414 | new = &((*new)->rb_right); |
097b8a7c | 415 | else if (cur->seq < tm->seq) |
bd989ba3 | 416 | new = &((*new)->rb_left); |
097b8a7c | 417 | else if (cur->seq > tm->seq) |
bd989ba3 | 418 | new = &((*new)->rb_right); |
5de865ee FDBM |
419 | else |
420 | return -EEXIST; | |
bd989ba3 JS |
421 | } |
422 | ||
423 | rb_link_node(&tm->node, parent, new); | |
424 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 425 | return 0; |
bd989ba3 JS |
426 | } |
427 | ||
097b8a7c JS |
428 | /* |
429 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
430 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
431 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 432 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 433 | */ |
e9b7fd4d JS |
434 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
435 | struct extent_buffer *eb) { | |
436 | smp_mb(); | |
437 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
438 | return 1; | |
097b8a7c JS |
439 | if (eb && btrfs_header_level(eb) == 0) |
440 | return 1; | |
5de865ee | 441 | |
b1a09f1e | 442 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 443 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 444 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
445 | return 1; |
446 | } | |
447 | ||
e9b7fd4d JS |
448 | return 0; |
449 | } | |
450 | ||
5de865ee FDBM |
451 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
452 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
453 | struct extent_buffer *eb) | |
454 | { | |
455 | smp_mb(); | |
456 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
457 | return 0; | |
458 | if (eb && btrfs_header_level(eb) == 0) | |
459 | return 0; | |
460 | ||
461 | return 1; | |
462 | } | |
463 | ||
464 | static struct tree_mod_elem * | |
465 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
466 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 467 | { |
097b8a7c | 468 | struct tree_mod_elem *tm; |
bd989ba3 | 469 | |
c8cc6341 JB |
470 | tm = kzalloc(sizeof(*tm), flags); |
471 | if (!tm) | |
5de865ee | 472 | return NULL; |
bd989ba3 | 473 | |
298cfd36 | 474 | tm->logical = eb->start; |
bd989ba3 JS |
475 | if (op != MOD_LOG_KEY_ADD) { |
476 | btrfs_node_key(eb, &tm->key, slot); | |
477 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
478 | } | |
479 | tm->op = op; | |
480 | tm->slot = slot; | |
481 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 482 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 483 | |
5de865ee | 484 | return tm; |
097b8a7c JS |
485 | } |
486 | ||
e09c2efe DS |
487 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
488 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 489 | { |
5de865ee FDBM |
490 | struct tree_mod_elem *tm; |
491 | int ret; | |
492 | ||
e09c2efe | 493 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
494 | return 0; |
495 | ||
496 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
497 | if (!tm) | |
498 | return -ENOMEM; | |
499 | ||
e09c2efe | 500 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 501 | kfree(tm); |
097b8a7c | 502 | return 0; |
5de865ee FDBM |
503 | } |
504 | ||
e09c2efe | 505 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 506 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
507 | if (ret) |
508 | kfree(tm); | |
097b8a7c | 509 | |
5de865ee | 510 | return ret; |
097b8a7c JS |
511 | } |
512 | ||
6074d45f DS |
513 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
514 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 515 | { |
5de865ee FDBM |
516 | struct tree_mod_elem *tm = NULL; |
517 | struct tree_mod_elem **tm_list = NULL; | |
518 | int ret = 0; | |
bd989ba3 | 519 | int i; |
5de865ee | 520 | int locked = 0; |
bd989ba3 | 521 | |
6074d45f | 522 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 523 | return 0; |
bd989ba3 | 524 | |
176ef8f5 | 525 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
526 | if (!tm_list) |
527 | return -ENOMEM; | |
528 | ||
176ef8f5 | 529 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
530 | if (!tm) { |
531 | ret = -ENOMEM; | |
532 | goto free_tms; | |
533 | } | |
534 | ||
298cfd36 | 535 | tm->logical = eb->start; |
5de865ee FDBM |
536 | tm->slot = src_slot; |
537 | tm->move.dst_slot = dst_slot; | |
538 | tm->move.nr_items = nr_items; | |
539 | tm->op = MOD_LOG_MOVE_KEYS; | |
540 | ||
541 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
542 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 543 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
544 | if (!tm_list[i]) { |
545 | ret = -ENOMEM; | |
546 | goto free_tms; | |
547 | } | |
548 | } | |
549 | ||
6074d45f | 550 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
551 | goto free_tms; |
552 | locked = 1; | |
553 | ||
01763a2e JS |
554 | /* |
555 | * When we override something during the move, we log these removals. | |
556 | * This can only happen when we move towards the beginning of the | |
557 | * buffer, i.e. dst_slot < src_slot. | |
558 | */ | |
bd989ba3 | 559 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 560 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
561 | if (ret) |
562 | goto free_tms; | |
bd989ba3 JS |
563 | } |
564 | ||
6074d45f | 565 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
566 | if (ret) |
567 | goto free_tms; | |
b1a09f1e | 568 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 569 | kfree(tm_list); |
f395694c | 570 | |
5de865ee FDBM |
571 | return 0; |
572 | free_tms: | |
573 | for (i = 0; i < nr_items; i++) { | |
574 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 575 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
576 | kfree(tm_list[i]); |
577 | } | |
578 | if (locked) | |
b1a09f1e | 579 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
580 | kfree(tm_list); |
581 | kfree(tm); | |
bd989ba3 | 582 | |
5de865ee | 583 | return ret; |
bd989ba3 JS |
584 | } |
585 | ||
5de865ee FDBM |
586 | static inline int |
587 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
588 | struct tree_mod_elem **tm_list, | |
589 | int nritems) | |
097b8a7c | 590 | { |
5de865ee | 591 | int i, j; |
097b8a7c JS |
592 | int ret; |
593 | ||
097b8a7c | 594 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
595 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
596 | if (ret) { | |
597 | for (j = nritems - 1; j > i; j--) | |
598 | rb_erase(&tm_list[j]->node, | |
599 | &fs_info->tree_mod_log); | |
600 | return ret; | |
601 | } | |
097b8a7c | 602 | } |
5de865ee FDBM |
603 | |
604 | return 0; | |
097b8a7c JS |
605 | } |
606 | ||
95b757c1 DS |
607 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
608 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 609 | { |
95b757c1 | 610 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
611 | struct tree_mod_elem *tm = NULL; |
612 | struct tree_mod_elem **tm_list = NULL; | |
613 | int nritems = 0; | |
614 | int ret = 0; | |
615 | int i; | |
bd989ba3 | 616 | |
5de865ee | 617 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
618 | return 0; |
619 | ||
5de865ee FDBM |
620 | if (log_removal && btrfs_header_level(old_root) > 0) { |
621 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 622 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 623 | GFP_NOFS); |
5de865ee FDBM |
624 | if (!tm_list) { |
625 | ret = -ENOMEM; | |
626 | goto free_tms; | |
627 | } | |
628 | for (i = 0; i < nritems; i++) { | |
629 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 630 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
631 | if (!tm_list[i]) { |
632 | ret = -ENOMEM; | |
633 | goto free_tms; | |
634 | } | |
635 | } | |
636 | } | |
d9abbf1c | 637 | |
bcc8e07f | 638 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
639 | if (!tm) { |
640 | ret = -ENOMEM; | |
641 | goto free_tms; | |
642 | } | |
bd989ba3 | 643 | |
298cfd36 | 644 | tm->logical = new_root->start; |
bd989ba3 JS |
645 | tm->old_root.logical = old_root->start; |
646 | tm->old_root.level = btrfs_header_level(old_root); | |
647 | tm->generation = btrfs_header_generation(old_root); | |
648 | tm->op = MOD_LOG_ROOT_REPLACE; | |
649 | ||
5de865ee FDBM |
650 | if (tree_mod_dont_log(fs_info, NULL)) |
651 | goto free_tms; | |
652 | ||
653 | if (tm_list) | |
654 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
655 | if (!ret) | |
656 | ret = __tree_mod_log_insert(fs_info, tm); | |
657 | ||
b1a09f1e | 658 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
659 | if (ret) |
660 | goto free_tms; | |
661 | kfree(tm_list); | |
662 | ||
663 | return ret; | |
664 | ||
665 | free_tms: | |
666 | if (tm_list) { | |
667 | for (i = 0; i < nritems; i++) | |
668 | kfree(tm_list[i]); | |
669 | kfree(tm_list); | |
670 | } | |
671 | kfree(tm); | |
672 | ||
673 | return ret; | |
bd989ba3 JS |
674 | } |
675 | ||
676 | static struct tree_mod_elem * | |
677 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
678 | int smallest) | |
679 | { | |
680 | struct rb_root *tm_root; | |
681 | struct rb_node *node; | |
682 | struct tree_mod_elem *cur = NULL; | |
683 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 684 | |
b1a09f1e | 685 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
686 | tm_root = &fs_info->tree_mod_log; |
687 | node = tm_root->rb_node; | |
688 | while (node) { | |
6b4df8b6 | 689 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 690 | if (cur->logical < start) { |
bd989ba3 | 691 | node = node->rb_left; |
298cfd36 | 692 | } else if (cur->logical > start) { |
bd989ba3 | 693 | node = node->rb_right; |
097b8a7c | 694 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
695 | node = node->rb_left; |
696 | } else if (!smallest) { | |
697 | /* we want the node with the highest seq */ | |
698 | if (found) | |
097b8a7c | 699 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
700 | found = cur; |
701 | node = node->rb_left; | |
097b8a7c | 702 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
703 | /* we want the node with the smallest seq */ |
704 | if (found) | |
097b8a7c | 705 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
706 | found = cur; |
707 | node = node->rb_right; | |
708 | } else { | |
709 | found = cur; | |
710 | break; | |
711 | } | |
712 | } | |
b1a09f1e | 713 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
714 | |
715 | return found; | |
716 | } | |
717 | ||
718 | /* | |
719 | * this returns the element from the log with the smallest time sequence | |
720 | * value that's in the log (the oldest log item). any element with a time | |
721 | * sequence lower than min_seq will be ignored. | |
722 | */ | |
723 | static struct tree_mod_elem * | |
724 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
725 | u64 min_seq) | |
726 | { | |
727 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
728 | } | |
729 | ||
730 | /* | |
731 | * this returns the element from the log with the largest time sequence | |
732 | * value that's in the log (the most recent log item). any element with | |
733 | * a time sequence lower than min_seq will be ignored. | |
734 | */ | |
735 | static struct tree_mod_elem * | |
736 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
737 | { | |
738 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
739 | } | |
740 | ||
ed874f0d | 741 | static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, |
bd989ba3 | 742 | struct extent_buffer *src, unsigned long dst_offset, |
90f8d62e | 743 | unsigned long src_offset, int nr_items) |
bd989ba3 | 744 | { |
ed874f0d | 745 | struct btrfs_fs_info *fs_info = dst->fs_info; |
5de865ee FDBM |
746 | int ret = 0; |
747 | struct tree_mod_elem **tm_list = NULL; | |
748 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 749 | int i; |
5de865ee | 750 | int locked = 0; |
bd989ba3 | 751 | |
5de865ee FDBM |
752 | if (!tree_mod_need_log(fs_info, NULL)) |
753 | return 0; | |
bd989ba3 | 754 | |
c8cc6341 | 755 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
756 | return 0; |
757 | ||
31e818fe | 758 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
759 | GFP_NOFS); |
760 | if (!tm_list) | |
761 | return -ENOMEM; | |
bd989ba3 | 762 | |
5de865ee FDBM |
763 | tm_list_add = tm_list; |
764 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 765 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
766 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
767 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
768 | if (!tm_list_rem[i]) { | |
769 | ret = -ENOMEM; | |
770 | goto free_tms; | |
771 | } | |
772 | ||
773 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
774 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
775 | if (!tm_list_add[i]) { | |
776 | ret = -ENOMEM; | |
777 | goto free_tms; | |
778 | } | |
779 | } | |
780 | ||
781 | if (tree_mod_dont_log(fs_info, NULL)) | |
782 | goto free_tms; | |
783 | locked = 1; | |
784 | ||
785 | for (i = 0; i < nr_items; i++) { | |
786 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
787 | if (ret) | |
788 | goto free_tms; | |
789 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
790 | if (ret) | |
791 | goto free_tms; | |
bd989ba3 | 792 | } |
5de865ee | 793 | |
b1a09f1e | 794 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
795 | kfree(tm_list); |
796 | ||
797 | return 0; | |
798 | ||
799 | free_tms: | |
800 | for (i = 0; i < nr_items * 2; i++) { | |
801 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
802 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
803 | kfree(tm_list[i]); | |
804 | } | |
805 | if (locked) | |
b1a09f1e | 806 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
807 | kfree(tm_list); |
808 | ||
809 | return ret; | |
bd989ba3 JS |
810 | } |
811 | ||
db7279a2 | 812 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 813 | { |
5de865ee FDBM |
814 | struct tree_mod_elem **tm_list = NULL; |
815 | int nritems = 0; | |
816 | int i; | |
817 | int ret = 0; | |
818 | ||
819 | if (btrfs_header_level(eb) == 0) | |
820 | return 0; | |
821 | ||
db7279a2 | 822 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
823 | return 0; |
824 | ||
825 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 826 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
827 | if (!tm_list) |
828 | return -ENOMEM; | |
829 | ||
830 | for (i = 0; i < nritems; i++) { | |
831 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
832 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
833 | if (!tm_list[i]) { | |
834 | ret = -ENOMEM; | |
835 | goto free_tms; | |
836 | } | |
837 | } | |
838 | ||
db7279a2 | 839 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
840 | goto free_tms; |
841 | ||
db7279a2 | 842 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 843 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
844 | if (ret) |
845 | goto free_tms; | |
846 | kfree(tm_list); | |
847 | ||
848 | return 0; | |
849 | ||
850 | free_tms: | |
851 | for (i = 0; i < nritems; i++) | |
852 | kfree(tm_list[i]); | |
853 | kfree(tm_list); | |
854 | ||
855 | return ret; | |
bd989ba3 JS |
856 | } |
857 | ||
5d4f98a2 YZ |
858 | /* |
859 | * check if the tree block can be shared by multiple trees | |
860 | */ | |
861 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
862 | struct extent_buffer *buf) | |
863 | { | |
864 | /* | |
01327610 | 865 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
866 | * are never shared. If a block was allocated after the last |
867 | * snapshot and the block was not allocated by tree relocation, | |
868 | * we know the block is not shared. | |
869 | */ | |
27cdeb70 | 870 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
871 | buf != root->node && buf != root->commit_root && |
872 | (btrfs_header_generation(buf) <= | |
873 | btrfs_root_last_snapshot(&root->root_item) || | |
874 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
875 | return 1; | |
a79865c6 | 876 | |
5d4f98a2 YZ |
877 | return 0; |
878 | } | |
879 | ||
880 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
881 | struct btrfs_root *root, | |
882 | struct extent_buffer *buf, | |
f0486c68 YZ |
883 | struct extent_buffer *cow, |
884 | int *last_ref) | |
5d4f98a2 | 885 | { |
0b246afa | 886 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
887 | u64 refs; |
888 | u64 owner; | |
889 | u64 flags; | |
890 | u64 new_flags = 0; | |
891 | int ret; | |
892 | ||
893 | /* | |
894 | * Backrefs update rules: | |
895 | * | |
896 | * Always use full backrefs for extent pointers in tree block | |
897 | * allocated by tree relocation. | |
898 | * | |
899 | * If a shared tree block is no longer referenced by its owner | |
900 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
901 | * use full backrefs for extent pointers in tree block. | |
902 | * | |
903 | * If a tree block is been relocating | |
904 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
905 | * use full backrefs for extent pointers in tree block. | |
906 | * The reason for this is some operations (such as drop tree) | |
907 | * are only allowed for blocks use full backrefs. | |
908 | */ | |
909 | ||
910 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 911 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
912 | btrfs_header_level(buf), 1, |
913 | &refs, &flags); | |
be1a5564 MF |
914 | if (ret) |
915 | return ret; | |
e5df9573 MF |
916 | if (refs == 0) { |
917 | ret = -EROFS; | |
0b246afa | 918 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
919 | return ret; |
920 | } | |
5d4f98a2 YZ |
921 | } else { |
922 | refs = 1; | |
923 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
924 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
925 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
926 | else | |
927 | flags = 0; | |
928 | } | |
929 | ||
930 | owner = btrfs_header_owner(buf); | |
931 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
932 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
933 | ||
934 | if (refs > 1) { | |
935 | if ((owner == root->root_key.objectid || | |
936 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
937 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 938 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
939 | if (ret) |
940 | return ret; | |
5d4f98a2 YZ |
941 | |
942 | if (root->root_key.objectid == | |
943 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 944 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
945 | if (ret) |
946 | return ret; | |
e339a6b0 | 947 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
948 | if (ret) |
949 | return ret; | |
5d4f98a2 YZ |
950 | } |
951 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
952 | } else { | |
953 | ||
954 | if (root->root_key.objectid == | |
955 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 956 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 957 | else |
e339a6b0 | 958 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
959 | if (ret) |
960 | return ret; | |
5d4f98a2 YZ |
961 | } |
962 | if (new_flags != 0) { | |
b1c79e09 JB |
963 | int level = btrfs_header_level(buf); |
964 | ||
f5c8daa5 | 965 | ret = btrfs_set_disk_extent_flags(trans, |
5d4f98a2 YZ |
966 | buf->start, |
967 | buf->len, | |
b1c79e09 | 968 | new_flags, level, 0); |
be1a5564 MF |
969 | if (ret) |
970 | return ret; | |
5d4f98a2 YZ |
971 | } |
972 | } else { | |
973 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
974 | if (root->root_key.objectid == | |
975 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 976 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 977 | else |
e339a6b0 | 978 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
979 | if (ret) |
980 | return ret; | |
e339a6b0 | 981 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
982 | if (ret) |
983 | return ret; | |
5d4f98a2 | 984 | } |
6a884d7d | 985 | btrfs_clean_tree_block(buf); |
f0486c68 | 986 | *last_ref = 1; |
5d4f98a2 YZ |
987 | } |
988 | return 0; | |
989 | } | |
990 | ||
a6279470 FM |
991 | static struct extent_buffer *alloc_tree_block_no_bg_flush( |
992 | struct btrfs_trans_handle *trans, | |
993 | struct btrfs_root *root, | |
994 | u64 parent_start, | |
995 | const struct btrfs_disk_key *disk_key, | |
996 | int level, | |
997 | u64 hint, | |
998 | u64 empty_size) | |
999 | { | |
1000 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1001 | struct extent_buffer *ret; | |
1002 | ||
1003 | /* | |
1004 | * If we are COWing a node/leaf from the extent, chunk, device or free | |
1005 | * space trees, make sure that we do not finish block group creation of | |
1006 | * pending block groups. We do this to avoid a deadlock. | |
1007 | * COWing can result in allocation of a new chunk, and flushing pending | |
1008 | * block groups (btrfs_create_pending_block_groups()) can be triggered | |
1009 | * when finishing allocation of a new chunk. Creation of a pending block | |
1010 | * group modifies the extent, chunk, device and free space trees, | |
1011 | * therefore we could deadlock with ourselves since we are holding a | |
1012 | * lock on an extent buffer that btrfs_create_pending_block_groups() may | |
1013 | * try to COW later. | |
1014 | * For similar reasons, we also need to delay flushing pending block | |
1015 | * groups when splitting a leaf or node, from one of those trees, since | |
1016 | * we are holding a write lock on it and its parent or when inserting a | |
1017 | * new root node for one of those trees. | |
1018 | */ | |
1019 | if (root == fs_info->extent_root || | |
1020 | root == fs_info->chunk_root || | |
1021 | root == fs_info->dev_root || | |
1022 | root == fs_info->free_space_root) | |
1023 | trans->can_flush_pending_bgs = false; | |
1024 | ||
1025 | ret = btrfs_alloc_tree_block(trans, root, parent_start, | |
1026 | root->root_key.objectid, disk_key, level, | |
1027 | hint, empty_size); | |
1028 | trans->can_flush_pending_bgs = true; | |
1029 | ||
1030 | return ret; | |
1031 | } | |
1032 | ||
d352ac68 | 1033 | /* |
d397712b CM |
1034 | * does the dirty work in cow of a single block. The parent block (if |
1035 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1036 | * dirty and returned locked. If you modify the block it needs to be marked | |
1037 | * dirty again. | |
d352ac68 CM |
1038 | * |
1039 | * search_start -- an allocation hint for the new block | |
1040 | * | |
d397712b CM |
1041 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1042 | * bytes the allocator should try to find free next to the block it returns. | |
1043 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1044 | */ |
d397712b | 1045 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1046 | struct btrfs_root *root, |
1047 | struct extent_buffer *buf, | |
1048 | struct extent_buffer *parent, int parent_slot, | |
1049 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1050 | u64 search_start, u64 empty_size) |
02217ed2 | 1051 | { |
0b246afa | 1052 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1053 | struct btrfs_disk_key disk_key; |
5f39d397 | 1054 | struct extent_buffer *cow; |
be1a5564 | 1055 | int level, ret; |
f0486c68 | 1056 | int last_ref = 0; |
925baedd | 1057 | int unlock_orig = 0; |
0f5053eb | 1058 | u64 parent_start = 0; |
7bb86316 | 1059 | |
925baedd CM |
1060 | if (*cow_ret == buf) |
1061 | unlock_orig = 1; | |
1062 | ||
b9447ef8 | 1063 | btrfs_assert_tree_locked(buf); |
925baedd | 1064 | |
27cdeb70 | 1065 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1066 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1067 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1068 | trans->transid != root->last_trans); | |
5f39d397 | 1069 | |
7bb86316 | 1070 | level = btrfs_header_level(buf); |
31840ae1 | 1071 | |
5d4f98a2 YZ |
1072 | if (level == 0) |
1073 | btrfs_item_key(buf, &disk_key, 0); | |
1074 | else | |
1075 | btrfs_node_key(buf, &disk_key, 0); | |
1076 | ||
0f5053eb GR |
1077 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1078 | parent_start = parent->start; | |
5d4f98a2 | 1079 | |
a6279470 FM |
1080 | cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, |
1081 | level, search_start, empty_size); | |
54aa1f4d CM |
1082 | if (IS_ERR(cow)) |
1083 | return PTR_ERR(cow); | |
6702ed49 | 1084 | |
b4ce94de CM |
1085 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1086 | ||
58e8012c | 1087 | copy_extent_buffer_full(cow, buf); |
db94535d | 1088 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1089 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1090 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1091 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1092 | BTRFS_HEADER_FLAG_RELOC); | |
