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 | ||
f7cea56c DS |
71 | size_t __const btrfs_get_num_csums(void) |
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 DS |
200 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
201 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 202 | if (IS_ERR(cow)) |
be20aa9d CM |
203 | return PTR_ERR(cow); |
204 | ||
58e8012c | 205 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
206 | btrfs_set_header_bytenr(cow, cow->start); |
207 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
208 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
209 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
210 | BTRFS_HEADER_FLAG_RELOC); | |
211 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
212 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
213 | else | |
214 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 215 | |
de37aa51 | 216 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 217 | |
be20aa9d | 218 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 219 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 220 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 221 | else |
e339a6b0 | 222 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 223 | |
be20aa9d CM |
224 | if (ret) |
225 | return ret; | |
226 | ||
227 | btrfs_mark_buffer_dirty(cow); | |
228 | *cow_ret = cow; | |
229 | return 0; | |
230 | } | |
231 | ||
bd989ba3 JS |
232 | enum mod_log_op { |
233 | MOD_LOG_KEY_REPLACE, | |
234 | MOD_LOG_KEY_ADD, | |
235 | MOD_LOG_KEY_REMOVE, | |
236 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
237 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
238 | MOD_LOG_MOVE_KEYS, | |
239 | MOD_LOG_ROOT_REPLACE, | |
240 | }; | |
241 | ||
bd989ba3 JS |
242 | struct tree_mod_root { |
243 | u64 logical; | |
244 | u8 level; | |
245 | }; | |
246 | ||
247 | struct tree_mod_elem { | |
248 | struct rb_node node; | |
298cfd36 | 249 | u64 logical; |
097b8a7c | 250 | u64 seq; |
bd989ba3 JS |
251 | enum mod_log_op op; |
252 | ||
253 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
254 | int slot; | |
255 | ||
256 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
257 | u64 generation; | |
258 | ||
259 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
260 | struct btrfs_disk_key key; | |
261 | u64 blockptr; | |
262 | ||
263 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
264 | struct { |
265 | int dst_slot; | |
266 | int nr_items; | |
267 | } move; | |
bd989ba3 JS |
268 | |
269 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
270 | struct tree_mod_root old_root; | |
271 | }; | |
272 | ||
fc36ed7e | 273 | /* |
fcebe456 | 274 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 275 | */ |
fcebe456 | 276 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
277 | { |
278 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
279 | } | |
280 | ||
097b8a7c JS |
281 | /* |
282 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
283 | * passed does not already have a sequence number set. So when a caller expects | |
284 | * to record tree modifications, it should ensure to set elem->seq to zero | |
285 | * before calling btrfs_get_tree_mod_seq. | |
286 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
287 | * blocker was added. | |
288 | */ | |
289 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
290 | struct seq_list *elem) | |
bd989ba3 | 291 | { |
b1a09f1e | 292 | write_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 293 | if (!elem->seq) { |
fcebe456 | 294 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
295 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
296 | } | |
b1a09f1e | 297 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 298 | |
fcebe456 | 299 | return elem->seq; |
bd989ba3 JS |
300 | } |
301 | ||
302 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
303 | struct seq_list *elem) | |
304 | { | |
305 | struct rb_root *tm_root; | |
306 | struct rb_node *node; | |
307 | struct rb_node *next; | |
bd989ba3 JS |
308 | struct tree_mod_elem *tm; |
309 | u64 min_seq = (u64)-1; | |
310 | u64 seq_putting = elem->seq; | |
311 | ||
312 | if (!seq_putting) | |
313 | return; | |
314 | ||
7227ff4d | 315 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 316 | list_del(&elem->list); |
097b8a7c | 317 | elem->seq = 0; |
bd989ba3 | 318 | |
42836cf4 FM |
