]>
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; | |
59a0fcdb DS |
34 | const char name[10]; |
35 | const char driver[12]; | |
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 */ | |
59a0fcdb DS |
66 | return btrfs_csums[csum_type].driver[0] ? |
67 | btrfs_csums[csum_type].driver : | |
b4e967be DS |
68 | btrfs_csums[csum_type].name; |
69 | } | |
70 | ||
604997b4 | 71 | size_t __attribute_const__ btrfs_get_num_csums(void) |
f7cea56c DS |
72 | { |
73 | return ARRAY_SIZE(btrfs_csums); | |
74 | } | |
75 | ||
df24a2b9 | 76 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 77 | { |
e2c89907 | 78 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
79 | } |
80 | ||
d352ac68 | 81 | /* this also releases the path */ |
df24a2b9 | 82 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 83 | { |
ff175d57 JJ |
84 | if (!p) |
85 | return; | |
b3b4aa74 | 86 | btrfs_release_path(p); |
df24a2b9 | 87 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
88 | } |
89 | ||
d352ac68 CM |
90 | /* |
91 | * path release drops references on the extent buffers in the path | |
92 | * and it drops any locks held by this path | |
93 | * | |
94 | * It is safe to call this on paths that no locks or extent buffers held. | |
95 | */ | |
b3b4aa74 | 96 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
97 | { |
98 | int i; | |
a2135011 | 99 | |
234b63a0 | 100 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 101 | p->slots[i] = 0; |
eb60ceac | 102 | if (!p->nodes[i]) |
925baedd CM |
103 | continue; |
104 | if (p->locks[i]) { | |
bd681513 | 105 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
106 | p->locks[i] = 0; |
107 | } | |
5f39d397 | 108 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 109 | p->nodes[i] = NULL; |
eb60ceac CM |
110 | } |
111 | } | |
112 | ||
d352ac68 CM |
113 | /* |
114 | * safely gets a reference on the root node of a tree. A lock | |
115 | * is not taken, so a concurrent writer may put a different node | |
116 | * at the root of the tree. See btrfs_lock_root_node for the | |
117 | * looping required. | |
118 | * | |
119 | * The extent buffer returned by this has a reference taken, so | |
120 | * it won't disappear. It may stop being the root of the tree | |
121 | * at any time because there are no locks held. | |
122 | */ | |
925baedd CM |
123 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
124 | { | |
125 | struct extent_buffer *eb; | |
240f62c8 | 126 | |
3083ee2e JB |
127 | while (1) { |
128 | rcu_read_lock(); | |
129 | eb = rcu_dereference(root->node); | |
130 | ||
131 | /* | |
132 | * RCU really hurts here, we could free up the root node because | |
01327610 | 133 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
134 | * the inc_not_zero dance and if it doesn't work then |
135 | * synchronize_rcu and try again. | |
136 | */ | |
137 | if (atomic_inc_not_zero(&eb->refs)) { | |
138 | rcu_read_unlock(); | |
139 | break; | |
140 | } | |
141 | rcu_read_unlock(); | |
142 | synchronize_rcu(); | |
143 | } | |
925baedd CM |
144 | return eb; |
145 | } | |
146 | ||
92a7cc42 QW |
147 | /* |
148 | * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), | |
149 | * just get put onto a simple dirty list. Transaction walks this list to make | |
150 | * sure they get properly updated on disk. | |
d352ac68 | 151 | */ |
0b86a832 CM |
152 | static void add_root_to_dirty_list(struct btrfs_root *root) |
153 | { | |
0b246afa JM |
154 | struct btrfs_fs_info *fs_info = root->fs_info; |
155 | ||
e7070be1 JB |
156 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
157 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
158 | return; | |
159 | ||
0b246afa | 160 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
161 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
162 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 163 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 164 | list_move_tail(&root->dirty_list, |
0b246afa | 165 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
166 | else |
167 | list_move(&root->dirty_list, | |
0b246afa | 168 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 169 | } |
0b246afa | 170 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
171 | } |
172 | ||
d352ac68 CM |
173 | /* |
174 | * used by snapshot creation to make a copy of a root for a tree with | |
175 | * a given objectid. The buffer with the new root node is returned in | |
176 | * cow_ret, and this func returns zero on success or a negative error code. | |
177 | */ | |
be20aa9d CM |
178 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
179 | struct btrfs_root *root, | |
180 | struct extent_buffer *buf, | |
181 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
182 | { | |
0b246afa | 183 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 184 | struct extent_buffer *cow; |
be20aa9d CM |
185 | int ret = 0; |
186 | int level; | |
5d4f98a2 | 187 | struct btrfs_disk_key disk_key; |
be20aa9d | 188 | |
92a7cc42 | 189 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 190 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 191 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 192 | trans->transid != root->last_trans); |
be20aa9d CM |
193 | |
194 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
195 | if (level == 0) |
196 | btrfs_item_key(buf, &disk_key, 0); | |
197 | else | |
198 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 199 | |
4d75f8a9 | 200 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
cf6f34aa JB |
201 | &disk_key, level, buf->start, 0, |
202 | BTRFS_NESTING_NEW_ROOT); | |
5d4f98a2 | 203 | if (IS_ERR(cow)) |
be20aa9d CM |
204 | return PTR_ERR(cow); |
205 | ||
58e8012c | 206 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
207 | btrfs_set_header_bytenr(cow, cow->start); |
208 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
209 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
210 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
211 | BTRFS_HEADER_FLAG_RELOC); | |
212 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
213 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
214 | else | |
215 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 216 | |
de37aa51 | 217 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 218 | |
be20aa9d | 219 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 220 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 221 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 222 | else |
e339a6b0 | 223 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 224 | |
be20aa9d CM |
225 | if (ret) |
226 | return ret; | |
227 | ||
228 | btrfs_mark_buffer_dirty(cow); | |
229 | *cow_ret = cow; | |
230 | return 0; | |
231 | } | |
232 | ||
bd989ba3 JS |
233 | enum mod_log_op { |
234 | MOD_LOG_KEY_REPLACE, | |
235 | MOD_LOG_KEY_ADD, | |
236 | MOD_LOG_KEY_REMOVE, | |
237 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
238 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
239 | MOD_LOG_MOVE_KEYS, | |
240 | MOD_LOG_ROOT_REPLACE, | |
241 | }; | |
242 | ||
bd989ba3 JS |
243 | struct tree_mod_root { |
244 | u64 logical; | |
245 | u8 level; | |
246 | }; | |
247 | ||
248 | struct tree_mod_elem { | |
249 | struct rb_node node; | |
298cfd36 | 250 | u64 logical; |
097b8a7c | 251 | u64 seq; |
bd989ba3 JS |
252 | enum mod_log_op op; |
253 | ||
254 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
255 | int slot; | |
256 | ||
257 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
258 | u64 generation; | |
259 | ||
260 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
261 | struct btrfs_disk_key key; | |
262 | u64 blockptr; | |
263 | ||
264 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
265 | struct { |
266 | int dst_slot; | |
267 | int nr_items; | |
268 | } move; | |
bd989ba3 JS |
269 | |
270 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
271 | struct tree_mod_root old_root; | |
272 | }; | |
273 | ||
fc36ed7e | 274 | /* |
fcebe456 | 275 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 276 | */ |
fcebe456 | 277 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
278 | { |
279 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
280 | } | |
281 | ||
097b8a7c JS |
282 | /* |
283 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
284 | * passed does not already have a sequence number set. So when a caller expects | |
285 | * to record tree modifications, it should ensure to set elem->seq to zero | |
286 | * before calling btrfs_get_tree_mod_seq. | |
287 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
288 | * blocker was added. | |
289 | */ | |
290 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
291 | struct seq_list *elem) | |
bd989ba3 | 292 | { |
b1a09f1e | 293 | write_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 294 | if (!elem->seq) { |
fcebe456 | 295 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
296 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
297 | } | |
b1a09f1e | 298 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 299 | |
fcebe456 | 300 | return elem->seq; |
bd989ba3 JS |
301 | } |
302 | ||
303 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
304 | struct seq_list *elem) | |
305 | { | |
306 | struct rb_root *tm_root; | |
307 | struct rb_node *node; | |
308 | struct rb_node *next; | |
bd989ba3 JS |
309 | struct tree_mod_elem *tm; |
310 | u64 min_seq = (u64)-1; | |
311 | u64 seq_putting = elem->seq; | |
312 | ||
313 | if (!seq_putting) | |
314 | return; | |
315 | ||
7227ff4d | 316 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 317 | list_del(&elem->list); |
097b8a7c | 318 | elem->seq = 0; |
bd989ba3 | 319 | |
42836cf4 FM |
320 | if (!list_empty(&fs_info->tree_mod_seq_list)) { |
321 | struct seq_list *first; | |
322 | ||
323 | first = list_first_entry(&fs_info->tree_mod_seq_list, | |
324 | struct seq_list, list); | |
325 | if (seq_putting > first->seq) { | |
326 | /* | |
327 | * Blocker with lower sequence number exists, we | |
328 | * cannot remove anything from the log. | |
329 | */ | |
330 | write_unlock(&fs_info->tree_mod_log_lock); | |
331 | return; | |
bd989ba3 | 332 | } |
42836cf4 | 333 | min_seq = first->seq; |
bd989ba3 | 334 | } |
097b8a7c | 335 | |
bd989ba3 JS |
336 | /* |
337 | * anything that's lower than the lowest existing (read: blocked) | |
338 | * sequence number can be removed from the tree. | |
339 | */ | |
bd989ba3 JS |
340 | tm_root = &fs_info->tree_mod_log; |
341 | for (node = rb_first(tm_root); node; node = next) { | |
342 | next = rb_next(node); | |
6b4df8b6 | 343 | tm = rb_entry(node, struct tree_mod_elem, node); |
6609fee8 | 344 | if (tm->seq >= min_seq) |
bd989ba3 JS |
345 | continue; |
346 | rb_erase(node, tm_root); | |
bd989ba3 JS |
347 | kfree(tm); |
348 | } | |
b1a09f1e | 349 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
350 | } |
351 | ||
352 | /* | |
353 | * key order of the log: | |
298cfd36 | 354 | * node/leaf start address -> sequence |
bd989ba3 | 355 | * |
298cfd36 CR |
356 | * The 'start address' is the logical address of the *new* root node |
357 | * for root replace operations, or the logical address of the affected | |
358 | * block for all other operations. | |
bd989ba3 JS |
359 | */ |
360 | static noinline int | |
361 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
362 | { | |
363 | struct rb_root *tm_root; | |
364 | struct rb_node **new; | |
365 | struct rb_node *parent = NULL; | |
366 | struct tree_mod_elem *cur; | |
c8cc6341 | 367 | |
73e82fe4 DS |
368 | lockdep_assert_held_write(&fs_info->tree_mod_log_lock); |
369 | ||
fcebe456 | 370 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 371 | |
bd989ba3 JS |
372 | tm_root = &fs_info->tree_mod_log; |
373 | new = &tm_root->rb_node; | |
374 | while (*new) { | |
6b4df8b6 | 375 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 376 | parent = *new; |
298cfd36 | 377 | if (cur->logical < tm->logical) |
bd989ba3 | 378 | new = &((*new)->rb_left); |
298cfd36 | 379 | else if (cur->logical > tm->logical) |
bd989ba3 | 380 | new = &((*new)->rb_right); |
097b8a7c | 381 | else if (cur->seq < tm->seq) |
bd989ba3 | 382 | new = &((*new)->rb_left); |
097b8a7c | 383 | else if (cur->seq > tm->seq) |
bd989ba3 | 384 | new = &((*new)->rb_right); |
5de865ee FDBM |
385 | else |
386 | return -EEXIST; | |
bd989ba3 JS |
387 | } |
388 | ||
389 | rb_link_node(&tm->node, parent, new); | |
390 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 391 | return 0; |
bd989ba3 JS |
392 | } |
393 | ||
097b8a7c JS |
394 | /* |
395 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
396 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
397 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 398 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 399 | */ |
e9b7fd4d JS |
400 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
401 | struct extent_buffer *eb) { | |
402 | smp_mb(); | |
403 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
404 | return 1; | |
097b8a7c JS |
405 | if (eb && btrfs_header_level(eb) == 0) |
406 | return 1; | |
5de865ee | 407 | |
b1a09f1e | 408 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 409 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 410 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
411 | return 1; |
412 | } | |
413 | ||
e9b7fd4d JS |
414 | return 0; |
415 | } | |
416 | ||
5de865ee FDBM |
417 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
418 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
419 | struct extent_buffer *eb) | |
420 | { | |
421 | smp_mb(); | |
422 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
423 | return 0; | |
424 | if (eb && btrfs_header_level(eb) == 0) | |
425 | return 0; | |
426 | ||
427 | return 1; | |
428 | } | |
429 | ||
430 | static struct tree_mod_elem * | |
431 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
432 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 433 | { |
097b8a7c | 434 | struct tree_mod_elem *tm; |
bd989ba3 | 435 | |
c8cc6341 JB |
436 | tm = kzalloc(sizeof(*tm), flags); |
437 | if (!tm) | |
5de865ee | 438 | return NULL; |
bd989ba3 | 439 | |
298cfd36 | 440 | tm->logical = eb->start; |
bd989ba3 JS |
441 | if (op != MOD_LOG_KEY_ADD) { |
442 | btrfs_node_key(eb, &tm->key, slot); | |
443 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
444 | } | |
445 | tm->op = op; | |
446 | tm->slot = slot; | |
447 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 448 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 449 | |
5de865ee | 450 | return tm; |
097b8a7c JS |
451 | } |
452 | ||
e09c2efe DS |
453 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
454 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 455 | { |
5de865ee FDBM |
456 | struct tree_mod_elem *tm; |
457 | int ret; | |
458 | ||
e09c2efe | 459 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
460 | return 0; |
461 | ||
462 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
463 | if (!tm) | |
464 | return -ENOMEM; | |
465 | ||
e09c2efe | 466 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 467 | kfree(tm); |
097b8a7c | 468 | return 0; |
5de865ee FDBM |
469 | } |
470 | ||
e09c2efe | 471 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 472 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
473 | if (ret) |
474 | kfree(tm); | |
097b8a7c | 475 | |
5de865ee | 476 | return ret; |
097b8a7c JS |
477 | } |
478 | ||
6074d45f DS |
479 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
480 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 481 | { |
5de865ee FDBM |
482 | struct tree_mod_elem *tm = NULL; |
483 | struct tree_mod_elem **tm_list = NULL; | |
484 | int ret = 0; | |
bd989ba3 | 485 | int i; |
5de865ee | 486 | int locked = 0; |
bd989ba3 | 487 | |
6074d45f | 488 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 489 | return 0; |
bd989ba3 | 490 | |
176ef8f5 | 491 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
492 | if (!tm_list) |
493 | return -ENOMEM; | |
494 | ||
176ef8f5 | 495 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
496 | if (!tm) { |
497 | ret = -ENOMEM; | |
498 | goto free_tms; | |
499 | } | |
500 | ||
298cfd36 | 501 | tm->logical = eb->start; |
5de865ee FDBM |
502 | tm->slot = src_slot; |
503 | tm->move.dst_slot = dst_slot; | |
504 | tm->move.nr_items = nr_items; | |
505 | tm->op = MOD_LOG_MOVE_KEYS; | |
506 | ||
507 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
508 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 509 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
510 | if (!tm_list[i]) { |
511 | ret = -ENOMEM; | |
512 | goto free_tms; | |
513 | } | |
514 | } | |
515 | ||
6074d45f | 516 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
517 | goto free_tms; |
518 | locked = 1; | |
519 | ||
01763a2e JS |
520 | /* |
521 | * When we override something during the move, we log these removals. | |
522 | * This can only happen when we move towards the beginning of the | |
523 | * buffer, i.e. dst_slot < src_slot. | |
524 | */ | |
bd989ba3 | 525 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 526 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
527 | if (ret) |
528 | goto free_tms; | |
bd989ba3 JS |
529 | } |
530 | ||
6074d45f | 531 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
532 | if (ret) |
533 | goto free_tms; | |
b1a09f1e | 534 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 535 | kfree(tm_list); |
f395694c | 536 | |
5de865ee FDBM |
537 | return 0; |
538 | free_tms: | |
539 | for (i = 0; i < nr_items; i++) { | |
540 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 541 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
542 | kfree(tm_list[i]); |
543 | } | |
544 | if (locked) | |
b1a09f1e | 545 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
546 | kfree(tm_list); |
547 | kfree(tm); | |
bd989ba3 | 548 | |
5de865ee | 549 | return ret; |
bd989ba3 JS |
550 | } |
551 | ||
5de865ee FDBM |
552 | static inline int |
553 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
554 | struct tree_mod_elem **tm_list, | |
555 | int nritems) | |
097b8a7c | 556 | { |
5de865ee | 557 | int i, j; |
097b8a7c JS |
558 | int ret; |
559 | ||
097b8a7c | 560 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
561 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
562 | if (ret) { | |
563 | for (j = nritems - 1; j > i; j--) | |
564 | rb_erase(&tm_list[j]->node, | |
565 | &fs_info->tree_mod_log); | |
566 | return ret; | |
567 | } | |
097b8a7c | 568 | } |
5de865ee FDBM |
569 | |
570 | return 0; | |
097b8a7c JS |
571 | } |
572 | ||
95b757c1 DS |
573 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
574 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 575 | { |
95b757c1 | 576 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
577 | struct tree_mod_elem *tm = NULL; |
578 | struct tree_mod_elem **tm_list = NULL; | |
579 | int nritems = 0; | |
580 | int ret = 0; | |
581 | int i; | |
bd989ba3 | 582 | |
5de865ee | 583 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
584 | return 0; |
585 | ||
5de865ee FDBM |
586 | if (log_removal && btrfs_header_level(old_root) > 0) { |
587 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 588 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 589 | GFP_NOFS); |
5de865ee FDBM |
590 | if (!tm_list) { |
591 | ret = -ENOMEM; | |
592 | goto free_tms; | |
593 | } | |
594 | for (i = 0; i < nritems; i++) { | |
595 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 596 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
597 | if (!tm_list[i]) { |
598 | ret = -ENOMEM; | |
599 | goto free_tms; | |
600 | } | |
601 | } | |
602 | } | |
d9abbf1c | 603 | |
bcc8e07f | 604 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
605 | if (!tm) { |
606 | ret = -ENOMEM; | |
607 | goto free_tms; | |
608 | } | |
bd989ba3 | 609 | |
298cfd36 | 610 | tm->logical = new_root->start; |
bd989ba3 JS |
611 | tm->old_root.logical = old_root->start; |
612 | tm->old_root.level = btrfs_header_level(old_root); | |
613 | tm->generation = btrfs_header_generation(old_root); | |
614 | tm->op = MOD_LOG_ROOT_REPLACE; | |
615 | ||
5de865ee FDBM |
616 | if (tree_mod_dont_log(fs_info, NULL)) |
617 | goto free_tms; | |
618 | ||
619 | if (tm_list) | |
620 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
621 | if (!ret) | |
622 | ret = __tree_mod_log_insert(fs_info, tm); | |
623 | ||
b1a09f1e | 624 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
625 | if (ret) |
626 | goto free_tms; | |
627 | kfree(tm_list); | |
628 | ||
629 | return ret; | |
630 | ||
631 | free_tms: | |
632 | if (tm_list) { | |
633 | for (i = 0; i < nritems; i++) | |
634 | kfree(tm_list[i]); | |
635 | kfree(tm_list); | |
636 | } | |
637 | kfree(tm); | |
638 | ||
639 | return ret; | |
bd989ba3 JS |
640 | } |
641 | ||
642 | static struct tree_mod_elem * | |
643 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
644 | int smallest) | |
645 | { | |
646 | struct rb_root *tm_root; | |
647 | struct rb_node *node; | |
648 | struct tree_mod_elem *cur = NULL; | |
649 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 650 | |
b1a09f1e | 651 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
652 | tm_root = &fs_info->tree_mod_log; |
653 | node = tm_root->rb_node; | |
654 | while (node) { | |
6b4df8b6 | 655 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 656 | if (cur->logical < start) { |
bd989ba3 | 657 | node = node->rb_left; |
298cfd36 | 658 | } else if (cur->logical > start) { |
bd989ba3 | 659 | node = node->rb_right; |
097b8a7c | 660 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
661 | node = node->rb_left; |
662 | } else if (!smallest) { | |
663 | /* we want the node with the highest seq */ | |
664 | if (found) | |
097b8a7c | 665 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
666 | found = cur; |
667 | node = node->rb_left; | |
097b8a7c | 668 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
669 | /* we want the node with the smallest seq */ |
670 | if (found) | |
097b8a7c | 671 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
672 | found = cur; |
673 | node = node->rb_right; | |
674 | } else { | |
675 | found = cur; | |
676 | break; | |
677 | } | |
678 | } | |
b1a09f1e | 679 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
680 | |
681 | return found; | |
682 | } | |
683 | ||
684 | /* | |
685 | * this returns the element from the log with the smallest time sequence | |
686 | * value that's in the log (the oldest log item). any element with a time | |
687 | * sequence lower than min_seq will be ignored. | |
688 | */ | |
689 | static struct tree_mod_elem * | |
690 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
691 | u64 min_seq) | |
692 | { | |
693 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
694 | } | |
695 | ||
696 | /* | |
697 | * this returns the element from the log with the largest time sequence | |
698 | * value that's in the log (the most recent log item). any element with | |
699 | * a time sequence lower than min_seq will be ignored. | |
700 | */ | |
701 | static struct tree_mod_elem * | |
702 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
703 | { | |
704 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
705 | } | |
706 | ||
ed874f0d | 707 | static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, |
bd989ba3 | 708 | struct extent_buffer *src, unsigned long dst_offset, |
90f8d62e | 709 | unsigned long src_offset, int nr_items) |
bd989ba3 | 710 | { |
ed874f0d | 711 | struct btrfs_fs_info *fs_info = dst->fs_info; |
5de865ee FDBM |
712 | int ret = 0; |
713 | struct tree_mod_elem **tm_list = NULL; | |
714 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 715 | int i; |
5de865ee | 716 | int locked = 0; |
bd989ba3 | 717 | |
5de865ee FDBM |
718 | if (!tree_mod_need_log(fs_info, NULL)) |
719 | return 0; | |
bd989ba3 | 720 | |
c8cc6341 | 721 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
722 | return 0; |
723 | ||
31e818fe | 724 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
725 | GFP_NOFS); |
726 | if (!tm_list) | |
727 | return -ENOMEM; | |
bd989ba3 | 728 | |
5de865ee FDBM |
729 | tm_list_add = tm_list; |
730 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 731 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
732 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
733 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
734 | if (!tm_list_rem[i]) { | |
735 | ret = -ENOMEM; | |
736 | goto free_tms; | |
737 | } | |
738 | ||
739 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
740 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
741 | if (!tm_list_add[i]) { | |
742 | ret = -ENOMEM; | |
743 | goto free_tms; | |
744 | } | |
745 | } | |
746 | ||
747 | if (tree_mod_dont_log(fs_info, NULL)) | |
748 | goto free_tms; | |
749 | locked = 1; | |
750 | ||
751 | for (i = 0; i < nr_items; i++) { | |
752 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
753 | if (ret) | |
754 | goto free_tms; | |
755 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
756 | if (ret) | |
757 | goto free_tms; | |
bd989ba3 | 758 | } |
5de865ee | 759 | |
b1a09f1e | 760 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
761 | kfree(tm_list); |
762 | ||
763 | return 0; | |
764 | ||
765 | free_tms: | |
766 | for (i = 0; i < nr_items * 2; i++) { | |
767 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
768 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
769 | kfree(tm_list[i]); | |
770 | } | |
771 | if (locked) | |
b1a09f1e | 772 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
773 | kfree(tm_list); |
774 | ||
775 | return ret; | |
bd989ba3 JS |
776 | } |
777 | ||
db7279a2 | 778 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 779 | { |
5de865ee FDBM |
780 | struct tree_mod_elem **tm_list = NULL; |
781 | int nritems = 0; | |
782 | int i; | |
783 | int ret = 0; | |
784 | ||
785 | if (btrfs_header_level(eb) == 0) | |
786 | return 0; | |
787 | ||
db7279a2 | 788 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
789 | return 0; |