1093 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1094 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1095 | else | |
1096 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1097 | |
de37aa51 | 1098 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 1099 | |
be1a5564 | 1100 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1101 | if (ret) { |
66642832 | 1102 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1103 | return ret; |
1104 | } | |
1a40e23b | 1105 | |
27cdeb70 | 1106 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1107 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1108 | if (ret) { |
66642832 | 1109 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1110 | return ret; |
93314e3b | 1111 | } |
83d4cfd4 | 1112 | } |
3fd0a558 | 1113 | |
02217ed2 | 1114 | if (buf == root->node) { |
925baedd | 1115 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1116 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1117 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1118 | parent_start = buf->start; | |
925baedd | 1119 | |
67439dad | 1120 | atomic_inc(&cow->refs); |
d9d19a01 DS |
1121 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1122 | BUG_ON(ret < 0); | |
240f62c8 | 1123 | rcu_assign_pointer(root->node, cow); |
925baedd | 1124 | |
f0486c68 | 1125 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1126 | last_ref); |
5f39d397 | 1127 | free_extent_buffer(buf); |
0b86a832 | 1128 | add_root_to_dirty_list(root); |
02217ed2 | 1129 | } else { |
5d4f98a2 | 1130 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1131 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1132 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1133 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1134 | cow->start); |
74493f7a CM |
1135 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1136 | trans->transid); | |
d6025579 | 1137 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1138 | if (last_ref) { |
db7279a2 | 1139 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1140 | if (ret) { |
66642832 | 1141 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1142 | return ret; |
1143 | } | |
1144 | } | |
f0486c68 | 1145 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1146 | last_ref); |
02217ed2 | 1147 | } |
925baedd CM |
1148 | if (unlock_orig) |
1149 | btrfs_tree_unlock(buf); | |
3083ee2e | 1150 | free_extent_buffer_stale(buf); |
ccd467d6 | 1151 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1152 | *cow_ret = cow; |
02217ed2 CM |
1153 | return 0; |
1154 | } | |
1155 | ||
5d9e75c4 JS |
1156 | /* |
1157 | * returns the logical address of the oldest predecessor of the given root. | |
1158 | * entries older than time_seq are ignored. | |
1159 | */ | |
bcd24dab DS |
1160 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1161 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1162 | { |
1163 | struct tree_mod_elem *tm; | |
1164 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1165 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1166 | int looped = 0; |
1167 | ||
1168 | if (!time_seq) | |
35a3621b | 1169 | return NULL; |
5d9e75c4 JS |
1170 | |
1171 | /* | |
298cfd36 CR |
1172 | * the very last operation that's logged for a root is the |
1173 | * replacement operation (if it is replaced at all). this has | |
1174 | * the logical address of the *new* root, making it the very | |
1175 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1176 | */ |
1177 | while (1) { | |
bcd24dab | 1178 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1179 | time_seq); |
1180 | if (!looped && !tm) | |
35a3621b | 1181 | return NULL; |
5d9e75c4 | 1182 | /* |
28da9fb4 JS |
1183 | * if there are no tree operation for the oldest root, we simply |
1184 | * return it. this should only happen if that (old) root is at | |
1185 | * level 0. | |
5d9e75c4 | 1186 | */ |
28da9fb4 JS |
1187 | if (!tm) |
1188 | break; | |
5d9e75c4 | 1189 | |
28da9fb4 JS |
1190 | /* |
1191 | * if there's an operation that's not a root replacement, we | |
1192 | * found the oldest version of our root. normally, we'll find a | |
1193 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1194 | */ | |
5d9e75c4 JS |
1195 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1196 | break; | |
1197 | ||
1198 | found = tm; | |
1199 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1200 | looped = 1; |
1201 | } | |
1202 | ||
a95236d9 JS |
1203 | /* if there's no old root to return, return what we found instead */ |
1204 | if (!found) | |
1205 | found = tm; | |
1206 | ||
5d9e75c4 JS |
1207 | return found; |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1212 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1213 | * time_seq). |
1214 | */ | |
1215 | static void | |
f1ca7e98 JB |
1216 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1217 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1218 | { |
1219 | u32 n; | |
1220 | struct rb_node *next; | |
1221 | struct tree_mod_elem *tm = first_tm; | |
1222 | unsigned long o_dst; | |
1223 | unsigned long o_src; | |
1224 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1225 | ||
1226 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1227 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1228 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1229 | /* |
1230 | * all the operations are recorded with the operator used for | |
1231 | * the modification. as we're going backwards, we do the | |
1232 | * opposite of each operation here. | |
1233 | */ | |
1234 | switch (tm->op) { | |
1235 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1236 | BUG_ON(tm->slot < n); | |
1c697d4a | 1237 | /* Fallthrough */ |
95c80bb1 | 1238 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1239 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1240 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1241 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1242 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1243 | tm->generation); | |
4c3e6969 | 1244 | n++; |
5d9e75c4 JS |
1245 | break; |
1246 | case MOD_LOG_KEY_REPLACE: | |
1247 | BUG_ON(tm->slot >= n); | |
1248 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1249 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1250 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1251 | tm->generation); | |
1252 | break; | |
1253 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1254 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1255 | n--; |
1256 | break; | |
1257 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1258 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1259 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1260 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1261 | tm->move.nr_items * p_size); |
1262 | break; | |
1263 | case MOD_LOG_ROOT_REPLACE: | |
1264 | /* | |
1265 | * this operation is special. for roots, this must be | |
1266 | * handled explicitly before rewinding. | |
1267 | * for non-roots, this operation may exist if the node | |
1268 | * was a root: root A -> child B; then A gets empty and | |
1269 | * B is promoted to the new root. in the mod log, we'll | |
1270 | * have a root-replace operation for B, a tree block | |
1271 | * that is no root. we simply ignore that operation. | |
1272 | */ | |
1273 | break; | |
1274 | } | |
1275 | next = rb_next(&tm->node); | |
1276 | if (!next) | |
1277 | break; | |
6b4df8b6 | 1278 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1279 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1280 | break; |
1281 | } | |
b1a09f1e | 1282 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1283 | btrfs_set_header_nritems(eb, n); |
1284 | } | |
1285 | ||
47fb091f | 1286 | /* |
01327610 | 1287 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1288 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1289 | * returned buffer is always read-locked. If the returned buffer is not the | |
1290 | * input buffer, the lock on the input buffer is released and the input buffer | |
1291 | * is freed (its refcount is decremented). | |
1292 | */ | |
5d9e75c4 | 1293 | static struct extent_buffer * |
9ec72677 JB |
1294 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1295 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1296 | { |
1297 | struct extent_buffer *eb_rewin; | |
1298 | struct tree_mod_elem *tm; | |
1299 | ||
1300 | if (!time_seq) | |
1301 | return eb; | |
1302 | ||
1303 | if (btrfs_header_level(eb) == 0) | |
1304 | return eb; | |
1305 | ||
1306 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1307 | if (!tm) | |
1308 | return eb; | |
1309 | ||
9ec72677 | 1310 | btrfs_set_path_blocking(path); |
300aa896 | 1311 | btrfs_set_lock_blocking_read(eb); |
9ec72677 | 1312 | |
5d9e75c4 JS |
1313 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1314 | BUG_ON(tm->slot != 0); | |
da17066c | 1315 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1316 | if (!eb_rewin) { |
9ec72677 | 1317 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1318 | free_extent_buffer(eb); |
1319 | return NULL; | |
1320 | } | |
5d9e75c4 JS |
1321 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1322 | btrfs_set_header_backref_rev(eb_rewin, | |
1323 | btrfs_header_backref_rev(eb)); | |
1324 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1325 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1326 | } else { |
1327 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1328 | if (!eb_rewin) { |
9ec72677 | 1329 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1330 | free_extent_buffer(eb); |
1331 | return NULL; | |
1332 | } | |
5d9e75c4 JS |
1333 | } |
1334 | ||
9ec72677 | 1335 | btrfs_tree_read_unlock_blocking(eb); |
5d9e75c4 JS |
1336 | free_extent_buffer(eb); |
1337 | ||
47fb091f | 1338 | btrfs_tree_read_lock(eb_rewin); |
f1ca7e98 | 1339 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1340 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1341 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1342 | |
1343 | return eb_rewin; | |
1344 | } | |
1345 | ||
8ba97a15 JS |
1346 | /* |
1347 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1348 | * value. If there are no changes, the current root->root_node is returned. If | |
1349 | * anything changed in between, there's a fresh buffer allocated on which the | |
1350 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1351 | * Returns NULL on error (with no locks held). | |
1352 | */ | |
5d9e75c4 JS |
1353 | static inline struct extent_buffer * |
1354 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1355 | { | |
0b246afa | 1356 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1357 | struct tree_mod_elem *tm; |
30b0463a JS |
1358 | struct extent_buffer *eb = NULL; |
1359 | struct extent_buffer *eb_root; | |
efad8a85 | 1360 | u64 eb_root_owner = 0; |
7bfdcf7f | 1361 | struct extent_buffer *old; |
a95236d9 | 1362 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1363 | u64 old_generation = 0; |
a95236d9 | 1364 | u64 logical; |
581c1760 | 1365 | int level; |
5d9e75c4 | 1366 | |
30b0463a | 1367 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1368 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1369 | if (!tm) |
30b0463a | 1370 | return eb_root; |
5d9e75c4 | 1371 | |
a95236d9 JS |
1372 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1373 | old_root = &tm->old_root; | |
1374 | old_generation = tm->generation; | |
1375 | logical = old_root->logical; | |
581c1760 | 1376 | level = old_root->level; |
a95236d9 | 1377 | } else { |
30b0463a | 1378 | logical = eb_root->start; |
581c1760 | 1379 | level = btrfs_header_level(eb_root); |
a95236d9 | 1380 | } |
5d9e75c4 | 1381 | |
0b246afa | 1382 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1383 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1384 | btrfs_tree_read_unlock(eb_root); |
1385 | free_extent_buffer(eb_root); | |
581c1760 | 1386 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1387 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1388 | if (!IS_ERR(old)) | |
1389 | free_extent_buffer(old); | |
0b246afa JM |
1390 | btrfs_warn(fs_info, |
1391 | "failed to read tree block %llu from get_old_root", | |
1392 | logical); | |
834328a8 | 1393 | } else { |
7bfdcf7f LB |
1394 | eb = btrfs_clone_extent_buffer(old); |
1395 | free_extent_buffer(old); | |
834328a8 JS |
1396 | } |
1397 | } else if (old_root) { | |
efad8a85 | 1398 | eb_root_owner = btrfs_header_owner(eb_root); |
30b0463a JS |
1399 | btrfs_tree_read_unlock(eb_root); |
1400 | free_extent_buffer(eb_root); | |
0b246afa | 1401 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1402 | } else { |
300aa896 | 1403 | btrfs_set_lock_blocking_read(eb_root); |
30b0463a | 1404 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1405 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1406 | free_extent_buffer(eb_root); |
834328a8 JS |
1407 | } |
1408 | ||
8ba97a15 JS |
1409 | if (!eb) |
1410 | return NULL; | |
1411 | btrfs_tree_read_lock(eb); | |
a95236d9 | 1412 | if (old_root) { |
5d9e75c4 JS |
1413 | btrfs_set_header_bytenr(eb, eb->start); |
1414 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
efad8a85 | 1415 | btrfs_set_header_owner(eb, eb_root_owner); |
a95236d9 JS |
1416 | btrfs_set_header_level(eb, old_root->level); |
1417 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1418 | } |
28da9fb4 | 1419 | if (tm) |
0b246afa | 1420 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1421 | else |
1422 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1423 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1424 | |
1425 | return eb; | |
1426 | } | |
1427 | ||
5b6602e7 JS |
1428 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1429 | { | |
1430 | struct tree_mod_elem *tm; | |
1431 | int level; | |
30b0463a | 1432 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1433 | |
bcd24dab | 1434 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1435 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1436 | level = tm->old_root.level; | |
1437 | } else { | |
30b0463a | 1438 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1439 | } |
30b0463a | 1440 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1441 | |
1442 | return level; | |
1443 | } | |
1444 | ||
5d4f98a2 YZ |
1445 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1446 | struct btrfs_root *root, | |
1447 | struct extent_buffer *buf) | |
1448 | { | |
f5ee5c9a | 1449 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1450 | return 0; |
fccb84c9 | 1451 | |
d1980131 DS |
1452 | /* Ensure we can see the FORCE_COW bit */ |
1453 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1454 | |
1455 | /* | |
1456 | * We do not need to cow a block if | |
1457 | * 1) this block is not created or changed in this transaction; | |
1458 | * 2) this block does not belong to TREE_RELOC tree; | |
1459 | * 3) the root is not forced COW. | |
1460 | * | |
1461 | * What is forced COW: | |
01327610 | 1462 | * when we create snapshot during committing the transaction, |
52042d8e | 1463 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
1464 | * block to ensure the metadata consistency. |
1465 | */ | |
5d4f98a2 YZ |
1466 | if (btrfs_header_generation(buf) == trans->transid && |
1467 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1468 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1469 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1470 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1471 | return 0; |
1472 | return 1; | |
1473 | } | |
1474 | ||
d352ac68 CM |
1475 | /* |
1476 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1477 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1478 | * once per transaction, as long as it hasn't been written yet |
1479 | */ | |
d397712b | 1480 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1481 | struct btrfs_root *root, struct extent_buffer *buf, |
1482 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1483 | struct extent_buffer **cow_ret) |
6702ed49 | 1484 | { |
0b246afa | 1485 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1486 | u64 search_start; |
f510cfec | 1487 | int ret; |
dc17ff8f | 1488 | |
83354f07 JB |
1489 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
1490 | btrfs_err(fs_info, | |
1491 | "COW'ing blocks on a fs root that's being dropped"); | |
1492 | ||
0b246afa | 1493 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1494 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1495 | trans->transid, |
0b246afa | 1496 | fs_info->running_transaction->transid); |
31b1a2bd | 1497 | |
0b246afa | 1498 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1499 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1500 | trans->transid, fs_info->generation); |
dc17ff8f | 1501 | |
5d4f98a2 | 1502 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1503 | trans->dirty = true; |
6702ed49 CM |
1504 | *cow_ret = buf; |
1505 | return 0; | |
1506 | } | |
c487685d | 1507 | |
ee22184b | 1508 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1509 | |
1510 | if (parent) | |
8bead258 DS |
1511 | btrfs_set_lock_blocking_write(parent); |
1512 | btrfs_set_lock_blocking_write(buf); | |
b4ce94de | 1513 | |
f616f5cd QW |
1514 | /* |
1515 | * Before CoWing this block for later modification, check if it's | |
1516 | * the subtree root and do the delayed subtree trace if needed. | |
1517 | * | |
1518 | * Also We don't care about the error, as it's handled internally. | |
1519 | */ | |
1520 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 1521 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1522 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1523 | |
1524 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1525 | ||
f510cfec | 1526 | return ret; |
6702ed49 CM |
1527 | } |
1528 | ||
d352ac68 CM |
1529 | /* |
1530 | * helper function for defrag to decide if two blocks pointed to by a | |
1531 | * node are actually close by | |
1532 | */ | |
6b80053d | 1533 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1534 | { |
6b80053d | 1535 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1536 | return 1; |
6b80053d | 1537 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1538 | return 1; |
1539 | return 0; | |
1540 | } | |
1541 | ||
081e9573 CM |
1542 | /* |
1543 | * compare two keys in a memcmp fashion | |
1544 | */ | |
310712b2 OS |
1545 | static int comp_keys(const struct btrfs_disk_key *disk, |
1546 | const struct btrfs_key *k2) | |
081e9573 CM |
1547 | { |
1548 | struct btrfs_key k1; | |
1549 | ||
1550 | btrfs_disk_key_to_cpu(&k1, disk); | |
1551 | ||
20736aba | 1552 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1553 | } |
1554 | ||
f3465ca4 JB |
1555 | /* |
1556 | * same as comp_keys only with two btrfs_key's | |
1557 | */ | |
e1f60a65 | 1558 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1559 | { |
1560 | if (k1->objectid > k2->objectid) | |
1561 | return 1; | |
1562 | if (k1->objectid < k2->objectid) | |
1563 | return -1; | |
1564 | if (k1->type > k2->type) | |
1565 | return 1; | |
1566 | if (k1->type < k2->type) | |
1567 | return -1; | |
1568 | if (k1->offset > k2->offset) | |
1569 | return 1; | |
1570 | if (k1->offset < k2->offset) | |
1571 | return -1; | |
1572 | return 0; | |
1573 | } | |
081e9573 | 1574 | |
d352ac68 CM |
1575 | /* |
1576 | * this is used by the defrag code to go through all the | |
1577 | * leaves pointed to by a node and reallocate them so that | |
1578 | * disk order is close to key order | |
1579 | */ | |
6702ed49 | 1580 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1581 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1582 | int start_slot, u64 *last_ret, |
a6b6e75e | 1583 | struct btrfs_key *progress) |
6702ed49 | 1584 | { |
0b246afa | 1585 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1586 | struct extent_buffer *cur; |
6702ed49 | 1587 | u64 blocknr; |
ca7a79ad | 1588 | u64 gen; |
e9d0b13b CM |
1589 | u64 search_start = *last_ret; |
1590 | u64 last_block = 0; | |
6702ed49 CM |
1591 | u64 other; |
1592 | u32 parent_nritems; | |
6702ed49 CM |
1593 | int end_slot; |
1594 | int i; | |
1595 | int err = 0; | |
f2183bde | 1596 | int parent_level; |
6b80053d CM |
1597 | int uptodate; |
1598 | u32 blocksize; | |
081e9573 CM |
1599 | int progress_passed = 0; |
1600 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1601 | |
5708b959 | 1602 | parent_level = btrfs_header_level(parent); |
5708b959 | 1603 | |
0b246afa JM |
1604 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1605 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1606 | |
6b80053d | 1607 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1608 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1609 | end_slot = parent_nritems - 1; |
6702ed49 | 1610 | |
5dfe2be7 | 1611 | if (parent_nritems <= 1) |
6702ed49 CM |
1612 | return 0; |
1613 | ||
8bead258 | 1614 | btrfs_set_lock_blocking_write(parent); |
b4ce94de | 1615 | |
5dfe2be7 | 1616 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1617 | struct btrfs_key first_key; |
6702ed49 | 1618 | int close = 1; |
a6b6e75e | 1619 | |
081e9573 CM |
1620 | btrfs_node_key(parent, &disk_key, i); |
1621 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1622 | continue; | |
1623 | ||
1624 | progress_passed = 1; | |
6b80053d | 1625 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1626 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1627 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1628 | if (last_block == 0) |
1629 | last_block = blocknr; | |
5708b959 | 1630 | |
6702ed49 | 1631 | if (i > 0) { |
6b80053d CM |
1632 | other = btrfs_node_blockptr(parent, i - 1); |
1633 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1634 | } |
5dfe2be7 | 1635 | if (!close && i < end_slot) { |
6b80053d CM |
1636 | other = btrfs_node_blockptr(parent, i + 1); |
1637 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1638 | } |
e9d0b13b CM |
1639 | if (close) { |
1640 | last_block = blocknr; | |
6702ed49 | 1641 | continue; |
e9d0b13b | 1642 | } |
6702ed49 | 1643 | |
0b246afa | 1644 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1645 | if (cur) |
b9fab919 | 1646 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1647 | else |
1648 | uptodate = 0; | |
5708b959 | 1649 | if (!cur || !uptodate) { |
6b80053d | 1650 | if (!cur) { |
581c1760 QW |
1651 | cur = read_tree_block(fs_info, blocknr, gen, |
1652 | parent_level - 1, | |
1653 | &first_key); | |
64c043de LB |
1654 | if (IS_ERR(cur)) { |
1655 | return PTR_ERR(cur); | |
1656 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1657 | free_extent_buffer(cur); |
97d9a8a4 | 1658 | return -EIO; |
416bc658 | 1659 | } |
6b80053d | 1660 | } else if (!uptodate) { |
581c1760 QW |
1661 | err = btrfs_read_buffer(cur, gen, |
1662 | parent_level - 1,&first_key); | |
018642a1 TI |
1663 | if (err) { |
1664 | free_extent_buffer(cur); | |
1665 | return err; | |
1666 | } | |
f2183bde | 1667 | } |
6702ed49 | 1668 | } |
e9d0b13b | 1669 | if (search_start == 0) |
6b80053d | 1670 | search_start = last_block; |
e9d0b13b | 1671 | |
e7a84565 | 1672 | btrfs_tree_lock(cur); |
8bead258 | 1673 | btrfs_set_lock_blocking_write(cur); |
6b80053d | 1674 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1675 | &cur, search_start, |
6b80053d | 1676 | min(16 * blocksize, |
9fa8cfe7 | 1677 | (end_slot - i) * blocksize)); |
252c38f0 | 1678 | if (err) { |
e7a84565 | 1679 | btrfs_tree_unlock(cur); |
6b80053d | 1680 | free_extent_buffer(cur); |
6702ed49 | 1681 | break; |
252c38f0 | 1682 | } |
e7a84565 CM |
1683 | search_start = cur->start; |
1684 | last_block = cur->start; | |
f2183bde | 1685 | *last_ret = search_start; |
e7a84565 CM |
1686 | btrfs_tree_unlock(cur); |
1687 | free_extent_buffer(cur); | |
6702ed49 CM |
1688 | } |
1689 | return err; | |
1690 | } | |
1691 | ||
74123bd7 | 1692 | /* |
5f39d397 CM |
1693 | * search for key in the extent_buffer. The items start at offset p, |
1694 | * and they are item_size apart. There are 'max' items in p. | |
1695 | * | |
74123bd7 CM |
1696 | * the slot in the array is returned via slot, and it points to |
1697 | * the place where you would insert key if it is not found in | |
1698 | * the array. | |
1699 | * | |
1700 | * slot may point to max if the key is bigger than all of the keys | |
1701 | */ | |