319 | if (!list_empty(&fs_info->tree_mod_seq_list)) { |
320 | struct seq_list *first; | |
321 | ||
322 | first = list_first_entry(&fs_info->tree_mod_seq_list, | |
323 | struct seq_list, list); | |
324 | if (seq_putting > first->seq) { | |
325 | /* | |
326 | * Blocker with lower sequence number exists, we | |
327 | * cannot remove anything from the log. | |
328 | */ | |
329 | write_unlock(&fs_info->tree_mod_log_lock); | |
330 | return; | |
bd989ba3 | 331 | } |
42836cf4 | 332 | min_seq = first->seq; |
bd989ba3 | 333 | } |
097b8a7c | 334 | |
bd989ba3 JS |
335 | /* |
336 | * anything that's lower than the lowest existing (read: blocked) | |
337 | * sequence number can be removed from the tree. | |
338 | */ | |
bd989ba3 JS |
339 | tm_root = &fs_info->tree_mod_log; |
340 | for (node = rb_first(tm_root); node; node = next) { | |
341 | next = rb_next(node); | |
6b4df8b6 | 342 | tm = rb_entry(node, struct tree_mod_elem, node); |
6609fee8 | 343 | if (tm->seq >= min_seq) |
bd989ba3 JS |
344 | continue; |
345 | rb_erase(node, tm_root); | |
bd989ba3 JS |
346 | kfree(tm); |
347 | } | |
b1a09f1e | 348 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
349 | } |
350 | ||
351 | /* | |
352 | * key order of the log: | |
298cfd36 | 353 | * node/leaf start address -> sequence |
bd989ba3 | 354 | * |
298cfd36 CR |
355 | * The 'start address' is the logical address of the *new* root node |
356 | * for root replace operations, or the logical address of the affected | |
357 | * block for all other operations. | |
bd989ba3 JS |
358 | */ |
359 | static noinline int | |
360 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
361 | { | |
362 | struct rb_root *tm_root; | |
363 | struct rb_node **new; | |
364 | struct rb_node *parent = NULL; | |
365 | struct tree_mod_elem *cur; | |
c8cc6341 | 366 | |
73e82fe4 DS |
367 | lockdep_assert_held_write(&fs_info->tree_mod_log_lock); |
368 | ||
fcebe456 | 369 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 370 | |
bd989ba3 JS |
371 | tm_root = &fs_info->tree_mod_log; |
372 | new = &tm_root->rb_node; | |
373 | while (*new) { | |
6b4df8b6 | 374 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 375 | parent = *new; |
298cfd36 | 376 | if (cur->logical < tm->logical) |
bd989ba3 | 377 | new = &((*new)->rb_left); |
298cfd36 | 378 | else if (cur->logical > tm->logical) |
bd989ba3 | 379 | new = &((*new)->rb_right); |
097b8a7c | 380 | else if (cur->seq < tm->seq) |
bd989ba3 | 381 | new = &((*new)->rb_left); |
097b8a7c | 382 | else if (cur->seq > tm->seq) |
bd989ba3 | 383 | new = &((*new)->rb_right); |
5de865ee FDBM |
384 | else |
385 | return -EEXIST; | |
bd989ba3 JS |
386 | } |
387 | ||
388 | rb_link_node(&tm->node, parent, new); | |
389 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 390 | return 0; |
bd989ba3 JS |
391 | } |
392 | ||
097b8a7c JS |
393 | /* |
394 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
395 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
396 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 397 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 398 | */ |
e9b7fd4d JS |
399 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
400 | struct extent_buffer *eb) { | |
401 | smp_mb(); | |
402 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
403 | return 1; | |
097b8a7c JS |
404 | if (eb && btrfs_header_level(eb) == 0) |
405 | return 1; | |
5de865ee | 406 | |
b1a09f1e | 407 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 408 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 409 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
410 | return 1; |
411 | } | |
412 | ||
e9b7fd4d JS |
413 | return 0; |
414 | } | |
415 | ||
5de865ee FDBM |
416 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
417 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
418 | struct extent_buffer *eb) | |
419 | { | |
420 | smp_mb(); | |
421 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
422 | return 0; | |
423 | if (eb && btrfs_header_level(eb) == 0) | |
424 | return 0; | |
425 | ||
426 | return 1; | |
427 | } | |
428 | ||
429 | static struct tree_mod_elem * | |
430 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
431 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 432 | { |
097b8a7c | 433 | struct tree_mod_elem *tm; |
bd989ba3 | 434 | |
c8cc6341 JB |
435 | tm = kzalloc(sizeof(*tm), flags); |
436 | if (!tm) | |
5de865ee | 437 | return NULL; |