790 | ||
791 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 792 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
793 | if (!tm_list) |
794 | return -ENOMEM; | |
795 | ||
796 | for (i = 0; i < nritems; i++) { | |
797 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
798 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
799 | if (!tm_list[i]) { | |
800 | ret = -ENOMEM; | |
801 | goto free_tms; | |
802 | } | |
803 | } | |
804 | ||
db7279a2 | 805 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
806 | goto free_tms; |
807 | ||
db7279a2 | 808 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 809 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
810 | if (ret) |
811 | goto free_tms; | |
812 | kfree(tm_list); | |
813 | ||
814 | return 0; | |
815 | ||
816 | free_tms: | |
817 | for (i = 0; i < nritems; i++) | |
818 | kfree(tm_list[i]); | |
819 | kfree(tm_list); | |
820 | ||
821 | return ret; | |
bd989ba3 JS |
822 | } |
823 | ||
5d4f98a2 YZ |
824 | /* |
825 | * check if the tree block can be shared by multiple trees | |
826 | */ | |
827 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
828 | struct extent_buffer *buf) | |
829 | { | |
830 | /* | |
92a7cc42 QW |
831 | * Tree blocks not in shareable trees and tree roots are never shared. |
832 | * If a block was allocated after the last snapshot and the block was | |
833 | * not allocated by tree relocation, we know the block is not shared. | |
5d4f98a2 | 834 | */ |
92a7cc42 | 835 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
5d4f98a2 YZ |
836 | buf != root->node && buf != root->commit_root && |
837 | (btrfs_header_generation(buf) <= | |
838 | btrfs_root_last_snapshot(&root->root_item) || | |
839 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
840 | return 1; | |
a79865c6 | 841 | |
5d4f98a2 YZ |
842 | return 0; |
843 | } | |
844 | ||
845 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
846 | struct btrfs_root *root, | |
847 | struct extent_buffer *buf, | |
f0486c68 YZ |
848 | struct extent_buffer *cow, |
849 | int *last_ref) | |
5d4f98a2 | 850 | { |
0b246afa | 851 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
852 | u64 refs; |
853 | u64 owner; | |
854 | u64 flags; | |
855 | u64 new_flags = 0; | |
856 | int ret; | |
857 | ||
858 | /* | |
859 | * Backrefs update rules: | |
860 | * | |
861 | * Always use full backrefs for extent pointers in tree block | |
862 | * allocated by tree relocation. | |
863 | * | |
864 | * If a shared tree block is no longer referenced by its owner | |
865 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
866 | * use full backrefs for extent pointers in tree block. | |
867 | * | |
868 | * If a tree block is been relocating | |
869 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
870 | * use full backrefs for extent pointers in tree block. | |
871 | * The reason for this is some operations (such as drop tree) | |
872 | * are only allowed for blocks use full backrefs. | |
873 | */ | |
874 | ||
875 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 876 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
877 | btrfs_header_level(buf), 1, |
878 | &refs, &flags); | |
be1a5564 MF |
879 | if (ret) |
880 | return ret; | |
e5df9573 MF |
881 | if (refs == 0) { |
882 | ret = -EROFS; | |
0b246afa | 883 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
884 | return ret; |
885 | } | |
5d4f98a2 YZ |
886 | } else { |
887 | refs = 1; | |
888 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
889 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
890 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
891 | else | |
892 | flags = 0; | |
893 | } | |
894 | ||
895 | owner = btrfs_header_owner(buf); | |
896 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
897 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
898 | ||
899 | if (refs > 1) { | |
900 | if ((owner == root->root_key.objectid || | |
901 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
902 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 903 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
904 | if (ret) |
905 | return ret; | |
5d4f98a2 YZ |
906 | |
907 | if (root->root_key.objectid == | |
908 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 909 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
910 | if (ret) |
911 | return ret; | |
e339a6b0 | 912 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
913 | if (ret) |
914 | return ret; | |
5d4f98a2 YZ |
915 | } |
916 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
917 | } else { | |
918 | ||
919 | if (root->root_key.objectid == | |
920 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 921 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 922 | else |
e339a6b0 | 923 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
924 | if (ret) |
925 | return ret; | |
5d4f98a2 YZ |
926 | } |
927 | if (new_flags != 0) { | |
b1c79e09 JB |
928 | int level = btrfs_header_level(buf); |
929 | ||
42c9d0b5 | 930 | ret = btrfs_set_disk_extent_flags(trans, buf, |
b1c79e09 | 931 | new_flags, level, 0); |
be1a5564 MF |
932 | if (ret) |
933 | return ret; | |
5d4f98a2 YZ |
934 | } |
935 | } else { | |
936 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
937 | if (root->root_key.objectid == | |
938 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 939 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 940 | else |
e339a6b0 | 941 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
942 | if (ret) |
943 | return ret; | |
e339a6b0 | 944 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
945 | if (ret) |
946 | return ret; | |
5d4f98a2 | 947 | } |
6a884d7d | 948 | btrfs_clean_tree_block(buf); |
f0486c68 | 949 | *last_ref = 1; |
5d4f98a2 YZ |
950 | } |
951 | return 0; | |
952 | } | |
953 | ||
a6279470 FM |
954 | static struct extent_buffer *alloc_tree_block_no_bg_flush( |
955 | struct btrfs_trans_handle *trans, | |
956 | struct btrfs_root *root, | |
957 | u64 parent_start, | |
958 | const struct btrfs_disk_key *disk_key, | |
959 | int level, | |
960 | u64 hint, | |
9631e4cc JB |
961 | u64 empty_size, |
962 | enum btrfs_lock_nesting nest) | |
a6279470 FM |
963 | { |
964 | struct btrfs_fs_info *fs_info = root->fs_info; | |
965 | struct extent_buffer *ret; | |
966 | ||
967 | /* | |
968 | * If we are COWing a node/leaf from the extent, chunk, device or free | |
969 | * space trees, make sure that we do not finish block group creation of | |
970 | * pending block groups. We do this to avoid a deadlock. | |
971 | * COWing can result in allocation of a new chunk, and flushing pending | |
972 | * block groups (btrfs_create_pending_block_groups()) can be triggered | |
973 | * when finishing allocation of a new chunk. Creation of a pending block | |
974 | * group modifies the extent, chunk, device and free space trees, | |
975 | * therefore we could deadlock with ourselves since we are holding a | |
976 | * lock on an extent buffer that btrfs_create_pending_block_groups() may | |
977 | * try to COW later. | |
978 | * For similar reasons, we also need to delay flushing pending block | |
979 | * groups when splitting a leaf or node, from one of those trees, since | |
980 | * we are holding a write lock on it and its parent or when inserting a | |
981 | * new root node for one of those trees. | |
982 | */ | |
983 | if (root == fs_info->extent_root || | |
984 | root == fs_info->chunk_root || | |
985 | root == fs_info->dev_root || | |
986 | root == fs_info->free_space_root) | |
987 | trans->can_flush_pending_bgs = false; | |
988 | ||
989 | ret = btrfs_alloc_tree_block(trans, root, parent_start, | |
990 | root->root_key.objectid, disk_key, level, | |
9631e4cc | 991 | hint, empty_size, nest); |
a6279470 FM |
992 | trans->can_flush_pending_bgs = true; |
993 | ||
994 | return ret; | |
995 | } | |
996 | ||
d352ac68 | 997 | /* |
d397712b CM |
998 | * does the dirty work in cow of a single block. The parent block (if |
999 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1000 | * dirty and returned locked. If you modify the block it needs to be marked | |
1001 | * dirty again. | |
d352ac68 CM |
1002 | * |
1003 | * search_start -- an allocation hint for the new block | |
1004 | * | |
d397712b CM |
1005 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1006 | * bytes the allocator should try to find free next to the block it returns. | |
1007 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1008 | */ |
d397712b | 1009 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1010 | struct btrfs_root *root, |
1011 | struct extent_buffer *buf, | |
1012 | struct extent_buffer *parent, int parent_slot, | |
1013 | struct extent_buffer **cow_ret, | |
9631e4cc JB |
1014 | u64 search_start, u64 empty_size, |
1015 | enum btrfs_lock_nesting nest) | |
02217ed2 | 1016 | { |
0b246afa | 1017 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1018 | struct btrfs_disk_key disk_key; |
5f39d397 | 1019 | struct extent_buffer *cow; |
be1a5564 | 1020 | int level, ret; |
f0486c68 | 1021 | int last_ref = 0; |
925baedd | 1022 | int unlock_orig = 0; |
0f5053eb | 1023 | u64 parent_start = 0; |
7bb86316 | 1024 | |
925baedd CM |
1025 | if (*cow_ret == buf) |
1026 | unlock_orig = 1; | |
1027 | ||
b9447ef8 | 1028 | btrfs_assert_tree_locked(buf); |
925baedd | 1029 | |
92a7cc42 | 1030 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 1031 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 1032 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 1033 | trans->transid != root->last_trans); |
5f39d397 | 1034 | |
7bb86316 | 1035 | level = btrfs_header_level(buf); |
31840ae1 | 1036 | |
5d4f98a2 YZ |
1037 | if (level == 0) |
1038 | btrfs_item_key(buf, &disk_key, 0); | |
1039 | else | |
1040 | btrfs_node_key(buf, &disk_key, 0); | |
1041 | ||
0f5053eb GR |
1042 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1043 | parent_start = parent->start; | |
5d4f98a2 | 1044 | |
a6279470 | 1045 | cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, |
9631e4cc | 1046 | level, search_start, empty_size, nest); |
54aa1f4d CM |
1047 | if (IS_ERR(cow)) |
1048 | return PTR_ERR(cow); | |
6702ed49 | 1049 | |
b4ce94de CM |
1050 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1051 | ||
58e8012c | 1052 | copy_extent_buffer_full(cow, buf); |
db94535d | 1053 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1054 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1055 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1056 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1057 | BTRFS_HEADER_FLAG_RELOC); | |
1058 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1059 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1060 | else | |
1061 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1062 | |
de37aa51 | 1063 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 1064 | |
be1a5564 | 1065 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1066 | if (ret) { |
572c83ac JB |
1067 | btrfs_tree_unlock(cow); |
1068 | free_extent_buffer(cow); | |
66642832 | 1069 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1070 | return ret; |
1071 | } | |
1a40e23b | 1072 | |
92a7cc42 | 1073 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
83d4cfd4 | 1074 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1075 | if (ret) { |
572c83ac JB |
1076 | btrfs_tree_unlock(cow); |
1077 | free_extent_buffer(cow); | |
66642832 | 1078 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1079 | return ret; |
93314e3b | 1080 | } |
83d4cfd4 | 1081 | } |
3fd0a558 | 1082 | |
02217ed2 | 1083 | if (buf == root->node) { |
925baedd | 1084 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1085 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1086 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1087 | parent_start = buf->start; | |
925baedd | 1088 | |
67439dad | 1089 | atomic_inc(&cow->refs); |
d9d19a01 DS |
1090 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1091 | BUG_ON(ret < 0); | |
240f62c8 | 1092 | rcu_assign_pointer(root->node, cow); |
925baedd | 1093 | |
f0486c68 | 1094 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1095 | last_ref); |
5f39d397 | 1096 | free_extent_buffer(buf); |
0b86a832 | 1097 | add_root_to_dirty_list(root); |
02217ed2 | 1098 | } else { |
5d4f98a2 | 1099 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1100 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1101 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1102 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1103 | cow->start); |
74493f7a CM |
1104 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1105 | trans->transid); | |
d6025579 | 1106 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1107 | if (last_ref) { |
db7279a2 | 1108 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1109 | if (ret) { |
572c83ac JB |
1110 | btrfs_tree_unlock(cow); |
1111 | free_extent_buffer(cow); | |
66642832 | 1112 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1113 | return ret; |
1114 | } | |
1115 | } | |
f0486c68 | 1116 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1117 | last_ref); |
02217ed2 | 1118 | } |
925baedd CM |
1119 | if (unlock_orig) |
1120 | btrfs_tree_unlock(buf); | |
3083ee2e | 1121 | free_extent_buffer_stale(buf); |
ccd467d6 | 1122 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1123 | *cow_ret = cow; |
02217ed2 CM |
1124 | return 0; |
1125 | } | |
1126 | ||
5d9e75c4 JS |
1127 | /* |
1128 | * returns the logical address of the oldest predecessor of the given root. | |
1129 | * entries older than time_seq are ignored. | |
1130 | */ | |
bcd24dab DS |
1131 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1132 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1133 | { |
1134 | struct tree_mod_elem *tm; | |
1135 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1136 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1137 | int looped = 0; |
1138 | ||
1139 | if (!time_seq) | |
35a3621b | 1140 | return NULL; |
5d9e75c4 JS |
1141 | |
1142 | /* | |
298cfd36 CR |
1143 | * the very last operation that's logged for a root is the |
1144 | * replacement operation (if it is replaced at all). this has | |
1145 | * the logical address of the *new* root, making it the very | |
1146 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1147 | */ |
1148 | while (1) { | |
bcd24dab | 1149 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1150 | time_seq); |
1151 | if (!looped && !tm) | |
35a3621b | 1152 | return NULL; |
5d9e75c4 | 1153 | /* |
28da9fb4 JS |
1154 | * if there are no tree operation for the oldest root, we simply |
1155 | * return it. this should only happen if that (old) root is at | |
1156 | * level 0. | |
5d9e75c4 | 1157 | */ |
28da9fb4 JS |
1158 | if (!tm) |
1159 | break; | |
5d9e75c4 | 1160 | |
28da9fb4 JS |
1161 | /* |
1162 | * if there's an operation that's not a root replacement, we | |
1163 | * found the oldest version of our root. normally, we'll find a | |
1164 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1165 | */ | |
5d9e75c4 JS |
1166 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1167 | break; | |
1168 | ||
1169 | found = tm; | |
1170 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1171 | looped = 1; |
1172 | } | |
1173 | ||
a95236d9 JS |
1174 | /* if there's no old root to return, return what we found instead */ |
1175 | if (!found) | |
1176 | found = tm; | |
1177 | ||
5d9e75c4 JS |
1178 | return found; |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1183 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1184 | * time_seq). |
1185 | */ | |
1186 | static void | |
f1ca7e98 JB |
1187 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1188 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1189 | { |
1190 | u32 n; | |
1191 | struct rb_node *next; | |
1192 | struct tree_mod_elem *tm = first_tm; | |
1193 | unsigned long o_dst; | |
1194 | unsigned long o_src; | |
1195 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1196 | ||
1197 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1198 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1199 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1200 | /* |
1201 | * all the operations are recorded with the operator used for | |
1202 | * the modification. as we're going backwards, we do the | |
1203 | * opposite of each operation here. | |
1204 | */ | |
1205 | switch (tm->op) { | |
1206 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1207 | BUG_ON(tm->slot < n); | |
c730ae0c | 1208 | fallthrough; |
95c80bb1 | 1209 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1210 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1211 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1212 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1213 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1214 | tm->generation); | |
4c3e6969 | 1215 | n++; |
5d9e75c4 JS |
1216 | break; |
1217 | case MOD_LOG_KEY_REPLACE: | |
1218 | BUG_ON(tm->slot >= n); | |
1219 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1220 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1221 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1222 | tm->generation); | |
1223 | break; | |
1224 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1225 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1226 | n--; |
1227 | break; | |
1228 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1229 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1230 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1231 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1232 | tm->move.nr_items * p_size); |
1233 | break; | |
1234 | case MOD_LOG_ROOT_REPLACE: | |
1235 | /* | |
1236 | * this operation is special. for roots, this must be | |
1237 | * handled explicitly before rewinding. | |
1238 | * for non-roots, this operation may exist if the node | |
1239 | * was a root: root A -> child B; then A gets empty and | |
1240 | * B is promoted to the new root. in the mod log, we'll | |
1241 | * have a root-replace operation for B, a tree block | |
1242 | * that is no root. we simply ignore that operation. | |
1243 | */ | |
1244 | break; | |
1245 | } | |
1246 | next = rb_next(&tm->node); | |
1247 | if (!next) | |
1248 | break; | |
6b4df8b6 | 1249 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1250 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1251 | break; |
1252 | } | |
b1a09f1e | 1253 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1254 | btrfs_set_header_nritems(eb, n); |
1255 | } | |
1256 | ||
47fb091f | 1257 | /* |
01327610 | 1258 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1259 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1260 | * returned buffer is always read-locked. If the returned buffer is not the | |
1261 | * input buffer, the lock on the input buffer is released and the input buffer | |
1262 | * is freed (its refcount is decremented). | |
1263 | */ | |
5d9e75c4 | 1264 | static struct extent_buffer * |
9ec72677 JB |
1265 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1266 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1267 | { |
1268 | struct extent_buffer *eb_rewin; | |
1269 | struct tree_mod_elem *tm; | |
1270 | ||
1271 | if (!time_seq) | |
1272 | return eb; | |
1273 | ||
1274 | if (btrfs_header_level(eb) == 0) | |
1275 | return eb; | |
1276 | ||
1277 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1278 | if (!tm) | |
1279 | return eb; | |
1280 | ||
1281 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { | |
1282 | BUG_ON(tm->slot != 0); | |
da17066c | 1283 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1284 | if (!eb_rewin) { |
ac5887c8 | 1285 | btrfs_tree_read_unlock(eb); |
db7f3436 JB |
1286 | free_extent_buffer(eb); |
1287 | return NULL; | |
1288 | } | |
5d9e75c4 JS |
1289 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1290 | btrfs_set_header_backref_rev(eb_rewin, | |
1291 | btrfs_header_backref_rev(eb)); | |
1292 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1293 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1294 | } else { |
1295 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1296 | if (!eb_rewin) { |
ac5887c8 | 1297 | btrfs_tree_read_unlock(eb); |
db7f3436 JB |
1298 | free_extent_buffer(eb); |
1299 | return NULL; | |
1300 | } | |
5d9e75c4 JS |
1301 | } |
1302 | ||
ac5887c8 | 1303 | btrfs_tree_read_unlock(eb); |
5d9e75c4 JS |
1304 | free_extent_buffer(eb); |
1305 | ||
d3beaa25 JB |
1306 | btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb_rewin), |
1307 | eb_rewin, btrfs_header_level(eb_rewin)); | |
47fb091f | 1308 | btrfs_tree_read_lock(eb_rewin); |
f1ca7e98 | 1309 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1310 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1311 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1312 | |
1313 | return eb_rewin; | |
1314 | } | |
1315 | ||
8ba97a15 JS |
1316 | /* |
1317 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1318 | * value. If there are no changes, the current root->root_node is returned. If | |
1319 | * anything changed in between, there's a fresh buffer allocated on which the | |
1320 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1321 | * Returns NULL on error (with no locks held). | |
1322 | */ | |
5d9e75c4 JS |
1323 | static inline struct extent_buffer * |
1324 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1325 | { | |
0b246afa | 1326 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1327 | struct tree_mod_elem *tm; |
30b0463a JS |
1328 | struct extent_buffer *eb = NULL; |
1329 | struct extent_buffer *eb_root; | |
efad8a85 | 1330 | u64 eb_root_owner = 0; |
7bfdcf7f | 1331 | struct extent_buffer *old; |
a95236d9 | 1332 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1333 | u64 old_generation = 0; |
a95236d9 | 1334 | u64 logical; |
581c1760 | 1335 | int level; |
5d9e75c4 | 1336 | |
30b0463a | 1337 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1338 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1339 | if (!tm) |
30b0463a | 1340 | return eb_root; |
5d9e75c4 | 1341 | |
a95236d9 JS |
1342 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1343 | old_root = &tm->old_root; | |
1344 | old_generation = tm->generation; | |
1345 | logical = old_root->logical; | |
581c1760 | 1346 | level = old_root->level; |
a95236d9 | 1347 | } else { |
30b0463a | 1348 | logical = eb_root->start; |
581c1760 | 1349 | level = btrfs_header_level(eb_root); |
a95236d9 | 1350 | } |
5d9e75c4 | 1351 | |
0b246afa | 1352 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1353 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1354 | btrfs_tree_read_unlock(eb_root); |
1355 | free_extent_buffer(eb_root); | |
1b7ec85e JB |
1356 | old = read_tree_block(fs_info, logical, root->root_key.objectid, |
1357 | 0, level, NULL); | |
64c043de LB |
1358 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1359 | if (!IS_ERR(old)) | |
1360 | free_extent_buffer(old); | |
0b246afa JM |
1361 | btrfs_warn(fs_info, |
1362 | "failed to read tree block %llu from get_old_root", | |
1363 | logical); | |
834328a8 | 1364 | } else { |
7bfdcf7f LB |
1365 | eb = btrfs_clone_extent_buffer(old); |
1366 | free_extent_buffer(old); | |
834328a8 JS |
1367 | } |
1368 | } else if (old_root) { | |
efad8a85 | 1369 | eb_root_owner = btrfs_header_owner(eb_root); |
30b0463a JS |
1370 | btrfs_tree_read_unlock(eb_root); |
1371 | free_extent_buffer(eb_root); | |
0b246afa | 1372 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1373 | } else { |
30b0463a | 1374 | eb = btrfs_clone_extent_buffer(eb_root); |
ac5887c8 | 1375 | btrfs_tree_read_unlock(eb_root); |
30b0463a | 1376 | free_extent_buffer(eb_root); |
834328a8 JS |
1377 | } |
1378 | ||
8ba97a15 JS |
1379 | if (!eb) |
1380 | return NULL; | |
a95236d9 | 1381 | if (old_root) { |
5d9e75c4 JS |
1382 | btrfs_set_header_bytenr(eb, eb->start); |
1383 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
efad8a85 | 1384 | btrfs_set_header_owner(eb, eb_root_owner); |
a95236d9 JS |
1385 | btrfs_set_header_level(eb, old_root->level); |
1386 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1387 | } |
d3beaa25 JB |
1388 | btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), eb, |
1389 | btrfs_header_level(eb)); | |
1390 | btrfs_tree_read_lock(eb); | |
28da9fb4 | 1391 | if (tm) |
0b246afa | 1392 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1393 | else |
1394 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1395 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1396 | |
1397 | return eb; | |
1398 | } | |
1399 | ||
5b6602e7 JS |
1400 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1401 | { | |
1402 | struct tree_mod_elem *tm; | |
1403 | int level; | |
30b0463a | 1404 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1405 | |
bcd24dab | 1406 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1407 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1408 | level = tm->old_root.level; | |
1409 | } else { | |
30b0463a | 1410 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1411 | } |
30b0463a | 1412 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1413 | |
1414 | return level; | |
1415 | } | |
1416 | ||
5d4f98a2 YZ |
1417 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1418 | struct btrfs_root *root, | |
1419 | struct extent_buffer *buf) | |
1420 | { | |
f5ee5c9a | 1421 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1422 | return 0; |
fccb84c9 | 1423 | |
d1980131 DS |
1424 | /* Ensure we can see the FORCE_COW bit */ |
1425 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1426 | |
1427 | /* | |
1428 | * We do not need to cow a block if | |
1429 | * 1) this block is not created or changed in this transaction; | |
1430 | * 2) this block does not belong to TREE_RELOC tree; | |
1431 | * 3) the root is not forced COW. | |
1432 | * | |
1433 | * What is forced COW: | |
01327610 | 1434 | * when we create snapshot during committing the transaction, |