e02119d5 | 1702 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1703 | unsigned long p, int item_size, |
1704 | const struct btrfs_key *key, | |
e02119d5 | 1705 | int max, int *slot) |
be0e5c09 CM |
1706 | { |
1707 | int low = 0; | |
1708 | int high = max; | |
1709 | int mid; | |
1710 | int ret; | |
479965d6 | 1711 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1712 | struct btrfs_disk_key unaligned; |
1713 | unsigned long offset; | |
5f39d397 CM |
1714 | char *kaddr = NULL; |
1715 | unsigned long map_start = 0; | |
1716 | unsigned long map_len = 0; | |
479965d6 | 1717 | int err; |
be0e5c09 | 1718 | |
5e24e9af LB |
1719 | if (low > high) { |
1720 | btrfs_err(eb->fs_info, | |
1721 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1722 | __func__, low, high, eb->start, | |
1723 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1724 | return -EINVAL; | |
1725 | } | |
1726 | ||
d397712b | 1727 | while (low < high) { |
be0e5c09 | 1728 | mid = (low + high) / 2; |
5f39d397 CM |
1729 | offset = p + mid * item_size; |
1730 | ||
a6591715 | 1731 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1732 | (offset + sizeof(struct btrfs_disk_key)) > |
1733 | map_start + map_len) { | |
934d375b CM |
1734 | |
1735 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1736 | sizeof(struct btrfs_disk_key), |
a6591715 | 1737 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1738 | |
1739 | if (!err) { | |
1740 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1741 | map_start); | |
415b35a5 | 1742 | } else if (err == 1) { |
479965d6 CM |
1743 | read_extent_buffer(eb, &unaligned, |
1744 | offset, sizeof(unaligned)); | |
1745 | tmp = &unaligned; | |
415b35a5 LB |
1746 | } else { |
1747 | return err; | |
479965d6 | 1748 | } |
5f39d397 | 1749 | |
5f39d397 CM |
1750 | } else { |
1751 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1752 | map_start); | |
1753 | } | |
be0e5c09 CM |
1754 | ret = comp_keys(tmp, key); |
1755 | ||
1756 | if (ret < 0) | |
1757 | low = mid + 1; | |
1758 | else if (ret > 0) | |
1759 | high = mid; | |
1760 | else { | |
1761 | *slot = mid; | |
1762 | return 0; | |
1763 | } | |
1764 | } | |
1765 | *slot = low; | |
1766 | return 1; | |
1767 | } | |
1768 | ||
97571fd0 CM |
1769 | /* |
1770 | * simple bin_search frontend that does the right thing for | |
1771 | * leaves vs nodes | |
1772 | */ | |
a74b35ec NB |
1773 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1774 | int level, int *slot) | |
be0e5c09 | 1775 | { |
f775738f | 1776 | if (level == 0) |
5f39d397 CM |
1777 | return generic_bin_search(eb, |
1778 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1779 | sizeof(struct btrfs_item), |
5f39d397 | 1780 | key, btrfs_header_nritems(eb), |
7518a238 | 1781 | slot); |
f775738f | 1782 | else |
5f39d397 CM |
1783 | return generic_bin_search(eb, |
1784 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1785 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1786 | key, btrfs_header_nritems(eb), |
7518a238 | 1787 | slot); |
be0e5c09 CM |
1788 | } |
1789 | ||
f0486c68 YZ |
1790 | static void root_add_used(struct btrfs_root *root, u32 size) |
1791 | { | |
1792 | spin_lock(&root->accounting_lock); | |
1793 | btrfs_set_root_used(&root->root_item, | |
1794 | btrfs_root_used(&root->root_item) + size); | |
1795 | spin_unlock(&root->accounting_lock); | |
1796 | } | |
1797 | ||
1798 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1799 | { | |
1800 | spin_lock(&root->accounting_lock); | |
1801 | btrfs_set_root_used(&root->root_item, | |
1802 | btrfs_root_used(&root->root_item) - size); | |
1803 | spin_unlock(&root->accounting_lock); | |
1804 | } | |
1805 | ||
d352ac68 CM |
1806 | /* given a node and slot number, this reads the blocks it points to. The |
1807 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1808 | */ |
4b231ae4 DS |
1809 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
1810 | int slot) | |
bb803951 | 1811 | { |
ca7a79ad | 1812 | int level = btrfs_header_level(parent); |
416bc658 | 1813 | struct extent_buffer *eb; |
581c1760 | 1814 | struct btrfs_key first_key; |
416bc658 | 1815 | |
fb770ae4 LB |
1816 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1817 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1818 | |
1819 | BUG_ON(level == 0); | |
1820 | ||
581c1760 | 1821 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 1822 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1823 | btrfs_node_ptr_generation(parent, slot), |
1824 | level - 1, &first_key); | |
fb770ae4 LB |
1825 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1826 | free_extent_buffer(eb); | |
1827 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1828 | } |
1829 | ||
1830 | return eb; | |
bb803951 CM |
1831 | } |
1832 | ||
d352ac68 CM |
1833 | /* |
1834 | * node level balancing, used to make sure nodes are in proper order for | |
1835 | * item deletion. We balance from the top down, so we have to make sure | |
1836 | * that a deletion won't leave an node completely empty later on. | |
1837 | */ | |
e02119d5 | 1838 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1839 | struct btrfs_root *root, |
1840 | struct btrfs_path *path, int level) | |
bb803951 | 1841 | { |
0b246afa | 1842 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1843 | struct extent_buffer *right = NULL; |
1844 | struct extent_buffer *mid; | |
1845 | struct extent_buffer *left = NULL; | |
1846 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1847 | int ret = 0; |
1848 | int wret; | |
1849 | int pslot; | |
bb803951 | 1850 | int orig_slot = path->slots[level]; |
79f95c82 | 1851 | u64 orig_ptr; |
bb803951 | 1852 | |
98e6b1eb | 1853 | ASSERT(level > 0); |
bb803951 | 1854 | |
5f39d397 | 1855 | mid = path->nodes[level]; |
b4ce94de | 1856 | |
bd681513 CM |
1857 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1858 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1859 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1860 | ||
1d4f8a0c | 1861 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1862 | |
a05a9bb1 | 1863 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1864 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1865 | pslot = path->slots[level + 1]; |
1866 | } | |
bb803951 | 1867 | |
40689478 CM |
1868 | /* |
1869 | * deal with the case where there is only one pointer in the root | |
1870 | * by promoting the node below to a root | |
1871 | */ | |
5f39d397 CM |
1872 | if (!parent) { |
1873 | struct extent_buffer *child; | |
bb803951 | 1874 | |
5f39d397 | 1875 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1876 | return 0; |
1877 | ||
1878 | /* promote the child to a root */ | |
4b231ae4 | 1879 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
1880 | if (IS_ERR(child)) { |
1881 | ret = PTR_ERR(child); | |
0b246afa | 1882 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1883 | goto enospc; |
1884 | } | |
1885 | ||
925baedd | 1886 | btrfs_tree_lock(child); |
8bead258 | 1887 | btrfs_set_lock_blocking_write(child); |
9fa8cfe7 | 1888 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1889 | if (ret) { |
1890 | btrfs_tree_unlock(child); | |
1891 | free_extent_buffer(child); | |
1892 | goto enospc; | |
1893 | } | |
2f375ab9 | 1894 | |
d9d19a01 DS |
1895 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1896 | BUG_ON(ret < 0); | |
240f62c8 | 1897 | rcu_assign_pointer(root->node, child); |
925baedd | 1898 | |
0b86a832 | 1899 | add_root_to_dirty_list(root); |
925baedd | 1900 | btrfs_tree_unlock(child); |
b4ce94de | 1901 | |
925baedd | 1902 | path->locks[level] = 0; |
bb803951 | 1903 | path->nodes[level] = NULL; |
6a884d7d | 1904 | btrfs_clean_tree_block(mid); |
925baedd | 1905 | btrfs_tree_unlock(mid); |
bb803951 | 1906 | /* once for the path */ |
5f39d397 | 1907 | free_extent_buffer(mid); |
f0486c68 YZ |
1908 | |
1909 | root_sub_used(root, mid->len); | |
5581a51a | 1910 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1911 | /* once for the root ptr */ |
3083ee2e | 1912 | free_extent_buffer_stale(mid); |
f0486c68 | 1913 | return 0; |
bb803951 | 1914 | } |
5f39d397 | 1915 | if (btrfs_header_nritems(mid) > |
0b246afa | 1916 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1917 | return 0; |
1918 | ||
4b231ae4 | 1919 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1920 | if (IS_ERR(left)) |
1921 | left = NULL; | |
1922 | ||
5f39d397 | 1923 | if (left) { |
925baedd | 1924 | btrfs_tree_lock(left); |
8bead258 | 1925 | btrfs_set_lock_blocking_write(left); |
5f39d397 | 1926 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1927 | parent, pslot - 1, &left); |
54aa1f4d CM |
1928 | if (wret) { |
1929 | ret = wret; | |
1930 | goto enospc; | |
1931 | } | |
2cc58cf2 | 1932 | } |
fb770ae4 | 1933 | |
4b231ae4 | 1934 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1935 | if (IS_ERR(right)) |
1936 | right = NULL; | |
1937 | ||
5f39d397 | 1938 | if (right) { |
925baedd | 1939 | btrfs_tree_lock(right); |
8bead258 | 1940 | btrfs_set_lock_blocking_write(right); |
5f39d397 | 1941 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1942 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1943 | if (wret) { |
1944 | ret = wret; | |
1945 | goto enospc; | |
1946 | } | |
1947 | } | |
1948 | ||
1949 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1950 | if (left) { |
1951 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 1952 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
1953 | if (wret < 0) |
1954 | ret = wret; | |
bb803951 | 1955 | } |
79f95c82 CM |
1956 | |
1957 | /* | |
1958 | * then try to empty the right most buffer into the middle | |
1959 | */ | |
5f39d397 | 1960 | if (right) { |
d30a668f | 1961 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1962 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1963 | ret = wret; |
5f39d397 | 1964 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 1965 | btrfs_clean_tree_block(right); |
925baedd | 1966 | btrfs_tree_unlock(right); |
afe5fea7 | 1967 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1968 | root_sub_used(root, right->len); |
5581a51a | 1969 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1970 | free_extent_buffer_stale(right); |
f0486c68 | 1971 | right = NULL; |
bb803951 | 1972 | } else { |
5f39d397 CM |
1973 | struct btrfs_disk_key right_key; |
1974 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1975 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1976 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1977 | BUG_ON(ret < 0); | |
5f39d397 CM |
1978 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1979 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1980 | } |
1981 | } | |
5f39d397 | 1982 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1983 | /* |
1984 | * we're not allowed to leave a node with one item in the | |
1985 | * tree during a delete. A deletion from lower in the tree | |
1986 | * could try to delete the only pointer in this node. | |
1987 | * So, pull some keys from the left. | |
1988 | * There has to be a left pointer at this point because | |
1989 | * otherwise we would have pulled some pointers from the | |
1990 | * right | |
1991 | */ | |
305a26af MF |
1992 | if (!left) { |
1993 | ret = -EROFS; | |
0b246afa | 1994 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1995 | goto enospc; |
1996 | } | |
55d32ed8 | 1997 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1998 | if (wret < 0) { |
79f95c82 | 1999 | ret = wret; |
54aa1f4d CM |
2000 | goto enospc; |
2001 | } | |
bce4eae9 | 2002 | if (wret == 1) { |
d30a668f | 2003 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
2004 | if (wret < 0) |
2005 | ret = wret; | |
2006 | } | |
79f95c82 CM |
2007 | BUG_ON(wret == 1); |
2008 | } | |
5f39d397 | 2009 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 2010 | btrfs_clean_tree_block(mid); |
925baedd | 2011 | btrfs_tree_unlock(mid); |
afe5fea7 | 2012 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2013 | root_sub_used(root, mid->len); |
5581a51a | 2014 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2015 | free_extent_buffer_stale(mid); |
f0486c68 | 2016 | mid = NULL; |
79f95c82 CM |
2017 | } else { |
2018 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2019 | struct btrfs_disk_key mid_key; |
2020 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
2021 | ret = tree_mod_log_insert_key(parent, pslot, |
2022 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2023 | BUG_ON(ret < 0); | |
5f39d397 CM |
2024 | btrfs_set_node_key(parent, &mid_key, pslot); |
2025 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2026 | } |
bb803951 | 2027 | |
79f95c82 | 2028 | /* update the path */ |
5f39d397 CM |
2029 | if (left) { |
2030 | if (btrfs_header_nritems(left) > orig_slot) { | |
67439dad | 2031 | atomic_inc(&left->refs); |
925baedd | 2032 | /* left was locked after cow */ |
5f39d397 | 2033 | path->nodes[level] = left; |
bb803951 CM |
2034 | path->slots[level + 1] -= 1; |
2035 | path->slots[level] = orig_slot; | |
925baedd CM |
2036 | if (mid) { |
2037 | btrfs_tree_unlock(mid); | |
5f39d397 | 2038 | free_extent_buffer(mid); |
925baedd | 2039 | } |
bb803951 | 2040 | } else { |
5f39d397 | 2041 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2042 | path->slots[level] = orig_slot; |
2043 | } | |
2044 | } | |
79f95c82 | 2045 | /* double check we haven't messed things up */ |
e20d96d6 | 2046 | if (orig_ptr != |
5f39d397 | 2047 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2048 | BUG(); |
54aa1f4d | 2049 | enospc: |
925baedd CM |
2050 | if (right) { |
2051 | btrfs_tree_unlock(right); | |
5f39d397 | 2052 | free_extent_buffer(right); |
925baedd CM |
2053 | } |
2054 | if (left) { | |
2055 | if (path->nodes[level] != left) | |
2056 | btrfs_tree_unlock(left); | |
5f39d397 | 2057 | free_extent_buffer(left); |
925baedd | 2058 | } |
bb803951 CM |
2059 | return ret; |
2060 | } | |
2061 | ||
d352ac68 CM |
2062 | /* Node balancing for insertion. Here we only split or push nodes around |
2063 | * when they are completely full. This is also done top down, so we | |
2064 | * have to be pessimistic. | |
2065 | */ | |
d397712b | 2066 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2067 | struct btrfs_root *root, |
2068 | struct btrfs_path *path, int level) | |
e66f709b | 2069 | { |
0b246afa | 2070 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2071 | struct extent_buffer *right = NULL; |
2072 | struct extent_buffer *mid; | |
2073 | struct extent_buffer *left = NULL; | |
2074 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2075 | int ret = 0; |
2076 | int wret; | |
2077 | int pslot; | |
2078 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2079 | |
2080 | if (level == 0) | |
2081 | return 1; | |
2082 | ||
5f39d397 | 2083 | mid = path->nodes[level]; |
7bb86316 | 2084 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2085 | |
a05a9bb1 | 2086 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2087 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2088 | pslot = path->slots[level + 1]; |
2089 | } | |
e66f709b | 2090 | |
5f39d397 | 2091 | if (!parent) |
e66f709b | 2092 | return 1; |
e66f709b | 2093 | |
4b231ae4 | 2094 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
2095 | if (IS_ERR(left)) |
2096 | left = NULL; | |
e66f709b CM |
2097 | |
2098 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2099 | if (left) { |
e66f709b | 2100 | u32 left_nr; |
925baedd CM |
2101 | |
2102 | btrfs_tree_lock(left); | |
8bead258 | 2103 | btrfs_set_lock_blocking_write(left); |
b4ce94de | 2104 | |
5f39d397 | 2105 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2106 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2107 | wret = 1; |
2108 | } else { | |
5f39d397 | 2109 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2110 | pslot - 1, &left); |
54aa1f4d CM |
2111 | if (ret) |
2112 | wret = 1; | |
2113 | else { | |
d30a668f | 2114 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 2115 | } |
33ade1f8 | 2116 | } |
e66f709b CM |
2117 | if (wret < 0) |
2118 | ret = wret; | |
2119 | if (wret == 0) { | |
5f39d397 | 2120 | struct btrfs_disk_key disk_key; |
e66f709b | 2121 | orig_slot += left_nr; |
5f39d397 | 2122 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2123 | ret = tree_mod_log_insert_key(parent, pslot, |
2124 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2125 | BUG_ON(ret < 0); | |
5f39d397 CM |
2126 | btrfs_set_node_key(parent, &disk_key, pslot); |
2127 | btrfs_mark_buffer_dirty(parent); | |
2128 | if (btrfs_header_nritems(left) > orig_slot) { | |
2129 | path->nodes[level] = left; | |
e66f709b CM |
2130 | path->slots[level + 1] -= 1; |
2131 | path->slots[level] = orig_slot; | |
925baedd | 2132 | btrfs_tree_unlock(mid); |
5f39d397 | 2133 | free_extent_buffer(mid); |
e66f709b CM |
2134 | } else { |
2135 | orig_slot -= | |
5f39d397 | 2136 | btrfs_header_nritems(left); |
e66f709b | 2137 | path->slots[level] = orig_slot; |
925baedd | 2138 | btrfs_tree_unlock(left); |
5f39d397 | 2139 | free_extent_buffer(left); |
e66f709b | 2140 | } |
e66f709b CM |
2141 | return 0; |
2142 | } | |
925baedd | 2143 | btrfs_tree_unlock(left); |
5f39d397 | 2144 | free_extent_buffer(left); |
e66f709b | 2145 | } |
4b231ae4 | 2146 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
2147 | if (IS_ERR(right)) |
2148 | right = NULL; | |
e66f709b CM |
2149 | |
2150 | /* | |
2151 | * then try to empty the right most buffer into the middle | |
2152 | */ | |
5f39d397 | 2153 | if (right) { |
33ade1f8 | 2154 | u32 right_nr; |
b4ce94de | 2155 | |
925baedd | 2156 | btrfs_tree_lock(right); |
8bead258 | 2157 | btrfs_set_lock_blocking_write(right); |
b4ce94de | 2158 | |
5f39d397 | 2159 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2160 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2161 | wret = 1; |
2162 | } else { | |
5f39d397 CM |
2163 | ret = btrfs_cow_block(trans, root, right, |
2164 | parent, pslot + 1, | |
9fa8cfe7 | 2165 | &right); |
54aa1f4d CM |
2166 | if (ret) |
2167 | wret = 1; | |
2168 | else { | |
55d32ed8 | 2169 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 2170 | } |
33ade1f8 | 2171 | } |
e66f709b CM |
2172 | if (wret < 0) |
2173 | ret = wret; | |
2174 | if (wret == 0) { | |
5f39d397 CM |
2175 | struct btrfs_disk_key disk_key; |
2176 | ||
2177 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2178 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2179 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2180 | BUG_ON(ret < 0); | |
5f39d397 CM |
2181 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2182 | btrfs_mark_buffer_dirty(parent); | |
2183 | ||
2184 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2185 | path->nodes[level] = right; | |
e66f709b CM |
2186 | path->slots[level + 1] += 1; |
2187 | path->slots[level] = orig_slot - | |
5f39d397 | 2188 | btrfs_header_nritems(mid); |
925baedd | 2189 | btrfs_tree_unlock(mid); |
5f39d397 | 2190 | free_extent_buffer(mid); |
e66f709b | 2191 | } else { |
925baedd | 2192 | btrfs_tree_unlock(right); |
5f39d397 | 2193 | free_extent_buffer(right); |
e66f709b | 2194 | } |
e66f709b CM |
2195 | return 0; |
2196 | } | |
925baedd | 2197 | btrfs_tree_unlock(right); |
5f39d397 | 2198 | free_extent_buffer(right); |
e66f709b | 2199 | } |
e66f709b CM |
2200 | return 1; |
2201 | } | |
2202 | ||
3c69faec | 2203 | /* |
d352ac68 CM |
2204 | * readahead one full node of leaves, finding things that are close |
2205 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2206 | */ |
2ff7e61e | 2207 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2208 | struct btrfs_path *path, |
2209 | int level, int slot, u64 objectid) | |
3c69faec | 2210 | { |
5f39d397 | 2211 | struct extent_buffer *node; |
01f46658 | 2212 | struct btrfs_disk_key disk_key; |
3c69faec | 2213 | u32 nritems; |
3c69faec | 2214 | u64 search; |
a7175319 | 2215 | u64 target; |
6b80053d | 2216 | u64 nread = 0; |
5f39d397 | 2217 | struct extent_buffer *eb; |
6b80053d CM |
2218 | u32 nr; |
2219 | u32 blocksize; | |
2220 | u32 nscan = 0; | |
db94535d | 2221 | |
a6b6e75e | 2222 | if (level != 1) |
6702ed49 CM |
2223 | return; |
2224 | ||
2225 | if (!path->nodes[level]) | |
3c69faec CM |
2226 | return; |
2227 | ||
5f39d397 | 2228 | node = path->nodes[level]; |
925baedd | 2229 | |
3c69faec | 2230 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2231 | blocksize = fs_info->nodesize; |
2232 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2233 | if (eb) { |
2234 | free_extent_buffer(eb); | |
3c69faec CM |
2235 | return; |
2236 | } | |
2237 | ||
a7175319 | 2238 | target = search; |
6b80053d | 2239 | |
5f39d397 | 2240 | nritems = btrfs_header_nritems(node); |
6b80053d | 2241 | nr = slot; |
25b8b936 | 2242 | |
d397712b | 2243 | while (1) { |
e4058b54 | 2244 | if (path->reada == READA_BACK) { |
6b80053d CM |
2245 | if (nr == 0) |
2246 | break; | |
2247 | nr--; | |
e4058b54 | 2248 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2249 | nr++; |
2250 | if (nr >= nritems) | |
2251 | break; | |
3c69faec | 2252 | } |
e4058b54 | 2253 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2254 | btrfs_node_key(node, &disk_key, nr); |
2255 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2256 | break; | |
2257 | } | |
6b80053d | 2258 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2259 | if ((search <= target && target - search <= 65536) || |
2260 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2261 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2262 | nread += blocksize; |
2263 | } | |
2264 | nscan++; | |
a7175319 | 2265 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2266 | break; |
3c69faec CM |
2267 | } |
2268 | } | |
925baedd | 2269 | |
2ff7e61e | 2270 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2271 | struct btrfs_path *path, int level) |
b4ce94de CM |
2272 | { |
2273 | int slot; | |
2274 | int nritems; | |
2275 | struct extent_buffer *parent; | |
2276 | struct extent_buffer *eb; | |
2277 | u64 gen; | |
2278 | u64 block1 = 0; | |
2279 | u64 block2 = 0; | |
b4ce94de | 2280 | |
8c594ea8 | 2281 | parent = path->nodes[level + 1]; |
b4ce94de | 2282 | if (!parent) |
0b08851f | 2283 | return; |
b4ce94de CM |
2284 | |
2285 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2286 | slot = path->slots[level + 1]; |
b4ce94de CM |
2287 | |
2288 | if (slot > 0) { | |
2289 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2290 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2291 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2292 | /* |
2293 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2294 | * don't want to return eagain here. That will loop | |
2295 | * forever | |
2296 | */ | |
2297 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2298 | block1 = 0; |
2299 | free_extent_buffer(eb); | |
2300 | } | |
8c594ea8 | 2301 | if (slot + 1 < nritems) { |
b4ce94de CM |
2302 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2303 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2304 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2305 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2306 | block2 = 0; |
2307 | free_extent_buffer(eb); | |
2308 | } | |
8c594ea8 | 2309 | |
0b08851f | 2310 | if (block1) |
2ff7e61e | 2311 | readahead_tree_block(fs_info, block1); |
0b08851f | 2312 | if (block2) |
2ff7e61e | 2313 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2314 | } |
2315 | ||
2316 | ||
d352ac68 | 2317 | /* |
d397712b CM |
2318 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2319 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2320 | * operations on the tree might require changing key pointers higher up in the | |
2321 | * tree. | |
d352ac68 | 2322 | * |
d397712b CM |
2323 | * callers might also have set path->keep_locks, which tells this code to keep |
2324 | * the lock if the path points to the last slot in the block. This is part of | |
2325 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2326 | * |
d397712b CM |
2327 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2328 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2329 | */ |
e02119d5 | 2330 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2331 | int lowest_unlock, int min_write_lock_level, |
2332 | int *write_lock_level) | |
925baedd CM |
2333 | { |
2334 | int i; | |
2335 | int skip_level = level; | |
051e1b9f | 2336 | int no_skips = 0; |
925baedd CM |
2337 | struct extent_buffer *t; |
2338 | ||