bd989ba3 | 438 | |
298cfd36 | 439 | tm->logical = eb->start; |
bd989ba3 JS |
440 | if (op != MOD_LOG_KEY_ADD) { |
441 | btrfs_node_key(eb, &tm->key, slot); | |
442 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
443 | } | |
444 | tm->op = op; | |
445 | tm->slot = slot; | |
446 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 447 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 448 | |
5de865ee | 449 | return tm; |
097b8a7c JS |
450 | } |
451 | ||
e09c2efe DS |
452 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
453 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 454 | { |
5de865ee FDBM |
455 | struct tree_mod_elem *tm; |
456 | int ret; | |
457 | ||
e09c2efe | 458 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
459 | return 0; |
460 | ||
461 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
462 | if (!tm) | |
463 | return -ENOMEM; | |
464 | ||
e09c2efe | 465 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 466 | kfree(tm); |
097b8a7c | 467 | return 0; |
5de865ee FDBM |
468 | } |
469 | ||
e09c2efe | 470 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 471 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
472 | if (ret) |
473 | kfree(tm); | |
097b8a7c | 474 | |
5de865ee | 475 | return ret; |
097b8a7c JS |
476 | } |
477 | ||
6074d45f DS |
478 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
479 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 480 | { |
5de865ee FDBM |
481 | struct tree_mod_elem *tm = NULL; |
482 | struct tree_mod_elem **tm_list = NULL; | |
483 | int ret = 0; | |
bd989ba3 | 484 | int i; |
5de865ee | 485 | int locked = 0; |
bd989ba3 | 486 | |
6074d45f | 487 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 488 | return 0; |
bd989ba3 | 489 | |
176ef8f5 | 490 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
491 | if (!tm_list) |
492 | return -ENOMEM; | |
493 | ||
176ef8f5 | 494 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
495 | if (!tm) { |
496 | ret = -ENOMEM; | |
497 | goto free_tms; | |
498 | } | |
499 | ||
298cfd36 | 500 | tm->logical = eb->start; |
5de865ee FDBM |
501 | tm->slot = src_slot; |
502 | tm->move.dst_slot = dst_slot; | |
503 | tm->move.nr_items = nr_items; | |
504 | tm->op = MOD_LOG_MOVE_KEYS; | |
505 | ||
506 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
507 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 508 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
509 | if (!tm_list[i]) { |
510 | ret = -ENOMEM; | |
511 | goto free_tms; | |
512 | } | |
513 | } | |
514 | ||
6074d45f | 515 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
516 | goto free_tms; |
517 | locked = 1; | |
518 | ||
01763a2e JS |
519 | /* |
520 | * When we override something during the move, we log these removals. | |
521 | * This can only happen when we move towards the beginning of the | |
522 | * buffer, i.e. dst_slot < src_slot. | |
523 | */ | |
bd989ba3 | 524 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 525 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
526 | if (ret) |
527 | goto free_tms; | |
bd989ba3 JS |
528 | } |
529 | ||
6074d45f | 530 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
531 | if (ret) |
532 | goto free_tms; | |
b1a09f1e | 533 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 534 | kfree(tm_list); |
f395694c | 535 | |
5de865ee FDBM |
536 | return 0; |
537 | free_tms: | |
538 | for (i = 0; i < nr_items; i++) { | |
539 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 540 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
541 | kfree(tm_list[i]); |
542 | } | |
543 | if (locked) | |
b1a09f1e | 544 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
545 | kfree(tm_list); |
546 | kfree(tm); | |
bd989ba3 | 547 | |
5de865ee | 548 | return ret; |
bd989ba3 JS |
549 | } |
550 | ||
5de865ee FDBM |
551 | static inline int |
552 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
553 | struct tree_mod_elem **tm_list, | |
554 | int nritems) | |
097b8a7c | 555 | { |
5de865ee | 556 | int i, j; |
097b8a7c JS |
557 | int ret; |
558 | ||
097b8a7c | 559 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
560 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
561 | if (ret) { | |
562 | for (j = nritems - 1; j > i; j--) | |
563 | rb_erase(&tm_list[j]->node, | |
564 | &fs_info->tree_mod_log); | |
565 | return ret; | |
566 | } | |
097b8a7c | 567 | } |