52042d8e | 1435 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
1436 | * block to ensure the metadata consistency. |
1437 | */ | |
5d4f98a2 YZ |
1438 | if (btrfs_header_generation(buf) == trans->transid && |
1439 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1440 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1441 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1442 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1443 | return 0; |
1444 | return 1; | |
1445 | } | |
1446 | ||
d352ac68 CM |
1447 | /* |
1448 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1449 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1450 | * once per transaction, as long as it hasn't been written yet |
1451 | */ | |
d397712b | 1452 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1453 | struct btrfs_root *root, struct extent_buffer *buf, |
1454 | struct extent_buffer *parent, int parent_slot, | |
9631e4cc JB |
1455 | struct extent_buffer **cow_ret, |
1456 | enum btrfs_lock_nesting nest) | |
6702ed49 | 1457 | { |
0b246afa | 1458 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1459 | u64 search_start; |
f510cfec | 1460 | int ret; |
dc17ff8f | 1461 | |
83354f07 JB |
1462 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
1463 | btrfs_err(fs_info, | |
1464 | "COW'ing blocks on a fs root that's being dropped"); | |
1465 | ||
0b246afa | 1466 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1467 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1468 | trans->transid, |
0b246afa | 1469 | fs_info->running_transaction->transid); |
31b1a2bd | 1470 | |
0b246afa | 1471 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1472 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1473 | trans->transid, fs_info->generation); |
dc17ff8f | 1474 | |
5d4f98a2 | 1475 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1476 | trans->dirty = true; |
6702ed49 CM |
1477 | *cow_ret = buf; |
1478 | return 0; | |
1479 | } | |
c487685d | 1480 | |
ee22184b | 1481 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de | 1482 | |
f616f5cd QW |
1483 | /* |
1484 | * Before CoWing this block for later modification, check if it's | |
1485 | * the subtree root and do the delayed subtree trace if needed. | |
1486 | * | |
1487 | * Also We don't care about the error, as it's handled internally. | |
1488 | */ | |
1489 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 1490 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9631e4cc | 1491 | parent_slot, cow_ret, search_start, 0, nest); |
1abe9b8a | 1492 | |
1493 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1494 | ||
f510cfec | 1495 | return ret; |
6702ed49 CM |
1496 | } |
1497 | ||
d352ac68 CM |
1498 | /* |
1499 | * helper function for defrag to decide if two blocks pointed to by a | |
1500 | * node are actually close by | |
1501 | */ | |
6b80053d | 1502 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1503 | { |
6b80053d | 1504 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1505 | return 1; |
6b80053d | 1506 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1507 | return 1; |
1508 | return 0; | |
1509 | } | |
1510 | ||
ce6ef5ab DS |
1511 | #ifdef __LITTLE_ENDIAN |
1512 | ||
1513 | /* | |
1514 | * Compare two keys, on little-endian the disk order is same as CPU order and | |
1515 | * we can avoid the conversion. | |
1516 | */ | |
1517 | static int comp_keys(const struct btrfs_disk_key *disk_key, | |
1518 | const struct btrfs_key *k2) | |
1519 | { | |
1520 | const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; | |
1521 | ||
1522 | return btrfs_comp_cpu_keys(k1, k2); | |
1523 | } | |
1524 | ||
1525 | #else | |
1526 | ||
081e9573 CM |
1527 | /* |
1528 | * compare two keys in a memcmp fashion | |
1529 | */ | |
310712b2 OS |
1530 | static int comp_keys(const struct btrfs_disk_key *disk, |
1531 | const struct btrfs_key *k2) | |
081e9573 CM |
1532 | { |
1533 | struct btrfs_key k1; | |
1534 | ||
1535 | btrfs_disk_key_to_cpu(&k1, disk); | |
1536 | ||
20736aba | 1537 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 | 1538 | } |
ce6ef5ab | 1539 | #endif |
081e9573 | 1540 | |
f3465ca4 JB |
1541 | /* |
1542 | * same as comp_keys only with two btrfs_key's | |
1543 | */ | |
e1f60a65 | 1544 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1545 | { |
1546 | if (k1->objectid > k2->objectid) | |
1547 | return 1; | |
1548 | if (k1->objectid < k2->objectid) | |
1549 | return -1; | |
1550 | if (k1->type > k2->type) | |
1551 | return 1; | |
1552 | if (k1->type < k2->type) | |
1553 | return -1; | |
1554 | if (k1->offset > k2->offset) | |
1555 | return 1; | |
1556 | if (k1->offset < k2->offset) | |
1557 | return -1; | |
1558 | return 0; | |
1559 | } | |
081e9573 | 1560 | |
d352ac68 CM |
1561 | /* |
1562 | * this is used by the defrag code to go through all the | |
1563 | * leaves pointed to by a node and reallocate them so that | |
1564 | * disk order is close to key order | |
1565 | */ | |
6702ed49 | 1566 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1567 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1568 | int start_slot, u64 *last_ret, |
a6b6e75e | 1569 | struct btrfs_key *progress) |
6702ed49 | 1570 | { |
0b246afa | 1571 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1572 | struct extent_buffer *cur; |
6702ed49 | 1573 | u64 blocknr; |
e9d0b13b CM |
1574 | u64 search_start = *last_ret; |
1575 | u64 last_block = 0; | |
6702ed49 CM |
1576 | u64 other; |
1577 | u32 parent_nritems; | |
6702ed49 CM |
1578 | int end_slot; |
1579 | int i; | |
1580 | int err = 0; | |
6b80053d | 1581 | u32 blocksize; |
081e9573 CM |
1582 | int progress_passed = 0; |
1583 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1584 | |
0b246afa JM |
1585 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1586 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1587 | |
6b80053d | 1588 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1589 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1590 | end_slot = parent_nritems - 1; |
6702ed49 | 1591 | |
5dfe2be7 | 1592 | if (parent_nritems <= 1) |
6702ed49 CM |
1593 | return 0; |
1594 | ||
5dfe2be7 | 1595 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 1596 | int close = 1; |
a6b6e75e | 1597 | |
081e9573 CM |
1598 | btrfs_node_key(parent, &disk_key, i); |
1599 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1600 | continue; | |
1601 | ||
1602 | progress_passed = 1; | |
6b80053d | 1603 | blocknr = btrfs_node_blockptr(parent, i); |
e9d0b13b CM |
1604 | if (last_block == 0) |
1605 | last_block = blocknr; | |
5708b959 | 1606 | |
6702ed49 | 1607 | if (i > 0) { |
6b80053d CM |
1608 | other = btrfs_node_blockptr(parent, i - 1); |
1609 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1610 | } |
5dfe2be7 | 1611 | if (!close && i < end_slot) { |
6b80053d CM |
1612 | other = btrfs_node_blockptr(parent, i + 1); |
1613 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1614 | } |
e9d0b13b CM |
1615 | if (close) { |
1616 | last_block = blocknr; | |
6702ed49 | 1617 | continue; |
e9d0b13b | 1618 | } |
6702ed49 | 1619 | |
206983b7 JB |
1620 | cur = btrfs_read_node_slot(parent, i); |
1621 | if (IS_ERR(cur)) | |
1622 | return PTR_ERR(cur); | |
e9d0b13b | 1623 | if (search_start == 0) |
6b80053d | 1624 | search_start = last_block; |
e9d0b13b | 1625 | |
e7a84565 | 1626 | btrfs_tree_lock(cur); |
6b80053d | 1627 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1628 | &cur, search_start, |
6b80053d | 1629 | min(16 * blocksize, |
9631e4cc JB |
1630 | (end_slot - i) * blocksize), |
1631 | BTRFS_NESTING_COW); | |
252c38f0 | 1632 | if (err) { |
e7a84565 | 1633 | btrfs_tree_unlock(cur); |
6b80053d | 1634 | free_extent_buffer(cur); |
6702ed49 | 1635 | break; |
252c38f0 | 1636 | } |
e7a84565 CM |
1637 | search_start = cur->start; |
1638 | last_block = cur->start; | |
f2183bde | 1639 | *last_ret = search_start; |
e7a84565 CM |
1640 | btrfs_tree_unlock(cur); |
1641 | free_extent_buffer(cur); | |
6702ed49 CM |
1642 | } |
1643 | return err; | |
1644 | } | |
1645 | ||
74123bd7 | 1646 | /* |
5f39d397 CM |
1647 | * search for key in the extent_buffer. The items start at offset p, |
1648 | * and they are item_size apart. There are 'max' items in p. | |
1649 | * | |
74123bd7 CM |
1650 | * the slot in the array is returned via slot, and it points to |
1651 | * the place where you would insert key if it is not found in | |
1652 | * the array. | |
1653 | * | |
1654 | * slot may point to max if the key is bigger than all of the keys | |
1655 | */ | |
e02119d5 | 1656 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1657 | unsigned long p, int item_size, |
1658 | const struct btrfs_key *key, | |
e02119d5 | 1659 | int max, int *slot) |
be0e5c09 CM |
1660 | { |
1661 | int low = 0; | |
1662 | int high = max; | |
be0e5c09 | 1663 | int ret; |
5cd17f34 | 1664 | const int key_size = sizeof(struct btrfs_disk_key); |
be0e5c09 | 1665 | |
5e24e9af LB |
1666 | if (low > high) { |
1667 | btrfs_err(eb->fs_info, | |
1668 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1669 | __func__, low, high, eb->start, | |
1670 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1671 | return -EINVAL; | |
1672 | } | |
1673 | ||
d397712b | 1674 | while (low < high) { |
5cd17f34 DS |
1675 | unsigned long oip; |
1676 | unsigned long offset; | |
1677 | struct btrfs_disk_key *tmp; | |
1678 | struct btrfs_disk_key unaligned; | |
1679 | int mid; | |
1680 | ||
be0e5c09 | 1681 | mid = (low + high) / 2; |
5f39d397 | 1682 | offset = p + mid * item_size; |
5cd17f34 | 1683 | oip = offset_in_page(offset); |
5f39d397 | 1684 | |
5cd17f34 DS |
1685 | if (oip + key_size <= PAGE_SIZE) { |
1686 | const unsigned long idx = offset >> PAGE_SHIFT; | |
1687 | char *kaddr = page_address(eb->pages[idx]); | |
5f39d397 | 1688 | |
5cd17f34 | 1689 | tmp = (struct btrfs_disk_key *)(kaddr + oip); |
5f39d397 | 1690 | } else { |
5cd17f34 DS |
1691 | read_extent_buffer(eb, &unaligned, offset, key_size); |
1692 | tmp = &unaligned; | |
5f39d397 | 1693 | } |
5cd17f34 | 1694 | |
be0e5c09 CM |
1695 | ret = comp_keys(tmp, key); |
1696 | ||
1697 | if (ret < 0) | |
1698 | low = mid + 1; | |
1699 | else if (ret > 0) | |
1700 | high = mid; | |
1701 | else { | |
1702 | *slot = mid; | |
1703 | return 0; | |
1704 | } | |
1705 | } | |
1706 | *slot = low; | |
1707 | return 1; | |
1708 | } | |
1709 | ||
97571fd0 CM |
1710 | /* |
1711 | * simple bin_search frontend that does the right thing for | |
1712 | * leaves vs nodes | |
1713 | */ | |
a74b35ec | 1714 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
e3b83361 | 1715 | int *slot) |
be0e5c09 | 1716 | { |
e3b83361 | 1717 | if (btrfs_header_level(eb) == 0) |
5f39d397 CM |
1718 | return generic_bin_search(eb, |
1719 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1720 | sizeof(struct btrfs_item), |
5f39d397 | 1721 | key, btrfs_header_nritems(eb), |
7518a238 | 1722 | slot); |
f775738f | 1723 | else |
5f39d397 CM |
1724 | return generic_bin_search(eb, |
1725 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1726 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1727 | key, btrfs_header_nritems(eb), |
7518a238 | 1728 | slot); |
be0e5c09 CM |
1729 | } |
1730 | ||
f0486c68 YZ |
1731 | static void root_add_used(struct btrfs_root *root, u32 size) |
1732 | { | |
1733 | spin_lock(&root->accounting_lock); | |
1734 | btrfs_set_root_used(&root->root_item, | |
1735 | btrfs_root_used(&root->root_item) + size); | |
1736 | spin_unlock(&root->accounting_lock); | |
1737 | } | |
1738 | ||
1739 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1740 | { | |
1741 | spin_lock(&root->accounting_lock); | |
1742 | btrfs_set_root_used(&root->root_item, | |
1743 | btrfs_root_used(&root->root_item) - size); | |
1744 | spin_unlock(&root->accounting_lock); | |
1745 | } | |
1746 | ||
d352ac68 CM |
1747 | /* given a node and slot number, this reads the blocks it points to. The |
1748 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1749 | */ |
4b231ae4 DS |
1750 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
1751 | int slot) | |
bb803951 | 1752 | { |
ca7a79ad | 1753 | int level = btrfs_header_level(parent); |
416bc658 | 1754 | struct extent_buffer *eb; |
581c1760 | 1755 | struct btrfs_key first_key; |
416bc658 | 1756 | |
fb770ae4 LB |
1757 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1758 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1759 | |
1760 | BUG_ON(level == 0); | |
1761 | ||
581c1760 | 1762 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 1763 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
1b7ec85e | 1764 | btrfs_header_owner(parent), |
581c1760 QW |
1765 | btrfs_node_ptr_generation(parent, slot), |
1766 | level - 1, &first_key); | |
fb770ae4 LB |
1767 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1768 | free_extent_buffer(eb); | |
1769 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1770 | } |
1771 | ||
1772 | return eb; | |
bb803951 CM |
1773 | } |
1774 | ||
d352ac68 CM |
1775 | /* |
1776 | * node level balancing, used to make sure nodes are in proper order for | |
1777 | * item deletion. We balance from the top down, so we have to make sure | |
1778 | * that a deletion won't leave an node completely empty later on. | |
1779 | */ | |
e02119d5 | 1780 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1781 | struct btrfs_root *root, |
1782 | struct btrfs_path *path, int level) | |
bb803951 | 1783 | { |
0b246afa | 1784 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1785 | struct extent_buffer *right = NULL; |
1786 | struct extent_buffer *mid; | |
1787 | struct extent_buffer *left = NULL; | |
1788 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1789 | int ret = 0; |
1790 | int wret; | |
1791 | int pslot; | |
bb803951 | 1792 | int orig_slot = path->slots[level]; |
79f95c82 | 1793 | u64 orig_ptr; |
bb803951 | 1794 | |
98e6b1eb | 1795 | ASSERT(level > 0); |
bb803951 | 1796 | |
5f39d397 | 1797 | mid = path->nodes[level]; |
b4ce94de | 1798 | |
ac5887c8 | 1799 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); |
7bb86316 CM |
1800 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1801 | ||
1d4f8a0c | 1802 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1803 | |
a05a9bb1 | 1804 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1805 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1806 | pslot = path->slots[level + 1]; |
1807 | } | |
bb803951 | 1808 | |
40689478 CM |
1809 | /* |
1810 | * deal with the case where there is only one pointer in the root | |
1811 | * by promoting the node below to a root | |
1812 | */ | |
5f39d397 CM |
1813 | if (!parent) { |
1814 | struct extent_buffer *child; | |
bb803951 | 1815 | |
5f39d397 | 1816 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1817 | return 0; |
1818 | ||
1819 | /* promote the child to a root */ | |
4b231ae4 | 1820 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
1821 | if (IS_ERR(child)) { |
1822 | ret = PTR_ERR(child); | |
0b246afa | 1823 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1824 | goto enospc; |
1825 | } | |
1826 | ||
925baedd | 1827 | btrfs_tree_lock(child); |
9631e4cc JB |
1828 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child, |
1829 | BTRFS_NESTING_COW); | |
f0486c68 YZ |
1830 | if (ret) { |
1831 | btrfs_tree_unlock(child); | |
1832 | free_extent_buffer(child); | |
1833 | goto enospc; | |
1834 | } | |
2f375ab9 | 1835 | |
d9d19a01 DS |
1836 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1837 | BUG_ON(ret < 0); | |
240f62c8 | 1838 | rcu_assign_pointer(root->node, child); |
925baedd | 1839 | |
0b86a832 | 1840 | add_root_to_dirty_list(root); |
925baedd | 1841 | btrfs_tree_unlock(child); |
b4ce94de | 1842 | |
925baedd | 1843 | path->locks[level] = 0; |
bb803951 | 1844 | path->nodes[level] = NULL; |
6a884d7d | 1845 | btrfs_clean_tree_block(mid); |
925baedd | 1846 | btrfs_tree_unlock(mid); |
bb803951 | 1847 | /* once for the path */ |
5f39d397 | 1848 | free_extent_buffer(mid); |
f0486c68 YZ |
1849 | |
1850 | root_sub_used(root, mid->len); | |
5581a51a | 1851 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1852 | /* once for the root ptr */ |
3083ee2e | 1853 | free_extent_buffer_stale(mid); |
f0486c68 | 1854 | return 0; |
bb803951 | 1855 | } |
5f39d397 | 1856 | if (btrfs_header_nritems(mid) > |
0b246afa | 1857 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1858 | return 0; |
1859 | ||
4b231ae4 | 1860 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1861 | if (IS_ERR(left)) |
1862 | left = NULL; | |
1863 | ||
5f39d397 | 1864 | if (left) { |
bf77467a | 1865 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
5f39d397 | 1866 | wret = btrfs_cow_block(trans, root, left, |
9631e4cc | 1867 | parent, pslot - 1, &left, |
bf59a5a2 | 1868 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
1869 | if (wret) { |
1870 | ret = wret; | |
1871 | goto enospc; | |
1872 | } | |
2cc58cf2 | 1873 | } |
fb770ae4 | 1874 | |
4b231ae4 | 1875 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1876 | if (IS_ERR(right)) |
1877 | right = NULL; | |
1878 | ||
5f39d397 | 1879 | if (right) { |
bf77467a | 1880 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
5f39d397 | 1881 | wret = btrfs_cow_block(trans, root, right, |
9631e4cc | 1882 | parent, pslot + 1, &right, |
bf59a5a2 | 1883 | BTRFS_NESTING_RIGHT_COW); |
2cc58cf2 CM |
1884 | if (wret) { |
1885 | ret = wret; | |
1886 | goto enospc; | |
1887 | } | |
1888 | } | |
1889 | ||
1890 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1891 | if (left) { |
1892 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 1893 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
1894 | if (wret < 0) |
1895 | ret = wret; | |
bb803951 | 1896 | } |
79f95c82 CM |
1897 | |
1898 | /* | |
1899 | * then try to empty the right most buffer into the middle | |
1900 | */ | |
5f39d397 | 1901 | if (right) { |
d30a668f | 1902 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1903 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1904 | ret = wret; |
5f39d397 | 1905 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 1906 | btrfs_clean_tree_block(right); |
925baedd | 1907 | btrfs_tree_unlock(right); |
afe5fea7 | 1908 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1909 | root_sub_used(root, right->len); |
5581a51a | 1910 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1911 | free_extent_buffer_stale(right); |
f0486c68 | 1912 | right = NULL; |
bb803951 | 1913 | } else { |
5f39d397 CM |
1914 | struct btrfs_disk_key right_key; |
1915 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1916 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1917 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1918 | BUG_ON(ret < 0); | |
5f39d397 CM |
1919 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1920 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1921 | } |
1922 | } | |
5f39d397 | 1923 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1924 | /* |
1925 | * we're not allowed to leave a node with one item in the | |
1926 | * tree during a delete. A deletion from lower in the tree | |
1927 | * could try to delete the only pointer in this node. | |
1928 | * So, pull some keys from the left. | |
1929 | * There has to be a left pointer at this point because | |
1930 | * otherwise we would have pulled some pointers from the | |
1931 | * right | |
1932 | */ | |
305a26af MF |
1933 | if (!left) { |
1934 | ret = -EROFS; | |
0b246afa | 1935 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1936 | goto enospc; |
1937 | } | |
55d32ed8 | 1938 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1939 | if (wret < 0) { |
79f95c82 | 1940 | ret = wret; |
54aa1f4d CM |
1941 | goto enospc; |
1942 | } | |
bce4eae9 | 1943 | if (wret == 1) { |
d30a668f | 1944 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1945 | if (wret < 0) |
1946 | ret = wret; | |
1947 | } | |
79f95c82 CM |
1948 | BUG_ON(wret == 1); |
1949 | } | |
5f39d397 | 1950 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 1951 | btrfs_clean_tree_block(mid); |
925baedd | 1952 | btrfs_tree_unlock(mid); |
afe5fea7 | 1953 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1954 | root_sub_used(root, mid->len); |
5581a51a | 1955 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1956 | free_extent_buffer_stale(mid); |
f0486c68 | 1957 | mid = NULL; |
79f95c82 CM |
1958 | } else { |
1959 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1960 | struct btrfs_disk_key mid_key; |
1961 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
1962 | ret = tree_mod_log_insert_key(parent, pslot, |
1963 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1964 | BUG_ON(ret < 0); | |
5f39d397 CM |
1965 | btrfs_set_node_key(parent, &mid_key, pslot); |
1966 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1967 | } |
bb803951 | 1968 | |
79f95c82 | 1969 | /* update the path */ |
5f39d397 CM |
1970 | if (left) { |
1971 | if (btrfs_header_nritems(left) > orig_slot) { | |
67439dad | 1972 | atomic_inc(&left->refs); |
925baedd | 1973 | /* left was locked after cow */ |
5f39d397 | 1974 | path->nodes[level] = left; |
bb803951 CM |
1975 | path->slots[level + 1] -= 1; |
1976 | path->slots[level] = orig_slot; | |
925baedd CM |
1977 | if (mid) { |
1978 | btrfs_tree_unlock(mid); | |
5f39d397 | 1979 | free_extent_buffer(mid); |
925baedd | 1980 | } |
bb803951 | 1981 | } else { |
5f39d397 | 1982 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1983 | path->slots[level] = orig_slot; |
1984 | } | |
1985 | } | |
79f95c82 | 1986 | /* double check we haven't messed things up */ |
e20d96d6 | 1987 | if (orig_ptr != |
5f39d397 | 1988 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1989 | BUG(); |
54aa1f4d | 1990 | enospc: |
925baedd CM |
1991 | if (right) { |
1992 | btrfs_tree_unlock(right); | |
5f39d397 | 1993 | free_extent_buffer(right); |
925baedd CM |
1994 | } |
1995 | if (left) { | |
1996 | if (path->nodes[level] != left) | |
1997 | btrfs_tree_unlock(left); | |
5f39d397 | 1998 | free_extent_buffer(left); |
925baedd | 1999 | } |
bb803951 CM |
2000 | return ret; |
2001 | } | |
2002 | ||
d352ac68 CM |
2003 | /* Node balancing for insertion. Here we only split or push nodes around |
2004 | * when they are completely full. This is also done top down, so we | |
2005 | * have to be pessimistic. | |
2006 | */ | |
d397712b | 2007 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2008 | struct btrfs_root *root, |
2009 | struct btrfs_path *path, int level) | |
e66f709b | 2010 | { |
0b246afa | 2011 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2012 | struct extent_buffer *right = NULL; |
2013 | struct extent_buffer *mid; | |
2014 | struct extent_buffer *left = NULL; | |
2015 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2016 | int ret = 0; |
2017 | int wret; | |
2018 | int pslot; | |
2019 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2020 | |
2021 | if (level == 0) | |
2022 | return 1; | |
2023 | ||
5f39d397 | 2024 | mid = path->nodes[level]; |
7bb86316 | 2025 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2026 | |
a05a9bb1 | 2027 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2028 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2029 | pslot = path->slots[level + 1]; |
2030 | } | |
e66f709b | 2031 | |
5f39d397 | 2032 | if (!parent) |
e66f709b | 2033 | return 1; |
e66f709b | 2034 | |
4b231ae4 | 2035 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
2036 | if (IS_ERR(left)) |
2037 | left = NULL; | |
e66f709b CM |
2038 | |
2039 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2040 | if (left) { |
e66f709b | 2041 | u32 left_nr; |
925baedd | 2042 | |
bf77467a | 2043 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
b4ce94de | 2044 | |
5f39d397 | 2045 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2046 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2047 | wret = 1; |
2048 | } else { | |
5f39d397 | 2049 | ret = btrfs_cow_block(trans, root, left, parent, |
9631e4cc | 2050 | pslot - 1, &left, |
bf59a5a2 | 2051 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
2052 | if (ret) |
2053 | wret = 1; | |
2054 | else { | |
d30a668f | 2055 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 2056 | } |
33ade1f8 | 2057 | } |
e66f709b CM |
2058 | if (wret < 0) |
2059 | ret = wret; | |
2060 | if (wret == 0) { | |
5f39d397 | 2061 | struct btrfs_disk_key disk_key; |
e66f709b | 2062 | orig_slot += left_nr; |
5f39d397 | 2063 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2064 | ret = tree_mod_log_insert_key(parent, pslot, |
2065 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2066 | BUG_ON(ret < 0); | |
5f39d397 CM |
2067 | btrfs_set_node_key(parent, &disk_key, pslot); |
2068 | btrfs_mark_buffer_dirty(parent); | |
2069 | if (btrfs_header_nritems(left) > orig_slot) { | |
2070 | path->nodes[level] = left; | |
e66f709b CM |
2071 | path->slots[level + 1] -= 1; |
2072 | path->slots[level] = orig_slot; | |
925baedd | 2073 | btrfs_tree_unlock(mid); |
5f39d397 | 2074 | free_extent_buffer(mid); |
e66f709b CM |
2075 | } else { |
2076 | orig_slot -= | |
5f39d397 | 2077 | btrfs_header_nritems(left); |
e66f709b | 2078 | path->slots[level] = orig_slot; |
925baedd | 2079 | btrfs_tree_unlock(left); |
5f39d397 | 2080 | free_extent_buffer(left); |
e66f709b | 2081 | } |
e66f709b CM |
2082 | return 0; |
2083 | } | |
925baedd | 2084 | btrfs_tree_unlock(left); |
5f39d397 | 2085 | free_extent_buffer(left); |
e66f709b | 2086 | } |
4b231ae4 | 2087 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
2088 | if (IS_ERR(right)) |
2089 | right = NULL; | |
e66f709b CM |
2090 | |
2091 | /* | |
2092 | * then try to empty the right most buffer into the middle | |
2093 | */ | |
5f39d397 | 2094 | if (right) { |
33ade1f8 | 2095 | u32 right_nr; |
b4ce94de | 2096 | |
bf77467a | 2097 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
b4ce94de | 2098 | |
5f39d397 | 2099 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2100 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2101 | wret = 1; |