2339 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2340 | if (!path->nodes[i]) | |
2341 | break; | |
2342 | if (!path->locks[i]) | |
2343 | break; | |
051e1b9f | 2344 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2345 | skip_level = i + 1; |
2346 | continue; | |
2347 | } | |
051e1b9f | 2348 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2349 | u32 nritems; |
2350 | t = path->nodes[i]; | |
2351 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2352 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2353 | skip_level = i + 1; |
2354 | continue; | |
2355 | } | |
2356 | } | |
051e1b9f CM |
2357 | if (skip_level < i && i >= lowest_unlock) |
2358 | no_skips = 1; | |
2359 | ||
925baedd | 2360 | t = path->nodes[i]; |
d80bb3f9 | 2361 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2362 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2363 | path->locks[i] = 0; |
f7c79f30 CM |
2364 | if (write_lock_level && |
2365 | i > min_write_lock_level && | |
2366 | i <= *write_lock_level) { | |
2367 | *write_lock_level = i - 1; | |
2368 | } | |
925baedd CM |
2369 | } |
2370 | } | |
2371 | } | |
2372 | ||
c8c42864 CM |
2373 | /* |
2374 | * helper function for btrfs_search_slot. The goal is to find a block | |
2375 | * in cache without setting the path to blocking. If we find the block | |
2376 | * we return zero and the path is unchanged. | |
2377 | * | |
2378 | * If we can't find the block, we set the path blocking and do some | |
2379 | * reada. -EAGAIN is returned and the search must be repeated. | |
2380 | */ | |
2381 | static int | |
d07b8528 LB |
2382 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2383 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2384 | const struct btrfs_key *key) |
c8c42864 | 2385 | { |
0b246afa | 2386 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2387 | u64 blocknr; |
2388 | u64 gen; | |
c8c42864 CM |
2389 | struct extent_buffer *b = *eb_ret; |
2390 | struct extent_buffer *tmp; | |
581c1760 | 2391 | struct btrfs_key first_key; |
76a05b35 | 2392 | int ret; |
581c1760 | 2393 | int parent_level; |
c8c42864 CM |
2394 | |
2395 | blocknr = btrfs_node_blockptr(b, slot); | |
2396 | gen = btrfs_node_ptr_generation(b, slot); | |
581c1760 QW |
2397 | parent_level = btrfs_header_level(b); |
2398 | btrfs_node_key_to_cpu(b, &first_key, slot); | |
c8c42864 | 2399 | |
0b246afa | 2400 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2401 | if (tmp) { |
b9fab919 | 2402 | /* first we do an atomic uptodate check */ |
bdf7c00e | 2403 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
2404 | /* |
2405 | * Do extra check for first_key, eb can be stale due to | |
2406 | * being cached, read from scrub, or have multiple | |
2407 | * parents (shared tree blocks). | |
2408 | */ | |
e064d5e9 | 2409 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
2410 | parent_level - 1, &first_key, gen)) { |
2411 | free_extent_buffer(tmp); | |
2412 | return -EUCLEAN; | |
2413 | } | |
bdf7c00e JB |
2414 | *eb_ret = tmp; |
2415 | return 0; | |
2416 | } | |
2417 | ||
2418 | /* the pages were up to date, but we failed | |
2419 | * the generation number check. Do a full | |
2420 | * read for the generation number that is correct. | |
2421 | * We must do this without dropping locks so | |
2422 | * we can trust our generation number | |
2423 | */ | |
2424 | btrfs_set_path_blocking(p); | |
2425 | ||
2426 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2427 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2428 | if (!ret) { |
2429 | *eb_ret = tmp; | |
2430 | return 0; | |
cb44921a | 2431 | } |
bdf7c00e JB |
2432 | free_extent_buffer(tmp); |
2433 | btrfs_release_path(p); | |
2434 | return -EIO; | |
c8c42864 CM |
2435 | } |
2436 | ||
2437 | /* | |
2438 | * reduce lock contention at high levels | |
2439 | * of the btree by dropping locks before | |
76a05b35 CM |
2440 | * we read. Don't release the lock on the current |
2441 | * level because we need to walk this node to figure | |
2442 | * out which blocks to read. | |
c8c42864 | 2443 | */ |
8c594ea8 CM |
2444 | btrfs_unlock_up_safe(p, level + 1); |
2445 | btrfs_set_path_blocking(p); | |
2446 | ||
e4058b54 | 2447 | if (p->reada != READA_NONE) |
2ff7e61e | 2448 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2449 | |
76a05b35 | 2450 | ret = -EAGAIN; |
02a3307a | 2451 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2452 | &first_key); |
64c043de | 2453 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2454 | /* |
2455 | * If the read above didn't mark this buffer up to date, | |
2456 | * it will never end up being up to date. Set ret to EIO now | |
2457 | * and give up so that our caller doesn't loop forever | |
2458 | * on our EAGAINs. | |
2459 | */ | |
e6a1d6fd | 2460 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2461 | ret = -EIO; |
c8c42864 | 2462 | free_extent_buffer(tmp); |
c871b0f2 LB |
2463 | } else { |
2464 | ret = PTR_ERR(tmp); | |
76a05b35 | 2465 | } |
02a3307a LB |
2466 | |
2467 | btrfs_release_path(p); | |
76a05b35 | 2468 | return ret; |
c8c42864 CM |
2469 | } |
2470 | ||
2471 | /* | |
2472 | * helper function for btrfs_search_slot. This does all of the checks | |
2473 | * for node-level blocks and does any balancing required based on | |
2474 | * the ins_len. | |
2475 | * | |
2476 | * If no extra work was required, zero is returned. If we had to | |
2477 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2478 | * start over | |
2479 | */ | |
2480 | static int | |
2481 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2482 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2483 | struct extent_buffer *b, int level, int ins_len, |
2484 | int *write_lock_level) | |
c8c42864 | 2485 | { |
0b246afa | 2486 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2487 | int ret; |
0b246afa | 2488 | |
c8c42864 | 2489 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2490 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2491 | int sret; |
2492 | ||
bd681513 CM |
2493 | if (*write_lock_level < level + 1) { |
2494 | *write_lock_level = level + 1; | |
2495 | btrfs_release_path(p); | |
2496 | goto again; | |
2497 | } | |
2498 | ||
c8c42864 | 2499 | btrfs_set_path_blocking(p); |
2ff7e61e | 2500 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2501 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2502 | |
2503 | BUG_ON(sret > 0); | |
2504 | if (sret) { | |
2505 | ret = sret; | |
2506 | goto done; | |
2507 | } | |
2508 | b = p->nodes[level]; | |
2509 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2510 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2511 | int sret; |
2512 | ||
bd681513 CM |
2513 | if (*write_lock_level < level + 1) { |
2514 | *write_lock_level = level + 1; | |
2515 | btrfs_release_path(p); | |
2516 | goto again; | |
2517 | } | |
2518 | ||
c8c42864 | 2519 | btrfs_set_path_blocking(p); |
2ff7e61e | 2520 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2521 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2522 | |
2523 | if (sret) { | |
2524 | ret = sret; | |
2525 | goto done; | |
2526 | } | |
2527 | b = p->nodes[level]; | |
2528 | if (!b) { | |
b3b4aa74 | 2529 | btrfs_release_path(p); |
c8c42864 CM |
2530 | goto again; |
2531 | } | |
2532 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2533 | } | |
2534 | return 0; | |
2535 | ||
2536 | again: | |
2537 | ret = -EAGAIN; | |
2538 | done: | |
2539 | return ret; | |
2540 | } | |
2541 | ||
310712b2 | 2542 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2543 | int level, int *prev_cmp, int *slot) |
2544 | { | |
2545 | if (*prev_cmp != 0) { | |
a74b35ec | 2546 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2547 | return *prev_cmp; |
2548 | } | |
2549 | ||
d7396f07 FDBM |
2550 | *slot = 0; |
2551 | ||
2552 | return 0; | |
2553 | } | |
2554 | ||
381cf658 | 2555 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2556 | u64 iobjectid, u64 ioff, u8 key_type, |
2557 | struct btrfs_key *found_key) | |
2558 | { | |
2559 | int ret; | |
2560 | struct btrfs_key key; | |
2561 | struct extent_buffer *eb; | |
381cf658 DS |
2562 | |
2563 | ASSERT(path); | |
1d4c08e0 | 2564 | ASSERT(found_key); |
e33d5c3d KN |
2565 | |
2566 | key.type = key_type; | |
2567 | key.objectid = iobjectid; | |
2568 | key.offset = ioff; | |
2569 | ||
2570 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2571 | if (ret < 0) |
e33d5c3d KN |
2572 | return ret; |
2573 | ||
2574 | eb = path->nodes[0]; | |
2575 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2576 | ret = btrfs_next_leaf(fs_root, path); | |
2577 | if (ret) | |
2578 | return ret; | |
2579 | eb = path->nodes[0]; | |
2580 | } | |
2581 | ||
2582 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2583 | if (found_key->type != key.type || | |
2584 | found_key->objectid != key.objectid) | |
2585 | return 1; | |
2586 | ||
2587 | return 0; | |
2588 | } | |
2589 | ||
1fc28d8e LB |
2590 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2591 | struct btrfs_path *p, | |
2592 | int write_lock_level) | |
2593 | { | |
2594 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2595 | struct extent_buffer *b; | |
2596 | int root_lock; | |
2597 | int level = 0; | |
2598 | ||
2599 | /* We try very hard to do read locks on the root */ | |
2600 | root_lock = BTRFS_READ_LOCK; | |
2601 | ||
2602 | if (p->search_commit_root) { | |
be6821f8 FM |
2603 | /* |
2604 | * The commit roots are read only so we always do read locks, | |
2605 | * and we always must hold the commit_root_sem when doing | |
2606 | * searches on them, the only exception is send where we don't | |
2607 | * want to block transaction commits for a long time, so | |
2608 | * we need to clone the commit root in order to avoid races | |
2609 | * with transaction commits that create a snapshot of one of | |
2610 | * the roots used by a send operation. | |
2611 | */ | |
2612 | if (p->need_commit_sem) { | |
1fc28d8e | 2613 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 2614 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 2615 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
2616 | if (!b) |
2617 | return ERR_PTR(-ENOMEM); | |
2618 | ||
2619 | } else { | |
2620 | b = root->commit_root; | |
67439dad | 2621 | atomic_inc(&b->refs); |
be6821f8 FM |
2622 | } |
2623 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
2624 | /* |
2625 | * Ensure that all callers have set skip_locking when | |
2626 | * p->search_commit_root = 1. | |
2627 | */ | |
2628 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2629 | |
2630 | goto out; | |
2631 | } | |
2632 | ||
2633 | if (p->skip_locking) { | |
2634 | b = btrfs_root_node(root); | |
2635 | level = btrfs_header_level(b); | |
2636 | goto out; | |
2637 | } | |
2638 | ||
2639 | /* | |
662c653b LB |
2640 | * If the level is set to maximum, we can skip trying to get the read |
2641 | * lock. | |
1fc28d8e | 2642 | */ |
662c653b LB |
2643 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2644 | /* | |
2645 | * We don't know the level of the root node until we actually | |
2646 | * have it read locked | |
2647 | */ | |
2648 | b = btrfs_read_lock_root_node(root); | |
2649 | level = btrfs_header_level(b); | |
2650 | if (level > write_lock_level) | |
2651 | goto out; | |
2652 | ||
2653 | /* Whoops, must trade for write lock */ | |
2654 | btrfs_tree_read_unlock(b); | |
2655 | free_extent_buffer(b); | |
2656 | } | |
1fc28d8e | 2657 | |
1fc28d8e LB |
2658 | b = btrfs_lock_root_node(root); |
2659 | root_lock = BTRFS_WRITE_LOCK; | |
2660 | ||
2661 | /* The level might have changed, check again */ | |
2662 | level = btrfs_header_level(b); | |
2663 | ||
2664 | out: | |
2665 | p->nodes[level] = b; | |
2666 | if (!p->skip_locking) | |
2667 | p->locks[level] = root_lock; | |
2668 | /* | |
2669 | * Callers are responsible for dropping b's references. | |
2670 | */ | |
2671 | return b; | |
2672 | } | |
2673 | ||
2674 | ||
74123bd7 | 2675 | /* |
4271ecea NB |
2676 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2677 | * modifications to preserve tree invariants. | |
74123bd7 | 2678 | * |
4271ecea NB |
2679 | * @trans: Handle of transaction, used when modifying the tree |
2680 | * @p: Holds all btree nodes along the search path | |
2681 | * @root: The root node of the tree | |
2682 | * @key: The key we are looking for | |
2683 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2684 | * deletions it's -1. 0 for plain searches | |
2685 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2686 | * when modifying the tree. | |
97571fd0 | 2687 | * |
4271ecea NB |
2688 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2689 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2690 | * | |
2691 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2692 | * of the path (level 0) | |
2693 | * | |
2694 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2695 | * points to the slot where it should be inserted | |
2696 | * | |
2697 | * If an error is encountered while searching the tree a negative error number | |
2698 | * is returned | |
74123bd7 | 2699 | */ |
310712b2 OS |
2700 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2701 | const struct btrfs_key *key, struct btrfs_path *p, | |
2702 | int ins_len, int cow) | |
be0e5c09 | 2703 | { |
5f39d397 | 2704 | struct extent_buffer *b; |
be0e5c09 CM |
2705 | int slot; |
2706 | int ret; | |
33c66f43 | 2707 | int err; |
be0e5c09 | 2708 | int level; |
925baedd | 2709 | int lowest_unlock = 1; |
bd681513 CM |
2710 | /* everything at write_lock_level or lower must be write locked */ |
2711 | int write_lock_level = 0; | |
9f3a7427 | 2712 | u8 lowest_level = 0; |
f7c79f30 | 2713 | int min_write_lock_level; |
d7396f07 | 2714 | int prev_cmp; |
9f3a7427 | 2715 | |
6702ed49 | 2716 | lowest_level = p->lowest_level; |
323ac95b | 2717 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2718 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2719 | BUG_ON(!cow && ins_len); |
25179201 | 2720 | |
bd681513 | 2721 | if (ins_len < 0) { |
925baedd | 2722 | lowest_unlock = 2; |
65b51a00 | 2723 | |
bd681513 CM |
2724 | /* when we are removing items, we might have to go up to level |
2725 | * two as we update tree pointers Make sure we keep write | |
2726 | * for those levels as well | |
2727 | */ | |
2728 | write_lock_level = 2; | |
2729 | } else if (ins_len > 0) { | |
2730 | /* | |
2731 | * for inserting items, make sure we have a write lock on | |
2732 | * level 1 so we can update keys | |
2733 | */ | |
2734 | write_lock_level = 1; | |
2735 | } | |
2736 | ||
2737 | if (!cow) | |
2738 | write_lock_level = -1; | |
2739 | ||
09a2a8f9 | 2740 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2741 | write_lock_level = BTRFS_MAX_LEVEL; |
2742 | ||
f7c79f30 CM |
2743 | min_write_lock_level = write_lock_level; |
2744 | ||
bb803951 | 2745 | again: |
d7396f07 | 2746 | prev_cmp = -1; |
1fc28d8e | 2747 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2748 | if (IS_ERR(b)) { |
2749 | ret = PTR_ERR(b); | |
2750 | goto done; | |
2751 | } | |
925baedd | 2752 | |
eb60ceac | 2753 | while (b) { |
f624d976 QW |
2754 | int dec = 0; |
2755 | ||
5f39d397 | 2756 | level = btrfs_header_level(b); |
65b51a00 | 2757 | |
02217ed2 | 2758 | if (cow) { |
9ea2c7c9 NB |
2759 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2760 | ||
c8c42864 CM |
2761 | /* |
2762 | * if we don't really need to cow this block | |
2763 | * then we don't want to set the path blocking, | |
2764 | * so we test it here | |
2765 | */ | |
64c12921 JM |
2766 | if (!should_cow_block(trans, root, b)) { |
2767 | trans->dirty = true; | |
65b51a00 | 2768 | goto cow_done; |
64c12921 | 2769 | } |
5d4f98a2 | 2770 | |
bd681513 CM |
2771 | /* |
2772 | * must have write locks on this node and the | |
2773 | * parent | |
2774 | */ | |
5124e00e JB |
2775 | if (level > write_lock_level || |
2776 | (level + 1 > write_lock_level && | |
2777 | level + 1 < BTRFS_MAX_LEVEL && | |
2778 | p->nodes[level + 1])) { | |
bd681513 CM |
2779 | write_lock_level = level + 1; |
2780 | btrfs_release_path(p); | |
2781 | goto again; | |
2782 | } | |
2783 | ||
160f4089 | 2784 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2785 | if (last_level) |
2786 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2787 | &b); | |
2788 | else | |
2789 | err = btrfs_cow_block(trans, root, b, | |
2790 | p->nodes[level + 1], | |
2791 | p->slots[level + 1], &b); | |
33c66f43 | 2792 | if (err) { |
33c66f43 | 2793 | ret = err; |
65b51a00 | 2794 | goto done; |
54aa1f4d | 2795 | } |
02217ed2 | 2796 | } |
65b51a00 | 2797 | cow_done: |
eb60ceac | 2798 | p->nodes[level] = b; |
52398340 LB |
2799 | /* |
2800 | * Leave path with blocking locks to avoid massive | |
2801 | * lock context switch, this is made on purpose. | |
2802 | */ | |
b4ce94de CM |
2803 | |
2804 | /* | |
2805 | * we have a lock on b and as long as we aren't changing | |
2806 | * the tree, there is no way to for the items in b to change. | |
2807 | * It is safe to drop the lock on our parent before we | |
2808 | * go through the expensive btree search on b. | |
2809 | * | |
eb653de1 FDBM |
2810 | * If we're inserting or deleting (ins_len != 0), then we might |
2811 | * be changing slot zero, which may require changing the parent. | |
2812 | * So, we can't drop the lock until after we know which slot | |
2813 | * we're operating on. | |
b4ce94de | 2814 | */ |
eb653de1 FDBM |
2815 | if (!ins_len && !p->keep_locks) { |
2816 | int u = level + 1; | |
2817 | ||
2818 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2819 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2820 | p->locks[u] = 0; | |
2821 | } | |
2822 | } | |
b4ce94de | 2823 | |
d7396f07 | 2824 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2825 | if (ret < 0) |
2826 | goto done; | |
b4ce94de | 2827 | |
f624d976 | 2828 | if (level == 0) { |
be0e5c09 | 2829 | p->slots[level] = slot; |
87b29b20 | 2830 | if (ins_len > 0 && |
e902baac | 2831 | btrfs_leaf_free_space(b) < ins_len) { |
bd681513 CM |
2832 | if (write_lock_level < 1) { |
2833 | write_lock_level = 1; | |
2834 | btrfs_release_path(p); | |
2835 | goto again; | |
2836 | } | |
2837 | ||
b4ce94de | 2838 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2839 | err = split_leaf(trans, root, key, |
2840 | p, ins_len, ret == 0); | |
b4ce94de | 2841 | |
33c66f43 YZ |
2842 | BUG_ON(err > 0); |
2843 | if (err) { | |
2844 | ret = err; | |
65b51a00 CM |
2845 | goto done; |
2846 | } | |
5c680ed6 | 2847 | } |
459931ec | 2848 | if (!p->search_for_split) |
f7c79f30 | 2849 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2850 | min_write_lock_level, NULL); |
65b51a00 | 2851 | goto done; |
be0e5c09 | 2852 | } |
f624d976 QW |
2853 | if (ret && slot > 0) { |
2854 | dec = 1; | |
2855 | slot--; | |
2856 | } | |
2857 | p->slots[level] = slot; | |
2858 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
2859 | &write_lock_level); | |
2860 | if (err == -EAGAIN) | |
2861 | goto again; | |
2862 | if (err) { | |
2863 | ret = err; | |
2864 | goto done; | |
2865 | } | |
2866 | b = p->nodes[level]; | |
2867 | slot = p->slots[level]; | |
2868 | ||
2869 | /* | |
2870 | * Slot 0 is special, if we change the key we have to update | |
2871 | * the parent pointer which means we must have a write lock on | |
2872 | * the parent | |
2873 | */ | |
2874 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
2875 | write_lock_level = level + 1; | |
2876 | btrfs_release_path(p); | |
2877 | goto again; | |
2878 | } | |
2879 | ||
2880 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
2881 | &write_lock_level); | |
2882 | ||
2883 | if (level == lowest_level) { | |
2884 | if (dec) | |
2885 | p->slots[level]++; | |
2886 | goto done; | |
2887 | } | |
2888 | ||
2889 | err = read_block_for_search(root, p, &b, level, slot, key); | |
2890 | if (err == -EAGAIN) | |
2891 | goto again; | |
2892 | if (err) { | |
2893 | ret = err; | |
2894 | goto done; | |
2895 | } | |
2896 | ||
2897 | if (!p->skip_locking) { | |
2898 | level = btrfs_header_level(b); | |
2899 | if (level <= write_lock_level) { | |
2900 | if (!btrfs_try_tree_write_lock(b)) { | |
2901 | btrfs_set_path_blocking(p); | |
2902 | btrfs_tree_lock(b); | |
2903 | } | |
2904 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2905 | } else { | |
2906 | if (!btrfs_tree_read_lock_atomic(b)) { | |
2907 | btrfs_set_path_blocking(p); | |
2908 | btrfs_tree_read_lock(b); | |
2909 | } | |
2910 | p->locks[level] = BTRFS_READ_LOCK; | |
2911 | } | |
2912 | p->nodes[level] = b; | |
2913 | } | |
be0e5c09 | 2914 | } |
65b51a00 CM |
2915 | ret = 1; |
2916 | done: | |
b4ce94de CM |
2917 | /* |
2918 | * we don't really know what they plan on doing with the path | |
2919 | * from here on, so for now just mark it as blocking | |
2920 | */ | |
b9473439 CM |
2921 | if (!p->leave_spinning) |
2922 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2923 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2924 | btrfs_release_path(p); |
65b51a00 | 2925 | return ret; |
be0e5c09 CM |
2926 | } |
2927 | ||
5d9e75c4 JS |
2928 | /* |
2929 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2930 | * current state of the tree together with the operations recorded in the tree | |
2931 | * modification log to search for the key in a previous version of this tree, as | |
2932 | * denoted by the time_seq parameter. | |
2933 | * | |
2934 | * Naturally, there is no support for insert, delete or cow operations. | |
2935 | * | |
2936 | * The resulting path and return value will be set up as if we called | |
2937 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2938 | */ | |
310712b2 | 2939 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2940 | struct btrfs_path *p, u64 time_seq) |
2941 | { | |
0b246afa | 2942 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2943 | struct extent_buffer *b; |
2944 | int slot; | |
2945 | int ret; | |
2946 | int err; | |
2947 | int level; | |
2948 | int lowest_unlock = 1; | |
2949 | u8 lowest_level = 0; | |
d4b4087c | 2950 | int prev_cmp = -1; |
5d9e75c4 JS |
2951 | |
2952 | lowest_level = p->lowest_level; | |
2953 | WARN_ON(p->nodes[0] != NULL); | |
2954 | ||
2955 | if (p->search_commit_root) { | |
2956 | BUG_ON(time_seq); | |
2957 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2958 | } | |
2959 | ||
2960 | again: | |
5d9e75c4 | 2961 | b = get_old_root(root, time_seq); |
315bed43 NB |
2962 | if (!b) { |
2963 | ret = -EIO; | |
2964 | goto done; | |
2965 | } | |
5d9e75c4 | 2966 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2967 | p->locks[level] = BTRFS_READ_LOCK; |
2968 | ||
2969 | while (b) { | |
abe9339d QW |
2970 | int dec = 0; |
2971 | ||
5d9e75c4 JS |
2972 | level = btrfs_header_level(b); |
2973 | p->nodes[level] = b; | |
5d9e75c4 JS |
2974 | |
2975 | /* | |
2976 | * we have a lock on b and as long as we aren't changing | |
2977 | * the tree, there is no way to for the items in b to change. | |
2978 | * It is safe to drop the lock on our parent before we | |
2979 | * go through the expensive btree search on b. | |
2980 | */ | |
2981 | btrfs_unlock_up_safe(p, level + 1); | |
2982 | ||
d4b4087c | 2983 | /* |
01327610 | 2984 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2985 | * time. |
2986 | */ | |
2987 | prev_cmp = -1; | |
d7396f07 | 2988 | ret = key_search(b, key, level, &prev_cmp, &slot); |
cbca7d59 FM |
2989 | if (ret < 0) |
2990 | goto done; | |
5d9e75c4 | 2991 | |
abe9339d | 2992 | if (level == 0) { |
5d9e75c4 JS |
2993 | p->slots[level] = slot; |
2994 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
2995 | goto done; |
2996 | } | |
5d9e75c4 | 2997 | |
abe9339d QW |
2998 | if (ret && slot > 0) { |
2999 | dec = 1; | |
3000 | slot--; | |
3001 | } | |
3002 | p->slots[level] = slot; | |
3003 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 3004 | |
abe9339d QW |
3005 | if (level == lowest_level) { |
3006 | if (dec) | |
3007 | p->slots[level]++; | |
3008 | goto done; | |
3009 | } | |
5d9e75c4 | 3010 | |
abe9339d QW |
3011 | err = read_block_for_search(root, p, &b, level, slot, key); |
3012 | if (err == -EAGAIN) | |
3013 | goto again; | |
3014 | if (err) { | |
3015 | ret = err; | |
5d9e75c4 JS |
3016 | goto done; |
3017 | } | |
abe9339d QW |
3018 | |
3019 | level = btrfs_header_level(b); | |
3020 | if (!btrfs_tree_read_lock_atomic(b)) { | |
3021 | btrfs_set_path_blocking(p); | |
3022 | btrfs_tree_read_lock(b); | |
3023 | } | |