5de865ee FDBM |
568 | |
569 | return 0; | |
097b8a7c JS |
570 | } |
571 | ||
95b757c1 DS |
572 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
573 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 574 | { |
95b757c1 | 575 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
576 | struct tree_mod_elem *tm = NULL; |
577 | struct tree_mod_elem **tm_list = NULL; | |
578 | int nritems = 0; | |
579 | int ret = 0; | |
580 | int i; | |
bd989ba3 | 581 | |
5de865ee | 582 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
583 | return 0; |
584 | ||
5de865ee FDBM |
585 | if (log_removal && btrfs_header_level(old_root) > 0) { |
586 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 587 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 588 | GFP_NOFS); |
5de865ee FDBM |
589 | if (!tm_list) { |
590 | ret = -ENOMEM; | |
591 | goto free_tms; | |
592 | } | |
593 | for (i = 0; i < nritems; i++) { | |
594 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 595 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
596 | if (!tm_list[i]) { |
597 | ret = -ENOMEM; | |
598 | goto free_tms; | |
599 | } | |
600 | } | |
601 | } | |
d9abbf1c | 602 | |
bcc8e07f | 603 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
604 | if (!tm) { |
605 | ret = -ENOMEM; | |
606 | goto free_tms; | |
607 | } | |
bd989ba3 | 608 | |
298cfd36 | 609 | tm->logical = new_root->start; |
bd989ba3 JS |
610 | tm->old_root.logical = old_root->start; |
611 | tm->old_root.level = btrfs_header_level(old_root); | |
612 | tm->generation = btrfs_header_generation(old_root); | |
613 | tm->op = MOD_LOG_ROOT_REPLACE; | |
614 | ||
5de865ee FDBM |
615 | if (tree_mod_dont_log(fs_info, NULL)) |
616 | goto free_tms; | |
617 | ||
618 | if (tm_list) | |
619 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
620 | if (!ret) | |
621 | ret = __tree_mod_log_insert(fs_info, tm); | |
622 | ||
b1a09f1e | 623 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
624 | if (ret) |
625 | goto free_tms; | |
626 | kfree(tm_list); | |
627 | ||
628 | return ret; | |
629 | ||
630 | free_tms: | |
631 | if (tm_list) { | |
632 | for (i = 0; i < nritems; i++) | |
633 | kfree(tm_list[i]); | |
634 | kfree(tm_list); | |
635 | } | |
636 | kfree(tm); | |
637 | ||
638 | return ret; | |
bd989ba3 JS |
639 | } |
640 | ||
641 | static struct tree_mod_elem * | |
642 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
643 | int smallest) | |
644 | { | |
645 | struct rb_root *tm_root; | |
646 | struct rb_node *node; | |
647 | struct tree_mod_elem *cur = NULL; | |
648 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 649 | |
b1a09f1e | 650 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
651 | tm_root = &fs_info->tree_mod_log; |
652 | node = tm_root->rb_node; | |
653 | while (node) { | |
6b4df8b6 | 654 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 655 | if (cur->logical < start) { |
bd989ba3 | 656 | node = node->rb_left; |
298cfd36 | 657 | } else if (cur->logical > start) { |
bd989ba3 | 658 | node = node->rb_right; |
097b8a7c | 659 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
660 | node = node->rb_left; |
661 | } else if (!smallest) { | |
662 | /* we want the node with the highest seq */ | |
663 | if (found) | |
097b8a7c | 664 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
665 | found = cur; |
666 | node = node->rb_left; | |
097b8a7c | 667 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
668 | /* we want the node with the smallest seq */ |
669 | if (found) | |
097b8a7c | 670 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
671 | found = cur; |
672 | node = node->rb_right; | |
673 | } else { | |
674 | found = cur; | |
675 | break; | |
676 | } | |
677 | } | |
b1a09f1e | 678 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
679 | |
680 | return found; | |
681 | } | |
682 | ||
683 | /* | |
684 | * this returns the element from the log with the smallest time sequence | |
685 | * value that's in the log (the oldest log item). any element with a time | |
686 | * sequence lower than min_seq will be ignored. | |
687 | */ | |
688 | static struct tree_mod_elem * | |
689 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
690 | u64 min_seq) | |
691 | { | |
692 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
693 | } | |
694 | ||
695 | /* | |
696 | * this returns the element from the log with the largest time sequence | |