2102 | } else { | |
5f39d397 CM |
2103 | ret = btrfs_cow_block(trans, root, right, |
2104 | parent, pslot + 1, | |
bf59a5a2 | 2105 | &right, BTRFS_NESTING_RIGHT_COW); |
54aa1f4d CM |
2106 | if (ret) |
2107 | wret = 1; | |
2108 | else { | |
55d32ed8 | 2109 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 2110 | } |
33ade1f8 | 2111 | } |
e66f709b CM |
2112 | if (wret < 0) |
2113 | ret = wret; | |
2114 | if (wret == 0) { | |
5f39d397 CM |
2115 | struct btrfs_disk_key disk_key; |
2116 | ||
2117 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2118 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2119 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2120 | BUG_ON(ret < 0); | |
5f39d397 CM |
2121 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2122 | btrfs_mark_buffer_dirty(parent); | |
2123 | ||
2124 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2125 | path->nodes[level] = right; | |
e66f709b CM |
2126 | path->slots[level + 1] += 1; |
2127 | path->slots[level] = orig_slot - | |
5f39d397 | 2128 | btrfs_header_nritems(mid); |
925baedd | 2129 | btrfs_tree_unlock(mid); |
5f39d397 | 2130 | free_extent_buffer(mid); |
e66f709b | 2131 | } else { |
925baedd | 2132 | btrfs_tree_unlock(right); |
5f39d397 | 2133 | free_extent_buffer(right); |
e66f709b | 2134 | } |
e66f709b CM |
2135 | return 0; |
2136 | } | |
925baedd | 2137 | btrfs_tree_unlock(right); |
5f39d397 | 2138 | free_extent_buffer(right); |
e66f709b | 2139 | } |
e66f709b CM |
2140 | return 1; |
2141 | } | |
2142 | ||
3c69faec | 2143 | /* |
d352ac68 CM |
2144 | * readahead one full node of leaves, finding things that are close |
2145 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2146 | */ |
2ff7e61e | 2147 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2148 | struct btrfs_path *path, |
2149 | int level, int slot, u64 objectid) | |
3c69faec | 2150 | { |
5f39d397 | 2151 | struct extent_buffer *node; |
01f46658 | 2152 | struct btrfs_disk_key disk_key; |
3c69faec | 2153 | u32 nritems; |
3c69faec | 2154 | u64 search; |
a7175319 | 2155 | u64 target; |
6b80053d | 2156 | u64 nread = 0; |
5f39d397 | 2157 | struct extent_buffer *eb; |
6b80053d CM |
2158 | u32 nr; |
2159 | u32 blocksize; | |
2160 | u32 nscan = 0; | |
db94535d | 2161 | |
a6b6e75e | 2162 | if (level != 1) |
6702ed49 CM |
2163 | return; |
2164 | ||
2165 | if (!path->nodes[level]) | |
3c69faec CM |
2166 | return; |
2167 | ||
5f39d397 | 2168 | node = path->nodes[level]; |
925baedd | 2169 | |
3c69faec | 2170 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2171 | blocksize = fs_info->nodesize; |
2172 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2173 | if (eb) { |
2174 | free_extent_buffer(eb); | |
3c69faec CM |
2175 | return; |
2176 | } | |
2177 | ||
a7175319 | 2178 | target = search; |
6b80053d | 2179 | |
5f39d397 | 2180 | nritems = btrfs_header_nritems(node); |
6b80053d | 2181 | nr = slot; |
25b8b936 | 2182 | |
d397712b | 2183 | while (1) { |
e4058b54 | 2184 | if (path->reada == READA_BACK) { |
6b80053d CM |
2185 | if (nr == 0) |
2186 | break; | |
2187 | nr--; | |
e4058b54 | 2188 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2189 | nr++; |
2190 | if (nr >= nritems) | |
2191 | break; | |
3c69faec | 2192 | } |
e4058b54 | 2193 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2194 | btrfs_node_key(node, &disk_key, nr); |
2195 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2196 | break; | |
2197 | } | |
6b80053d | 2198 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2199 | if ((search <= target && target - search <= 65536) || |
2200 | (search > target && search - target <= 65536)) { | |
bfb484d9 | 2201 | btrfs_readahead_node_child(node, nr); |
6b80053d CM |
2202 | nread += blocksize; |
2203 | } | |
2204 | nscan++; | |
a7175319 | 2205 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2206 | break; |
3c69faec CM |
2207 | } |
2208 | } | |
925baedd | 2209 | |
bfb484d9 | 2210 | static noinline void reada_for_balance(struct btrfs_path *path, int level) |
b4ce94de | 2211 | { |
bfb484d9 | 2212 | struct extent_buffer *parent; |
b4ce94de CM |
2213 | int slot; |
2214 | int nritems; | |
b4ce94de | 2215 | |
8c594ea8 | 2216 | parent = path->nodes[level + 1]; |
b4ce94de | 2217 | if (!parent) |
0b08851f | 2218 | return; |
b4ce94de CM |
2219 | |
2220 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2221 | slot = path->slots[level + 1]; |
b4ce94de | 2222 | |
bfb484d9 JB |
2223 | if (slot > 0) |
2224 | btrfs_readahead_node_child(parent, slot - 1); | |
2225 | if (slot + 1 < nritems) | |
2226 | btrfs_readahead_node_child(parent, slot + 1); | |
b4ce94de CM |
2227 | } |
2228 | ||
2229 | ||
d352ac68 | 2230 | /* |
d397712b CM |
2231 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2232 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2233 | * operations on the tree might require changing key pointers higher up in the | |
2234 | * tree. | |
d352ac68 | 2235 | * |
d397712b CM |
2236 | * callers might also have set path->keep_locks, which tells this code to keep |
2237 | * the lock if the path points to the last slot in the block. This is part of | |
2238 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2239 | * |
d397712b CM |
2240 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2241 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2242 | */ |
e02119d5 | 2243 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2244 | int lowest_unlock, int min_write_lock_level, |
2245 | int *write_lock_level) | |
925baedd CM |
2246 | { |
2247 | int i; | |
2248 | int skip_level = level; | |
051e1b9f | 2249 | int no_skips = 0; |
925baedd CM |
2250 | struct extent_buffer *t; |
2251 | ||
2252 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2253 | if (!path->nodes[i]) | |
2254 | break; | |
2255 | if (!path->locks[i]) | |
2256 | break; | |
051e1b9f | 2257 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2258 | skip_level = i + 1; |
2259 | continue; | |
2260 | } | |
051e1b9f | 2261 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2262 | u32 nritems; |
2263 | t = path->nodes[i]; | |
2264 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2265 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2266 | skip_level = i + 1; |
2267 | continue; | |
2268 | } | |
2269 | } | |
051e1b9f CM |
2270 | if (skip_level < i && i >= lowest_unlock) |
2271 | no_skips = 1; | |
2272 | ||
925baedd | 2273 | t = path->nodes[i]; |
d80bb3f9 | 2274 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2275 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2276 | path->locks[i] = 0; |
f7c79f30 CM |
2277 | if (write_lock_level && |
2278 | i > min_write_lock_level && | |
2279 | i <= *write_lock_level) { | |
2280 | *write_lock_level = i - 1; | |
2281 | } | |
925baedd CM |
2282 | } |
2283 | } | |
2284 | } | |
2285 | ||
c8c42864 CM |
2286 | /* |
2287 | * helper function for btrfs_search_slot. The goal is to find a block | |
2288 | * in cache without setting the path to blocking. If we find the block | |
2289 | * we return zero and the path is unchanged. | |
2290 | * | |
2291 | * If we can't find the block, we set the path blocking and do some | |
2292 | * reada. -EAGAIN is returned and the search must be repeated. | |
2293 | */ | |
2294 | static int | |
d07b8528 LB |
2295 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2296 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2297 | const struct btrfs_key *key) |
c8c42864 | 2298 | { |
0b246afa | 2299 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2300 | u64 blocknr; |
2301 | u64 gen; | |
c8c42864 | 2302 | struct extent_buffer *tmp; |
581c1760 | 2303 | struct btrfs_key first_key; |
76a05b35 | 2304 | int ret; |
581c1760 | 2305 | int parent_level; |
c8c42864 | 2306 | |
213ff4b7 NB |
2307 | blocknr = btrfs_node_blockptr(*eb_ret, slot); |
2308 | gen = btrfs_node_ptr_generation(*eb_ret, slot); | |
2309 | parent_level = btrfs_header_level(*eb_ret); | |
2310 | btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); | |
c8c42864 | 2311 | |
0b246afa | 2312 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2313 | if (tmp) { |
b9fab919 | 2314 | /* first we do an atomic uptodate check */ |
bdf7c00e | 2315 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
2316 | /* |
2317 | * Do extra check for first_key, eb can be stale due to | |
2318 | * being cached, read from scrub, or have multiple | |
2319 | * parents (shared tree blocks). | |
2320 | */ | |
e064d5e9 | 2321 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
2322 | parent_level - 1, &first_key, gen)) { |
2323 | free_extent_buffer(tmp); | |
2324 | return -EUCLEAN; | |
2325 | } | |
bdf7c00e JB |
2326 | *eb_ret = tmp; |
2327 | return 0; | |
2328 | } | |
2329 | ||
bdf7c00e | 2330 | /* now we're allowed to do a blocking uptodate check */ |
581c1760 | 2331 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2332 | if (!ret) { |
2333 | *eb_ret = tmp; | |
2334 | return 0; | |
cb44921a | 2335 | } |
bdf7c00e JB |
2336 | free_extent_buffer(tmp); |
2337 | btrfs_release_path(p); | |
2338 | return -EIO; | |
c8c42864 CM |
2339 | } |
2340 | ||
2341 | /* | |
2342 | * reduce lock contention at high levels | |
2343 | * of the btree by dropping locks before | |
76a05b35 CM |
2344 | * we read. Don't release the lock on the current |
2345 | * level because we need to walk this node to figure | |
2346 | * out which blocks to read. | |
c8c42864 | 2347 | */ |
8c594ea8 | 2348 | btrfs_unlock_up_safe(p, level + 1); |
8c594ea8 | 2349 | |
e4058b54 | 2350 | if (p->reada != READA_NONE) |
2ff7e61e | 2351 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2352 | |
76a05b35 | 2353 | ret = -EAGAIN; |
1b7ec85e JB |
2354 | tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, |
2355 | gen, parent_level - 1, &first_key); | |
64c043de | 2356 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2357 | /* |
2358 | * If the read above didn't mark this buffer up to date, | |
2359 | * it will never end up being up to date. Set ret to EIO now | |
2360 | * and give up so that our caller doesn't loop forever | |
2361 | * on our EAGAINs. | |
2362 | */ | |
e6a1d6fd | 2363 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2364 | ret = -EIO; |
c8c42864 | 2365 | free_extent_buffer(tmp); |
c871b0f2 LB |
2366 | } else { |
2367 | ret = PTR_ERR(tmp); | |
76a05b35 | 2368 | } |
02a3307a LB |
2369 | |
2370 | btrfs_release_path(p); | |
76a05b35 | 2371 | return ret; |
c8c42864 CM |
2372 | } |
2373 | ||
2374 | /* | |
2375 | * helper function for btrfs_search_slot. This does all of the checks | |
2376 | * for node-level blocks and does any balancing required based on | |
2377 | * the ins_len. | |
2378 | * | |
2379 | * If no extra work was required, zero is returned. If we had to | |
2380 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2381 | * start over | |
2382 | */ | |
2383 | static int | |
2384 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2385 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2386 | struct extent_buffer *b, int level, int ins_len, |
2387 | int *write_lock_level) | |
c8c42864 | 2388 | { |
0b246afa | 2389 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2390 | int ret; |
0b246afa | 2391 | |
c8c42864 | 2392 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2393 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2394 | int sret; |
2395 | ||
bd681513 CM |
2396 | if (*write_lock_level < level + 1) { |
2397 | *write_lock_level = level + 1; | |
2398 | btrfs_release_path(p); | |
2399 | goto again; | |
2400 | } | |
2401 | ||
bfb484d9 | 2402 | reada_for_balance(p, level); |
c8c42864 | 2403 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2404 | |
2405 | BUG_ON(sret > 0); | |
2406 | if (sret) { | |
2407 | ret = sret; | |
2408 | goto done; | |
2409 | } | |
2410 | b = p->nodes[level]; | |
2411 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2412 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2413 | int sret; |
2414 | ||
bd681513 CM |
2415 | if (*write_lock_level < level + 1) { |
2416 | *write_lock_level = level + 1; | |
2417 | btrfs_release_path(p); | |
2418 | goto again; | |
2419 | } | |
2420 | ||
bfb484d9 | 2421 | reada_for_balance(p, level); |
c8c42864 | 2422 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2423 | |
2424 | if (sret) { | |
2425 | ret = sret; | |
2426 | goto done; | |
2427 | } | |
2428 | b = p->nodes[level]; | |
2429 | if (!b) { | |
b3b4aa74 | 2430 | btrfs_release_path(p); |
c8c42864 CM |
2431 | goto again; |
2432 | } | |
2433 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2434 | } | |
2435 | return 0; | |
2436 | ||
2437 | again: | |
2438 | ret = -EAGAIN; | |
2439 | done: | |
2440 | return ret; | |
2441 | } | |
2442 | ||
381cf658 | 2443 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2444 | u64 iobjectid, u64 ioff, u8 key_type, |
2445 | struct btrfs_key *found_key) | |
2446 | { | |
2447 | int ret; | |
2448 | struct btrfs_key key; | |
2449 | struct extent_buffer *eb; | |
381cf658 DS |
2450 | |
2451 | ASSERT(path); | |
1d4c08e0 | 2452 | ASSERT(found_key); |
e33d5c3d KN |
2453 | |
2454 | key.type = key_type; | |
2455 | key.objectid = iobjectid; | |
2456 | key.offset = ioff; | |
2457 | ||
2458 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2459 | if (ret < 0) |
e33d5c3d KN |
2460 | return ret; |
2461 | ||
2462 | eb = path->nodes[0]; | |
2463 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2464 | ret = btrfs_next_leaf(fs_root, path); | |
2465 | if (ret) | |
2466 | return ret; | |
2467 | eb = path->nodes[0]; | |
2468 | } | |
2469 | ||
2470 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2471 | if (found_key->type != key.type || | |
2472 | found_key->objectid != key.objectid) | |
2473 | return 1; | |
2474 | ||
2475 | return 0; | |
2476 | } | |
2477 | ||
1fc28d8e LB |
2478 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2479 | struct btrfs_path *p, | |
2480 | int write_lock_level) | |
2481 | { | |
2482 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2483 | struct extent_buffer *b; | |
2484 | int root_lock; | |
2485 | int level = 0; | |
2486 | ||
2487 | /* We try very hard to do read locks on the root */ | |
2488 | root_lock = BTRFS_READ_LOCK; | |
2489 | ||
2490 | if (p->search_commit_root) { | |
be6821f8 FM |
2491 | /* |
2492 | * The commit roots are read only so we always do read locks, | |
2493 | * and we always must hold the commit_root_sem when doing | |
2494 | * searches on them, the only exception is send where we don't | |
2495 | * want to block transaction commits for a long time, so | |
2496 | * we need to clone the commit root in order to avoid races | |
2497 | * with transaction commits that create a snapshot of one of | |
2498 | * the roots used by a send operation. | |
2499 | */ | |
2500 | if (p->need_commit_sem) { | |
1fc28d8e | 2501 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 2502 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 2503 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
2504 | if (!b) |
2505 | return ERR_PTR(-ENOMEM); | |
2506 | ||
2507 | } else { | |
2508 | b = root->commit_root; | |
67439dad | 2509 | atomic_inc(&b->refs); |
be6821f8 FM |
2510 | } |
2511 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
2512 | /* |
2513 | * Ensure that all callers have set skip_locking when | |
2514 | * p->search_commit_root = 1. | |
2515 | */ | |
2516 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2517 | |
2518 | goto out; | |
2519 | } | |
2520 | ||
2521 | if (p->skip_locking) { | |
2522 | b = btrfs_root_node(root); | |
2523 | level = btrfs_header_level(b); | |
2524 | goto out; | |
2525 | } | |
2526 | ||
2527 | /* | |
662c653b LB |
2528 | * If the level is set to maximum, we can skip trying to get the read |
2529 | * lock. | |
1fc28d8e | 2530 | */ |
662c653b LB |
2531 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2532 | /* | |
2533 | * We don't know the level of the root node until we actually | |
2534 | * have it read locked | |
2535 | */ | |
51899412 | 2536 | b = __btrfs_read_lock_root_node(root, p->recurse); |
662c653b LB |
2537 | level = btrfs_header_level(b); |
2538 | if (level > write_lock_level) | |
2539 | goto out; | |
2540 | ||
2541 | /* Whoops, must trade for write lock */ | |
2542 | btrfs_tree_read_unlock(b); | |
2543 | free_extent_buffer(b); | |
2544 | } | |
1fc28d8e | 2545 | |
1fc28d8e LB |
2546 | b = btrfs_lock_root_node(root); |
2547 | root_lock = BTRFS_WRITE_LOCK; | |
2548 | ||
2549 | /* The level might have changed, check again */ | |
2550 | level = btrfs_header_level(b); | |
2551 | ||
2552 | out: | |
2553 | p->nodes[level] = b; | |
2554 | if (!p->skip_locking) | |
2555 | p->locks[level] = root_lock; | |
2556 | /* | |
2557 | * Callers are responsible for dropping b's references. | |
2558 | */ | |
2559 | return b; | |
2560 | } | |
2561 | ||
2562 | ||
74123bd7 | 2563 | /* |
4271ecea NB |
2564 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2565 | * modifications to preserve tree invariants. | |
74123bd7 | 2566 | * |
4271ecea NB |
2567 | * @trans: Handle of transaction, used when modifying the tree |
2568 | * @p: Holds all btree nodes along the search path | |
2569 | * @root: The root node of the tree | |
2570 | * @key: The key we are looking for | |
2571 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2572 | * deletions it's -1. 0 for plain searches | |
2573 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2574 | * when modifying the tree. | |
97571fd0 | 2575 | * |
4271ecea NB |
2576 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2577 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2578 | * | |
2579 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2580 | * of the path (level 0) | |
2581 | * | |
2582 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2583 | * points to the slot where it should be inserted | |
2584 | * | |
2585 | * If an error is encountered while searching the tree a negative error number | |
2586 | * is returned | |
74123bd7 | 2587 | */ |
310712b2 OS |
2588 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2589 | const struct btrfs_key *key, struct btrfs_path *p, | |
2590 | int ins_len, int cow) | |
be0e5c09 | 2591 | { |
5f39d397 | 2592 | struct extent_buffer *b; |
be0e5c09 CM |
2593 | int slot; |
2594 | int ret; | |
33c66f43 | 2595 | int err; |
be0e5c09 | 2596 | int level; |
925baedd | 2597 | int lowest_unlock = 1; |
bd681513 CM |
2598 | /* everything at write_lock_level or lower must be write locked */ |
2599 | int write_lock_level = 0; | |
9f3a7427 | 2600 | u8 lowest_level = 0; |
f7c79f30 | 2601 | int min_write_lock_level; |
d7396f07 | 2602 | int prev_cmp; |
9f3a7427 | 2603 | |
6702ed49 | 2604 | lowest_level = p->lowest_level; |
323ac95b | 2605 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2606 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2607 | BUG_ON(!cow && ins_len); |
25179201 | 2608 | |
bd681513 | 2609 | if (ins_len < 0) { |
925baedd | 2610 | lowest_unlock = 2; |
65b51a00 | 2611 | |
bd681513 CM |
2612 | /* when we are removing items, we might have to go up to level |
2613 | * two as we update tree pointers Make sure we keep write | |
2614 | * for those levels as well | |
2615 | */ | |
2616 | write_lock_level = 2; | |
2617 | } else if (ins_len > 0) { | |
2618 | /* | |
2619 | * for inserting items, make sure we have a write lock on | |
2620 | * level 1 so we can update keys | |
2621 | */ | |
2622 | write_lock_level = 1; | |
2623 | } | |
2624 | ||
2625 | if (!cow) | |
2626 | write_lock_level = -1; | |
2627 | ||
09a2a8f9 | 2628 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2629 | write_lock_level = BTRFS_MAX_LEVEL; |
2630 | ||
f7c79f30 CM |
2631 | min_write_lock_level = write_lock_level; |
2632 | ||
bb803951 | 2633 | again: |
d7396f07 | 2634 | prev_cmp = -1; |
1fc28d8e | 2635 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2636 | if (IS_ERR(b)) { |
2637 | ret = PTR_ERR(b); | |
2638 | goto done; | |
2639 | } | |
925baedd | 2640 | |
eb60ceac | 2641 | while (b) { |
f624d976 QW |
2642 | int dec = 0; |
2643 | ||
5f39d397 | 2644 | level = btrfs_header_level(b); |
65b51a00 | 2645 | |
02217ed2 | 2646 | if (cow) { |
9ea2c7c9 NB |
2647 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2648 | ||
c8c42864 CM |
2649 | /* |
2650 | * if we don't really need to cow this block | |
2651 | * then we don't want to set the path blocking, | |
2652 | * so we test it here | |
2653 | */ | |
64c12921 JM |
2654 | if (!should_cow_block(trans, root, b)) { |
2655 | trans->dirty = true; | |
65b51a00 | 2656 | goto cow_done; |
64c12921 | 2657 | } |
5d4f98a2 | 2658 | |
bd681513 CM |
2659 | /* |
2660 | * must have write locks on this node and the | |
2661 | * parent | |
2662 | */ | |
5124e00e JB |
2663 | if (level > write_lock_level || |
2664 | (level + 1 > write_lock_level && | |
2665 | level + 1 < BTRFS_MAX_LEVEL && | |
2666 | p->nodes[level + 1])) { | |
bd681513 CM |
2667 | write_lock_level = level + 1; |
2668 | btrfs_release_path(p); | |
2669 | goto again; | |
2670 | } | |
2671 | ||
9ea2c7c9 NB |
2672 | if (last_level) |
2673 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
9631e4cc JB |
2674 | &b, |
2675 | BTRFS_NESTING_COW); | |
9ea2c7c9 NB |
2676 | else |
2677 | err = btrfs_cow_block(trans, root, b, | |
2678 | p->nodes[level + 1], | |
9631e4cc JB |
2679 | p->slots[level + 1], &b, |
2680 | BTRFS_NESTING_COW); | |
33c66f43 | 2681 | if (err) { |
33c66f43 | 2682 | ret = err; |
65b51a00 | 2683 | goto done; |
54aa1f4d | 2684 | } |
02217ed2 | 2685 | } |
65b51a00 | 2686 | cow_done: |
eb60ceac | 2687 | p->nodes[level] = b; |
52398340 LB |
2688 | /* |
2689 | * Leave path with blocking locks to avoid massive | |
2690 | * lock context switch, this is made on purpose. | |
2691 | */ | |
b4ce94de CM |
2692 | |
2693 | /* | |
2694 | * we have a lock on b and as long as we aren't changing | |
2695 | * the tree, there is no way to for the items in b to change. | |
2696 | * It is safe to drop the lock on our parent before we | |
2697 | * go through the expensive btree search on b. | |
2698 | * | |
eb653de1 FDBM |
2699 | * If we're inserting or deleting (ins_len != 0), then we might |
2700 | * be changing slot zero, which may require changing the parent. | |
2701 | * So, we can't drop the lock until after we know which slot | |
2702 | * we're operating on. | |
b4ce94de | 2703 | */ |
eb653de1 FDBM |
2704 | if (!ins_len && !p->keep_locks) { |
2705 | int u = level + 1; | |
2706 | ||
2707 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2708 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2709 | p->locks[u] = 0; | |
2710 | } | |
2711 | } | |
b4ce94de | 2712 | |
995e9a16 NB |
2713 | /* |
2714 | * If btrfs_bin_search returns an exact match (prev_cmp == 0) | |
2715 | * we can safely assume the target key will always be in slot 0 | |
2716 | * on lower levels due to the invariants BTRFS' btree provides, | |
2717 | * namely that a btrfs_key_ptr entry always points to the | |
2718 | * lowest key in the child node, thus we can skip searching | |
2719 | * lower levels | |
2720 | */ | |
2721 | if (prev_cmp == 0) { | |
2722 | slot = 0; | |
2723 | ret = 0; | |
2724 | } else { | |
2725 | ret = btrfs_bin_search(b, key, &slot); | |
2726 | prev_cmp = ret; | |
2727 | if (ret < 0) | |
2728 | goto done; | |
2729 | } | |
b4ce94de | 2730 | |
f624d976 | 2731 | if (level == 0) { |
be0e5c09 | 2732 | p->slots[level] = slot; |
87b29b20 | 2733 | if (ins_len > 0 && |
e902baac | 2734 | btrfs_leaf_free_space(b) < ins_len) { |
bd681513 CM |
2735 | if (write_lock_level < 1) { |
2736 | write_lock_level = 1; | |
2737 | btrfs_release_path(p); | |
2738 | goto again; | |
2739 | } | |
2740 | ||
33c66f43 YZ |
2741 | err = split_leaf(trans, root, key, |
2742 | p, ins_len, ret == 0); | |
b4ce94de | 2743 | |
33c66f43 YZ |
2744 | BUG_ON(err > 0); |
2745 | if (err) { | |
2746 | ret = err; | |
65b51a00 CM |
2747 | goto done; |
2748 | } | |
5c680ed6 | 2749 | } |
459931ec | 2750 | if (!p->search_for_split) |
f7c79f30 | 2751 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2752 | min_write_lock_level, NULL); |
65b51a00 | 2753 | goto done; |
be0e5c09 | 2754 | } |
f624d976 QW |
2755 | if (ret && slot > 0) { |
2756 | dec = 1; | |
2757 | slot--; | |
2758 | } | |
2759 | p->slots[level] = slot; | |
2760 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
2761 | &write_lock_level); | |
2762 | if (err == -EAGAIN) | |
2763 | goto again; | |
2764 | if (err) { | |
2765 | ret = err; | |
2766 | goto done; | |
2767 | } | |
2768 | b = p->nodes[level]; | |
2769 | slot = p->slots[level]; | |
2770 | ||
2771 | /* | |
2772 | * Slot 0 is special, if we change the key we have to update | |
2773 | * the parent pointer which means we must have a write lock on | |
2774 | * the parent | |
2775 | */ | |
2776 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
2777 | write_lock_level = level + 1; | |
2778 | btrfs_release_path(p); | |
2779 | goto again; | |
2780 | } | |
2781 | ||
2782 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
2783 | &write_lock_level); | |
2784 | ||
2785 | if (level == lowest_level) { | |
2786 | if (dec) | |
2787 | p->slots[level]++; | |
2788 | goto done; | |
2789 | } | |
2790 | ||
2791 | err = read_block_for_search(root, p, &b, level, slot, key); | |
2792 | if (err == -EAGAIN) | |
2793 | goto again; | |
2794 | if (err) { | |
2795 | ret = err; | |
2796 | goto done; | |
2797 | } | |
2798 | ||
2799 | if (!p->skip_locking) { | |
2800 | level = btrfs_header_level(b); | |
2801 | if (level <= write_lock_level) { | |
ac5887c8 | 2802 | btrfs_tree_lock(b); |
f624d976 QW |
2803 | p->locks[level] = BTRFS_WRITE_LOCK; |
2804 | } else { | |
ac5887c8 JB |
2805 | __btrfs_tree_read_lock(b, BTRFS_NESTING_NORMAL, |
2806 | p->recurse); | |
f624d976 QW |
2807 | p->locks[level] = BTRFS_READ_LOCK; |
2808 | } | |
2809 | p->nodes[level] = b; | |
2810 | } | |
be0e5c09 | 2811 | } |
65b51a00 CM |
2812 | ret = 1; |
2813 | done: | |
5f5bc6b1 | 2814 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2815 | btrfs_release_path(p); |