3024 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); | |
3025 | if (!b) { | |
3026 | ret = -ENOMEM; | |
3027 | goto done; | |
3028 | } | |
3029 | p->locks[level] = BTRFS_READ_LOCK; | |
3030 | p->nodes[level] = b; | |
5d9e75c4 JS |
3031 | } |
3032 | ret = 1; | |
3033 | done: | |
3034 | if (!p->leave_spinning) | |
3035 | btrfs_set_path_blocking(p); | |
3036 | if (ret < 0) | |
3037 | btrfs_release_path(p); | |
3038 | ||
3039 | return ret; | |
3040 | } | |
3041 | ||
2f38b3e1 AJ |
3042 | /* |
3043 | * helper to use instead of search slot if no exact match is needed but | |
3044 | * instead the next or previous item should be returned. | |
3045 | * When find_higher is true, the next higher item is returned, the next lower | |
3046 | * otherwise. | |
3047 | * When return_any and find_higher are both true, and no higher item is found, | |
3048 | * return the next lower instead. | |
3049 | * When return_any is true and find_higher is false, and no lower item is found, | |
3050 | * return the next higher instead. | |
3051 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3052 | * < 0 on error | |
3053 | */ | |
3054 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3055 | const struct btrfs_key *key, |
3056 | struct btrfs_path *p, int find_higher, | |
3057 | int return_any) | |
2f38b3e1 AJ |
3058 | { |
3059 | int ret; | |
3060 | struct extent_buffer *leaf; | |
3061 | ||
3062 | again: | |
3063 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3064 | if (ret <= 0) | |
3065 | return ret; | |
3066 | /* | |
3067 | * a return value of 1 means the path is at the position where the | |
3068 | * item should be inserted. Normally this is the next bigger item, | |
3069 | * but in case the previous item is the last in a leaf, path points | |
3070 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3071 | * item. | |
3072 | */ | |
3073 | leaf = p->nodes[0]; | |
3074 | ||
3075 | if (find_higher) { | |
3076 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3077 | ret = btrfs_next_leaf(root, p); | |
3078 | if (ret <= 0) | |
3079 | return ret; | |
3080 | if (!return_any) | |
3081 | return 1; | |
3082 | /* | |
3083 | * no higher item found, return the next | |
3084 | * lower instead | |
3085 | */ | |
3086 | return_any = 0; | |
3087 | find_higher = 0; | |
3088 | btrfs_release_path(p); | |
3089 | goto again; | |
3090 | } | |
3091 | } else { | |
e6793769 AJ |
3092 | if (p->slots[0] == 0) { |
3093 | ret = btrfs_prev_leaf(root, p); | |
3094 | if (ret < 0) | |
3095 | return ret; | |
3096 | if (!ret) { | |
23c6bf6a FDBM |
3097 | leaf = p->nodes[0]; |
3098 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3099 | p->slots[0]--; | |
e6793769 | 3100 | return 0; |
2f38b3e1 | 3101 | } |
e6793769 AJ |
3102 | if (!return_any) |
3103 | return 1; | |
3104 | /* | |
3105 | * no lower item found, return the next | |
3106 | * higher instead | |
3107 | */ | |
3108 | return_any = 0; | |
3109 | find_higher = 1; | |
3110 | btrfs_release_path(p); | |
3111 | goto again; | |
3112 | } else { | |
2f38b3e1 AJ |
3113 | --p->slots[0]; |
3114 | } | |
3115 | } | |
3116 | return 0; | |
3117 | } | |
3118 | ||
74123bd7 CM |
3119 | /* |
3120 | * adjust the pointers going up the tree, starting at level | |
3121 | * making sure the right key of each node is points to 'key'. | |
3122 | * This is used after shifting pointers to the left, so it stops | |
3123 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3124 | * higher levels | |
aa5d6bed | 3125 | * |
74123bd7 | 3126 | */ |
b167fa91 | 3127 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3128 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3129 | { |
3130 | int i; | |
5f39d397 | 3131 | struct extent_buffer *t; |
0e82bcfe | 3132 | int ret; |
5f39d397 | 3133 | |
234b63a0 | 3134 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3135 | int tslot = path->slots[i]; |
0e82bcfe | 3136 | |
eb60ceac | 3137 | if (!path->nodes[i]) |
be0e5c09 | 3138 | break; |
5f39d397 | 3139 | t = path->nodes[i]; |
0e82bcfe DS |
3140 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3141 | GFP_ATOMIC); | |
3142 | BUG_ON(ret < 0); | |
5f39d397 | 3143 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3144 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3145 | if (tslot != 0) |
3146 | break; | |
3147 | } | |
3148 | } | |
3149 | ||
31840ae1 ZY |
3150 | /* |
3151 | * update item key. | |
3152 | * | |
3153 | * This function isn't completely safe. It's the caller's responsibility | |
3154 | * that the new key won't break the order | |
3155 | */ | |
b7a0365e DD |
3156 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3157 | struct btrfs_path *path, | |
310712b2 | 3158 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3159 | { |
3160 | struct btrfs_disk_key disk_key; | |
3161 | struct extent_buffer *eb; | |
3162 | int slot; | |
3163 | ||
3164 | eb = path->nodes[0]; | |
3165 | slot = path->slots[0]; | |
3166 | if (slot > 0) { | |
3167 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
3168 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
3169 | btrfs_crit(fs_info, | |
3170 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3171 | slot, btrfs_disk_key_objectid(&disk_key), | |
3172 | btrfs_disk_key_type(&disk_key), | |
3173 | btrfs_disk_key_offset(&disk_key), | |
3174 | new_key->objectid, new_key->type, | |
3175 | new_key->offset); | |
3176 | btrfs_print_leaf(eb); | |
3177 | BUG(); | |
3178 | } | |
31840ae1 ZY |
3179 | } |
3180 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3181 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
3182 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
3183 | btrfs_crit(fs_info, | |
3184 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3185 | slot, btrfs_disk_key_objectid(&disk_key), | |
3186 | btrfs_disk_key_type(&disk_key), | |
3187 | btrfs_disk_key_offset(&disk_key), | |
3188 | new_key->objectid, new_key->type, | |
3189 | new_key->offset); | |
3190 | btrfs_print_leaf(eb); | |
3191 | BUG(); | |
3192 | } | |
31840ae1 ZY |
3193 | } |
3194 | ||
3195 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3196 | btrfs_set_item_key(eb, &disk_key, slot); | |
3197 | btrfs_mark_buffer_dirty(eb); | |
3198 | if (slot == 0) | |
b167fa91 | 3199 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3200 | } |
3201 | ||
74123bd7 CM |
3202 | /* |
3203 | * try to push data from one node into the next node left in the | |
79f95c82 | 3204 | * tree. |
aa5d6bed CM |
3205 | * |
3206 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3207 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3208 | */ |
98ed5174 | 3209 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 3210 | struct extent_buffer *dst, |
971a1f66 | 3211 | struct extent_buffer *src, int empty) |
be0e5c09 | 3212 | { |
d30a668f | 3213 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 3214 | int push_items = 0; |
bb803951 CM |
3215 | int src_nritems; |
3216 | int dst_nritems; | |
aa5d6bed | 3217 | int ret = 0; |
be0e5c09 | 3218 | |
5f39d397 CM |
3219 | src_nritems = btrfs_header_nritems(src); |
3220 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3221 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3222 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3223 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3224 | |
bce4eae9 | 3225 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3226 | return 1; |
3227 | ||
d397712b | 3228 | if (push_items <= 0) |
be0e5c09 CM |
3229 | return 1; |
3230 | ||
bce4eae9 | 3231 | if (empty) { |
971a1f66 | 3232 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3233 | if (push_items < src_nritems) { |
3234 | /* leave at least 8 pointers in the node if | |
3235 | * we aren't going to empty it | |
3236 | */ | |
3237 | if (src_nritems - push_items < 8) { | |
3238 | if (push_items <= 8) | |
3239 | return 1; | |
3240 | push_items -= 8; | |
3241 | } | |
3242 | } | |
3243 | } else | |
3244 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3245 | |
ed874f0d | 3246 | ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 3247 | if (ret) { |
66642832 | 3248 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3249 | return ret; |
3250 | } | |
5f39d397 CM |
3251 | copy_extent_buffer(dst, src, |
3252 | btrfs_node_key_ptr_offset(dst_nritems), | |
3253 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3254 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3255 | |
bb803951 | 3256 | if (push_items < src_nritems) { |
57911b8b | 3257 | /* |
bf1d3425 DS |
3258 | * Don't call tree_mod_log_insert_move here, key removal was |
3259 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3260 | */ |
5f39d397 CM |
3261 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3262 | btrfs_node_key_ptr_offset(push_items), | |
3263 | (src_nritems - push_items) * | |
3264 | sizeof(struct btrfs_key_ptr)); | |
3265 | } | |
3266 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3267 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3268 | btrfs_mark_buffer_dirty(src); | |
3269 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3270 | |
79f95c82 CM |
3271 | return ret; |
3272 | } | |
3273 | ||
3274 | /* | |
3275 | * try to push data from one node into the next node right in the | |
3276 | * tree. | |
3277 | * | |
3278 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3279 | * error, and > 0 if there was no room in the right hand block. | |
3280 | * | |
3281 | * this will only push up to 1/2 the contents of the left node over | |
3282 | */ | |
5f39d397 | 3283 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3284 | struct extent_buffer *dst, |
3285 | struct extent_buffer *src) | |
79f95c82 | 3286 | { |
55d32ed8 | 3287 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
3288 | int push_items = 0; |
3289 | int max_push; | |
3290 | int src_nritems; | |
3291 | int dst_nritems; | |
3292 | int ret = 0; | |
79f95c82 | 3293 | |
7bb86316 CM |
3294 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3295 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3296 | ||
5f39d397 CM |
3297 | src_nritems = btrfs_header_nritems(src); |
3298 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3299 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3300 | if (push_items <= 0) |
79f95c82 | 3301 | return 1; |
bce4eae9 | 3302 | |
d397712b | 3303 | if (src_nritems < 4) |
bce4eae9 | 3304 | return 1; |
79f95c82 CM |
3305 | |
3306 | max_push = src_nritems / 2 + 1; | |
3307 | /* don't try to empty the node */ | |
d397712b | 3308 | if (max_push >= src_nritems) |
79f95c82 | 3309 | return 1; |
252c38f0 | 3310 | |
79f95c82 CM |
3311 | if (max_push < push_items) |
3312 | push_items = max_push; | |
3313 | ||
bf1d3425 DS |
3314 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3315 | BUG_ON(ret < 0); | |
5f39d397 CM |
3316 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3317 | btrfs_node_key_ptr_offset(0), | |
3318 | (dst_nritems) * | |
3319 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3320 | |
ed874f0d DS |
3321 | ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
3322 | push_items); | |
5de865ee | 3323 | if (ret) { |
66642832 | 3324 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3325 | return ret; |
3326 | } | |
5f39d397 CM |
3327 | copy_extent_buffer(dst, src, |
3328 | btrfs_node_key_ptr_offset(0), | |
3329 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3330 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3331 | |
5f39d397 CM |
3332 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3333 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3334 | |
5f39d397 CM |
3335 | btrfs_mark_buffer_dirty(src); |
3336 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3337 | |
aa5d6bed | 3338 | return ret; |
be0e5c09 CM |
3339 | } |
3340 | ||
97571fd0 CM |
3341 | /* |
3342 | * helper function to insert a new root level in the tree. | |
3343 | * A new node is allocated, and a single item is inserted to | |
3344 | * point to the existing root | |
aa5d6bed CM |
3345 | * |
3346 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3347 | */ |
d397712b | 3348 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3349 | struct btrfs_root *root, |
fdd99c72 | 3350 | struct btrfs_path *path, int level) |
5c680ed6 | 3351 | { |
0b246afa | 3352 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3353 | u64 lower_gen; |
5f39d397 CM |
3354 | struct extent_buffer *lower; |
3355 | struct extent_buffer *c; | |
925baedd | 3356 | struct extent_buffer *old; |
5f39d397 | 3357 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3358 | int ret; |
5c680ed6 CM |
3359 | |
3360 | BUG_ON(path->nodes[level]); | |
3361 | BUG_ON(path->nodes[level-1] != root->node); | |
3362 | ||
7bb86316 CM |
3363 | lower = path->nodes[level-1]; |
3364 | if (level == 1) | |
3365 | btrfs_item_key(lower, &lower_key, 0); | |
3366 | else | |
3367 | btrfs_node_key(lower, &lower_key, 0); | |
3368 | ||
a6279470 FM |
3369 | c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, |
3370 | root->node->start, 0); | |
5f39d397 CM |
3371 | if (IS_ERR(c)) |
3372 | return PTR_ERR(c); | |
925baedd | 3373 | |
0b246afa | 3374 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3375 | |
5f39d397 | 3376 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3377 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3378 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3379 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3380 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3381 | |
3382 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3383 | |
5f39d397 | 3384 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3385 | |
925baedd | 3386 | old = root->node; |
d9d19a01 DS |
3387 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3388 | BUG_ON(ret < 0); | |
240f62c8 | 3389 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3390 | |
3391 | /* the super has an extra ref to root->node */ | |
3392 | free_extent_buffer(old); | |
3393 | ||
0b86a832 | 3394 | add_root_to_dirty_list(root); |
67439dad | 3395 | atomic_inc(&c->refs); |
5f39d397 | 3396 | path->nodes[level] = c; |
95449a16 | 3397 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3398 | path->slots[level] = 0; |
3399 | return 0; | |
3400 | } | |
3401 | ||
74123bd7 CM |
3402 | /* |
3403 | * worker function to insert a single pointer in a node. | |
3404 | * the node should have enough room for the pointer already | |
97571fd0 | 3405 | * |
74123bd7 CM |
3406 | * slot and level indicate where you want the key to go, and |
3407 | * blocknr is the block the key points to. | |
3408 | */ | |
143bede5 | 3409 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 3410 | struct btrfs_path *path, |
143bede5 | 3411 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3412 | int slot, int level) |
74123bd7 | 3413 | { |
5f39d397 | 3414 | struct extent_buffer *lower; |
74123bd7 | 3415 | int nritems; |
f3ea38da | 3416 | int ret; |
5c680ed6 CM |
3417 | |
3418 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3419 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3420 | lower = path->nodes[level]; |
3421 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3422 | BUG_ON(slot > nritems); |
6ad3cf6d | 3423 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 3424 | if (slot != nritems) { |
bf1d3425 DS |
3425 | if (level) { |
3426 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3427 | nritems - slot); |
bf1d3425 DS |
3428 | BUG_ON(ret < 0); |
3429 | } | |
5f39d397 CM |
3430 | memmove_extent_buffer(lower, |
3431 | btrfs_node_key_ptr_offset(slot + 1), | |
3432 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3433 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3434 | } |
c3e06965 | 3435 | if (level) { |
e09c2efe DS |
3436 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3437 | GFP_NOFS); | |
f3ea38da JS |
3438 | BUG_ON(ret < 0); |
3439 | } | |
5f39d397 | 3440 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3441 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3442 | WARN_ON(trans->transid == 0); |
3443 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3444 | btrfs_set_header_nritems(lower, nritems + 1); |
3445 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3446 | } |
3447 | ||
97571fd0 CM |
3448 | /* |
3449 | * split the node at the specified level in path in two. | |
3450 | * The path is corrected to point to the appropriate node after the split | |
3451 | * | |
3452 | * Before splitting this tries to make some room in the node by pushing | |
3453 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3454 | * |
3455 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3456 | */ |
e02119d5 CM |
3457 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3458 | struct btrfs_root *root, | |
3459 | struct btrfs_path *path, int level) | |
be0e5c09 | 3460 | { |
0b246afa | 3461 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3462 | struct extent_buffer *c; |
3463 | struct extent_buffer *split; | |
3464 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3465 | int mid; |
5c680ed6 | 3466 | int ret; |
7518a238 | 3467 | u32 c_nritems; |
eb60ceac | 3468 | |
5f39d397 | 3469 | c = path->nodes[level]; |
7bb86316 | 3470 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3471 | if (c == root->node) { |
d9abbf1c | 3472 | /* |
90f8d62e JS |
3473 | * trying to split the root, lets make a new one |
3474 | * | |
fdd99c72 | 3475 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3476 | * insert_new_root, because that root buffer will be kept as a |
3477 | * normal node. We are going to log removal of half of the | |
3478 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3479 | * tree lock on the buffer, which is why we cannot race with | |
3480 | * other tree_mod_log users. | |
d9abbf1c | 3481 | */ |
fdd99c72 | 3482 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3483 | if (ret) |
3484 | return ret; | |
b3612421 | 3485 | } else { |
e66f709b | 3486 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3487 | c = path->nodes[level]; |
3488 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3489 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3490 | return 0; |
54aa1f4d CM |
3491 | if (ret < 0) |
3492 | return ret; | |
be0e5c09 | 3493 | } |
e66f709b | 3494 | |
5f39d397 | 3495 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3496 | mid = (c_nritems + 1) / 2; |
3497 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3498 | |
a6279470 FM |
3499 | split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, |
3500 | c->start, 0); | |
5f39d397 CM |
3501 | if (IS_ERR(split)) |
3502 | return PTR_ERR(split); | |
3503 | ||
0b246afa | 3504 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3505 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3506 | |
ed874f0d | 3507 | ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 3508 | if (ret) { |
66642832 | 3509 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3510 | return ret; |
3511 | } | |
5f39d397 CM |
3512 | copy_extent_buffer(split, c, |
3513 | btrfs_node_key_ptr_offset(0), | |
3514 | btrfs_node_key_ptr_offset(mid), | |
3515 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3516 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3517 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3518 | ret = 0; |
3519 | ||
5f39d397 CM |
3520 | btrfs_mark_buffer_dirty(c); |
3521 | btrfs_mark_buffer_dirty(split); | |
3522 | ||
6ad3cf6d | 3523 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 3524 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3525 | |
5de08d7d | 3526 | if (path->slots[level] >= mid) { |
5c680ed6 | 3527 | path->slots[level] -= mid; |
925baedd | 3528 | btrfs_tree_unlock(c); |
5f39d397 CM |
3529 | free_extent_buffer(c); |
3530 | path->nodes[level] = split; | |
5c680ed6 CM |
3531 | path->slots[level + 1] += 1; |
3532 | } else { | |
925baedd | 3533 | btrfs_tree_unlock(split); |
5f39d397 | 3534 | free_extent_buffer(split); |
be0e5c09 | 3535 | } |
aa5d6bed | 3536 | return ret; |
be0e5c09 CM |
3537 | } |
3538 | ||
74123bd7 CM |
3539 | /* |
3540 | * how many bytes are required to store the items in a leaf. start | |
3541 | * and nr indicate which items in the leaf to check. This totals up the | |
3542 | * space used both by the item structs and the item data | |
3543 | */ | |
5f39d397 | 3544 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3545 | { |
41be1f3b JB |
3546 | struct btrfs_item *start_item; |
3547 | struct btrfs_item *end_item; | |
3548 | struct btrfs_map_token token; | |
be0e5c09 | 3549 | int data_len; |
5f39d397 | 3550 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3551 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3552 | |
3553 | if (!nr) | |
3554 | return 0; | |
c82f823c | 3555 | btrfs_init_map_token(&token, l); |
dd3cc16b RK |
3556 | start_item = btrfs_item_nr(start); |
3557 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3558 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3559 | btrfs_token_item_size(l, start_item, &token); | |
3560 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3561 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3562 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3563 | return data_len; |
3564 | } | |
3565 | ||
d4dbff95 CM |
3566 | /* |
3567 | * The space between the end of the leaf items and | |
3568 | * the start of the leaf data. IOW, how much room | |
3569 | * the leaf has left for both items and data | |
3570 | */ | |
e902baac | 3571 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 3572 | { |
e902baac | 3573 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
3574 | int nritems = btrfs_header_nritems(leaf); |
3575 | int ret; | |
0b246afa JM |
3576 | |
3577 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3578 | if (ret < 0) { |
0b246afa JM |
3579 | btrfs_crit(fs_info, |
3580 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3581 | ret, | |
3582 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3583 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3584 | } |
3585 | return ret; | |
d4dbff95 CM |
3586 | } |
3587 | ||
99d8f83c CM |
3588 | /* |
3589 | * min slot controls the lowest index we're willing to push to the | |
3590 | * right. We'll push up to and including min_slot, but no lower | |
3591 | */ | |
f72f0010 | 3592 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
3593 | int data_size, int empty, |
3594 | struct extent_buffer *right, | |
99d8f83c CM |
3595 | int free_space, u32 left_nritems, |
3596 | u32 min_slot) | |
00ec4c51 | 3597 | { |
f72f0010 | 3598 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 3599 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3600 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3601 | struct btrfs_map_token token; |
5f39d397 | 3602 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3603 | int slot; |
34a38218 | 3604 | u32 i; |
00ec4c51 CM |
3605 | int push_space = 0; |
3606 | int push_items = 0; | |
0783fcfc | 3607 | struct btrfs_item *item; |
34a38218 | 3608 | u32 nr; |
7518a238 | 3609 | u32 right_nritems; |
5f39d397 | 3610 | u32 data_end; |
db94535d | 3611 | u32 this_item_size; |
00ec4c51 | 3612 | |
34a38218 CM |
3613 | if (empty) |
3614 | nr = 0; | |
3615 | else | |
99d8f83c | 3616 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3617 | |
31840ae1 | 3618 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3619 | push_space += data_size; |
31840ae1 | 3620 | |
44871b1b | 3621 | slot = path->slots[1]; |
34a38218 CM |
3622 | i = left_nritems - 1; |
3623 | while (i >= nr) { | |
dd3cc16b | 3624 | item = btrfs_item_nr(i); |
db94535d | 3625 | |
31840ae1 ZY |
3626 | if (!empty && push_items > 0) { |
3627 | if (path->slots[0] > i) | |
3628 | break; | |
3629 | if (path->slots[0] == i) { | |
e902baac DS |
3630 | int space = btrfs_leaf_free_space(left); |
3631 | ||
31840ae1 ZY |
3632 | if (space + push_space * 2 > free_space) |
3633 | break; | |
3634 | } | |
3635 | } | |
3636 | ||
00ec4c51 | 3637 | if (path->slots[0] == i) |
87b29b20 | 3638 | push_space += data_size; |
db94535d | 3639 | |
db94535d CM |
3640 | this_item_size = btrfs_item_size(left, item); |
3641 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3642 | break; |
31840ae1 | 3643 | |
00ec4c51 | 3644 | push_items++; |
db94535d | 3645 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3646 | if (i == 0) |
3647 | break; | |
3648 | i--; | |
db94535d | 3649 | } |
5f39d397 | 3650 | |
925baedd CM |