697 | * value that's in the log (the most recent log item). any element with | |
698 | * a time sequence lower than min_seq will be ignored. | |
699 | */ | |
700 | static struct tree_mod_elem * | |
701 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
702 | { | |
703 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
704 | } | |
705 | ||
ed874f0d | 706 | static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, |
bd989ba3 | 707 | struct extent_buffer *src, unsigned long dst_offset, |
90f8d62e | 708 | unsigned long src_offset, int nr_items) |
bd989ba3 | 709 | { |
ed874f0d | 710 | struct btrfs_fs_info *fs_info = dst->fs_info; |
5de865ee FDBM |
711 | int ret = 0; |
712 | struct tree_mod_elem **tm_list = NULL; | |
713 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 714 | int i; |
5de865ee | 715 | int locked = 0; |
bd989ba3 | 716 | |
5de865ee FDBM |
717 | if (!tree_mod_need_log(fs_info, NULL)) |
718 | return 0; | |
bd989ba3 | 719 | |
c8cc6341 | 720 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
721 | return 0; |
722 | ||
31e818fe | 723 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
724 | GFP_NOFS); |
725 | if (!tm_list) | |
726 | return -ENOMEM; | |
bd989ba3 | 727 | |
5de865ee FDBM |
728 | tm_list_add = tm_list; |
729 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 730 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
731 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
732 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
733 | if (!tm_list_rem[i]) { | |
734 | ret = -ENOMEM; | |
735 | goto free_tms; | |
736 | } | |
737 | ||
738 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
739 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
740 | if (!tm_list_add[i]) { | |
741 | ret = -ENOMEM; | |
742 | goto free_tms; | |
743 | } | |
744 | } | |
745 | ||
746 | if (tree_mod_dont_log(fs_info, NULL)) | |
747 | goto free_tms; | |
748 | locked = 1; | |
749 | ||
750 | for (i = 0; i < nr_items; i++) { | |
751 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
752 | if (ret) | |
753 | goto free_tms; | |
754 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
755 | if (ret) | |
756 | goto free_tms; | |
bd989ba3 | 757 | } |
5de865ee | 758 | |
b1a09f1e | 759 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
760 | kfree(tm_list); |
761 | ||
762 | return 0; | |
763 | ||
764 | free_tms: | |
765 | for (i = 0; i < nr_items * 2; i++) { | |
766 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
767 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
768 | kfree(tm_list[i]); | |
769 | } | |
770 | if (locked) | |
b1a09f1e | 771 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
772 | kfree(tm_list); |
773 | ||
774 | return ret; | |
bd989ba3 JS |
775 | } |
776 | ||
db7279a2 | 777 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 778 | { |
5de865ee FDBM |
779 | struct tree_mod_elem **tm_list = NULL; |
780 | int nritems = 0; | |
781 | int i; | |
782 | int ret = 0; | |
783 | ||
784 | if (btrfs_header_level(eb) == 0) | |
785 | return 0; | |
786 | ||
db7279a2 | 787 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
788 | return 0; |
789 | ||
790 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 791 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
792 | if (!tm_list) |
793 | return -ENOMEM; | |
794 | ||
795 | for (i = 0; i < nritems; i++) { | |
796 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
797 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
798 | if (!tm_list[i]) { | |
799 | ret = -ENOMEM; | |
800 | goto free_tms; | |
801 | } | |
802 | } | |
803 | ||
db7279a2 | 804 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
805 | goto free_tms; |
806 | ||
db7279a2 | 807 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 808 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
809 | if (ret) |
810 | goto free_tms; | |
811 | kfree(tm_list); | |
812 | ||
813 | return 0; | |
814 | ||
815 | free_tms: | |
816 | for (i = 0; i < nritems; i++) | |
817 | kfree(tm_list[i]); | |
818 | kfree(tm_list); | |
819 | ||
820 | return ret; | |
bd989ba3 JS |
821 | } |
822 | ||
5d4f98a2 YZ |
823 | /* |
824 | * check if the tree block can be shared by multiple trees | |
825 | */ | |
826 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
827 | struct extent_buffer *buf) | |
828 | { | |
829 | /* | |
92a7cc42 QW |