65b51a00 | 2816 | return ret; |
be0e5c09 CM |
2817 | } |
2818 | ||
5d9e75c4 JS |
2819 | /* |
2820 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2821 | * current state of the tree together with the operations recorded in the tree | |
2822 | * modification log to search for the key in a previous version of this tree, as | |
2823 | * denoted by the time_seq parameter. | |
2824 | * | |
2825 | * Naturally, there is no support for insert, delete or cow operations. | |
2826 | * | |
2827 | * The resulting path and return value will be set up as if we called | |
2828 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2829 | */ | |
310712b2 | 2830 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2831 | struct btrfs_path *p, u64 time_seq) |
2832 | { | |
0b246afa | 2833 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2834 | struct extent_buffer *b; |
2835 | int slot; | |
2836 | int ret; | |
2837 | int err; | |
2838 | int level; | |
2839 | int lowest_unlock = 1; | |
2840 | u8 lowest_level = 0; | |
2841 | ||
2842 | lowest_level = p->lowest_level; | |
2843 | WARN_ON(p->nodes[0] != NULL); | |
2844 | ||
2845 | if (p->search_commit_root) { | |
2846 | BUG_ON(time_seq); | |
2847 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2848 | } | |
2849 | ||
2850 | again: | |
5d9e75c4 | 2851 | b = get_old_root(root, time_seq); |
315bed43 NB |
2852 | if (!b) { |
2853 | ret = -EIO; | |
2854 | goto done; | |
2855 | } | |
5d9e75c4 | 2856 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2857 | p->locks[level] = BTRFS_READ_LOCK; |
2858 | ||
2859 | while (b) { | |
abe9339d QW |
2860 | int dec = 0; |
2861 | ||
5d9e75c4 JS |
2862 | level = btrfs_header_level(b); |
2863 | p->nodes[level] = b; | |
5d9e75c4 JS |
2864 | |
2865 | /* | |
2866 | * we have a lock on b and as long as we aren't changing | |
2867 | * the tree, there is no way to for the items in b to change. | |
2868 | * It is safe to drop the lock on our parent before we | |
2869 | * go through the expensive btree search on b. | |
2870 | */ | |
2871 | btrfs_unlock_up_safe(p, level + 1); | |
2872 | ||
995e9a16 | 2873 | ret = btrfs_bin_search(b, key, &slot); |
cbca7d59 FM |
2874 | if (ret < 0) |
2875 | goto done; | |
5d9e75c4 | 2876 | |
abe9339d | 2877 | if (level == 0) { |
5d9e75c4 JS |
2878 | p->slots[level] = slot; |
2879 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
2880 | goto done; |
2881 | } | |
5d9e75c4 | 2882 | |
abe9339d QW |
2883 | if (ret && slot > 0) { |
2884 | dec = 1; | |
2885 | slot--; | |
2886 | } | |
2887 | p->slots[level] = slot; | |
2888 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 2889 | |
abe9339d QW |
2890 | if (level == lowest_level) { |
2891 | if (dec) | |
2892 | p->slots[level]++; | |
2893 | goto done; | |
2894 | } | |
5d9e75c4 | 2895 | |
abe9339d QW |
2896 | err = read_block_for_search(root, p, &b, level, slot, key); |
2897 | if (err == -EAGAIN) | |
2898 | goto again; | |
2899 | if (err) { | |
2900 | ret = err; | |
5d9e75c4 JS |
2901 | goto done; |
2902 | } | |
abe9339d QW |
2903 | |
2904 | level = btrfs_header_level(b); | |
ac5887c8 | 2905 | btrfs_tree_read_lock(b); |
abe9339d QW |
2906 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
2907 | if (!b) { | |
2908 | ret = -ENOMEM; | |
2909 | goto done; | |
2910 | } | |
2911 | p->locks[level] = BTRFS_READ_LOCK; | |
2912 | p->nodes[level] = b; | |
5d9e75c4 JS |
2913 | } |
2914 | ret = 1; | |
2915 | done: | |
5d9e75c4 JS |
2916 | if (ret < 0) |
2917 | btrfs_release_path(p); | |
2918 | ||
2919 | return ret; | |
2920 | } | |
2921 | ||
2f38b3e1 AJ |
2922 | /* |
2923 | * helper to use instead of search slot if no exact match is needed but | |
2924 | * instead the next or previous item should be returned. | |
2925 | * When find_higher is true, the next higher item is returned, the next lower | |
2926 | * otherwise. | |
2927 | * When return_any and find_higher are both true, and no higher item is found, | |
2928 | * return the next lower instead. | |
2929 | * When return_any is true and find_higher is false, and no lower item is found, | |
2930 | * return the next higher instead. | |
2931 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2932 | * < 0 on error | |
2933 | */ | |
2934 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
2935 | const struct btrfs_key *key, |
2936 | struct btrfs_path *p, int find_higher, | |
2937 | int return_any) | |
2f38b3e1 AJ |
2938 | { |
2939 | int ret; | |
2940 | struct extent_buffer *leaf; | |
2941 | ||
2942 | again: | |
2943 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2944 | if (ret <= 0) | |
2945 | return ret; | |
2946 | /* | |
2947 | * a return value of 1 means the path is at the position where the | |
2948 | * item should be inserted. Normally this is the next bigger item, | |
2949 | * but in case the previous item is the last in a leaf, path points | |
2950 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2951 | * item. | |
2952 | */ | |
2953 | leaf = p->nodes[0]; | |
2954 | ||
2955 | if (find_higher) { | |
2956 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2957 | ret = btrfs_next_leaf(root, p); | |
2958 | if (ret <= 0) | |
2959 | return ret; | |
2960 | if (!return_any) | |
2961 | return 1; | |
2962 | /* | |
2963 | * no higher item found, return the next | |
2964 | * lower instead | |
2965 | */ | |
2966 | return_any = 0; | |
2967 | find_higher = 0; | |
2968 | btrfs_release_path(p); | |
2969 | goto again; | |
2970 | } | |
2971 | } else { | |
e6793769 AJ |
2972 | if (p->slots[0] == 0) { |
2973 | ret = btrfs_prev_leaf(root, p); | |
2974 | if (ret < 0) | |
2975 | return ret; | |
2976 | if (!ret) { | |
23c6bf6a FDBM |
2977 | leaf = p->nodes[0]; |
2978 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
2979 | p->slots[0]--; | |
e6793769 | 2980 | return 0; |
2f38b3e1 | 2981 | } |
e6793769 AJ |
2982 | if (!return_any) |
2983 | return 1; | |
2984 | /* | |
2985 | * no lower item found, return the next | |
2986 | * higher instead | |
2987 | */ | |
2988 | return_any = 0; | |
2989 | find_higher = 1; | |
2990 | btrfs_release_path(p); | |
2991 | goto again; | |
2992 | } else { | |
2f38b3e1 AJ |
2993 | --p->slots[0]; |
2994 | } | |
2995 | } | |
2996 | return 0; | |
2997 | } | |
2998 | ||
74123bd7 CM |
2999 | /* |
3000 | * adjust the pointers going up the tree, starting at level | |
3001 | * making sure the right key of each node is points to 'key'. | |
3002 | * This is used after shifting pointers to the left, so it stops | |
3003 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3004 | * higher levels | |
aa5d6bed | 3005 | * |
74123bd7 | 3006 | */ |
b167fa91 | 3007 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3008 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3009 | { |
3010 | int i; | |
5f39d397 | 3011 | struct extent_buffer *t; |
0e82bcfe | 3012 | int ret; |
5f39d397 | 3013 | |
234b63a0 | 3014 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3015 | int tslot = path->slots[i]; |
0e82bcfe | 3016 | |
eb60ceac | 3017 | if (!path->nodes[i]) |
be0e5c09 | 3018 | break; |
5f39d397 | 3019 | t = path->nodes[i]; |
0e82bcfe DS |
3020 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3021 | GFP_ATOMIC); | |
3022 | BUG_ON(ret < 0); | |
5f39d397 | 3023 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3024 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3025 | if (tslot != 0) |
3026 | break; | |
3027 | } | |
3028 | } | |
3029 | ||
31840ae1 ZY |
3030 | /* |
3031 | * update item key. | |
3032 | * | |
3033 | * This function isn't completely safe. It's the caller's responsibility | |
3034 | * that the new key won't break the order | |
3035 | */ | |
b7a0365e DD |
3036 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3037 | struct btrfs_path *path, | |
310712b2 | 3038 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3039 | { |
3040 | struct btrfs_disk_key disk_key; | |
3041 | struct extent_buffer *eb; | |
3042 | int slot; | |
3043 | ||
3044 | eb = path->nodes[0]; | |
3045 | slot = path->slots[0]; | |
3046 | if (slot > 0) { | |
3047 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
3048 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
3049 | btrfs_crit(fs_info, | |
3050 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3051 | slot, btrfs_disk_key_objectid(&disk_key), | |
3052 | btrfs_disk_key_type(&disk_key), | |
3053 | btrfs_disk_key_offset(&disk_key), | |
3054 | new_key->objectid, new_key->type, | |
3055 | new_key->offset); | |
3056 | btrfs_print_leaf(eb); | |
3057 | BUG(); | |
3058 | } | |
31840ae1 ZY |
3059 | } |
3060 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3061 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
3062 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
3063 | btrfs_crit(fs_info, | |
3064 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3065 | slot, btrfs_disk_key_objectid(&disk_key), | |
3066 | btrfs_disk_key_type(&disk_key), | |
3067 | btrfs_disk_key_offset(&disk_key), | |
3068 | new_key->objectid, new_key->type, | |
3069 | new_key->offset); | |
3070 | btrfs_print_leaf(eb); | |
3071 | BUG(); | |
3072 | } | |
31840ae1 ZY |
3073 | } |
3074 | ||
3075 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3076 | btrfs_set_item_key(eb, &disk_key, slot); | |
3077 | btrfs_mark_buffer_dirty(eb); | |
3078 | if (slot == 0) | |
b167fa91 | 3079 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3080 | } |
3081 | ||
d16c702f QW |
3082 | /* |
3083 | * Check key order of two sibling extent buffers. | |
3084 | * | |
3085 | * Return true if something is wrong. | |
3086 | * Return false if everything is fine. | |
3087 | * | |
3088 | * Tree-checker only works inside one tree block, thus the following | |
3089 | * corruption can not be detected by tree-checker: | |
3090 | * | |
3091 | * Leaf @left | Leaf @right | |
3092 | * -------------------------------------------------------------- | |
3093 | * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | | |
3094 | * | |
3095 | * Key f6 in leaf @left itself is valid, but not valid when the next | |
3096 | * key in leaf @right is 7. | |
3097 | * This can only be checked at tree block merge time. | |
3098 | * And since tree checker has ensured all key order in each tree block | |
3099 | * is correct, we only need to bother the last key of @left and the first | |
3100 | * key of @right. | |
3101 | */ | |
3102 | static bool check_sibling_keys(struct extent_buffer *left, | |
3103 | struct extent_buffer *right) | |
3104 | { | |
3105 | struct btrfs_key left_last; | |
3106 | struct btrfs_key right_first; | |
3107 | int level = btrfs_header_level(left); | |
3108 | int nr_left = btrfs_header_nritems(left); | |
3109 | int nr_right = btrfs_header_nritems(right); | |
3110 | ||
3111 | /* No key to check in one of the tree blocks */ | |
3112 | if (!nr_left || !nr_right) | |
3113 | return false; | |
3114 | ||
3115 | if (level) { | |
3116 | btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); | |
3117 | btrfs_node_key_to_cpu(right, &right_first, 0); | |
3118 | } else { | |
3119 | btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); | |
3120 | btrfs_item_key_to_cpu(right, &right_first, 0); | |
3121 | } | |
3122 | ||
3123 | if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { | |
3124 | btrfs_crit(left->fs_info, | |
3125 | "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", | |
3126 | left_last.objectid, left_last.type, | |
3127 | left_last.offset, right_first.objectid, | |
3128 | right_first.type, right_first.offset); | |
3129 | return true; | |
3130 | } | |
3131 | return false; | |
3132 | } | |
3133 | ||
74123bd7 CM |
3134 | /* |
3135 | * try to push data from one node into the next node left in the | |
79f95c82 | 3136 | * tree. |
aa5d6bed CM |
3137 | * |
3138 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3139 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3140 | */ |
98ed5174 | 3141 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 3142 | struct extent_buffer *dst, |
971a1f66 | 3143 | struct extent_buffer *src, int empty) |
be0e5c09 | 3144 | { |
d30a668f | 3145 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 3146 | int push_items = 0; |
bb803951 CM |
3147 | int src_nritems; |
3148 | int dst_nritems; | |
aa5d6bed | 3149 | int ret = 0; |
be0e5c09 | 3150 | |
5f39d397 CM |
3151 | src_nritems = btrfs_header_nritems(src); |
3152 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3153 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3154 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3155 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3156 | |
bce4eae9 | 3157 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3158 | return 1; |
3159 | ||
d397712b | 3160 | if (push_items <= 0) |
be0e5c09 CM |
3161 | return 1; |
3162 | ||
bce4eae9 | 3163 | if (empty) { |
971a1f66 | 3164 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3165 | if (push_items < src_nritems) { |
3166 | /* leave at least 8 pointers in the node if | |
3167 | * we aren't going to empty it | |
3168 | */ | |
3169 | if (src_nritems - push_items < 8) { | |
3170 | if (push_items <= 8) | |
3171 | return 1; | |
3172 | push_items -= 8; | |
3173 | } | |
3174 | } | |
3175 | } else | |
3176 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3177 | |
d16c702f QW |
3178 | /* dst is the left eb, src is the middle eb */ |
3179 | if (check_sibling_keys(dst, src)) { | |
3180 | ret = -EUCLEAN; | |
3181 | btrfs_abort_transaction(trans, ret); | |
3182 | return ret; | |
3183 | } | |
ed874f0d | 3184 | ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 3185 | if (ret) { |
66642832 | 3186 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3187 | return ret; |
3188 | } | |
5f39d397 CM |
3189 | copy_extent_buffer(dst, src, |
3190 | btrfs_node_key_ptr_offset(dst_nritems), | |
3191 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3192 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3193 | |
bb803951 | 3194 | if (push_items < src_nritems) { |
57911b8b | 3195 | /* |
bf1d3425 DS |
3196 | * Don't call tree_mod_log_insert_move here, key removal was |
3197 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3198 | */ |
5f39d397 CM |
3199 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3200 | btrfs_node_key_ptr_offset(push_items), | |
3201 | (src_nritems - push_items) * | |
3202 | sizeof(struct btrfs_key_ptr)); | |
3203 | } | |
3204 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3205 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3206 | btrfs_mark_buffer_dirty(src); | |
3207 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3208 | |
79f95c82 CM |
3209 | return ret; |
3210 | } | |
3211 | ||
3212 | /* | |
3213 | * try to push data from one node into the next node right in the | |
3214 | * tree. | |
3215 | * | |
3216 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3217 | * error, and > 0 if there was no room in the right hand block. | |
3218 | * | |
3219 | * this will only push up to 1/2 the contents of the left node over | |
3220 | */ | |
5f39d397 | 3221 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3222 | struct extent_buffer *dst, |
3223 | struct extent_buffer *src) | |
79f95c82 | 3224 | { |
55d32ed8 | 3225 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
3226 | int push_items = 0; |
3227 | int max_push; | |
3228 | int src_nritems; | |
3229 | int dst_nritems; | |
3230 | int ret = 0; | |
79f95c82 | 3231 | |
7bb86316 CM |
3232 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3233 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3234 | ||
5f39d397 CM |
3235 | src_nritems = btrfs_header_nritems(src); |
3236 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3237 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3238 | if (push_items <= 0) |
79f95c82 | 3239 | return 1; |
bce4eae9 | 3240 | |
d397712b | 3241 | if (src_nritems < 4) |
bce4eae9 | 3242 | return 1; |
79f95c82 CM |
3243 | |
3244 | max_push = src_nritems / 2 + 1; | |
3245 | /* don't try to empty the node */ | |
d397712b | 3246 | if (max_push >= src_nritems) |
79f95c82 | 3247 | return 1; |
252c38f0 | 3248 | |
79f95c82 CM |
3249 | if (max_push < push_items) |
3250 | push_items = max_push; | |
3251 | ||
d16c702f QW |
3252 | /* dst is the right eb, src is the middle eb */ |
3253 | if (check_sibling_keys(src, dst)) { | |
3254 | ret = -EUCLEAN; | |
3255 | btrfs_abort_transaction(trans, ret); | |
3256 | return ret; | |
3257 | } | |
bf1d3425 DS |
3258 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3259 | BUG_ON(ret < 0); | |
5f39d397 CM |
3260 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3261 | btrfs_node_key_ptr_offset(0), | |
3262 | (dst_nritems) * | |
3263 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3264 | |
ed874f0d DS |
3265 | ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
3266 | push_items); | |
5de865ee | 3267 | if (ret) { |
66642832 | 3268 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3269 | return ret; |
3270 | } | |
5f39d397 CM |
3271 | copy_extent_buffer(dst, src, |
3272 | btrfs_node_key_ptr_offset(0), | |
3273 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3274 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3275 | |
5f39d397 CM |
3276 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3277 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3278 | |
5f39d397 CM |
3279 | btrfs_mark_buffer_dirty(src); |
3280 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3281 | |
aa5d6bed | 3282 | return ret; |
be0e5c09 CM |
3283 | } |
3284 | ||
97571fd0 CM |
3285 | /* |
3286 | * helper function to insert a new root level in the tree. | |
3287 | * A new node is allocated, and a single item is inserted to | |
3288 | * point to the existing root | |
aa5d6bed CM |
3289 | * |
3290 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3291 | */ |
d397712b | 3292 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3293 | struct btrfs_root *root, |
fdd99c72 | 3294 | struct btrfs_path *path, int level) |
5c680ed6 | 3295 | { |
0b246afa | 3296 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3297 | u64 lower_gen; |
5f39d397 CM |
3298 | struct extent_buffer *lower; |
3299 | struct extent_buffer *c; | |
925baedd | 3300 | struct extent_buffer *old; |
5f39d397 | 3301 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3302 | int ret; |
5c680ed6 CM |
3303 | |
3304 | BUG_ON(path->nodes[level]); | |
3305 | BUG_ON(path->nodes[level-1] != root->node); | |
3306 | ||
7bb86316 CM |
3307 | lower = path->nodes[level-1]; |
3308 | if (level == 1) | |
3309 | btrfs_item_key(lower, &lower_key, 0); | |
3310 | else | |
3311 | btrfs_node_key(lower, &lower_key, 0); | |
3312 | ||
a6279470 | 3313 | c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, |
9631e4cc | 3314 | root->node->start, 0, |
cf6f34aa | 3315 | BTRFS_NESTING_NEW_ROOT); |
5f39d397 CM |
3316 | if (IS_ERR(c)) |
3317 | return PTR_ERR(c); | |
925baedd | 3318 | |
0b246afa | 3319 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3320 | |
5f39d397 | 3321 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3322 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3323 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3324 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3325 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3326 | |
3327 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3328 | |
5f39d397 | 3329 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3330 | |
925baedd | 3331 | old = root->node; |
d9d19a01 DS |
3332 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3333 | BUG_ON(ret < 0); | |
240f62c8 | 3334 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3335 | |
3336 | /* the super has an extra ref to root->node */ | |
3337 | free_extent_buffer(old); | |
3338 | ||
0b86a832 | 3339 | add_root_to_dirty_list(root); |
67439dad | 3340 | atomic_inc(&c->refs); |
5f39d397 | 3341 | path->nodes[level] = c; |
ac5887c8 | 3342 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3343 | path->slots[level] = 0; |
3344 | return 0; | |
3345 | } | |
3346 | ||
74123bd7 CM |
3347 | /* |
3348 | * worker function to insert a single pointer in a node. | |
3349 | * the node should have enough room for the pointer already | |
97571fd0 | 3350 | * |
74123bd7 CM |
3351 | * slot and level indicate where you want the key to go, and |
3352 | * blocknr is the block the key points to. | |
3353 | */ | |
143bede5 | 3354 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 3355 | struct btrfs_path *path, |
143bede5 | 3356 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3357 | int slot, int level) |
74123bd7 | 3358 | { |
5f39d397 | 3359 | struct extent_buffer *lower; |
74123bd7 | 3360 | int nritems; |
f3ea38da | 3361 | int ret; |
5c680ed6 CM |
3362 | |
3363 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3364 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3365 | lower = path->nodes[level]; |
3366 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3367 | BUG_ON(slot > nritems); |
6ad3cf6d | 3368 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 3369 | if (slot != nritems) { |
bf1d3425 DS |
3370 | if (level) { |
3371 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3372 | nritems - slot); |
bf1d3425 DS |
3373 | BUG_ON(ret < 0); |
3374 | } | |
5f39d397 CM |
3375 | memmove_extent_buffer(lower, |
3376 | btrfs_node_key_ptr_offset(slot + 1), | |
3377 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3378 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3379 | } |
c3e06965 | 3380 | if (level) { |
e09c2efe DS |
3381 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3382 | GFP_NOFS); | |
f3ea38da JS |
3383 | BUG_ON(ret < 0); |
3384 | } | |
5f39d397 | 3385 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3386 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3387 | WARN_ON(trans->transid == 0); |
3388 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3389 | btrfs_set_header_nritems(lower, nritems + 1); |
3390 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3391 | } |
3392 | ||
97571fd0 CM |
3393 | /* |
3394 | * split the node at the specified level in path in two. | |
3395 | * The path is corrected to point to the appropriate node after the split | |
3396 | * | |
3397 | * Before splitting this tries to make some room in the node by pushing | |
3398 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3399 | * |
3400 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3401 | */ |
e02119d5 CM |
3402 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3403 | struct btrfs_root *root, | |
3404 | struct btrfs_path *path, int level) | |
be0e5c09 | 3405 | { |
0b246afa | 3406 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3407 | struct extent_buffer *c; |
3408 | struct extent_buffer *split; | |
3409 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3410 | int mid; |
5c680ed6 | 3411 | int ret; |
7518a238 | 3412 | u32 c_nritems; |
eb60ceac | 3413 | |
5f39d397 | 3414 | c = path->nodes[level]; |
7bb86316 | 3415 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3416 | if (c == root->node) { |
d9abbf1c | 3417 | /* |
90f8d62e JS |
3418 | * trying to split the root, lets make a new one |
3419 | * | |
fdd99c72 | 3420 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3421 | * insert_new_root, because that root buffer will be kept as a |
3422 | * normal node. We are going to log removal of half of the | |
3423 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3424 | * tree lock on the buffer, which is why we cannot race with | |
3425 | * other tree_mod_log users. | |
d9abbf1c | 3426 | */ |
fdd99c72 | 3427 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3428 | if (ret) |
3429 | return ret; | |
b3612421 | 3430 | } else { |
e66f709b | 3431 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3432 | c = path->nodes[level]; |
3433 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3434 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3435 | return 0; |
54aa1f4d CM |
3436 | if (ret < 0) |
3437 | return ret; | |
be0e5c09 | 3438 | } |
e66f709b | 3439 | |
5f39d397 | 3440 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3441 | mid = (c_nritems + 1) / 2; |
3442 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3443 | |
a6279470 | 3444 | split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, |
4dff97e6 | 3445 | c->start, 0, BTRFS_NESTING_SPLIT); |
5f39d397 CM |
3446 | if (IS_ERR(split)) |
3447 | return PTR_ERR(split); | |
3448 | ||
0b246afa | 3449 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3450 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3451 | |
ed874f0d | 3452 | ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 3453 | if (ret) { |
66642832 | 3454 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3455 | return ret; |
3456 | } | |
5f39d397 CM |
3457 | copy_extent_buffer(split, c, |
3458 | btrfs_node_key_ptr_offset(0), | |
3459 | btrfs_node_key_ptr_offset(mid), | |
3460 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3461 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3462 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3463 | ret = 0; |
3464 | ||
5f39d397 CM |
3465 | btrfs_mark_buffer_dirty(c); |
3466 | btrfs_mark_buffer_dirty(split); | |
3467 | ||
6ad3cf6d | 3468 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 3469 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3470 | |
5de08d7d | 3471 | if (path->slots[level] >= mid) { |
5c680ed6 | 3472 | path->slots[level] -= mid; |
925baedd | 3473 | btrfs_tree_unlock(c); |
5f39d397 CM |
3474 | free_extent_buffer(c); |
3475 | path->nodes[level] = split; | |
5c680ed6 CM |
3476 | path->slots[level + 1] += 1; |
3477 | } else { | |
925baedd | 3478 | btrfs_tree_unlock(split); |
5f39d397 | 3479 | free_extent_buffer(split); |
be0e5c09 | 3480 | } |
aa5d6bed | 3481 | return ret; |
be0e5c09 CM |
3482 | } |
3483 | ||
74123bd7 CM |
3484 | /* |
3485 | * how many bytes are required to store the items in a leaf. start | |
3486 | * and nr indicate which items in the leaf to check. This totals up the | |