3651 | if (push_items == 0) |
3652 | goto out_unlock; | |
5f39d397 | 3653 | |
6c1500f2 | 3654 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3655 | |
00ec4c51 | 3656 | /* push left to right */ |
5f39d397 | 3657 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3658 | |
5f39d397 | 3659 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
8f881e8c | 3660 | push_space -= leaf_data_end(left); |
5f39d397 | 3661 | |
00ec4c51 | 3662 | /* make room in the right data area */ |
8f881e8c | 3663 | data_end = leaf_data_end(right); |
5f39d397 | 3664 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3665 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3666 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3667 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3668 | |
00ec4c51 | 3669 | /* copy from the left data area */ |
3d9ec8c4 | 3670 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3671 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 3672 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 3673 | push_space); |
5f39d397 CM |
3674 | |
3675 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3676 | btrfs_item_nr_offset(0), | |
3677 | right_nritems * sizeof(struct btrfs_item)); | |
3678 | ||
00ec4c51 | 3679 | /* copy the items from left to right */ |
5f39d397 CM |
3680 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3681 | btrfs_item_nr_offset(left_nritems - push_items), | |
3682 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3683 | |
3684 | /* update the item pointers */ | |
c82f823c | 3685 | btrfs_init_map_token(&token, right); |
7518a238 | 3686 | right_nritems += push_items; |
5f39d397 | 3687 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3688 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3689 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3690 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3691 | push_space -= btrfs_token_item_size(right, item, &token); |
3692 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3693 | } |
3694 | ||
7518a238 | 3695 | left_nritems -= push_items; |
5f39d397 | 3696 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3697 | |
34a38218 CM |
3698 | if (left_nritems) |
3699 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3700 | else |
6a884d7d | 3701 | btrfs_clean_tree_block(left); |
f0486c68 | 3702 | |
5f39d397 | 3703 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3704 | |
5f39d397 CM |
3705 | btrfs_item_key(right, &disk_key, 0); |
3706 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3707 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3708 | |
00ec4c51 | 3709 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3710 | if (path->slots[0] >= left_nritems) { |
3711 | path->slots[0] -= left_nritems; | |
925baedd | 3712 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 3713 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 3714 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3715 | free_extent_buffer(path->nodes[0]); |
3716 | path->nodes[0] = right; | |
00ec4c51 CM |
3717 | path->slots[1] += 1; |
3718 | } else { | |
925baedd | 3719 | btrfs_tree_unlock(right); |
5f39d397 | 3720 | free_extent_buffer(right); |
00ec4c51 CM |
3721 | } |
3722 | return 0; | |
925baedd CM |
3723 | |
3724 | out_unlock: | |
3725 | btrfs_tree_unlock(right); | |
3726 | free_extent_buffer(right); | |
3727 | return 1; | |
00ec4c51 | 3728 | } |
925baedd | 3729 | |
44871b1b CM |
3730 | /* |
3731 | * push some data in the path leaf to the right, trying to free up at | |
3732 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3733 | * | |
3734 | * returns 1 if the push failed because the other node didn't have enough | |
3735 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3736 | * |
3737 | * this will push starting from min_slot to the end of the leaf. It won't | |
3738 | * push any slot lower than min_slot | |
44871b1b CM |
3739 | */ |
3740 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3741 | *root, struct btrfs_path *path, |
3742 | int min_data_size, int data_size, | |
3743 | int empty, u32 min_slot) | |
44871b1b CM |
3744 | { |
3745 | struct extent_buffer *left = path->nodes[0]; | |
3746 | struct extent_buffer *right; | |
3747 | struct extent_buffer *upper; | |
3748 | int slot; | |
3749 | int free_space; | |
3750 | u32 left_nritems; | |
3751 | int ret; | |
3752 | ||
3753 | if (!path->nodes[1]) | |
3754 | return 1; | |
3755 | ||
3756 | slot = path->slots[1]; | |
3757 | upper = path->nodes[1]; | |
3758 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3759 | return 1; | |
3760 | ||
3761 | btrfs_assert_tree_locked(path->nodes[1]); | |
3762 | ||
4b231ae4 | 3763 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3764 | /* |
3765 | * slot + 1 is not valid or we fail to read the right node, | |
3766 | * no big deal, just return. | |
3767 | */ | |
3768 | if (IS_ERR(right)) | |
91ca338d TI |
3769 | return 1; |
3770 | ||
44871b1b | 3771 | btrfs_tree_lock(right); |
8bead258 | 3772 | btrfs_set_lock_blocking_write(right); |
44871b1b | 3773 | |
e902baac | 3774 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3775 | if (free_space < data_size) |
3776 | goto out_unlock; | |
3777 | ||
3778 | /* cow and double check */ | |
3779 | ret = btrfs_cow_block(trans, root, right, upper, | |
3780 | slot + 1, &right); | |
3781 | if (ret) | |
3782 | goto out_unlock; | |
3783 | ||
e902baac | 3784 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3785 | if (free_space < data_size) |
3786 | goto out_unlock; | |
3787 | ||
3788 | left_nritems = btrfs_header_nritems(left); | |
3789 | if (left_nritems == 0) | |
3790 | goto out_unlock; | |
3791 | ||
2ef1fed2 FDBM |
3792 | if (path->slots[0] == left_nritems && !empty) { |
3793 | /* Key greater than all keys in the leaf, right neighbor has | |
3794 | * enough room for it and we're not emptying our leaf to delete | |
3795 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3796 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3797 | btrfs_tree_unlock(left); |
3798 | free_extent_buffer(left); | |
3799 | path->nodes[0] = right; | |
3800 | path->slots[0] = 0; | |
3801 | path->slots[1]++; | |
3802 | return 0; | |
3803 | } | |
3804 | ||
f72f0010 | 3805 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3806 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3807 | out_unlock: |
3808 | btrfs_tree_unlock(right); | |
3809 | free_extent_buffer(right); | |
3810 | return 1; | |
3811 | } | |
3812 | ||
74123bd7 CM |
3813 | /* |
3814 | * push some data in the path leaf to the left, trying to free up at | |
3815 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3816 | * |
3817 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3818 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3819 | * items | |
74123bd7 | 3820 | */ |
8087c193 | 3821 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3822 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3823 | int free_space, u32 right_nritems, |
3824 | u32 max_slot) | |
be0e5c09 | 3825 | { |
8087c193 | 3826 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3827 | struct btrfs_disk_key disk_key; |
3828 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3829 | int i; |
be0e5c09 CM |
3830 | int push_space = 0; |
3831 | int push_items = 0; | |
0783fcfc | 3832 | struct btrfs_item *item; |
7518a238 | 3833 | u32 old_left_nritems; |
34a38218 | 3834 | u32 nr; |
aa5d6bed | 3835 | int ret = 0; |
db94535d CM |
3836 | u32 this_item_size; |
3837 | u32 old_left_item_size; | |
cfed81a0 CM |
3838 | struct btrfs_map_token token; |
3839 | ||
34a38218 | 3840 | if (empty) |
99d8f83c | 3841 | nr = min(right_nritems, max_slot); |
34a38218 | 3842 | else |
99d8f83c | 3843 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3844 | |
3845 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3846 | item = btrfs_item_nr(i); |
db94535d | 3847 | |
31840ae1 ZY |
3848 | if (!empty && push_items > 0) { |
3849 | if (path->slots[0] < i) | |
3850 | break; | |
3851 | if (path->slots[0] == i) { | |
e902baac DS |
3852 | int space = btrfs_leaf_free_space(right); |
3853 | ||
31840ae1 ZY |
3854 | if (space + push_space * 2 > free_space) |
3855 | break; | |
3856 | } | |
3857 | } | |
3858 | ||
be0e5c09 | 3859 | if (path->slots[0] == i) |
87b29b20 | 3860 | push_space += data_size; |
db94535d CM |
3861 | |
3862 | this_item_size = btrfs_item_size(right, item); | |
3863 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3864 | break; |
db94535d | 3865 | |
be0e5c09 | 3866 | push_items++; |
db94535d CM |
3867 | push_space += this_item_size + sizeof(*item); |
3868 | } | |
3869 | ||
be0e5c09 | 3870 | if (push_items == 0) { |
925baedd CM |
3871 | ret = 1; |
3872 | goto out; | |
be0e5c09 | 3873 | } |
fae7f21c | 3874 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3875 | |
be0e5c09 | 3876 | /* push data from right to left */ |
5f39d397 CM |
3877 | copy_extent_buffer(left, right, |
3878 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3879 | btrfs_item_nr_offset(0), | |
3880 | push_items * sizeof(struct btrfs_item)); | |
3881 | ||
0b246afa | 3882 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3883 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3884 | |
3d9ec8c4 | 3885 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3886 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3887 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3888 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3889 | push_space); |
5f39d397 | 3890 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3891 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3892 | |
c82f823c | 3893 | btrfs_init_map_token(&token, left); |
db94535d | 3894 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3895 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3896 | u32 ioff; |
db94535d | 3897 | |
dd3cc16b | 3898 | item = btrfs_item_nr(i); |
db94535d | 3899 | |
cfed81a0 CM |
3900 | ioff = btrfs_token_item_offset(left, item, &token); |
3901 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3902 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3903 | &token); |
be0e5c09 | 3904 | } |
5f39d397 | 3905 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3906 | |
3907 | /* fixup right node */ | |
31b1a2bd JL |
3908 | if (push_items > right_nritems) |
3909 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3910 | right_nritems); |
34a38218 CM |
3911 | |
3912 | if (push_items < right_nritems) { | |
3913 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
8f881e8c | 3914 | leaf_data_end(right); |
3d9ec8c4 | 3915 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3916 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3917 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3918 | leaf_data_end(right), push_space); |
34a38218 CM |
3919 | |
3920 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3921 | btrfs_item_nr_offset(push_items), |
3922 | (btrfs_header_nritems(right) - push_items) * | |
3923 | sizeof(struct btrfs_item)); | |
34a38218 | 3924 | } |
c82f823c DS |
3925 | |
3926 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3927 | right_nritems -= push_items; |
3928 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3929 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3930 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3931 | item = btrfs_item_nr(i); |
db94535d | 3932 | |
cfed81a0 CM |
3933 | push_space = push_space - btrfs_token_item_size(right, |
3934 | item, &token); | |
3935 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3936 | } |
eb60ceac | 3937 | |
5f39d397 | 3938 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3939 | if (right_nritems) |
3940 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3941 | else |
6a884d7d | 3942 | btrfs_clean_tree_block(right); |
098f59c2 | 3943 | |
5f39d397 | 3944 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3945 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3946 | |
3947 | /* then fixup the leaf pointer in the path */ | |
3948 | if (path->slots[0] < push_items) { | |
3949 | path->slots[0] += old_left_nritems; | |
925baedd | 3950 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3951 | free_extent_buffer(path->nodes[0]); |
3952 | path->nodes[0] = left; | |
be0e5c09 CM |
3953 | path->slots[1] -= 1; |
3954 | } else { | |
925baedd | 3955 | btrfs_tree_unlock(left); |
5f39d397 | 3956 | free_extent_buffer(left); |
be0e5c09 CM |
3957 | path->slots[0] -= push_items; |
3958 | } | |
eb60ceac | 3959 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3960 | return ret; |
925baedd CM |
3961 | out: |
3962 | btrfs_tree_unlock(left); | |
3963 | free_extent_buffer(left); | |
3964 | return ret; | |
be0e5c09 CM |
3965 | } |
3966 | ||
44871b1b CM |
3967 | /* |
3968 | * push some data in the path leaf to the left, trying to free up at | |
3969 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3970 | * |
3971 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3972 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3973 | * items | |
44871b1b CM |
3974 | */ |
3975 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3976 | *root, struct btrfs_path *path, int min_data_size, |
3977 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3978 | { |
3979 | struct extent_buffer *right = path->nodes[0]; | |
3980 | struct extent_buffer *left; | |
3981 | int slot; | |
3982 | int free_space; | |
3983 | u32 right_nritems; | |
3984 | int ret = 0; | |
3985 | ||
3986 | slot = path->slots[1]; | |
3987 | if (slot == 0) | |
3988 | return 1; | |
3989 | if (!path->nodes[1]) | |
3990 | return 1; | |
3991 | ||
3992 | right_nritems = btrfs_header_nritems(right); | |
3993 | if (right_nritems == 0) | |
3994 | return 1; | |
3995 | ||
3996 | btrfs_assert_tree_locked(path->nodes[1]); | |
3997 | ||
4b231ae4 | 3998 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
3999 | /* |
4000 | * slot - 1 is not valid or we fail to read the left node, | |
4001 | * no big deal, just return. | |
4002 | */ | |
4003 | if (IS_ERR(left)) | |
91ca338d TI |
4004 | return 1; |
4005 | ||
44871b1b | 4006 | btrfs_tree_lock(left); |
8bead258 | 4007 | btrfs_set_lock_blocking_write(left); |
44871b1b | 4008 | |
e902baac | 4009 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
4010 | if (free_space < data_size) { |
4011 | ret = 1; | |
4012 | goto out; | |
4013 | } | |
4014 | ||
4015 | /* cow and double check */ | |
4016 | ret = btrfs_cow_block(trans, root, left, | |
4017 | path->nodes[1], slot - 1, &left); | |
4018 | if (ret) { | |
4019 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4020 | if (ret == -ENOSPC) |
4021 | ret = 1; | |
44871b1b CM |
4022 | goto out; |
4023 | } | |
4024 | ||
e902baac | 4025 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
4026 | if (free_space < data_size) { |
4027 | ret = 1; | |
4028 | goto out; | |
4029 | } | |
4030 | ||
8087c193 | 4031 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
4032 | empty, left, free_space, right_nritems, |
4033 | max_slot); | |
44871b1b CM |
4034 | out: |
4035 | btrfs_tree_unlock(left); | |
4036 | free_extent_buffer(left); | |
4037 | return ret; | |
4038 | } | |
4039 | ||
4040 | /* | |
4041 | * split the path's leaf in two, making sure there is at least data_size | |
4042 | * available for the resulting leaf level of the path. | |
44871b1b | 4043 | */ |
143bede5 | 4044 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
4045 | struct btrfs_path *path, |
4046 | struct extent_buffer *l, | |
4047 | struct extent_buffer *right, | |
4048 | int slot, int mid, int nritems) | |
44871b1b | 4049 | { |
94f94ad9 | 4050 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
4051 | int data_copy_size; |
4052 | int rt_data_off; | |
4053 | int i; | |
44871b1b | 4054 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4055 | struct btrfs_map_token token; |
4056 | ||
44871b1b CM |
4057 | nritems = nritems - mid; |
4058 | btrfs_set_header_nritems(right, nritems); | |
8f881e8c | 4059 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); |
44871b1b CM |
4060 | |
4061 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4062 | btrfs_item_nr_offset(mid), | |
4063 | nritems * sizeof(struct btrfs_item)); | |
4064 | ||
4065 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4066 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4067 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 4068 | leaf_data_end(l), data_copy_size); |
44871b1b | 4069 | |
0b246afa | 4070 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b | 4071 | |
c82f823c | 4072 | btrfs_init_map_token(&token, right); |
44871b1b | 4073 | for (i = 0; i < nritems; i++) { |
dd3cc16b | 4074 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4075 | u32 ioff; |
4076 | ||
cfed81a0 CM |
4077 | ioff = btrfs_token_item_offset(right, item, &token); |
4078 | btrfs_set_token_item_offset(right, item, | |
4079 | ioff + rt_data_off, &token); | |
44871b1b CM |
4080 | } |
4081 | ||
44871b1b | 4082 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4083 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 4084 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
4085 | |
4086 | btrfs_mark_buffer_dirty(right); | |
4087 | btrfs_mark_buffer_dirty(l); | |
4088 | BUG_ON(path->slots[0] != slot); | |
4089 | ||
44871b1b CM |
4090 | if (mid <= slot) { |
4091 | btrfs_tree_unlock(path->nodes[0]); | |
4092 | free_extent_buffer(path->nodes[0]); | |
4093 | path->nodes[0] = right; | |
4094 | path->slots[0] -= mid; | |
4095 | path->slots[1] += 1; | |
4096 | } else { | |
4097 | btrfs_tree_unlock(right); | |
4098 | free_extent_buffer(right); | |
4099 | } | |
4100 | ||
4101 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4102 | } |
4103 | ||
99d8f83c CM |
4104 | /* |
4105 | * double splits happen when we need to insert a big item in the middle | |
4106 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4107 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4108 | * A B C | |
4109 | * | |
4110 | * We avoid this by trying to push the items on either side of our target | |
4111 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4112 | * completely. | |
4113 | */ | |
4114 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4115 | struct btrfs_root *root, | |
4116 | struct btrfs_path *path, | |
4117 | int data_size) | |
4118 | { | |
4119 | int ret; | |
4120 | int progress = 0; | |
4121 | int slot; | |
4122 | u32 nritems; | |
5a4267ca | 4123 | int space_needed = data_size; |
99d8f83c CM |
4124 | |
4125 | slot = path->slots[0]; | |
5a4267ca | 4126 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 4127 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
4128 | |
4129 | /* | |
4130 | * try to push all the items after our slot into the | |
4131 | * right leaf | |
4132 | */ | |
5a4267ca | 4133 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4134 | if (ret < 0) |
4135 | return ret; | |
4136 | ||
4137 | if (ret == 0) | |
4138 | progress++; | |
4139 | ||
4140 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4141 | /* | |
4142 | * our goal is to get our slot at the start or end of a leaf. If | |
4143 | * we've done so we're done | |
4144 | */ | |
4145 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4146 | return 0; | |
4147 | ||
e902baac | 4148 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4149 | return 0; |
4150 | ||
4151 | /* try to push all the items before our slot into the next leaf */ | |
4152 | slot = path->slots[0]; | |
263d3995 FM |
4153 | space_needed = data_size; |
4154 | if (slot > 0) | |
e902baac | 4155 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 4156 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4157 | if (ret < 0) |
4158 | return ret; | |
4159 | ||
4160 | if (ret == 0) | |
4161 | progress++; | |
4162 | ||
4163 | if (progress) | |
4164 | return 0; | |
4165 | return 1; | |
4166 | } | |
4167 | ||
74123bd7 CM |
4168 | /* |
4169 | * split the path's leaf in two, making sure there is at least data_size | |
4170 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4171 | * |
4172 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4173 | */ |
e02119d5 CM |
4174 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4175 | struct btrfs_root *root, | |
310712b2 | 4176 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4177 | struct btrfs_path *path, int data_size, |
4178 | int extend) | |
be0e5c09 | 4179 | { |
5d4f98a2 | 4180 | struct btrfs_disk_key disk_key; |
5f39d397 | 4181 | struct extent_buffer *l; |
7518a238 | 4182 | u32 nritems; |
eb60ceac CM |
4183 | int mid; |
4184 | int slot; | |
5f39d397 | 4185 | struct extent_buffer *right; |
b7a0365e | 4186 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4187 | int ret = 0; |
aa5d6bed | 4188 | int wret; |
5d4f98a2 | 4189 | int split; |
cc0c5538 | 4190 | int num_doubles = 0; |
99d8f83c | 4191 | int tried_avoid_double = 0; |
aa5d6bed | 4192 | |
a5719521 YZ |
4193 | l = path->nodes[0]; |
4194 | slot = path->slots[0]; | |
4195 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4196 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4197 | return -EOVERFLOW; |
4198 | ||
40689478 | 4199 | /* first try to make some room by pushing left and right */ |
33157e05 | 4200 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4201 | int space_needed = data_size; |
4202 | ||
4203 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 4204 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4205 | |
4206 | wret = push_leaf_right(trans, root, path, space_needed, | |
4207 | space_needed, 0, 0); | |
d397712b | 4208 | if (wret < 0) |
eaee50e8 | 4209 | return wret; |
3685f791 | 4210 | if (wret) { |
263d3995 FM |
4211 | space_needed = data_size; |
4212 | if (slot > 0) | |
e902baac | 4213 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4214 | wret = push_leaf_left(trans, root, path, space_needed, |
4215 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4216 | if (wret < 0) |
4217 | return wret; | |
4218 | } | |
4219 | l = path->nodes[0]; | |
aa5d6bed | 4220 | |
3685f791 | 4221 | /* did the pushes work? */ |
e902baac | 4222 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 4223 | return 0; |
3326d1b0 | 4224 | } |
aa5d6bed | 4225 | |
5c680ed6 | 4226 | if (!path->nodes[1]) { |
fdd99c72 | 4227 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4228 | if (ret) |
4229 | return ret; | |
4230 | } | |
cc0c5538 | 4231 | again: |
5d4f98a2 | 4232 | split = 1; |
cc0c5538 | 4233 | l = path->nodes[0]; |
eb60ceac | 4234 | slot = path->slots[0]; |
5f39d397 | 4235 | nritems = btrfs_header_nritems(l); |
d397712b | 4236 | mid = (nritems + 1) / 2; |
54aa1f4d | 4237 | |
5d4f98a2 YZ |
4238 | if (mid <= slot) { |
4239 | if (nritems == 1 || | |
4240 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4241 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4242 | if (slot >= nritems) { |
4243 | split = 0; | |
4244 | } else { | |
4245 | mid = slot; | |
4246 | if (mid != nritems && | |
4247 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4248 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4249 | if (data_size && !tried_avoid_double) |
4250 | goto push_for_double; | |
5d4f98a2 YZ |
4251 | split = 2; |
4252 | } | |
4253 | } | |
4254 | } | |
4255 | } else { | |
4256 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4257 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4258 | if (!extend && data_size && slot == 0) { |
4259 | split = 0; | |
4260 | } else if ((extend || !data_size) && slot == 0) { | |
4261 | mid = 1; | |
4262 | } else { | |
4263 | mid = slot; | |
4264 | if (mid != nritems && | |
4265 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4266 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4267 | if (data_size && !tried_avoid_double) |
4268 | goto push_for_double; | |
67871254 | 4269 | split = 2; |
5d4f98a2 YZ |
4270 | } |
4271 | } | |
4272 | } | |
4273 | } | |
4274 | ||
4275 | if (split == 0) | |
4276 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4277 | else | |
4278 | btrfs_item_key(l, &disk_key, mid); | |
4279 | ||
a6279470 FM |
4280 | right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, |
4281 | l->start, 0); | |
f0486c68 | 4282 | if (IS_ERR(right)) |
5f39d397 | 4283 | return PTR_ERR(right); |
f0486c68 | 4284 | |
0b246afa | 4285 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4286 | |
5d4f98a2 YZ |
4287 | if (split == 0) { |
4288 | if (mid <= slot) { | |