3487 | * space used both by the item structs and the item data | |
3488 | */ | |
5f39d397 | 3489 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3490 | { |
41be1f3b JB |
3491 | struct btrfs_item *start_item; |
3492 | struct btrfs_item *end_item; | |
be0e5c09 | 3493 | int data_len; |
5f39d397 | 3494 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3495 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3496 | |
3497 | if (!nr) | |
3498 | return 0; | |
dd3cc16b RK |
3499 | start_item = btrfs_item_nr(start); |
3500 | end_item = btrfs_item_nr(end); | |
a31356b9 DS |
3501 | data_len = btrfs_item_offset(l, start_item) + |
3502 | btrfs_item_size(l, start_item); | |
3503 | data_len = data_len - btrfs_item_offset(l, end_item); | |
0783fcfc | 3504 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3505 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3506 | return data_len; |
3507 | } | |
3508 | ||
d4dbff95 CM |
3509 | /* |
3510 | * The space between the end of the leaf items and | |
3511 | * the start of the leaf data. IOW, how much room | |
3512 | * the leaf has left for both items and data | |
3513 | */ | |
e902baac | 3514 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 3515 | { |
e902baac | 3516 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
3517 | int nritems = btrfs_header_nritems(leaf); |
3518 | int ret; | |
0b246afa JM |
3519 | |
3520 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3521 | if (ret < 0) { |
0b246afa JM |
3522 | btrfs_crit(fs_info, |
3523 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3524 | ret, | |
3525 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3526 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3527 | } |
3528 | return ret; | |
d4dbff95 CM |
3529 | } |
3530 | ||
99d8f83c CM |
3531 | /* |
3532 | * min slot controls the lowest index we're willing to push to the | |
3533 | * right. We'll push up to and including min_slot, but no lower | |
3534 | */ | |
f72f0010 | 3535 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
3536 | int data_size, int empty, |
3537 | struct extent_buffer *right, | |
99d8f83c CM |
3538 | int free_space, u32 left_nritems, |
3539 | u32 min_slot) | |
00ec4c51 | 3540 | { |
f72f0010 | 3541 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 3542 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3543 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3544 | struct btrfs_map_token token; |
5f39d397 | 3545 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3546 | int slot; |
34a38218 | 3547 | u32 i; |
00ec4c51 CM |
3548 | int push_space = 0; |
3549 | int push_items = 0; | |
0783fcfc | 3550 | struct btrfs_item *item; |
34a38218 | 3551 | u32 nr; |
7518a238 | 3552 | u32 right_nritems; |
5f39d397 | 3553 | u32 data_end; |
db94535d | 3554 | u32 this_item_size; |
00ec4c51 | 3555 | |
34a38218 CM |
3556 | if (empty) |
3557 | nr = 0; | |
3558 | else | |
99d8f83c | 3559 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3560 | |
31840ae1 | 3561 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3562 | push_space += data_size; |
31840ae1 | 3563 | |
44871b1b | 3564 | slot = path->slots[1]; |
34a38218 CM |
3565 | i = left_nritems - 1; |
3566 | while (i >= nr) { | |
dd3cc16b | 3567 | item = btrfs_item_nr(i); |
db94535d | 3568 | |
31840ae1 ZY |
3569 | if (!empty && push_items > 0) { |
3570 | if (path->slots[0] > i) | |
3571 | break; | |
3572 | if (path->slots[0] == i) { | |
e902baac DS |
3573 | int space = btrfs_leaf_free_space(left); |
3574 | ||
31840ae1 ZY |
3575 | if (space + push_space * 2 > free_space) |
3576 | break; | |
3577 | } | |
3578 | } | |
3579 | ||
00ec4c51 | 3580 | if (path->slots[0] == i) |
87b29b20 | 3581 | push_space += data_size; |
db94535d | 3582 | |
db94535d CM |
3583 | this_item_size = btrfs_item_size(left, item); |
3584 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3585 | break; |
31840ae1 | 3586 | |
00ec4c51 | 3587 | push_items++; |
db94535d | 3588 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3589 | if (i == 0) |
3590 | break; | |
3591 | i--; | |
db94535d | 3592 | } |
5f39d397 | 3593 | |
925baedd CM |
3594 | if (push_items == 0) |
3595 | goto out_unlock; | |
5f39d397 | 3596 | |
6c1500f2 | 3597 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3598 | |
00ec4c51 | 3599 | /* push left to right */ |
5f39d397 | 3600 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3601 | |
5f39d397 | 3602 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
8f881e8c | 3603 | push_space -= leaf_data_end(left); |
5f39d397 | 3604 | |
00ec4c51 | 3605 | /* make room in the right data area */ |
8f881e8c | 3606 | data_end = leaf_data_end(right); |
5f39d397 | 3607 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3608 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3609 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3610 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3611 | |
00ec4c51 | 3612 | /* copy from the left data area */ |
3d9ec8c4 | 3613 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3614 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 3615 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 3616 | push_space); |
5f39d397 CM |
3617 | |
3618 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3619 | btrfs_item_nr_offset(0), | |
3620 | right_nritems * sizeof(struct btrfs_item)); | |
3621 | ||
00ec4c51 | 3622 | /* copy the items from left to right */ |
5f39d397 CM |
3623 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3624 | btrfs_item_nr_offset(left_nritems - push_items), | |
3625 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3626 | |
3627 | /* update the item pointers */ | |
c82f823c | 3628 | btrfs_init_map_token(&token, right); |
7518a238 | 3629 | right_nritems += push_items; |
5f39d397 | 3630 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3631 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3632 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3633 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
3634 | push_space -= btrfs_token_item_size(&token, item); |
3635 | btrfs_set_token_item_offset(&token, item, push_space); | |
db94535d CM |
3636 | } |
3637 | ||
7518a238 | 3638 | left_nritems -= push_items; |
5f39d397 | 3639 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3640 | |
34a38218 CM |
3641 | if (left_nritems) |
3642 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3643 | else |
6a884d7d | 3644 | btrfs_clean_tree_block(left); |
f0486c68 | 3645 | |
5f39d397 | 3646 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3647 | |
5f39d397 CM |
3648 | btrfs_item_key(right, &disk_key, 0); |
3649 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3650 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3651 | |
00ec4c51 | 3652 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3653 | if (path->slots[0] >= left_nritems) { |
3654 | path->slots[0] -= left_nritems; | |
925baedd | 3655 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 3656 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 3657 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3658 | free_extent_buffer(path->nodes[0]); |
3659 | path->nodes[0] = right; | |
00ec4c51 CM |
3660 | path->slots[1] += 1; |
3661 | } else { | |
925baedd | 3662 | btrfs_tree_unlock(right); |
5f39d397 | 3663 | free_extent_buffer(right); |
00ec4c51 CM |
3664 | } |
3665 | return 0; | |
925baedd CM |
3666 | |
3667 | out_unlock: | |
3668 | btrfs_tree_unlock(right); | |
3669 | free_extent_buffer(right); | |
3670 | return 1; | |
00ec4c51 | 3671 | } |
925baedd | 3672 | |
44871b1b CM |
3673 | /* |
3674 | * push some data in the path leaf to the right, trying to free up at | |
3675 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3676 | * | |
3677 | * returns 1 if the push failed because the other node didn't have enough | |
3678 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3679 | * |
3680 | * this will push starting from min_slot to the end of the leaf. It won't | |
3681 | * push any slot lower than min_slot | |
44871b1b CM |
3682 | */ |
3683 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3684 | *root, struct btrfs_path *path, |
3685 | int min_data_size, int data_size, | |
3686 | int empty, u32 min_slot) | |
44871b1b CM |
3687 | { |
3688 | struct extent_buffer *left = path->nodes[0]; | |
3689 | struct extent_buffer *right; | |
3690 | struct extent_buffer *upper; | |
3691 | int slot; | |
3692 | int free_space; | |
3693 | u32 left_nritems; | |
3694 | int ret; | |
3695 | ||
3696 | if (!path->nodes[1]) | |
3697 | return 1; | |
3698 | ||
3699 | slot = path->slots[1]; | |
3700 | upper = path->nodes[1]; | |
3701 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3702 | return 1; | |
3703 | ||
3704 | btrfs_assert_tree_locked(path->nodes[1]); | |
3705 | ||
4b231ae4 | 3706 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3707 | /* |
3708 | * slot + 1 is not valid or we fail to read the right node, | |
3709 | * no big deal, just return. | |
3710 | */ | |
3711 | if (IS_ERR(right)) | |
91ca338d TI |
3712 | return 1; |
3713 | ||
bf77467a | 3714 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
44871b1b | 3715 | |
e902baac | 3716 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3717 | if (free_space < data_size) |
3718 | goto out_unlock; | |
3719 | ||
3720 | /* cow and double check */ | |
3721 | ret = btrfs_cow_block(trans, root, right, upper, | |
bf59a5a2 | 3722 | slot + 1, &right, BTRFS_NESTING_RIGHT_COW); |
44871b1b CM |
3723 | if (ret) |
3724 | goto out_unlock; | |
3725 | ||
e902baac | 3726 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3727 | if (free_space < data_size) |
3728 | goto out_unlock; | |
3729 | ||
3730 | left_nritems = btrfs_header_nritems(left); | |
3731 | if (left_nritems == 0) | |
3732 | goto out_unlock; | |
3733 | ||
d16c702f QW |
3734 | if (check_sibling_keys(left, right)) { |
3735 | ret = -EUCLEAN; | |
3736 | btrfs_tree_unlock(right); | |
3737 | free_extent_buffer(right); | |
3738 | return ret; | |
3739 | } | |
2ef1fed2 FDBM |
3740 | if (path->slots[0] == left_nritems && !empty) { |
3741 | /* Key greater than all keys in the leaf, right neighbor has | |
3742 | * enough room for it and we're not emptying our leaf to delete | |
3743 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3744 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3745 | btrfs_tree_unlock(left); |
3746 | free_extent_buffer(left); | |
3747 | path->nodes[0] = right; | |
3748 | path->slots[0] = 0; | |
3749 | path->slots[1]++; | |
3750 | return 0; | |
3751 | } | |
3752 | ||
f72f0010 | 3753 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3754 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3755 | out_unlock: |
3756 | btrfs_tree_unlock(right); | |
3757 | free_extent_buffer(right); | |
3758 | return 1; | |
3759 | } | |
3760 | ||
74123bd7 CM |
3761 | /* |
3762 | * push some data in the path leaf to the left, trying to free up at | |
3763 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3764 | * |
3765 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3766 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3767 | * items | |
74123bd7 | 3768 | */ |
8087c193 | 3769 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3770 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3771 | int free_space, u32 right_nritems, |
3772 | u32 max_slot) | |
be0e5c09 | 3773 | { |
8087c193 | 3774 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3775 | struct btrfs_disk_key disk_key; |
3776 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3777 | int i; |
be0e5c09 CM |
3778 | int push_space = 0; |
3779 | int push_items = 0; | |
0783fcfc | 3780 | struct btrfs_item *item; |
7518a238 | 3781 | u32 old_left_nritems; |
34a38218 | 3782 | u32 nr; |
aa5d6bed | 3783 | int ret = 0; |
db94535d CM |
3784 | u32 this_item_size; |
3785 | u32 old_left_item_size; | |
cfed81a0 CM |
3786 | struct btrfs_map_token token; |
3787 | ||
34a38218 | 3788 | if (empty) |
99d8f83c | 3789 | nr = min(right_nritems, max_slot); |
34a38218 | 3790 | else |
99d8f83c | 3791 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3792 | |
3793 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3794 | item = btrfs_item_nr(i); |
db94535d | 3795 | |
31840ae1 ZY |
3796 | if (!empty && push_items > 0) { |
3797 | if (path->slots[0] < i) | |
3798 | break; | |
3799 | if (path->slots[0] == i) { | |
e902baac DS |
3800 | int space = btrfs_leaf_free_space(right); |
3801 | ||
31840ae1 ZY |
3802 | if (space + push_space * 2 > free_space) |
3803 | break; | |
3804 | } | |
3805 | } | |
3806 | ||
be0e5c09 | 3807 | if (path->slots[0] == i) |
87b29b20 | 3808 | push_space += data_size; |
db94535d CM |
3809 | |
3810 | this_item_size = btrfs_item_size(right, item); | |
3811 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3812 | break; |
db94535d | 3813 | |
be0e5c09 | 3814 | push_items++; |
db94535d CM |
3815 | push_space += this_item_size + sizeof(*item); |
3816 | } | |
3817 | ||
be0e5c09 | 3818 | if (push_items == 0) { |
925baedd CM |
3819 | ret = 1; |
3820 | goto out; | |
be0e5c09 | 3821 | } |
fae7f21c | 3822 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3823 | |
be0e5c09 | 3824 | /* push data from right to left */ |
5f39d397 CM |
3825 | copy_extent_buffer(left, right, |
3826 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3827 | btrfs_item_nr_offset(0), | |
3828 | push_items * sizeof(struct btrfs_item)); | |
3829 | ||
0b246afa | 3830 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3831 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3832 | |
3d9ec8c4 | 3833 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3834 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3835 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3836 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3837 | push_space); |
5f39d397 | 3838 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3839 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3840 | |
c82f823c | 3841 | btrfs_init_map_token(&token, left); |
db94535d | 3842 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3843 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3844 | u32 ioff; |
db94535d | 3845 | |
dd3cc16b | 3846 | item = btrfs_item_nr(i); |
db94535d | 3847 | |
cc4c13d5 DS |
3848 | ioff = btrfs_token_item_offset(&token, item); |
3849 | btrfs_set_token_item_offset(&token, item, | |
3850 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); | |
be0e5c09 | 3851 | } |
5f39d397 | 3852 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3853 | |
3854 | /* fixup right node */ | |
31b1a2bd JL |
3855 | if (push_items > right_nritems) |
3856 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3857 | right_nritems); |
34a38218 CM |
3858 | |
3859 | if (push_items < right_nritems) { | |
3860 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
8f881e8c | 3861 | leaf_data_end(right); |
3d9ec8c4 | 3862 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3863 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3864 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3865 | leaf_data_end(right), push_space); |
34a38218 CM |
3866 | |
3867 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3868 | btrfs_item_nr_offset(push_items), |
3869 | (btrfs_header_nritems(right) - push_items) * | |
3870 | sizeof(struct btrfs_item)); | |
34a38218 | 3871 | } |
c82f823c DS |
3872 | |
3873 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3874 | right_nritems -= push_items; |
3875 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3876 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3877 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3878 | item = btrfs_item_nr(i); |
db94535d | 3879 | |
cc4c13d5 DS |
3880 | push_space = push_space - btrfs_token_item_size(&token, item); |
3881 | btrfs_set_token_item_offset(&token, item, push_space); | |
db94535d | 3882 | } |
eb60ceac | 3883 | |
5f39d397 | 3884 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3885 | if (right_nritems) |
3886 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3887 | else |
6a884d7d | 3888 | btrfs_clean_tree_block(right); |
098f59c2 | 3889 | |
5f39d397 | 3890 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3891 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3892 | |
3893 | /* then fixup the leaf pointer in the path */ | |
3894 | if (path->slots[0] < push_items) { | |
3895 | path->slots[0] += old_left_nritems; | |
925baedd | 3896 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3897 | free_extent_buffer(path->nodes[0]); |
3898 | path->nodes[0] = left; | |
be0e5c09 CM |
3899 | path->slots[1] -= 1; |
3900 | } else { | |
925baedd | 3901 | btrfs_tree_unlock(left); |
5f39d397 | 3902 | free_extent_buffer(left); |
be0e5c09 CM |
3903 | path->slots[0] -= push_items; |
3904 | } | |
eb60ceac | 3905 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3906 | return ret; |
925baedd CM |
3907 | out: |
3908 | btrfs_tree_unlock(left); | |
3909 | free_extent_buffer(left); | |
3910 | return ret; | |
be0e5c09 CM |
3911 | } |
3912 | ||
44871b1b CM |
3913 | /* |
3914 | * push some data in the path leaf to the left, trying to free up at | |
3915 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3916 | * |
3917 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3918 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3919 | * items | |
44871b1b CM |
3920 | */ |
3921 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3922 | *root, struct btrfs_path *path, int min_data_size, |
3923 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3924 | { |
3925 | struct extent_buffer *right = path->nodes[0]; | |
3926 | struct extent_buffer *left; | |
3927 | int slot; | |
3928 | int free_space; | |
3929 | u32 right_nritems; | |
3930 | int ret = 0; | |
3931 | ||
3932 | slot = path->slots[1]; | |
3933 | if (slot == 0) | |
3934 | return 1; | |
3935 | if (!path->nodes[1]) | |
3936 | return 1; | |
3937 | ||
3938 | right_nritems = btrfs_header_nritems(right); | |
3939 | if (right_nritems == 0) | |
3940 | return 1; | |
3941 | ||
3942 | btrfs_assert_tree_locked(path->nodes[1]); | |
3943 | ||
4b231ae4 | 3944 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
3945 | /* |
3946 | * slot - 1 is not valid or we fail to read the left node, | |
3947 | * no big deal, just return. | |
3948 | */ | |
3949 | if (IS_ERR(left)) | |
91ca338d TI |
3950 | return 1; |
3951 | ||
bf77467a | 3952 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
44871b1b | 3953 | |
e902baac | 3954 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3955 | if (free_space < data_size) { |
3956 | ret = 1; | |
3957 | goto out; | |
3958 | } | |
3959 | ||
3960 | /* cow and double check */ | |
3961 | ret = btrfs_cow_block(trans, root, left, | |
9631e4cc | 3962 | path->nodes[1], slot - 1, &left, |
bf59a5a2 | 3963 | BTRFS_NESTING_LEFT_COW); |
44871b1b CM |
3964 | if (ret) { |
3965 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3966 | if (ret == -ENOSPC) |
3967 | ret = 1; | |
44871b1b CM |
3968 | goto out; |
3969 | } | |
3970 | ||
e902baac | 3971 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3972 | if (free_space < data_size) { |
3973 | ret = 1; | |
3974 | goto out; | |
3975 | } | |
3976 | ||
d16c702f QW |
3977 | if (check_sibling_keys(left, right)) { |
3978 | ret = -EUCLEAN; | |
3979 | goto out; | |
3980 | } | |
8087c193 | 3981 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
3982 | empty, left, free_space, right_nritems, |
3983 | max_slot); | |
44871b1b CM |
3984 | out: |
3985 | btrfs_tree_unlock(left); | |
3986 | free_extent_buffer(left); | |
3987 | return ret; | |
3988 | } | |
3989 | ||
3990 | /* | |
3991 | * split the path's leaf in two, making sure there is at least data_size | |
3992 | * available for the resulting leaf level of the path. | |
44871b1b | 3993 | */ |
143bede5 | 3994 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
3995 | struct btrfs_path *path, |
3996 | struct extent_buffer *l, | |
3997 | struct extent_buffer *right, | |
3998 | int slot, int mid, int nritems) | |
44871b1b | 3999 | { |
94f94ad9 | 4000 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
4001 | int data_copy_size; |
4002 | int rt_data_off; | |
4003 | int i; | |
44871b1b | 4004 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4005 | struct btrfs_map_token token; |
4006 | ||
44871b1b CM |
4007 | nritems = nritems - mid; |
4008 | btrfs_set_header_nritems(right, nritems); | |
8f881e8c | 4009 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); |
44871b1b CM |
4010 | |
4011 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4012 | btrfs_item_nr_offset(mid), | |
4013 | nritems * sizeof(struct btrfs_item)); | |
4014 | ||
4015 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4016 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4017 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 4018 | leaf_data_end(l), data_copy_size); |
44871b1b | 4019 | |
0b246afa | 4020 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b | 4021 | |
c82f823c | 4022 | btrfs_init_map_token(&token, right); |
44871b1b | 4023 | for (i = 0; i < nritems; i++) { |
dd3cc16b | 4024 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4025 | u32 ioff; |
4026 | ||
cc4c13d5 DS |
4027 | ioff = btrfs_token_item_offset(&token, item); |
4028 | btrfs_set_token_item_offset(&token, item, ioff + rt_data_off); | |
44871b1b CM |
4029 | } |
4030 | ||
44871b1b | 4031 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4032 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 4033 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
4034 | |
4035 | btrfs_mark_buffer_dirty(right); | |
4036 | btrfs_mark_buffer_dirty(l); | |
4037 | BUG_ON(path->slots[0] != slot); | |
4038 | ||
44871b1b CM |
4039 | if (mid <= slot) { |
4040 | btrfs_tree_unlock(path->nodes[0]); | |
4041 | free_extent_buffer(path->nodes[0]); | |
4042 | path->nodes[0] = right; | |
4043 | path->slots[0] -= mid; | |
4044 | path->slots[1] += 1; | |
4045 | } else { | |
4046 | btrfs_tree_unlock(right); | |
4047 | free_extent_buffer(right); | |
4048 | } | |
4049 | ||
4050 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4051 | } |
4052 | ||
99d8f83c CM |
4053 | /* |
4054 | * double splits happen when we need to insert a big item in the middle | |
4055 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4056 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4057 | * A B C | |
4058 | * | |
4059 | * We avoid this by trying to push the items on either side of our target | |
4060 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4061 | * completely. | |
4062 | */ | |
4063 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4064 | struct btrfs_root *root, | |
4065 | struct btrfs_path *path, | |
4066 | int data_size) | |
4067 | { | |
4068 | int ret; | |
4069 | int progress = 0; | |
4070 | int slot; | |
4071 | u32 nritems; | |
5a4267ca | 4072 | int space_needed = data_size; |
99d8f83c CM |
4073 | |
4074 | slot = path->slots[0]; | |
5a4267ca | 4075 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 4076 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
4077 | |
4078 | /* | |
4079 | * try to push all the items after our slot into the | |
4080 | * right leaf | |
4081 | */ | |
5a4267ca | 4082 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4083 | if (ret < 0) |
4084 | return ret; | |
4085 | ||
4086 | if (ret == 0) | |
4087 | progress++; | |
4088 | ||
4089 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4090 | /* | |
4091 | * our goal is to get our slot at the start or end of a leaf. If | |
4092 | * we've done so we're done | |
4093 | */ | |
4094 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4095 | return 0; | |
4096 | ||
e902baac | 4097 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4098 | return 0; |
4099 | ||
4100 | /* try to push all the items before our slot into the next leaf */ | |
4101 | slot = path->slots[0]; | |
263d3995 FM |
4102 | space_needed = data_size; |
4103 | if (slot > 0) | |
e902baac | 4104 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 4105 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4106 | if (ret < 0) |
4107 | return ret; | |
4108 | ||
4109 | if (ret == 0) | |
4110 | progress++; | |
4111 | ||
4112 | if (progress) | |
4113 | return 0; | |
4114 | return 1; | |
4115 | } | |
4116 | ||
74123bd7 CM |
4117 | /* |
4118 | * split the path's leaf in two, making sure there is at least data_size | |
4119 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4120 | * |
4121 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4122 | */ |
e02119d5 CM |
4123 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4124 | struct btrfs_root *root, | |
310712b2 | 4125 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4126 | struct btrfs_path *path, int data_size, |
4127 | int extend) | |
be0e5c09 | 4128 | { |
5d4f98a2 | 4129 | struct btrfs_disk_key disk_key; |
5f39d397 | 4130 | struct extent_buffer *l; |
7518a238 | 4131 | u32 nritems; |
eb60ceac CM |
4132 | int mid; |
4133 | int slot; | |
5f39d397 | 4134 | struct extent_buffer *right; |
b7a0365e | 4135 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4136 | int ret = 0; |
aa5d6bed | 4137 | int wret; |
5d4f98a2 | 4138 | int split; |
cc0c5538 | 4139 | int num_doubles = 0; |
99d8f83c | 4140 | int tried_avoid_double = 0; |
aa5d6bed | 4141 | |
a5719521 YZ |
4142 | l = path->nodes[0]; |
4143 | slot = path->slots[0]; | |
4144 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4145 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4146 | return -EOVERFLOW; |
4147 | ||
40689478 | 4148 | /* first try to make some room by pushing left and right */ |
33157e05 | 4149 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4150 | int space_needed = data_size; |
4151 | ||
4152 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 4153 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4154 | |
4155 | wret = push_leaf_right(trans, root, path, space_needed, | |
4156 | space_needed, 0, 0); | |
d397712b | 4157 | if (wret < 0) |
eaee50e8 | 4158 | return wret; |
3685f791 | 4159 | if (wret) { |
263d3995 FM |
4160 | space_needed = data_size; |
4161 | if (slot > 0) | |
e902baac | 4162 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4163 | wret = push_leaf_left(trans, root, path, space_needed, |
4164 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4165 | if (wret < 0) |
4166 | return wret; | |
4167 | } | |
4168 | l = path->nodes[0]; | |
aa5d6bed | 4169 | |
3685f791 | 4170 | /* did the pushes work? */ |
e902baac | 4171 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 4172 | return 0; |
3326d1b0 | 4173 | } |
aa5d6bed | 4174 | |
5c680ed6 | 4175 | if (!path->nodes[1]) { |
fdd99c72 | 4176 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4177 | if (ret) |
4178 | return ret; | |
4179 | } | |
cc0c5538 | 4180 | again: |
5d4f98a2 | 4181 | split = 1; |
cc0c5538 | 4182 | l = path->nodes[0]; |
eb60ceac | 4183 | slot = path->slots[0]; |
5f39d397 | 4184 | nritems = btrfs_header_nritems(l); |
d397712b | 4185 | mid = (nritems + 1) / 2; |
54aa1f4d | 4186 | |
5d4f98a2 YZ |
4187 | if (mid <= slot) { |
4188 | if (nritems == 1 || | |
4189 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4190 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4191 | if (slot >= nritems) { |
4192 | split = 0; | |
4193 | } else { | |
4194 | mid = slot; | |
4195 | if (mid != nritems && | |
4196 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4197 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4198 | if (data_size && !tried_avoid_double) |
4199 | goto push_for_double; | |
5d4f98a2 YZ |
4200 | split = 2; |
4201 | } | |
4202 | } | |
4203 | } | |
4204 | } else { | |
4205 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4206 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4207 | if (!extend && data_size && slot == 0) { |
4208 | split = 0; | |
4209 | } else if ((extend || !data_size) && slot == 0) { | |
4210 | mid = 1; | |
4211 | } else { | |
4212 | mid = slot; | |
4213 | if (mid != nritems && | |
4214 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4215 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4216 | if (data_size && !tried_avoid_double) |
4217 | goto push_for_double; | |
67871254 | 4218 | split = 2; |
5d4f98a2 YZ |
4219 | } |
4220 | } | |
4221 | } | |
4222 | } | |
4223 | ||
4224 | if (split == 0) | |
4225 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4226 | else | |
4227 | btrfs_item_key(l, &disk_key, mid); | |
4228 | ||
ca9d473a JB |
4229 | /* |
4230 | * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double | |
4231 | * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES | |
4232 | * subclasses, which is 8 at the time of this patch, and we've maxed it | |
4233 | * out. In the future we could add a | |
4234 | * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just | |
4235 | * use BTRFS_NESTING_NEW_ROOT. | |
4236 | */ | |
a6279470 | 4237 | right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, |
ca9d473a JB |
4238 | l->start, 0, num_doubles ? |
4239 | BTRFS_NESTING_NEW_ROOT : | |
4240 | BTRFS_NESTING_SPLIT); | |
f0486c68 | 4241 | if (IS_ERR(right)) |
5f39d397 | 4242 | return PTR_ERR(right); |
f0486c68 | 4243 | |
0b246afa | 4244 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4245 | |
5d4f98a2 YZ |
4246 | if (split == 0) { |
4247 | if (mid <= slot) { | |
4248 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4249 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4250 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4251 | btrfs_tree_unlock(path->nodes[0]); |
4252 | free_extent_buffer(path->nodes[0]); | |
4253 | path->nodes[0] = right; | |
4254 | path->slots[0] = 0; | |
4255 | path->slots[1] += 1; | |
4256 | } else { | |
4257 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4258 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4259 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
4260 | btrfs_tree_unlock(path->nodes[0]); |
4261 | free_extent_buffer(path->nodes[0]); | |
4262 | path->nodes[0] = right; | |
4263 | path->slots[0] = 0; | |
143bede5 | 4264 | if (path->slots[1] == 0) |
b167fa91 | 4265 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4266 | } |
196e0249 LB |
4267 | /* |
4268 | * We create a new leaf 'right' for the required ins_len and | |
4269 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4270 | * the content of ins_len to 'right'. | |
4271 | */ | |
5d4f98a2 | 4272 | return ret; |
d4dbff95 | 4273 | } |
74123bd7 | 4274 | |
94f94ad9 | 4275 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 4276 | |
5d4f98a2 | 4277 | if (split == 2) { |
cc0c5538 CM |
4278 | BUG_ON(num_doubles != 0); |
4279 | num_doubles++; | |
4280 | goto again; | |
a429e513 | 4281 | } |
44871b1b | 4282 | |
143bede5 | 4283 | return 0; |
99d8f83c CM |
4284 | |
4285 | push_for_double: | |
4286 | push_for_double_split(trans, root, path, data_size); | |
4287 | tried_avoid_double = 1; | |
e902baac | 4288 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4289 | return 0; |
4290 | goto again; | |
be0e5c09 CM |
4291 | } |
4292 | ||
ad48fd75 YZ |
4293 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4294 | struct btrfs_root *root, | |
4295 | struct btrfs_path *path, int ins_len) | |
459931ec | 4296 | { |
ad48fd75 | 4297 | struct btrfs_key key; |
459931ec | 4298 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4299 | struct btrfs_file_extent_item *fi; |
4300 | u64 extent_len = 0; | |
4301 | u32 item_size; | |
4302 | int ret; | |
459931ec CM |
4303 | |
4304 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4305 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4306 | ||
4307 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4308 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4309 | ||
e902baac | 4310 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 4311 | return 0; |
459931ec CM |
4312 | |
4313 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4314 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4315 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4316 | struct btrfs_file_extent_item); | |
4317 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4318 | } | |
b3b4aa74 | 4319 | btrfs_release_path(path); |
459931ec | 4320 | |
459931ec | 4321 | path->keep_locks = 1; |
ad48fd75 YZ |
4322 | path->search_for_split = 1; |
4323 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4324 | path->search_for_split = 0; |
a8df6fe6 FM |
4325 | if (ret > 0) |
4326 | ret = -EAGAIN; | |
ad48fd75 YZ |
4327 | if (ret < 0) |
4328 | goto err; | |
459931ec | 4329 | |
ad48fd75 YZ |
4330 | ret = -EAGAIN; |
4331 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4332 | /* if our item isn't there, return now */ |
4333 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4334 | goto err; |
4335 | ||
109f6aef | 4336 | /* the leaf has changed, it now has room. return now */ |
e902baac | 4337 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
4338 | goto err; |
4339 | ||
ad48fd75 YZ |
4340 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4341 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4342 | struct btrfs_file_extent_item); | |
4343 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4344 | goto err; | |
459931ec CM |
4345 | } |
4346 | ||
ad48fd75 | 4347 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4348 | if (ret) |
4349 | goto err; | |
459931ec | 4350 | |
ad48fd75 | 4351 | path->keep_locks = 0; |
b9473439 | 4352 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4353 | return 0; |
4354 | err: | |
4355 | path->keep_locks = 0; | |
4356 | return ret; | |
4357 | } | |
4358 | ||
25263cd7 | 4359 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 4360 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4361 | unsigned long split_offset) |
4362 | { | |
4363 | struct extent_buffer *leaf; | |
4364 | struct btrfs_item *item; | |
4365 | struct btrfs_item *new_item; | |
4366 | int slot; | |
4367 | char *buf; | |
4368 | u32 nritems; | |
4369 | u32 item_size; | |
4370 | u32 orig_offset; | |
4371 | struct btrfs_disk_key disk_key; | |
4372 | ||
b9473439 | 4373 | leaf = path->nodes[0]; |
e902baac | 4374 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4375 | |
dd3cc16b | 4376 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4377 | orig_offset = btrfs_item_offset(leaf, item); |
4378 | item_size = btrfs_item_size(leaf, item); | |
4379 | ||
459931ec | 4380 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4381 | if (!buf) |
4382 | return -ENOMEM; | |
4383 | ||
459931ec CM |
4384 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4385 | path->slots[0]), item_size); | |
459931ec | 4386 | |
ad48fd75 | 4387 | slot = path->slots[0] + 1; |
459931ec | 4388 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4389 | if (slot != nritems) { |
4390 | /* shift the items */ | |
4391 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4392 | btrfs_item_nr_offset(slot), |
4393 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4394 | } |
4395 | ||
4396 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4397 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4398 | ||
dd3cc16b | 4399 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4400 | |
4401 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4402 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4403 | ||
4404 | btrfs_set_item_offset(leaf, item, | |
4405 | orig_offset + item_size - split_offset); | |
4406 | btrfs_set_item_size(leaf, item, split_offset); | |
4407 | ||
4408 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4409 | ||
4410 | /* write the data for the start of the original item */ | |
4411 | write_extent_buffer(leaf, buf, | |
4412 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4413 | split_offset); | |
4414 | ||
4415 | /* write the data for the new item */ | |
4416 | write_extent_buffer(leaf, buf + split_offset, | |
4417 | btrfs_item_ptr_offset(leaf, slot), | |
4418 | item_size - split_offset); | |
4419 | btrfs_mark_buffer_dirty(leaf); | |
4420 | ||
e902baac | 4421 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 4422 | kfree(buf); |
ad48fd75 YZ |
4423 | return 0; |
4424 | } | |
4425 | ||
4426 | /* | |
4427 | * This function splits a single item into two items, | |
4428 | * giving 'new_key' to the new item and splitting the | |
4429 | * old one at split_offset (from the start of the item). | |
4430 | * | |
4431 | * The path may be released by this operation. After | |
4432 | * the split, the path is pointing to the old item. The | |
4433 | * new item is going to be in the same node as the old one. | |
4434 | * | |
4435 | * Note, the item being split must be smaller enough to live alone on | |
4436 | * a tree block with room for one extra struct btrfs_item | |
4437 | * | |
4438 | * This allows us to split the item in place, keeping a lock on the | |
4439 | * leaf the entire time. | |
4440 | */ | |
4441 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4442 | struct btrfs_root *root, | |
4443 | struct btrfs_path *path, | |
310712b2 | 4444 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4445 | unsigned long split_offset) |
4446 | { | |
4447 | int ret; | |
4448 | ret = setup_leaf_for_split(trans, root, path, | |
4449 | sizeof(struct btrfs_item)); | |
4450 | if (ret) | |
4451 | return ret; | |
4452 | ||
25263cd7 | 4453 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
4454 | return ret; |
4455 | } | |
4456 | ||
ad48fd75 YZ |
4457 | /* |
4458 | * This function duplicate a item, giving 'new_key' to the new item. | |
4459 | * It guarantees both items live in the same tree leaf and the new item | |
4460 | * is contiguous with the original item. | |
4461 | * | |
4462 | * This allows us to split file extent in place, keeping a lock on the | |
4463 | * leaf the entire time. | |
4464 | */ | |
4465 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4466 | struct btrfs_root *root, | |
4467 | struct btrfs_path *path, | |
310712b2 | 4468 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4469 | { |
4470 | struct extent_buffer *leaf; | |
4471 | int ret; | |
4472 | u32 item_size; | |
4473 | ||
4474 | leaf = path->nodes[0]; | |
4475 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4476 | ret = setup_leaf_for_split(trans, root, path, | |
4477 | item_size + sizeof(struct btrfs_item)); | |
4478 | if (ret) | |
4479 | return ret; | |
4480 | ||
4481 | path->slots[0]++; | |
fc0d82e1 | 4482 | setup_items_for_insert(root, path, new_key, &item_size, 1); |
ad48fd75 YZ |
4483 | leaf = path->nodes[0]; |
4484 | memcpy_extent_buffer(leaf, | |
4485 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4486 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4487 | item_size); | |
4488 | return 0; | |
4489 | } | |
4490 | ||
d352ac68 CM |
4491 | /* |
4492 | * make the item pointed to by the path smaller. new_size indicates | |
4493 | * how small to make it, and from_end tells us if we just chop bytes | |
4494 | * off the end of the item or if we shift the item to chop bytes off | |
4495 | * the front. | |
4496 | */ | |
78ac4f9e | 4497 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 4498 | { |
b18c6685 | 4499 | int slot; |
5f39d397 CM |
4500 | struct extent_buffer *leaf; |
4501 | struct btrfs_item *item; | |
b18c6685 CM |
4502 | u32 nritems; |
4503 | unsigned int data_end; | |
4504 | unsigned int old_data_start; | |
4505 | unsigned int old_size; | |
4506 | unsigned int size_diff; | |
4507 | int i; | |
cfed81a0 CM |
4508 | struct btrfs_map_token token; |
4509 | ||
5f39d397 | 4510 | leaf = path->nodes[0]; |
179e29e4 CM |
4511 | slot = path->slots[0]; |
4512 | ||
4513 | old_size = btrfs_item_size_nr(leaf, slot); | |
4514 | if (old_size == new_size) | |
143bede5 | 4515 | return; |
b18c6685 | 4516 | |
5f39d397 | 4517 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4518 | data_end = leaf_data_end(leaf); |
b18c6685 | 4519 | |
5f39d397 | 4520 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4521 | |
b18c6685 CM |
4522 | size_diff = old_size - new_size; |
4523 | ||
4524 | BUG_ON(slot < 0); | |
4525 | BUG_ON(slot >= nritems); | |
4526 | ||
4527 | /* | |
4528 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4529 | */ | |
4530 | /* first correct the data pointers */ | |
c82f823c | 4531 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 4532 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4533 | u32 ioff; |
dd3cc16b | 4534 | item = btrfs_item_nr(i); |
db94535d | 4535 | |
cc4c13d5 DS |
4536 | ioff = btrfs_token_item_offset(&token, item); |
4537 | btrfs_set_token_item_offset(&token, item, ioff + size_diff); | |
b18c6685 | 4538 | } |
db94535d | 4539 | |
b18c6685 | 4540 | /* shift the data */ |
179e29e4 | 4541 | if (from_end) { |
3d9ec8c4 NB |
4542 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4543 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4544 | data_end, old_data_start + new_size - data_end); |
4545 | } else { | |
4546 | struct btrfs_disk_key disk_key; | |
4547 | u64 offset; | |
4548 | ||
4549 | btrfs_item_key(leaf, &disk_key, slot); | |
4550 | ||
4551 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4552 | unsigned long ptr; | |
4553 | struct btrfs_file_extent_item *fi; | |
4554 | ||
4555 | fi = btrfs_item_ptr(leaf, slot, | |
4556 | struct btrfs_file_extent_item); | |
4557 | fi = (struct btrfs_file_extent_item *)( | |
4558 | (unsigned long)fi - size_diff); | |
4559 | ||
4560 | if (btrfs_file_extent_type(leaf, fi) == | |
4561 | BTRFS_FILE_EXTENT_INLINE) { | |
4562 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4563 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4564 | (unsigned long)fi, |
7ec20afb | 4565 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4566 | } |
4567 | } | |
4568 | ||
3d9ec8c4 NB |
4569 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4570 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4571 | data_end, old_data_start - data_end); |
4572 | ||
4573 | offset = btrfs_disk_key_offset(&disk_key); | |
4574 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4575 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4576 | if (slot == 0) | |
b167fa91 | 4577 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4578 | } |
5f39d397 | 4579 | |
dd3cc16b | 4580 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4581 | btrfs_set_item_size(leaf, item, new_size); |
4582 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4583 | |
e902baac | 4584 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4585 | btrfs_print_leaf(leaf); |
b18c6685 | 4586 | BUG(); |
5f39d397 | 4587 | } |
b18c6685 CM |
4588 | } |
4589 | ||
d352ac68 | 4590 | /* |
8f69dbd2 | 4591 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4592 | */ |
c71dd880 | 4593 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 4594 | { |
6567e837 | 4595 | int slot; |
5f39d397 CM |
4596 | struct extent_buffer *leaf; |
4597 | struct btrfs_item *item; | |
6567e837 CM |
4598 | u32 nritems; |
4599 | unsigned int data_end; | |
4600 | unsigned int old_data; | |
4601 | unsigned int old_size; | |
4602 | int i; | |
cfed81a0 CM |
4603 | struct btrfs_map_token token; |
4604 | ||
5f39d397 | 4605 | leaf = path->nodes[0]; |
6567e837 | 4606 | |
5f39d397 | 4607 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4608 | data_end = leaf_data_end(leaf); |
6567e837 | 4609 | |
e902baac | 4610 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 4611 | btrfs_print_leaf(leaf); |
6567e837 | 4612 | BUG(); |
5f39d397 | 4613 | } |
6567e837 | 4614 | slot = path->slots[0]; |
5f39d397 | 4615 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4616 | |
4617 | BUG_ON(slot < 0); | |
3326d1b0 | 4618 | if (slot >= nritems) { |
a4f78750 | 4619 | btrfs_print_leaf(leaf); |
c71dd880 | 4620 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 4621 | slot, nritems); |
290342f6 | 4622 | BUG(); |
3326d1b0 | 4623 | } |
6567e837 CM |
4624 | |
4625 | /* | |
4626 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4627 | */ | |
4628 | /* first correct the data pointers */ | |
c82f823c | 4629 | btrfs_init_map_token(&token, leaf); |
6567e837 | 4630 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4631 | u32 ioff; |
dd3cc16b | 4632 | item = btrfs_item_nr(i); |
db94535d | 4633 | |
cc4c13d5 DS |
4634 | ioff = btrfs_token_item_offset(&token, item); |
4635 | btrfs_set_token_item_offset(&token, item, ioff - data_size); | |
6567e837 | 4636 | } |
5f39d397 | 4637 | |
6567e837 | 4638 | /* shift the data */ |
3d9ec8c4 NB |
4639 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4640 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4641 | data_end, old_data - data_end); |
5f39d397 | 4642 | |
6567e837 | 4643 | data_end = old_data; |
5f39d397 | 4644 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4645 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4646 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4647 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4648 | |
e902baac | 4649 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4650 | btrfs_print_leaf(leaf); |
6567e837 | 4651 | BUG(); |
5f39d397 | 4652 | } |
6567e837 CM |
4653 | } |
4654 | ||
da9ffb24 NB |
4655 | /** |
4656 | * setup_items_for_insert - Helper called before inserting one or more items | |
4657 | * to a leaf. Main purpose is to save stack depth by doing the bulk of the work | |
4658 | * in a function that doesn't call btrfs_search_slot | |
4659 | * | |
4660 | * @root: root we are inserting items to | |
4661 | * @path: points to the leaf/slot where we are going to insert new items | |
4662 | * @cpu_key: array of keys for items to be inserted | |
4663 | * @data_size: size of the body of each item we are going to insert | |
4664 | * @nr: size of @cpu_key/@data_size arrays | |
74123bd7 | 4665 | */ |
afe5fea7 | 4666 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4667 | const struct btrfs_key *cpu_key, u32 *data_size, |
fc0d82e1 | 4668 | int nr) |
be0e5c09 | 4669 | { |
0b246afa | 4670 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4671 | struct btrfs_item *item; |
9c58309d | 4672 | int i; |
7518a238 | 4673 | u32 nritems; |
be0e5c09 | 4674 | unsigned int data_end; |
e2fa7227 | 4675 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4676 | struct extent_buffer *leaf; |
4677 | int slot; | |
cfed81a0 | 4678 | struct btrfs_map_token token; |
fc0d82e1 NB |
4679 | u32 total_size; |
4680 | u32 total_data = 0; | |
4681 | ||
4682 | for (i = 0; i < nr; i++) | |
4683 | total_data += data_size[i]; | |
4684 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
cfed81a0 | 4685 | |
24cdc847 FM |
4686 | if (path->slots[0] == 0) { |
4687 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4688 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4689 | } |
4690 | btrfs_unlock_up_safe(path, 1); | |
4691 | ||
5f39d397 | 4692 | leaf = path->nodes[0]; |
44871b1b | 4693 | slot = path->slots[0]; |
74123bd7 | 4694 | |
5f39d397 | 4695 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4696 | data_end = leaf_data_end(leaf); |
eb60ceac | 4697 | |
e902baac | 4698 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4699 | btrfs_print_leaf(leaf); |
0b246afa | 4700 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4701 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4702 | BUG(); |
d4dbff95 | 4703 | } |
5f39d397 | 4704 | |
c82f823c | 4705 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4706 | if (slot != nritems) { |
5f39d397 | 4707 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4708 | |
5f39d397 | 4709 | if (old_data < data_end) { |
a4f78750 | 4710 | btrfs_print_leaf(leaf); |
7269ddd2 NB |
4711 | btrfs_crit(fs_info, |
4712 | "item at slot %d with data offset %u beyond data end of leaf %u", | |
5d163e0e | 4713 | slot, old_data, data_end); |
290342f6 | 4714 | BUG(); |
5f39d397 | 4715 | } |
be0e5c09 CM |
4716 | /* |
4717 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4718 | */ | |
4719 | /* first correct the data pointers */ | |
0783fcfc | 4720 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4721 | u32 ioff; |
db94535d | 4722 | |
62e85577 | 4723 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
4724 | ioff = btrfs_token_item_offset(&token, item); |
4725 | btrfs_set_token_item_offset(&token, item, | |
4726 | ioff - total_data); | |
0783fcfc | 4727 | } |
be0e5c09 | 4728 | /* shift the items */ |
9c58309d | 4729 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4730 | btrfs_item_nr_offset(slot), |
d6025579 | 4731 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4732 | |
4733 | /* shift the data */ | |
3d9ec8c4 NB |
4734 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4735 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4736 | data_end, old_data - data_end); |
be0e5c09 CM |
4737 | data_end = old_data; |
4738 | } | |
5f39d397 | 4739 | |
62e2749e | 4740 | /* setup the item for the new data */ |
9c58309d CM |
4741 | for (i = 0; i < nr; i++) { |
4742 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4743 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4744 | item = btrfs_item_nr(slot + i); |
9c58309d | 4745 | data_end -= data_size[i]; |
fc0716c2 | 4746 | btrfs_set_token_item_offset(&token, item, data_end); |
cc4c13d5 | 4747 | btrfs_set_token_item_size(&token, item, data_size[i]); |
9c58309d | 4748 | } |
44871b1b | 4749 | |
9c58309d | 4750 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4751 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4752 | |
e902baac | 4753 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4754 | btrfs_print_leaf(leaf); |
be0e5c09 | 4755 | BUG(); |
5f39d397 | 4756 | } |
44871b1b CM |
4757 | } |
4758 | ||
4759 | /* | |
4760 | * Given a key and some data, insert items into the tree. | |
4761 | * This does all the path init required, making room in the tree if needed. | |
4762 | */ | |
4763 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4764 | struct btrfs_root *root, | |
4765 | struct btrfs_path *path, | |
310712b2 | 4766 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4767 | int nr) |
4768 | { | |
44871b1b CM |
4769 | int ret = 0; |
4770 | int slot; | |
4771 | int i; | |
4772 | u32 total_size = 0; | |
4773 | u32 total_data = 0; | |
4774 | ||
4775 | for (i = 0; i < nr; i++) | |
4776 | total_data += data_size[i]; | |
4777 | ||
4778 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4779 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4780 | if (ret == 0) | |
4781 | return -EEXIST; | |
4782 | if (ret < 0) | |
143bede5 | 4783 | return ret; |
44871b1b | 4784 | |
44871b1b CM |
4785 | slot = path->slots[0]; |
4786 | BUG_ON(slot < 0); | |
4787 | ||
fc0d82e1 | 4788 | setup_items_for_insert(root, path, cpu_key, data_size, nr); |
143bede5 | 4789 | return 0; |
62e2749e CM |
4790 | } |
4791 | ||
4792 | /* | |
4793 | * Given a key and some data, insert an item into the tree. | |
4794 | * This does all the path init required, making room in the tree if needed. | |
4795 | */ | |
310712b2 OS |
4796 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4797 | const struct btrfs_key *cpu_key, void *data, | |
4798 | u32 data_size) | |
62e2749e CM |
4799 | { |
4800 | int ret = 0; | |
2c90e5d6 | 4801 | struct btrfs_path *path; |
5f39d397 CM |
4802 | struct extent_buffer *leaf; |
4803 | unsigned long ptr; | |
62e2749e | 4804 | |
2c90e5d6 | 4805 | path = btrfs_alloc_path(); |
db5b493a TI |
4806 | if (!path) |
4807 | return -ENOMEM; | |
2c90e5d6 | 4808 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4809 | if (!ret) { |
5f39d397 CM |
4810 | leaf = path->nodes[0]; |
4811 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4812 | write_extent_buffer(leaf, data, ptr, data_size); | |
4813 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4814 | } |
2c90e5d6 | 4815 | btrfs_free_path(path); |
aa5d6bed | 4816 | return ret; |
be0e5c09 CM |
4817 | } |
4818 | ||
74123bd7 | 4819 | /* |
5de08d7d | 4820 | * delete the pointer from a given node. |
74123bd7 | 4821 | * |
d352ac68 CM |