4289 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4290 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4291 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4292 | btrfs_tree_unlock(path->nodes[0]); |
4293 | free_extent_buffer(path->nodes[0]); | |
4294 | path->nodes[0] = right; | |
4295 | path->slots[0] = 0; | |
4296 | path->slots[1] += 1; | |
4297 | } else { | |
4298 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4299 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4300 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
4301 | btrfs_tree_unlock(path->nodes[0]); |
4302 | free_extent_buffer(path->nodes[0]); | |
4303 | path->nodes[0] = right; | |
4304 | path->slots[0] = 0; | |
143bede5 | 4305 | if (path->slots[1] == 0) |
b167fa91 | 4306 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4307 | } |
196e0249 LB |
4308 | /* |
4309 | * We create a new leaf 'right' for the required ins_len and | |
4310 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4311 | * the content of ins_len to 'right'. | |
4312 | */ | |
5d4f98a2 | 4313 | return ret; |
d4dbff95 | 4314 | } |
74123bd7 | 4315 | |
94f94ad9 | 4316 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 4317 | |
5d4f98a2 | 4318 | if (split == 2) { |
cc0c5538 CM |
4319 | BUG_ON(num_doubles != 0); |
4320 | num_doubles++; | |
4321 | goto again; | |
a429e513 | 4322 | } |
44871b1b | 4323 | |
143bede5 | 4324 | return 0; |
99d8f83c CM |
4325 | |
4326 | push_for_double: | |
4327 | push_for_double_split(trans, root, path, data_size); | |
4328 | tried_avoid_double = 1; | |
e902baac | 4329 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4330 | return 0; |
4331 | goto again; | |
be0e5c09 CM |
4332 | } |
4333 | ||
ad48fd75 YZ |
4334 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4335 | struct btrfs_root *root, | |
4336 | struct btrfs_path *path, int ins_len) | |
459931ec | 4337 | { |
ad48fd75 | 4338 | struct btrfs_key key; |
459931ec | 4339 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4340 | struct btrfs_file_extent_item *fi; |
4341 | u64 extent_len = 0; | |
4342 | u32 item_size; | |
4343 | int ret; | |
459931ec CM |
4344 | |
4345 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4346 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4347 | ||
4348 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4349 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4350 | ||
e902baac | 4351 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 4352 | return 0; |
459931ec CM |
4353 | |
4354 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4355 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4356 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4357 | struct btrfs_file_extent_item); | |
4358 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4359 | } | |
b3b4aa74 | 4360 | btrfs_release_path(path); |
459931ec | 4361 | |
459931ec | 4362 | path->keep_locks = 1; |
ad48fd75 YZ |
4363 | path->search_for_split = 1; |
4364 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4365 | path->search_for_split = 0; |
a8df6fe6 FM |
4366 | if (ret > 0) |
4367 | ret = -EAGAIN; | |
ad48fd75 YZ |
4368 | if (ret < 0) |
4369 | goto err; | |
459931ec | 4370 | |
ad48fd75 YZ |
4371 | ret = -EAGAIN; |
4372 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4373 | /* if our item isn't there, return now */ |
4374 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4375 | goto err; |
4376 | ||
109f6aef | 4377 | /* the leaf has changed, it now has room. return now */ |
e902baac | 4378 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
4379 | goto err; |
4380 | ||
ad48fd75 YZ |
4381 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4382 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4383 | struct btrfs_file_extent_item); | |
4384 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4385 | goto err; | |
459931ec CM |
4386 | } |
4387 | ||
b9473439 | 4388 | btrfs_set_path_blocking(path); |
ad48fd75 | 4389 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4390 | if (ret) |
4391 | goto err; | |
459931ec | 4392 | |
ad48fd75 | 4393 | path->keep_locks = 0; |
b9473439 | 4394 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4395 | return 0; |
4396 | err: | |
4397 | path->keep_locks = 0; | |
4398 | return ret; | |
4399 | } | |
4400 | ||
25263cd7 | 4401 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 4402 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4403 | unsigned long split_offset) |
4404 | { | |
4405 | struct extent_buffer *leaf; | |
4406 | struct btrfs_item *item; | |
4407 | struct btrfs_item *new_item; | |
4408 | int slot; | |
4409 | char *buf; | |
4410 | u32 nritems; | |
4411 | u32 item_size; | |
4412 | u32 orig_offset; | |
4413 | struct btrfs_disk_key disk_key; | |
4414 | ||
b9473439 | 4415 | leaf = path->nodes[0]; |
e902baac | 4416 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4417 | |
b4ce94de CM |
4418 | btrfs_set_path_blocking(path); |
4419 | ||
dd3cc16b | 4420 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4421 | orig_offset = btrfs_item_offset(leaf, item); |
4422 | item_size = btrfs_item_size(leaf, item); | |
4423 | ||
459931ec | 4424 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4425 | if (!buf) |
4426 | return -ENOMEM; | |
4427 | ||
459931ec CM |
4428 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4429 | path->slots[0]), item_size); | |
459931ec | 4430 | |
ad48fd75 | 4431 | slot = path->slots[0] + 1; |
459931ec | 4432 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4433 | if (slot != nritems) { |
4434 | /* shift the items */ | |
4435 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4436 | btrfs_item_nr_offset(slot), |
4437 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4438 | } |
4439 | ||
4440 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4441 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4442 | ||
dd3cc16b | 4443 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4444 | |
4445 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4446 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4447 | ||
4448 | btrfs_set_item_offset(leaf, item, | |
4449 | orig_offset + item_size - split_offset); | |
4450 | btrfs_set_item_size(leaf, item, split_offset); | |
4451 | ||
4452 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4453 | ||
4454 | /* write the data for the start of the original item */ | |
4455 | write_extent_buffer(leaf, buf, | |
4456 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4457 | split_offset); | |
4458 | ||
4459 | /* write the data for the new item */ | |
4460 | write_extent_buffer(leaf, buf + split_offset, | |
4461 | btrfs_item_ptr_offset(leaf, slot), | |
4462 | item_size - split_offset); | |
4463 | btrfs_mark_buffer_dirty(leaf); | |
4464 | ||
e902baac | 4465 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 4466 | kfree(buf); |
ad48fd75 YZ |
4467 | return 0; |
4468 | } | |
4469 | ||
4470 | /* | |
4471 | * This function splits a single item into two items, | |
4472 | * giving 'new_key' to the new item and splitting the | |
4473 | * old one at split_offset (from the start of the item). | |
4474 | * | |
4475 | * The path may be released by this operation. After | |
4476 | * the split, the path is pointing to the old item. The | |
4477 | * new item is going to be in the same node as the old one. | |
4478 | * | |
4479 | * Note, the item being split must be smaller enough to live alone on | |
4480 | * a tree block with room for one extra struct btrfs_item | |
4481 | * | |
4482 | * This allows us to split the item in place, keeping a lock on the | |
4483 | * leaf the entire time. | |
4484 | */ | |
4485 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4486 | struct btrfs_root *root, | |
4487 | struct btrfs_path *path, | |
310712b2 | 4488 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4489 | unsigned long split_offset) |
4490 | { | |
4491 | int ret; | |
4492 | ret = setup_leaf_for_split(trans, root, path, | |
4493 | sizeof(struct btrfs_item)); | |
4494 | if (ret) | |
4495 | return ret; | |
4496 | ||
25263cd7 | 4497 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
4498 | return ret; |
4499 | } | |
4500 | ||
ad48fd75 YZ |
4501 | /* |
4502 | * This function duplicate a item, giving 'new_key' to the new item. | |
4503 | * It guarantees both items live in the same tree leaf and the new item | |
4504 | * is contiguous with the original item. | |
4505 | * | |
4506 | * This allows us to split file extent in place, keeping a lock on the | |
4507 | * leaf the entire time. | |
4508 | */ | |
4509 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4510 | struct btrfs_root *root, | |
4511 | struct btrfs_path *path, | |
310712b2 | 4512 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4513 | { |
4514 | struct extent_buffer *leaf; | |
4515 | int ret; | |
4516 | u32 item_size; | |
4517 | ||
4518 | leaf = path->nodes[0]; | |
4519 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4520 | ret = setup_leaf_for_split(trans, root, path, | |
4521 | item_size + sizeof(struct btrfs_item)); | |
4522 | if (ret) | |
4523 | return ret; | |
4524 | ||
4525 | path->slots[0]++; | |
afe5fea7 | 4526 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4527 | item_size, item_size + |
4528 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4529 | leaf = path->nodes[0]; |
4530 | memcpy_extent_buffer(leaf, | |
4531 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4532 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4533 | item_size); | |
4534 | return 0; | |
4535 | } | |
4536 | ||
d352ac68 CM |
4537 | /* |
4538 | * make the item pointed to by the path smaller. new_size indicates | |
4539 | * how small to make it, and from_end tells us if we just chop bytes | |
4540 | * off the end of the item or if we shift the item to chop bytes off | |
4541 | * the front. | |
4542 | */ | |
78ac4f9e | 4543 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 4544 | { |
b18c6685 | 4545 | int slot; |
5f39d397 CM |
4546 | struct extent_buffer *leaf; |
4547 | struct btrfs_item *item; | |
b18c6685 CM |
4548 | u32 nritems; |
4549 | unsigned int data_end; | |
4550 | unsigned int old_data_start; | |
4551 | unsigned int old_size; | |
4552 | unsigned int size_diff; | |
4553 | int i; | |
cfed81a0 CM |
4554 | struct btrfs_map_token token; |
4555 | ||
5f39d397 | 4556 | leaf = path->nodes[0]; |
179e29e4 CM |
4557 | slot = path->slots[0]; |
4558 | ||
4559 | old_size = btrfs_item_size_nr(leaf, slot); | |
4560 | if (old_size == new_size) | |
143bede5 | 4561 | return; |
b18c6685 | 4562 | |
5f39d397 | 4563 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4564 | data_end = leaf_data_end(leaf); |
b18c6685 | 4565 | |
5f39d397 | 4566 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4567 | |
b18c6685 CM |
4568 | size_diff = old_size - new_size; |
4569 | ||
4570 | BUG_ON(slot < 0); | |
4571 | BUG_ON(slot >= nritems); | |
4572 | ||
4573 | /* | |
4574 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4575 | */ | |
4576 | /* first correct the data pointers */ | |
c82f823c | 4577 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 4578 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4579 | u32 ioff; |
dd3cc16b | 4580 | item = btrfs_item_nr(i); |
db94535d | 4581 | |
cfed81a0 CM |
4582 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4583 | btrfs_set_token_item_offset(leaf, item, | |
4584 | ioff + size_diff, &token); | |
b18c6685 | 4585 | } |
db94535d | 4586 | |
b18c6685 | 4587 | /* shift the data */ |
179e29e4 | 4588 | if (from_end) { |
3d9ec8c4 NB |
4589 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4590 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4591 | data_end, old_data_start + new_size - data_end); |
4592 | } else { | |
4593 | struct btrfs_disk_key disk_key; | |
4594 | u64 offset; | |
4595 | ||
4596 | btrfs_item_key(leaf, &disk_key, slot); | |
4597 | ||
4598 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4599 | unsigned long ptr; | |
4600 | struct btrfs_file_extent_item *fi; | |
4601 | ||
4602 | fi = btrfs_item_ptr(leaf, slot, | |
4603 | struct btrfs_file_extent_item); | |
4604 | fi = (struct btrfs_file_extent_item *)( | |
4605 | (unsigned long)fi - size_diff); | |
4606 | ||
4607 | if (btrfs_file_extent_type(leaf, fi) == | |
4608 | BTRFS_FILE_EXTENT_INLINE) { | |
4609 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4610 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4611 | (unsigned long)fi, |
7ec20afb | 4612 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4613 | } |
4614 | } | |
4615 | ||
3d9ec8c4 NB |
4616 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4617 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4618 | data_end, old_data_start - data_end); |
4619 | ||
4620 | offset = btrfs_disk_key_offset(&disk_key); | |
4621 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4622 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4623 | if (slot == 0) | |
b167fa91 | 4624 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4625 | } |
5f39d397 | 4626 | |
dd3cc16b | 4627 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4628 | btrfs_set_item_size(leaf, item, new_size); |
4629 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4630 | |
e902baac | 4631 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4632 | btrfs_print_leaf(leaf); |
b18c6685 | 4633 | BUG(); |
5f39d397 | 4634 | } |
b18c6685 CM |
4635 | } |
4636 | ||
d352ac68 | 4637 | /* |
8f69dbd2 | 4638 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4639 | */ |
c71dd880 | 4640 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 4641 | { |
6567e837 | 4642 | int slot; |
5f39d397 CM |
4643 | struct extent_buffer *leaf; |
4644 | struct btrfs_item *item; | |
6567e837 CM |
4645 | u32 nritems; |
4646 | unsigned int data_end; | |
4647 | unsigned int old_data; | |
4648 | unsigned int old_size; | |
4649 | int i; | |
cfed81a0 CM |
4650 | struct btrfs_map_token token; |
4651 | ||
5f39d397 | 4652 | leaf = path->nodes[0]; |
6567e837 | 4653 | |
5f39d397 | 4654 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4655 | data_end = leaf_data_end(leaf); |
6567e837 | 4656 | |
e902baac | 4657 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 4658 | btrfs_print_leaf(leaf); |
6567e837 | 4659 | BUG(); |
5f39d397 | 4660 | } |
6567e837 | 4661 | slot = path->slots[0]; |
5f39d397 | 4662 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4663 | |
4664 | BUG_ON(slot < 0); | |
3326d1b0 | 4665 | if (slot >= nritems) { |
a4f78750 | 4666 | btrfs_print_leaf(leaf); |
c71dd880 | 4667 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 4668 | slot, nritems); |
290342f6 | 4669 | BUG(); |
3326d1b0 | 4670 | } |
6567e837 CM |
4671 | |
4672 | /* | |
4673 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4674 | */ | |
4675 | /* first correct the data pointers */ | |
c82f823c | 4676 | btrfs_init_map_token(&token, leaf); |
6567e837 | 4677 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4678 | u32 ioff; |
dd3cc16b | 4679 | item = btrfs_item_nr(i); |
db94535d | 4680 | |
cfed81a0 CM |
4681 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4682 | btrfs_set_token_item_offset(leaf, item, | |
4683 | ioff - data_size, &token); | |
6567e837 | 4684 | } |
5f39d397 | 4685 | |
6567e837 | 4686 | /* shift the data */ |
3d9ec8c4 NB |
4687 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4688 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4689 | data_end, old_data - data_end); |
5f39d397 | 4690 | |
6567e837 | 4691 | data_end = old_data; |
5f39d397 | 4692 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4693 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4694 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4695 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4696 | |
e902baac | 4697 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4698 | btrfs_print_leaf(leaf); |
6567e837 | 4699 | BUG(); |
5f39d397 | 4700 | } |
6567e837 CM |
4701 | } |
4702 | ||
74123bd7 | 4703 | /* |
44871b1b CM |
4704 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4705 | * to save stack depth by doing the bulk of the work in a function | |
4706 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4707 | */ |
afe5fea7 | 4708 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4709 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4710 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4711 | { |
0b246afa | 4712 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4713 | struct btrfs_item *item; |
9c58309d | 4714 | int i; |
7518a238 | 4715 | u32 nritems; |
be0e5c09 | 4716 | unsigned int data_end; |
e2fa7227 | 4717 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4718 | struct extent_buffer *leaf; |
4719 | int slot; | |
cfed81a0 CM |
4720 | struct btrfs_map_token token; |
4721 | ||
24cdc847 FM |
4722 | if (path->slots[0] == 0) { |
4723 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4724 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4725 | } |
4726 | btrfs_unlock_up_safe(path, 1); | |
4727 | ||
5f39d397 | 4728 | leaf = path->nodes[0]; |
44871b1b | 4729 | slot = path->slots[0]; |
74123bd7 | 4730 | |
5f39d397 | 4731 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4732 | data_end = leaf_data_end(leaf); |
eb60ceac | 4733 | |
e902baac | 4734 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4735 | btrfs_print_leaf(leaf); |
0b246afa | 4736 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4737 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4738 | BUG(); |
d4dbff95 | 4739 | } |
5f39d397 | 4740 | |
c82f823c | 4741 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4742 | if (slot != nritems) { |
5f39d397 | 4743 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4744 | |
5f39d397 | 4745 | if (old_data < data_end) { |
a4f78750 | 4746 | btrfs_print_leaf(leaf); |
0b246afa | 4747 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4748 | slot, old_data, data_end); |
290342f6 | 4749 | BUG(); |
5f39d397 | 4750 | } |
be0e5c09 CM |
4751 | /* |
4752 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4753 | */ | |
4754 | /* first correct the data pointers */ | |
0783fcfc | 4755 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4756 | u32 ioff; |
db94535d | 4757 | |
62e85577 | 4758 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4759 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4760 | btrfs_set_token_item_offset(leaf, item, | |
4761 | ioff - total_data, &token); | |
0783fcfc | 4762 | } |
be0e5c09 | 4763 | /* shift the items */ |
9c58309d | 4764 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4765 | btrfs_item_nr_offset(slot), |
d6025579 | 4766 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4767 | |
4768 | /* shift the data */ | |
3d9ec8c4 NB |
4769 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4770 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4771 | data_end, old_data - data_end); |
be0e5c09 CM |
4772 | data_end = old_data; |
4773 | } | |
5f39d397 | 4774 | |
62e2749e | 4775 | /* setup the item for the new data */ |
9c58309d CM |
4776 | for (i = 0; i < nr; i++) { |
4777 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4778 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4779 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4780 | btrfs_set_token_item_offset(leaf, item, |
4781 | data_end - data_size[i], &token); | |
9c58309d | 4782 | data_end -= data_size[i]; |
cfed81a0 | 4783 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4784 | } |
44871b1b | 4785 | |
9c58309d | 4786 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4787 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4788 | |
e902baac | 4789 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4790 | btrfs_print_leaf(leaf); |
be0e5c09 | 4791 | BUG(); |
5f39d397 | 4792 | } |
44871b1b CM |
4793 | } |
4794 | ||
4795 | /* | |
4796 | * Given a key and some data, insert items into the tree. | |
4797 | * This does all the path init required, making room in the tree if needed. | |
4798 | */ | |
4799 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4800 | struct btrfs_root *root, | |
4801 | struct btrfs_path *path, | |
310712b2 | 4802 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4803 | int nr) |
4804 | { | |
44871b1b CM |
4805 | int ret = 0; |
4806 | int slot; | |
4807 | int i; | |
4808 | u32 total_size = 0; | |
4809 | u32 total_data = 0; | |
4810 | ||
4811 | for (i = 0; i < nr; i++) | |
4812 | total_data += data_size[i]; | |
4813 | ||
4814 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4815 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4816 | if (ret == 0) | |
4817 | return -EEXIST; | |
4818 | if (ret < 0) | |
143bede5 | 4819 | return ret; |
44871b1b | 4820 | |
44871b1b CM |
4821 | slot = path->slots[0]; |
4822 | BUG_ON(slot < 0); | |
4823 | ||
afe5fea7 | 4824 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4825 | total_data, total_size, nr); |
143bede5 | 4826 | return 0; |
62e2749e CM |
4827 | } |
4828 | ||
4829 | /* | |
4830 | * Given a key and some data, insert an item into the tree. | |
4831 | * This does all the path init required, making room in the tree if needed. | |
4832 | */ | |
310712b2 OS |
4833 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4834 | const struct btrfs_key *cpu_key, void *data, | |
4835 | u32 data_size) | |
62e2749e CM |
4836 | { |
4837 | int ret = 0; | |
2c90e5d6 | 4838 | struct btrfs_path *path; |
5f39d397 CM |
4839 | struct extent_buffer *leaf; |
4840 | unsigned long ptr; | |
62e2749e | 4841 | |
2c90e5d6 | 4842 | path = btrfs_alloc_path(); |
db5b493a TI |
4843 | if (!path) |
4844 | return -ENOMEM; | |
2c90e5d6 | 4845 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4846 | if (!ret) { |
5f39d397 CM |
4847 | leaf = path->nodes[0]; |
4848 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4849 | write_extent_buffer(leaf, data, ptr, data_size); | |
4850 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4851 | } |
2c90e5d6 | 4852 | btrfs_free_path(path); |
aa5d6bed | 4853 | return ret; |
be0e5c09 CM |
4854 | } |
4855 | ||
74123bd7 | 4856 | /* |
5de08d7d | 4857 | * delete the pointer from a given node. |
74123bd7 | 4858 | * |
d352ac68 CM |
4859 | * the tree should have been previously balanced so the deletion does not |
4860 | * empty a node. | |
74123bd7 | 4861 | */ |
afe5fea7 TI |
4862 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4863 | int level, int slot) | |
be0e5c09 | 4864 | { |
5f39d397 | 4865 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4866 | u32 nritems; |
f3ea38da | 4867 | int ret; |
be0e5c09 | 4868 | |
5f39d397 | 4869 | nritems = btrfs_header_nritems(parent); |
d397712b | 4870 | if (slot != nritems - 1) { |
bf1d3425 DS |
4871 | if (level) { |
4872 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4873 | nritems - slot - 1); |
bf1d3425 DS |
4874 | BUG_ON(ret < 0); |
4875 | } | |
5f39d397 CM |
4876 | memmove_extent_buffer(parent, |
4877 | btrfs_node_key_ptr_offset(slot), | |
4878 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4879 | sizeof(struct btrfs_key_ptr) * |
4880 | (nritems - slot - 1)); | |
57ba86c0 | 4881 | } else if (level) { |
e09c2efe DS |
4882 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4883 | GFP_NOFS); | |
57ba86c0 | 4884 | BUG_ON(ret < 0); |
bb803951 | 4885 | } |
f3ea38da | 4886 | |
7518a238 | 4887 | nritems--; |
5f39d397 | 4888 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4889 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4890 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4891 | /* just turn the root into a leaf and break */ |
5f39d397 | 4892 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4893 | } else if (slot == 0) { |
5f39d397 CM |
4894 | struct btrfs_disk_key disk_key; |
4895 | ||
4896 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4897 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4898 | } |
d6025579 | 4899 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4900 | } |
4901 | ||
323ac95b CM |
4902 | /* |
4903 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4904 | * path->nodes[1]. |
323ac95b CM |
4905 | * |
4906 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4907 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4908 | * | |
4909 | * The path must have already been setup for deleting the leaf, including | |
4910 | * all the proper balancing. path->nodes[1] must be locked. | |
4911 | */ | |