4822 | * the tree should have been previously balanced so the deletion does not |
4823 | * empty a node. | |
74123bd7 | 4824 | */ |
afe5fea7 TI |
4825 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4826 | int level, int slot) | |
be0e5c09 | 4827 | { |
5f39d397 | 4828 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4829 | u32 nritems; |
f3ea38da | 4830 | int ret; |
be0e5c09 | 4831 | |
5f39d397 | 4832 | nritems = btrfs_header_nritems(parent); |
d397712b | 4833 | if (slot != nritems - 1) { |
bf1d3425 DS |
4834 | if (level) { |
4835 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4836 | nritems - slot - 1); |
bf1d3425 DS |
4837 | BUG_ON(ret < 0); |
4838 | } | |
5f39d397 CM |
4839 | memmove_extent_buffer(parent, |
4840 | btrfs_node_key_ptr_offset(slot), | |
4841 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4842 | sizeof(struct btrfs_key_ptr) * |
4843 | (nritems - slot - 1)); | |
57ba86c0 | 4844 | } else if (level) { |
e09c2efe DS |
4845 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4846 | GFP_NOFS); | |
57ba86c0 | 4847 | BUG_ON(ret < 0); |
bb803951 | 4848 | } |
f3ea38da | 4849 | |
7518a238 | 4850 | nritems--; |
5f39d397 | 4851 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4852 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4853 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4854 | /* just turn the root into a leaf and break */ |
5f39d397 | 4855 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4856 | } else if (slot == 0) { |
5f39d397 CM |
4857 | struct btrfs_disk_key disk_key; |
4858 | ||
4859 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4860 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4861 | } |
d6025579 | 4862 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4863 | } |
4864 | ||
323ac95b CM |
4865 | /* |
4866 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4867 | * path->nodes[1]. |
323ac95b CM |
4868 | * |
4869 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4870 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4871 | * | |
4872 | * The path must have already been setup for deleting the leaf, including | |
4873 | * all the proper balancing. path->nodes[1] must be locked. | |
4874 | */ | |
143bede5 JM |
4875 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4876 | struct btrfs_root *root, | |
4877 | struct btrfs_path *path, | |
4878 | struct extent_buffer *leaf) | |
323ac95b | 4879 | { |
5d4f98a2 | 4880 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4881 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4882 | |
4d081c41 CM |
4883 | /* |
4884 | * btrfs_free_extent is expensive, we want to make sure we | |
4885 | * aren't holding any locks when we call it | |
4886 | */ | |
4887 | btrfs_unlock_up_safe(path, 0); | |
4888 | ||
f0486c68 YZ |
4889 | root_sub_used(root, leaf->len); |
4890 | ||
67439dad | 4891 | atomic_inc(&leaf->refs); |
5581a51a | 4892 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4893 | free_extent_buffer_stale(leaf); |
323ac95b | 4894 | } |
74123bd7 CM |
4895 | /* |
4896 | * delete the item at the leaf level in path. If that empties | |
4897 | * the leaf, remove it from the tree | |
4898 | */ | |
85e21bac CM |
4899 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4900 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4901 | { |
0b246afa | 4902 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4903 | struct extent_buffer *leaf; |
4904 | struct btrfs_item *item; | |
ce0eac2a AM |
4905 | u32 last_off; |
4906 | u32 dsize = 0; | |
aa5d6bed CM |
4907 | int ret = 0; |
4908 | int wret; | |
85e21bac | 4909 | int i; |
7518a238 | 4910 | u32 nritems; |
be0e5c09 | 4911 | |
5f39d397 | 4912 | leaf = path->nodes[0]; |
85e21bac CM |
4913 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4914 | ||
4915 | for (i = 0; i < nr; i++) | |
4916 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4917 | ||
5f39d397 | 4918 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4919 | |
85e21bac | 4920 | if (slot + nr != nritems) { |
8f881e8c | 4921 | int data_end = leaf_data_end(leaf); |
c82f823c | 4922 | struct btrfs_map_token token; |
5f39d397 | 4923 | |
3d9ec8c4 | 4924 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4925 | data_end + dsize, |
3d9ec8c4 | 4926 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4927 | last_off - data_end); |
5f39d397 | 4928 | |
c82f823c | 4929 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4930 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4931 | u32 ioff; |
db94535d | 4932 | |
dd3cc16b | 4933 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
4934 | ioff = btrfs_token_item_offset(&token, item); |
4935 | btrfs_set_token_item_offset(&token, item, ioff + dsize); | |
0783fcfc | 4936 | } |
db94535d | 4937 | |
5f39d397 | 4938 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4939 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4940 | sizeof(struct btrfs_item) * |
85e21bac | 4941 | (nritems - slot - nr)); |
be0e5c09 | 4942 | } |
85e21bac CM |
4943 | btrfs_set_header_nritems(leaf, nritems - nr); |
4944 | nritems -= nr; | |
5f39d397 | 4945 | |
74123bd7 | 4946 | /* delete the leaf if we've emptied it */ |
7518a238 | 4947 | if (nritems == 0) { |
5f39d397 CM |
4948 | if (leaf == root->node) { |
4949 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4950 | } else { |
6a884d7d | 4951 | btrfs_clean_tree_block(leaf); |
143bede5 | 4952 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4953 | } |
be0e5c09 | 4954 | } else { |
7518a238 | 4955 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4956 | if (slot == 0) { |
5f39d397 CM |
4957 | struct btrfs_disk_key disk_key; |
4958 | ||
4959 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4960 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 4961 | } |
aa5d6bed | 4962 | |
74123bd7 | 4963 | /* delete the leaf if it is mostly empty */ |
0b246afa | 4964 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
4965 | /* push_leaf_left fixes the path. |
4966 | * make sure the path still points to our leaf | |
4967 | * for possible call to del_ptr below | |
4968 | */ | |
4920c9ac | 4969 | slot = path->slots[1]; |
67439dad | 4970 | atomic_inc(&leaf->refs); |
5f39d397 | 4971 | |
99d8f83c CM |
4972 | wret = push_leaf_left(trans, root, path, 1, 1, |
4973 | 1, (u32)-1); | |
54aa1f4d | 4974 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4975 | ret = wret; |
5f39d397 CM |
4976 | |
4977 | if (path->nodes[0] == leaf && | |
4978 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4979 | wret = push_leaf_right(trans, root, path, 1, |
4980 | 1, 1, 0); | |
54aa1f4d | 4981 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4982 | ret = wret; |
4983 | } | |
5f39d397 CM |
4984 | |
4985 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4986 | path->slots[1] = slot; |
143bede5 | 4987 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4988 | free_extent_buffer(leaf); |
143bede5 | 4989 | ret = 0; |
5de08d7d | 4990 | } else { |
925baedd CM |
4991 | /* if we're still in the path, make sure |
4992 | * we're dirty. Otherwise, one of the | |
4993 | * push_leaf functions must have already | |
4994 | * dirtied this buffer | |
4995 | */ | |
4996 | if (path->nodes[0] == leaf) | |
4997 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4998 | free_extent_buffer(leaf); |
be0e5c09 | 4999 | } |
d5719762 | 5000 | } else { |
5f39d397 | 5001 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5002 | } |
5003 | } | |
aa5d6bed | 5004 | return ret; |
be0e5c09 CM |
5005 | } |
5006 | ||
7bb86316 | 5007 | /* |
925baedd | 5008 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5009 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5010 | * returns < 0 on io errors. | |
d352ac68 CM |
5011 | * |
5012 | * This may release the path, and so you may lose any locks held at the | |
5013 | * time you call it. | |
7bb86316 | 5014 | */ |
16e7549f | 5015 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5016 | { |
925baedd CM |
5017 | struct btrfs_key key; |
5018 | struct btrfs_disk_key found_key; | |
5019 | int ret; | |
7bb86316 | 5020 | |
925baedd | 5021 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5022 | |
e8b0d724 | 5023 | if (key.offset > 0) { |
925baedd | 5024 | key.offset--; |
e8b0d724 | 5025 | } else if (key.type > 0) { |
925baedd | 5026 | key.type--; |
e8b0d724 FDBM |
5027 | key.offset = (u64)-1; |
5028 | } else if (key.objectid > 0) { | |
925baedd | 5029 | key.objectid--; |
e8b0d724 FDBM |
5030 | key.type = (u8)-1; |
5031 | key.offset = (u64)-1; | |
5032 | } else { | |
925baedd | 5033 | return 1; |
e8b0d724 | 5034 | } |
7bb86316 | 5035 | |
b3b4aa74 | 5036 | btrfs_release_path(path); |
925baedd CM |
5037 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5038 | if (ret < 0) | |
5039 | return ret; | |
5040 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5041 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5042 | /* |
5043 | * We might have had an item with the previous key in the tree right | |
5044 | * before we released our path. And after we released our path, that | |
5045 | * item might have been pushed to the first slot (0) of the leaf we | |
5046 | * were holding due to a tree balance. Alternatively, an item with the | |
5047 | * previous key can exist as the only element of a leaf (big fat item). | |
5048 | * Therefore account for these 2 cases, so that our callers (like | |
5049 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5050 | * the previous key we computed above. | |
5051 | */ | |
5052 | if (ret <= 0) | |
925baedd CM |
5053 | return 0; |
5054 | return 1; | |
7bb86316 CM |
5055 | } |
5056 | ||
3f157a2f CM |
5057 | /* |
5058 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5059 | * for nodes or leaves that are have a minimum transaction id. |
5060 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5061 | * |
5062 | * This does not cow, but it does stuff the starting key it finds back | |
5063 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5064 | * key and get a writable path. | |
5065 | * | |
3f157a2f CM |
5066 | * This honors path->lowest_level to prevent descent past a given level |
5067 | * of the tree. | |
5068 | * | |
d352ac68 CM |
5069 | * min_trans indicates the oldest transaction that you are interested |
5070 | * in walking through. Any nodes or leaves older than min_trans are | |
5071 | * skipped over (without reading them). | |
5072 | * | |
3f157a2f CM |
5073 | * returns zero if something useful was found, < 0 on error and 1 if there |
5074 | * was nothing in the tree that matched the search criteria. | |
5075 | */ | |
5076 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5077 | struct btrfs_path *path, |
3f157a2f CM |
5078 | u64 min_trans) |
5079 | { | |
5080 | struct extent_buffer *cur; | |
5081 | struct btrfs_key found_key; | |
5082 | int slot; | |
9652480b | 5083 | int sret; |
3f157a2f CM |
5084 | u32 nritems; |
5085 | int level; | |
5086 | int ret = 1; | |
f98de9b9 | 5087 | int keep_locks = path->keep_locks; |
3f157a2f | 5088 | |
f98de9b9 | 5089 | path->keep_locks = 1; |
3f157a2f | 5090 | again: |
bd681513 | 5091 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5092 | level = btrfs_header_level(cur); |
e02119d5 | 5093 | WARN_ON(path->nodes[level]); |
3f157a2f | 5094 | path->nodes[level] = cur; |
bd681513 | 5095 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5096 | |
5097 | if (btrfs_header_generation(cur) < min_trans) { | |
5098 | ret = 1; | |
5099 | goto out; | |
5100 | } | |
d397712b | 5101 | while (1) { |
3f157a2f CM |
5102 | nritems = btrfs_header_nritems(cur); |
5103 | level = btrfs_header_level(cur); | |
e3b83361 | 5104 | sret = btrfs_bin_search(cur, min_key, &slot); |
cbca7d59 FM |
5105 | if (sret < 0) { |
5106 | ret = sret; | |
5107 | goto out; | |
5108 | } | |
3f157a2f | 5109 | |
323ac95b CM |
5110 | /* at the lowest level, we're done, setup the path and exit */ |
5111 | if (level == path->lowest_level) { | |
e02119d5 CM |
5112 | if (slot >= nritems) |
5113 | goto find_next_key; | |
3f157a2f CM |
5114 | ret = 0; |
5115 | path->slots[level] = slot; | |
5116 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5117 | goto out; | |
5118 | } | |
9652480b Y |
5119 | if (sret && slot > 0) |
5120 | slot--; | |
3f157a2f | 5121 | /* |
de78b51a | 5122 | * check this node pointer against the min_trans parameters. |
260db43c | 5123 | * If it is too old, skip to the next one. |
3f157a2f | 5124 | */ |
d397712b | 5125 | while (slot < nritems) { |
3f157a2f | 5126 | u64 gen; |
e02119d5 | 5127 | |
3f157a2f CM |
5128 | gen = btrfs_node_ptr_generation(cur, slot); |
5129 | if (gen < min_trans) { | |
5130 | slot++; | |
5131 | continue; | |
5132 | } | |
de78b51a | 5133 | break; |
3f157a2f | 5134 | } |
e02119d5 | 5135 | find_next_key: |
3f157a2f CM |
5136 | /* |
5137 | * we didn't find a candidate key in this node, walk forward | |
5138 | * and find another one | |
5139 | */ | |
5140 | if (slot >= nritems) { | |
e02119d5 CM |
5141 | path->slots[level] = slot; |
5142 | sret = btrfs_find_next_key(root, path, min_key, level, | |
de78b51a | 5143 | min_trans); |
e02119d5 | 5144 | if (sret == 0) { |
b3b4aa74 | 5145 | btrfs_release_path(path); |
3f157a2f CM |
5146 | goto again; |
5147 | } else { | |
5148 | goto out; | |
5149 | } | |
5150 | } | |
5151 | /* save our key for returning back */ | |
5152 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5153 | path->slots[level] = slot; | |
5154 | if (level == path->lowest_level) { | |
5155 | ret = 0; | |
3f157a2f CM |
5156 | goto out; |
5157 | } | |
4b231ae4 | 5158 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
5159 | if (IS_ERR(cur)) { |
5160 | ret = PTR_ERR(cur); | |
5161 | goto out; | |
5162 | } | |
3f157a2f | 5163 | |
bd681513 | 5164 | btrfs_tree_read_lock(cur); |
b4ce94de | 5165 | |
bd681513 | 5166 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5167 | path->nodes[level - 1] = cur; |
f7c79f30 | 5168 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5169 | } |
5170 | out: | |
f98de9b9 FM |
5171 | path->keep_locks = keep_locks; |
5172 | if (ret == 0) { | |
5173 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
3f157a2f | 5174 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5175 | } |
3f157a2f CM |
5176 | return ret; |
5177 | } | |
5178 | ||
5179 | /* | |
5180 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5181 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5182 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5183 | * |
5184 | * 0 is returned if another key is found, < 0 if there are any errors | |
5185 | * and 1 is returned if there are no higher keys in the tree | |
5186 | * | |
5187 | * path->keep_locks should be set to 1 on the search made before | |
5188 | * calling this function. | |
5189 | */ | |
e7a84565 | 5190 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5191 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5192 | { |
e7a84565 CM |
5193 | int slot; |
5194 | struct extent_buffer *c; | |
5195 | ||
6a9fb468 | 5196 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 5197 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5198 | if (!path->nodes[level]) |
5199 | return 1; | |
5200 | ||
5201 | slot = path->slots[level] + 1; | |
5202 | c = path->nodes[level]; | |
3f157a2f | 5203 | next: |
e7a84565 | 5204 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5205 | int ret; |
5206 | int orig_lowest; | |
5207 | struct btrfs_key cur_key; | |
5208 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5209 | !path->nodes[level + 1]) | |
e7a84565 | 5210 | return 1; |
33c66f43 | 5211 | |
6a9fb468 | 5212 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
5213 | level++; |
5214 | continue; | |
5215 | } | |
5216 | ||
5217 | slot = btrfs_header_nritems(c) - 1; | |
5218 | if (level == 0) | |
5219 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5220 | else | |
5221 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5222 | ||
5223 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5224 | btrfs_release_path(path); |
33c66f43 YZ |
5225 | path->lowest_level = level; |
5226 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5227 | 0, 0); | |
5228 | path->lowest_level = orig_lowest; | |
5229 | if (ret < 0) | |
5230 | return ret; | |
5231 | ||
5232 | c = path->nodes[level]; | |
5233 | slot = path->slots[level]; | |
5234 | if (ret == 0) | |
5235 | slot++; | |
5236 | goto next; | |
e7a84565 | 5237 | } |
33c66f43 | 5238 | |
e7a84565 CM |
5239 | if (level == 0) |
5240 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5241 | else { |
3f157a2f CM |
5242 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5243 | ||
3f157a2f CM |
5244 | if (gen < min_trans) { |
5245 | slot++; | |
5246 | goto next; | |
5247 | } | |
e7a84565 | 5248 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5249 | } |
e7a84565 CM |
5250 | return 0; |
5251 | } | |
5252 | return 1; | |
5253 | } | |
5254 | ||
97571fd0 | 5255 | /* |
925baedd | 5256 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5257 | * returns 0 if it found something or 1 if there are no greater leaves. |
5258 | * returns < 0 on io errors. | |
97571fd0 | 5259 | */ |
234b63a0 | 5260 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5261 | { |
5262 | return btrfs_next_old_leaf(root, path, 0); | |
5263 | } | |
5264 | ||
5265 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5266 | u64 time_seq) | |
d97e63b6 CM |
5267 | { |
5268 | int slot; | |
8e73f275 | 5269 | int level; |
5f39d397 | 5270 | struct extent_buffer *c; |
8e73f275 | 5271 | struct extent_buffer *next; |
925baedd CM |
5272 | struct btrfs_key key; |
5273 | u32 nritems; | |
5274 | int ret; | |
bd681513 | 5275 | int next_rw_lock = 0; |
925baedd CM |
5276 | |
5277 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5278 | if (nritems == 0) |
925baedd | 5279 | return 1; |
925baedd | 5280 | |
8e73f275 CM |
5281 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5282 | again: | |
5283 | level = 1; | |
5284 | next = NULL; | |
bd681513 | 5285 | next_rw_lock = 0; |
b3b4aa74 | 5286 | btrfs_release_path(path); |
8e73f275 | 5287 | |
a2135011 | 5288 | path->keep_locks = 1; |
8e73f275 | 5289 | |
3d7806ec JS |
5290 | if (time_seq) |
5291 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5292 | else | |
5293 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5294 | path->keep_locks = 0; |
5295 | ||
5296 | if (ret < 0) | |
5297 | return ret; | |
5298 | ||
a2135011 | 5299 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5300 | /* |
5301 | * by releasing the path above we dropped all our locks. A balance | |
5302 | * could have added more items next to the key that used to be | |
5303 | * at the very end of the block. So, check again here and | |
5304 | * advance the path if there are now more items available. | |
5305 | */ | |
a2135011 | 5306 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5307 | if (ret == 0) |
5308 | path->slots[0]++; | |
8e73f275 | 5309 | ret = 0; |
925baedd CM |
5310 | goto done; |
5311 | } | |
0b43e04f LB |
5312 | /* |
5313 | * So the above check misses one case: | |
5314 | * - after releasing the path above, someone has removed the item that | |
5315 | * used to be at the very end of the block, and balance between leafs | |
5316 | * gets another one with bigger key.offset to replace it. | |
5317 | * | |
5318 | * This one should be returned as well, or we can get leaf corruption | |
5319 | * later(esp. in __btrfs_drop_extents()). | |
5320 | * | |
5321 | * And a bit more explanation about this check, | |
5322 | * with ret > 0, the key isn't found, the path points to the slot | |
5323 | * where it should be inserted, so the path->slots[0] item must be the | |
5324 | * bigger one. | |
5325 | */ | |
5326 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5327 | ret = 0; | |
5328 | goto done; | |
5329 | } | |
d97e63b6 | 5330 | |
d397712b | 5331 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5332 | if (!path->nodes[level]) { |
5333 | ret = 1; | |
5334 | goto done; | |
5335 | } | |
5f39d397 | 5336 | |
d97e63b6 CM |
5337 | slot = path->slots[level] + 1; |
5338 | c = path->nodes[level]; | |
5f39d397 | 5339 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5340 | level++; |
8e73f275 CM |
5341 | if (level == BTRFS_MAX_LEVEL) { |
5342 | ret = 1; | |
5343 | goto done; | |
5344 | } | |
d97e63b6 CM |
5345 | continue; |
5346 | } | |
5f39d397 | 5347 | |
925baedd | 5348 | if (next) { |
bd681513 | 5349 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5350 | free_extent_buffer(next); |
925baedd | 5351 | } |
5f39d397 | 5352 | |
8e73f275 | 5353 | next = c; |
bd681513 | 5354 | next_rw_lock = path->locks[level]; |
d07b8528 | 5355 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5356 | slot, &key); |
8e73f275 CM |
5357 | if (ret == -EAGAIN) |
5358 | goto again; | |
5f39d397 | 5359 | |
76a05b35 | 5360 | if (ret < 0) { |
b3b4aa74 | 5361 | btrfs_release_path(path); |
76a05b35 CM |
5362 | goto done; |
5363 | } | |
5364 | ||
5cd57b2c | 5365 | if (!path->skip_locking) { |
bd681513 | 5366 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5367 | if (!ret && time_seq) { |
5368 | /* | |
5369 | * If we don't get the lock, we may be racing | |
5370 | * with push_leaf_left, holding that lock while | |
5371 | * itself waiting for the leaf we've currently | |
5372 | * locked. To solve this situation, we give up | |
5373 | * on our lock and cycle. | |
5374 | */ | |
cf538830 | 5375 | free_extent_buffer(next); |
d42244a0 JS |
5376 | btrfs_release_path(path); |
5377 | cond_resched(); | |
5378 | goto again; | |
5379 | } | |
8e73f275 | 5380 | if (!ret) { |
fd7ba1c1 | 5381 | __btrfs_tree_read_lock(next, |
bf77467a | 5382 | BTRFS_NESTING_RIGHT, |
fd7ba1c1 | 5383 | path->recurse); |
8e73f275 | 5384 | } |
31533fb2 | 5385 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5386 | } |
d97e63b6 CM |
5387 | break; |
5388 | } | |
5389 | path->slots[level] = slot; | |
d397712b | 5390 | while (1) { |
d97e63b6 CM |
5391 | level--; |
5392 | c = path->nodes[level]; | |
925baedd | 5393 | if (path->locks[level]) |
bd681513 | 5394 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5395 | |
5f39d397 | 5396 | free_extent_buffer(c); |
d97e63b6 CM |
5397 | path->nodes[level] = next; |
5398 | path->slots[level] = 0; | |
a74a4b97 | 5399 | if (!path->skip_locking) |
bd681513 | 5400 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5401 | if (!level) |
5402 | break; | |
b4ce94de | 5403 | |
d07b8528 | 5404 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5405 | 0, &key); |
8e73f275 CM |
5406 | if (ret == -EAGAIN) |
5407 | goto again; | |
5408 | ||
76a05b35 | 5409 | if (ret < 0) { |
b3b4aa74 | 5410 | btrfs_release_path(path); |
76a05b35 CM |
5411 | goto done; |
5412 | } | |
5413 | ||
5cd57b2c | 5414 | if (!path->skip_locking) { |
ac5887c8 JB |
5415 | __btrfs_tree_read_lock(next, BTRFS_NESTING_RIGHT, |
5416 | path->recurse); | |
31533fb2 | 5417 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5418 | } |
d97e63b6 | 5419 | } |
8e73f275 | 5420 | ret = 0; |
925baedd | 5421 | done: |
f7c79f30 | 5422 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5423 | |
5424 | return ret; | |
d97e63b6 | 5425 | } |
0b86a832 | 5426 | |
3f157a2f CM |
5427 | /* |
5428 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5429 | * searching until it gets past min_objectid or finds an item of 'type' | |
5430 | * | |
5431 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5432 | */ | |
0b86a832 CM |
5433 | int btrfs_previous_item(struct btrfs_root *root, |
5434 | struct btrfs_path *path, u64 min_objectid, | |
5435 | int type) | |
5436 | { | |
5437 | struct btrfs_key found_key; | |
5438 | struct extent_buffer *leaf; | |
e02119d5 | 5439 | u32 nritems; |
0b86a832 CM |
5440 | int ret; |
5441 | ||
d397712b | 5442 | while (1) { |
0b86a832 CM |
5443 | if (path->slots[0] == 0) { |
5444 | ret = btrfs_prev_leaf(root, path); | |
5445 | if (ret != 0) | |
5446 | return ret; | |
5447 | } else { | |
5448 | path->slots[0]--; | |
5449 | } | |
5450 | leaf = path->nodes[0]; | |
e02119d5 CM |
5451 | nritems = btrfs_header_nritems(leaf); |
5452 | if (nritems == 0) | |
5453 | return 1; | |
5454 | if (path->slots[0] == nritems) | |
5455 | path->slots[0]--; | |
5456 | ||
0b86a832 | 5457 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5458 | if (found_key.objectid < min_objectid) |
5459 | break; | |
0a4eefbb YZ |
5460 | if (found_key.type == type) |
5461 | return 0; | |
e02119d5 CM |
5462 | if (found_key.objectid == min_objectid && |
5463 | found_key.type < type) | |
5464 | break; | |
0b86a832 CM |
5465 | } |
5466 | return 1; | |
5467 | } | |
ade2e0b3 WS |
5468 | |
5469 | /* | |
5470 | * search in extent tree to find a previous Metadata/Data extent item with | |
5471 | * min objecitd. | |
5472 | * | |
5473 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5474 | */ | |
5475 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5476 | struct btrfs_path *path, u64 min_objectid) | |
5477 | { | |
5478 | struct btrfs_key found_key; | |
5479 | struct extent_buffer *leaf; | |
5480 | u32 nritems; | |
5481 | int ret; | |
5482 | ||
5483 | while (1) { | |
5484 | if (path->slots[0] == 0) { | |
ade2e0b3 WS |
5485 | ret = btrfs_prev_leaf(root, path); |
5486 | if (ret != 0) | |
5487 | return ret; | |
5488 | } else { | |
5489 | path->slots[0]--; | |
5490 | } | |
5491 | leaf = path->nodes[0]; | |
5492 | nritems = btrfs_header_nritems(leaf); | |
5493 | if (nritems == 0) | |
5494 | return 1; | |
5495 | if (path->slots[0] == nritems) | |
5496 | path->slots[0]--; | |
5497 | ||
5498 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5499 | if (found_key.objectid < min_objectid) | |
5500 | break; | |
5501 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5502 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5503 | return 0; | |
5504 | if (found_key.objectid == min_objectid && | |
5505 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5506 | break; | |
5507 | } | |
5508 | return 1; | |
5509 | } |