143bede5 JM |
4912 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4913 | struct btrfs_root *root, | |
4914 | struct btrfs_path *path, | |
4915 | struct extent_buffer *leaf) | |
323ac95b | 4916 | { |
5d4f98a2 | 4917 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4918 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4919 | |
4d081c41 CM |
4920 | /* |
4921 | * btrfs_free_extent is expensive, we want to make sure we | |
4922 | * aren't holding any locks when we call it | |
4923 | */ | |
4924 | btrfs_unlock_up_safe(path, 0); | |
4925 | ||
f0486c68 YZ |
4926 | root_sub_used(root, leaf->len); |
4927 | ||
67439dad | 4928 | atomic_inc(&leaf->refs); |
5581a51a | 4929 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4930 | free_extent_buffer_stale(leaf); |
323ac95b | 4931 | } |
74123bd7 CM |
4932 | /* |
4933 | * delete the item at the leaf level in path. If that empties | |
4934 | * the leaf, remove it from the tree | |
4935 | */ | |
85e21bac CM |
4936 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4937 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4938 | { |
0b246afa | 4939 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4940 | struct extent_buffer *leaf; |
4941 | struct btrfs_item *item; | |
ce0eac2a AM |
4942 | u32 last_off; |
4943 | u32 dsize = 0; | |
aa5d6bed CM |
4944 | int ret = 0; |
4945 | int wret; | |
85e21bac | 4946 | int i; |
7518a238 | 4947 | u32 nritems; |
be0e5c09 | 4948 | |
5f39d397 | 4949 | leaf = path->nodes[0]; |
85e21bac CM |
4950 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4951 | ||
4952 | for (i = 0; i < nr; i++) | |
4953 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4954 | ||
5f39d397 | 4955 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4956 | |
85e21bac | 4957 | if (slot + nr != nritems) { |
8f881e8c | 4958 | int data_end = leaf_data_end(leaf); |
c82f823c | 4959 | struct btrfs_map_token token; |
5f39d397 | 4960 | |
3d9ec8c4 | 4961 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4962 | data_end + dsize, |
3d9ec8c4 | 4963 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4964 | last_off - data_end); |
5f39d397 | 4965 | |
c82f823c | 4966 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4967 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4968 | u32 ioff; |
db94535d | 4969 | |
dd3cc16b | 4970 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4971 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4972 | btrfs_set_token_item_offset(leaf, item, | |
4973 | ioff + dsize, &token); | |
0783fcfc | 4974 | } |
db94535d | 4975 | |
5f39d397 | 4976 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4977 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4978 | sizeof(struct btrfs_item) * |
85e21bac | 4979 | (nritems - slot - nr)); |
be0e5c09 | 4980 | } |
85e21bac CM |
4981 | btrfs_set_header_nritems(leaf, nritems - nr); |
4982 | nritems -= nr; | |
5f39d397 | 4983 | |
74123bd7 | 4984 | /* delete the leaf if we've emptied it */ |
7518a238 | 4985 | if (nritems == 0) { |
5f39d397 CM |
4986 | if (leaf == root->node) { |
4987 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4988 | } else { |
f0486c68 | 4989 | btrfs_set_path_blocking(path); |
6a884d7d | 4990 | btrfs_clean_tree_block(leaf); |
143bede5 | 4991 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4992 | } |
be0e5c09 | 4993 | } else { |
7518a238 | 4994 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4995 | if (slot == 0) { |
5f39d397 CM |
4996 | struct btrfs_disk_key disk_key; |
4997 | ||
4998 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4999 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 5000 | } |
aa5d6bed | 5001 | |
74123bd7 | 5002 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5003 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5004 | /* push_leaf_left fixes the path. |
5005 | * make sure the path still points to our leaf | |
5006 | * for possible call to del_ptr below | |
5007 | */ | |
4920c9ac | 5008 | slot = path->slots[1]; |
67439dad | 5009 | atomic_inc(&leaf->refs); |
5f39d397 | 5010 | |
b9473439 | 5011 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5012 | wret = push_leaf_left(trans, root, path, 1, 1, |
5013 | 1, (u32)-1); | |
54aa1f4d | 5014 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5015 | ret = wret; |
5f39d397 CM |
5016 | |
5017 | if (path->nodes[0] == leaf && | |
5018 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5019 | wret = push_leaf_right(trans, root, path, 1, |
5020 | 1, 1, 0); | |
54aa1f4d | 5021 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5022 | ret = wret; |
5023 | } | |
5f39d397 CM |
5024 | |
5025 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5026 | path->slots[1] = slot; |
143bede5 | 5027 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5028 | free_extent_buffer(leaf); |
143bede5 | 5029 | ret = 0; |
5de08d7d | 5030 | } else { |
925baedd CM |
5031 | /* if we're still in the path, make sure |
5032 | * we're dirty. Otherwise, one of the | |
5033 | * push_leaf functions must have already | |
5034 | * dirtied this buffer | |
5035 | */ | |
5036 | if (path->nodes[0] == leaf) | |
5037 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5038 | free_extent_buffer(leaf); |
be0e5c09 | 5039 | } |
d5719762 | 5040 | } else { |
5f39d397 | 5041 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5042 | } |
5043 | } | |
aa5d6bed | 5044 | return ret; |
be0e5c09 CM |
5045 | } |
5046 | ||
7bb86316 | 5047 | /* |
925baedd | 5048 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5049 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5050 | * returns < 0 on io errors. | |
d352ac68 CM |
5051 | * |
5052 | * This may release the path, and so you may lose any locks held at the | |
5053 | * time you call it. | |
7bb86316 | 5054 | */ |
16e7549f | 5055 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5056 | { |
925baedd CM |
5057 | struct btrfs_key key; |
5058 | struct btrfs_disk_key found_key; | |
5059 | int ret; | |
7bb86316 | 5060 | |
925baedd | 5061 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5062 | |
e8b0d724 | 5063 | if (key.offset > 0) { |
925baedd | 5064 | key.offset--; |
e8b0d724 | 5065 | } else if (key.type > 0) { |
925baedd | 5066 | key.type--; |
e8b0d724 FDBM |
5067 | key.offset = (u64)-1; |
5068 | } else if (key.objectid > 0) { | |
925baedd | 5069 | key.objectid--; |
e8b0d724 FDBM |
5070 | key.type = (u8)-1; |
5071 | key.offset = (u64)-1; | |
5072 | } else { | |
925baedd | 5073 | return 1; |
e8b0d724 | 5074 | } |
7bb86316 | 5075 | |
b3b4aa74 | 5076 | btrfs_release_path(path); |
925baedd CM |
5077 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5078 | if (ret < 0) | |
5079 | return ret; | |
5080 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5081 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5082 | /* |
5083 | * We might have had an item with the previous key in the tree right | |
5084 | * before we released our path. And after we released our path, that | |
5085 | * item might have been pushed to the first slot (0) of the leaf we | |
5086 | * were holding due to a tree balance. Alternatively, an item with the | |
5087 | * previous key can exist as the only element of a leaf (big fat item). | |
5088 | * Therefore account for these 2 cases, so that our callers (like | |
5089 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5090 | * the previous key we computed above. | |
5091 | */ | |
5092 | if (ret <= 0) | |
925baedd CM |
5093 | return 0; |
5094 | return 1; | |
7bb86316 CM |
5095 | } |
5096 | ||
3f157a2f CM |
5097 | /* |
5098 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5099 | * for nodes or leaves that are have a minimum transaction id. |
5100 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5101 | * |
5102 | * This does not cow, but it does stuff the starting key it finds back | |
5103 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5104 | * key and get a writable path. | |
5105 | * | |
3f157a2f CM |
5106 | * This honors path->lowest_level to prevent descent past a given level |
5107 | * of the tree. | |
5108 | * | |
d352ac68 CM |
5109 | * min_trans indicates the oldest transaction that you are interested |
5110 | * in walking through. Any nodes or leaves older than min_trans are | |
5111 | * skipped over (without reading them). | |
5112 | * | |
3f157a2f CM |
5113 | * returns zero if something useful was found, < 0 on error and 1 if there |
5114 | * was nothing in the tree that matched the search criteria. | |
5115 | */ | |
5116 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5117 | struct btrfs_path *path, |
3f157a2f CM |
5118 | u64 min_trans) |
5119 | { | |
5120 | struct extent_buffer *cur; | |
5121 | struct btrfs_key found_key; | |
5122 | int slot; | |
9652480b | 5123 | int sret; |
3f157a2f CM |
5124 | u32 nritems; |
5125 | int level; | |
5126 | int ret = 1; | |
f98de9b9 | 5127 | int keep_locks = path->keep_locks; |
3f157a2f | 5128 | |
f98de9b9 | 5129 | path->keep_locks = 1; |
3f157a2f | 5130 | again: |
bd681513 | 5131 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5132 | level = btrfs_header_level(cur); |
e02119d5 | 5133 | WARN_ON(path->nodes[level]); |
3f157a2f | 5134 | path->nodes[level] = cur; |
bd681513 | 5135 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5136 | |
5137 | if (btrfs_header_generation(cur) < min_trans) { | |
5138 | ret = 1; | |
5139 | goto out; | |
5140 | } | |
d397712b | 5141 | while (1) { |
3f157a2f CM |
5142 | nritems = btrfs_header_nritems(cur); |
5143 | level = btrfs_header_level(cur); | |
a74b35ec | 5144 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
cbca7d59 FM |
5145 | if (sret < 0) { |
5146 | ret = sret; | |
5147 | goto out; | |
5148 | } | |
3f157a2f | 5149 | |
323ac95b CM |
5150 | /* at the lowest level, we're done, setup the path and exit */ |
5151 | if (level == path->lowest_level) { | |
e02119d5 CM |
5152 | if (slot >= nritems) |
5153 | goto find_next_key; | |
3f157a2f CM |
5154 | ret = 0; |
5155 | path->slots[level] = slot; | |
5156 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5157 | goto out; | |
5158 | } | |
9652480b Y |
5159 | if (sret && slot > 0) |
5160 | slot--; | |
3f157a2f | 5161 | /* |
de78b51a ES |
5162 | * check this node pointer against the min_trans parameters. |
5163 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5164 | */ |
d397712b | 5165 | while (slot < nritems) { |
3f157a2f | 5166 | u64 gen; |
e02119d5 | 5167 | |
3f157a2f CM |
5168 | gen = btrfs_node_ptr_generation(cur, slot); |
5169 | if (gen < min_trans) { | |
5170 | slot++; | |
5171 | continue; | |
5172 | } | |
de78b51a | 5173 | break; |
3f157a2f | 5174 | } |
e02119d5 | 5175 | find_next_key: |
3f157a2f CM |
5176 | /* |
5177 | * we didn't find a candidate key in this node, walk forward | |
5178 | * and find another one | |
5179 | */ | |
5180 | if (slot >= nritems) { | |
e02119d5 | 5181 | path->slots[level] = slot; |
b4ce94de | 5182 | btrfs_set_path_blocking(path); |
e02119d5 | 5183 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5184 | min_trans); |
e02119d5 | 5185 | if (sret == 0) { |
b3b4aa74 | 5186 | btrfs_release_path(path); |
3f157a2f CM |
5187 | goto again; |
5188 | } else { | |
5189 | goto out; | |
5190 | } | |
5191 | } | |
5192 | /* save our key for returning back */ | |
5193 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5194 | path->slots[level] = slot; | |
5195 | if (level == path->lowest_level) { | |
5196 | ret = 0; | |
3f157a2f CM |
5197 | goto out; |
5198 | } | |
b4ce94de | 5199 | btrfs_set_path_blocking(path); |
4b231ae4 | 5200 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
5201 | if (IS_ERR(cur)) { |
5202 | ret = PTR_ERR(cur); | |
5203 | goto out; | |
5204 | } | |
3f157a2f | 5205 | |
bd681513 | 5206 | btrfs_tree_read_lock(cur); |
b4ce94de | 5207 | |
bd681513 | 5208 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5209 | path->nodes[level - 1] = cur; |
f7c79f30 | 5210 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5211 | } |
5212 | out: | |
f98de9b9 FM |
5213 | path->keep_locks = keep_locks; |
5214 | if (ret == 0) { | |
5215 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5216 | btrfs_set_path_blocking(path); | |
3f157a2f | 5217 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5218 | } |
3f157a2f CM |
5219 | return ret; |
5220 | } | |
5221 | ||
5222 | /* | |
5223 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5224 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5225 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5226 | * |
5227 | * 0 is returned if another key is found, < 0 if there are any errors | |
5228 | * and 1 is returned if there are no higher keys in the tree | |
5229 | * | |
5230 | * path->keep_locks should be set to 1 on the search made before | |
5231 | * calling this function. | |
5232 | */ | |
e7a84565 | 5233 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5234 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5235 | { |
e7a84565 CM |
5236 | int slot; |
5237 | struct extent_buffer *c; | |
5238 | ||
6a9fb468 | 5239 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 5240 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5241 | if (!path->nodes[level]) |
5242 | return 1; | |
5243 | ||
5244 | slot = path->slots[level] + 1; | |
5245 | c = path->nodes[level]; | |
3f157a2f | 5246 | next: |
e7a84565 | 5247 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5248 | int ret; |
5249 | int orig_lowest; | |
5250 | struct btrfs_key cur_key; | |
5251 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5252 | !path->nodes[level + 1]) | |
e7a84565 | 5253 | return 1; |
33c66f43 | 5254 | |
6a9fb468 | 5255 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
5256 | level++; |
5257 | continue; | |
5258 | } | |
5259 | ||
5260 | slot = btrfs_header_nritems(c) - 1; | |
5261 | if (level == 0) | |
5262 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5263 | else | |
5264 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5265 | ||
5266 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5267 | btrfs_release_path(path); |
33c66f43 YZ |
5268 | path->lowest_level = level; |
5269 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5270 | 0, 0); | |
5271 | path->lowest_level = orig_lowest; | |
5272 | if (ret < 0) | |
5273 | return ret; | |
5274 | ||
5275 | c = path->nodes[level]; | |
5276 | slot = path->slots[level]; | |
5277 | if (ret == 0) | |
5278 | slot++; | |
5279 | goto next; | |
e7a84565 | 5280 | } |
33c66f43 | 5281 | |
e7a84565 CM |
5282 | if (level == 0) |
5283 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5284 | else { |
3f157a2f CM |
5285 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5286 | ||
3f157a2f CM |
5287 | if (gen < min_trans) { |
5288 | slot++; | |
5289 | goto next; | |
5290 | } | |
e7a84565 | 5291 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5292 | } |
e7a84565 CM |
5293 | return 0; |
5294 | } | |
5295 | return 1; | |
5296 | } | |
5297 | ||
97571fd0 | 5298 | /* |
925baedd | 5299 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5300 | * returns 0 if it found something or 1 if there are no greater leaves. |
5301 | * returns < 0 on io errors. | |
97571fd0 | 5302 | */ |
234b63a0 | 5303 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5304 | { |
5305 | return btrfs_next_old_leaf(root, path, 0); | |
5306 | } | |
5307 | ||
5308 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5309 | u64 time_seq) | |
d97e63b6 CM |
5310 | { |
5311 | int slot; | |
8e73f275 | 5312 | int level; |
5f39d397 | 5313 | struct extent_buffer *c; |
8e73f275 | 5314 | struct extent_buffer *next; |
925baedd CM |
5315 | struct btrfs_key key; |
5316 | u32 nritems; | |
5317 | int ret; | |
8e73f275 | 5318 | int old_spinning = path->leave_spinning; |
bd681513 | 5319 | int next_rw_lock = 0; |
925baedd CM |
5320 | |
5321 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5322 | if (nritems == 0) |
925baedd | 5323 | return 1; |
925baedd | 5324 | |
8e73f275 CM |
5325 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5326 | again: | |
5327 | level = 1; | |
5328 | next = NULL; | |
bd681513 | 5329 | next_rw_lock = 0; |
b3b4aa74 | 5330 | btrfs_release_path(path); |
8e73f275 | 5331 | |
a2135011 | 5332 | path->keep_locks = 1; |
31533fb2 | 5333 | path->leave_spinning = 1; |
8e73f275 | 5334 | |
3d7806ec JS |
5335 | if (time_seq) |
5336 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5337 | else | |
5338 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5339 | path->keep_locks = 0; |
5340 | ||
5341 | if (ret < 0) | |
5342 | return ret; | |
5343 | ||
a2135011 | 5344 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5345 | /* |
5346 | * by releasing the path above we dropped all our locks. A balance | |
5347 | * could have added more items next to the key that used to be | |
5348 | * at the very end of the block. So, check again here and | |
5349 | * advance the path if there are now more items available. | |
5350 | */ | |
a2135011 | 5351 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5352 | if (ret == 0) |
5353 | path->slots[0]++; | |
8e73f275 | 5354 | ret = 0; |
925baedd CM |
5355 | goto done; |
5356 | } | |
0b43e04f LB |
5357 | /* |
5358 | * So the above check misses one case: | |
5359 | * - after releasing the path above, someone has removed the item that | |
5360 | * used to be at the very end of the block, and balance between leafs | |
5361 | * gets another one with bigger key.offset to replace it. | |
5362 | * | |
5363 | * This one should be returned as well, or we can get leaf corruption | |
5364 | * later(esp. in __btrfs_drop_extents()). | |
5365 | * | |
5366 | * And a bit more explanation about this check, | |
5367 | * with ret > 0, the key isn't found, the path points to the slot | |
5368 | * where it should be inserted, so the path->slots[0] item must be the | |
5369 | * bigger one. | |
5370 | */ | |
5371 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5372 | ret = 0; | |
5373 | goto done; | |
5374 | } | |
d97e63b6 | 5375 | |
d397712b | 5376 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5377 | if (!path->nodes[level]) { |
5378 | ret = 1; | |
5379 | goto done; | |
5380 | } | |
5f39d397 | 5381 | |
d97e63b6 CM |
5382 | slot = path->slots[level] + 1; |
5383 | c = path->nodes[level]; | |
5f39d397 | 5384 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5385 | level++; |
8e73f275 CM |
5386 | if (level == BTRFS_MAX_LEVEL) { |
5387 | ret = 1; | |
5388 | goto done; | |
5389 | } | |
d97e63b6 CM |
5390 | continue; |
5391 | } | |
5f39d397 | 5392 | |
925baedd | 5393 | if (next) { |
bd681513 | 5394 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5395 | free_extent_buffer(next); |
925baedd | 5396 | } |
5f39d397 | 5397 | |
8e73f275 | 5398 | next = c; |
bd681513 | 5399 | next_rw_lock = path->locks[level]; |
d07b8528 | 5400 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5401 | slot, &key); |
8e73f275 CM |
5402 | if (ret == -EAGAIN) |
5403 | goto again; | |
5f39d397 | 5404 | |
76a05b35 | 5405 | if (ret < 0) { |
b3b4aa74 | 5406 | btrfs_release_path(path); |
76a05b35 CM |
5407 | goto done; |
5408 | } | |
5409 | ||
5cd57b2c | 5410 | if (!path->skip_locking) { |
bd681513 | 5411 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5412 | if (!ret && time_seq) { |
5413 | /* | |
5414 | * If we don't get the lock, we may be racing | |
5415 | * with push_leaf_left, holding that lock while | |
5416 | * itself waiting for the leaf we've currently | |
5417 | * locked. To solve this situation, we give up | |
5418 | * on our lock and cycle. | |
5419 | */ | |
cf538830 | 5420 | free_extent_buffer(next); |
d42244a0 JS |
5421 | btrfs_release_path(path); |
5422 | cond_resched(); | |
5423 | goto again; | |
5424 | } | |
8e73f275 CM |
5425 | if (!ret) { |
5426 | btrfs_set_path_blocking(path); | |
bd681513 | 5427 | btrfs_tree_read_lock(next); |
8e73f275 | 5428 | } |
31533fb2 | 5429 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5430 | } |
d97e63b6 CM |
5431 | break; |
5432 | } | |
5433 | path->slots[level] = slot; | |
d397712b | 5434 | while (1) { |
d97e63b6 CM |
5435 | level--; |
5436 | c = path->nodes[level]; | |
925baedd | 5437 | if (path->locks[level]) |
bd681513 | 5438 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5439 | |
5f39d397 | 5440 | free_extent_buffer(c); |
d97e63b6 CM |
5441 | path->nodes[level] = next; |
5442 | path->slots[level] = 0; | |
a74a4b97 | 5443 | if (!path->skip_locking) |
bd681513 | 5444 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5445 | if (!level) |
5446 | break; | |
b4ce94de | 5447 | |
d07b8528 | 5448 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5449 | 0, &key); |
8e73f275 CM |
5450 | if (ret == -EAGAIN) |
5451 | goto again; | |
5452 | ||
76a05b35 | 5453 | if (ret < 0) { |
b3b4aa74 | 5454 | btrfs_release_path(path); |
76a05b35 CM |
5455 | goto done; |
5456 | } | |
5457 | ||
5cd57b2c | 5458 | if (!path->skip_locking) { |
bd681513 | 5459 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5460 | if (!ret) { |
5461 | btrfs_set_path_blocking(path); | |
bd681513 | 5462 | btrfs_tree_read_lock(next); |
bd681513 | 5463 | } |
31533fb2 | 5464 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5465 | } |
d97e63b6 | 5466 | } |
8e73f275 | 5467 | ret = 0; |
925baedd | 5468 | done: |
f7c79f30 | 5469 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5470 | path->leave_spinning = old_spinning; |
5471 | if (!old_spinning) | |
5472 | btrfs_set_path_blocking(path); | |
5473 | ||
5474 | return ret; | |
d97e63b6 | 5475 | } |
0b86a832 | 5476 | |
3f157a2f CM |
5477 | /* |
5478 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5479 | * searching until it gets past min_objectid or finds an item of 'type' | |
5480 | * | |
5481 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5482 | */ | |
0b86a832 CM |
5483 | int btrfs_previous_item(struct btrfs_root *root, |
5484 | struct btrfs_path *path, u64 min_objectid, | |
5485 | int type) | |
5486 | { | |
5487 | struct btrfs_key found_key; | |
5488 | struct extent_buffer *leaf; | |
e02119d5 | 5489 | u32 nritems; |
0b86a832 CM |
5490 | int ret; |
5491 | ||
d397712b | 5492 | while (1) { |
0b86a832 | 5493 | if (path->slots[0] == 0) { |
b4ce94de | 5494 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5495 | ret = btrfs_prev_leaf(root, path); |
5496 | if (ret != 0) | |
5497 | return ret; | |
5498 | } else { | |
5499 | path->slots[0]--; | |
5500 | } | |
5501 | leaf = path->nodes[0]; | |
e02119d5 CM |
5502 | nritems = btrfs_header_nritems(leaf); |
5503 | if (nritems == 0) | |
5504 | return 1; | |
5505 | if (path->slots[0] == nritems) | |
5506 | path->slots[0]--; | |
5507 | ||
0b86a832 | 5508 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5509 | if (found_key.objectid < min_objectid) |
5510 | break; | |
0a4eefbb YZ |
5511 | if (found_key.type == type) |
5512 | return 0; | |
e02119d5 CM |
5513 | if (found_key.objectid == min_objectid && |
5514 | found_key.type < type) | |
5515 | break; | |
0b86a832 CM |
5516 | } |
5517 | return 1; | |
5518 | } | |
ade2e0b3 WS |
5519 | |
5520 | /* | |
5521 | * search in extent tree to find a previous Metadata/Data extent item with | |
5522 | * min objecitd. | |
5523 | * | |
5524 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5525 | */ | |
5526 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5527 | struct btrfs_path *path, u64 min_objectid) | |
5528 | { | |
5529 | struct btrfs_key found_key; | |
5530 | struct extent_buffer *leaf; | |
5531 | u32 nritems; | |
5532 | int ret; | |
5533 | ||
5534 | while (1) { | |
5535 | if (path->slots[0] == 0) { | |
5536 | btrfs_set_path_blocking(path); | |
5537 | ret = btrfs_prev_leaf(root, path); | |
5538 | if (ret != 0) | |
5539 | return ret; | |
5540 | } else { | |
5541 | path->slots[0]--; | |
5542 | } | |
5543 | leaf = path->nodes[0]; | |
5544 | nritems = btrfs_header_nritems(leaf); | |
5545 | if (nritems == 0) | |
5546 | return 1; | |
5547 | if (path->slots[0] == nritems) | |
5548 | path->slots[0]--; | |
5549 | ||
5550 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5551 | if (found_key.objectid < min_objectid) | |
5552 | break; | |
5553 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5554 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5555 | return 0; | |
5556 | if (found_key.objectid == min_objectid && | |
5557 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5558 | break; | |
5559 | } | |
5560 | return 1; | |
5561 | } |