830 | * Tree blocks not in shareable trees and tree roots are never shared. |
831 | * If a block was allocated after the last snapshot and the block was | |
832 | * not allocated by tree relocation, we know the block is not shared. | |
5d4f98a2 | 833 | */ |
92a7cc42 | 834 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
5d4f98a2 YZ |
835 | buf != root->node && buf != root->commit_root && |
836 | (btrfs_header_generation(buf) <= | |
837 | btrfs_root_last_snapshot(&root->root_item) || | |
838 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
839 | return 1; | |
a79865c6 | 840 | |
5d4f98a2 YZ |
841 | return 0; |
842 | } | |
843 | ||
844 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
845 | struct btrfs_root *root, | |
846 | struct extent_buffer *buf, | |
f0486c68 YZ |
847 | struct extent_buffer *cow, |
848 | int *last_ref) | |
5d4f98a2 | 849 | { |
0b246afa | 850 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
851 | u64 refs; |
852 | u64 owner; | |
853 | u64 flags; | |
854 | u64 new_flags = 0; | |
855 | int ret; | |
856 | ||
857 | /* | |
858 | * Backrefs update rules: | |
859 | * | |
860 | * Always use full backrefs for extent pointers in tree block | |
861 | * allocated by tree relocation. | |
862 | * | |
863 | * If a shared tree block is no longer referenced by its owner | |
864 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
865 | * use full backrefs for extent pointers in tree block. | |
866 | * | |
867 | * If a tree block is been relocating | |
868 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
869 | * use full backrefs for extent pointers in tree block. | |
870 | * The reason for this is some operations (such as drop tree) | |
871 | * are only allowed for blocks use full backrefs. | |
872 | */ | |
873 | ||
874 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 875 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
876 | btrfs_header_level(buf), 1, |
877 | &refs, &flags); | |
be1a5564 MF |
878 | if (ret) |
879 | return ret; | |
e5df9573 MF |
880 | if (refs == 0) { |
881 | ret = -EROFS; | |
0b246afa | 882 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
883 | return ret; |
884 | } | |
5d4f98a2 YZ |
885 | } else { |
886 | refs = 1; | |
887 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
888 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
889 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
890 | else | |
891 | flags = 0; | |
892 | } | |
893 | ||
894 | owner = btrfs_header_owner(buf); | |
895 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
896 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
897 | ||
898 | if (refs > 1) { | |
899 | if ((owner == root->root_key.objectid || | |
900 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
901 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 902 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
903 | if (ret) |
904 | return ret; | |
5d4f98a2 YZ |
905 | |
906 | if (root->root_key.objectid == | |
907 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 908 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
909 | if (ret) |
910 | return ret; | |
e339a6b0 | 911 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
912 | if (ret) |
913 | return ret; | |
5d4f98a2 YZ |
914 | } |
915 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
916 | } else { | |
917 | ||
918 | if (root->root_key.objectid == | |
919 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 920 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 921 | else |
e339a6b0 | 922 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
923 | if (ret) |
924 | return ret; | |
5d4f98a2 YZ |
925 | } |
926 | if (new_flags != 0) { | |
b1c79e09 JB |
927 | int level = btrfs_header_level(buf); |
928 | ||
42c9d0b5 | 929 | ret = btrfs_set_disk_extent_flags(trans, buf, |
b1c79e09 | 930 | new_flags, level, 0); |
be1a5564 MF |
931 | if (ret) |
932 | return ret; | |
5d4f98a2 YZ |
933 | } |
934 | } else { | |
935 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
936 | if (root->root_key.objectid == | |
937 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 938 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 939 | else |
e339a6b0 | 940 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
941 | if (ret) |
942 | return ret; | |
e339a6b0 | 943 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
944 | if (ret) |
945 | return ret; | |
5d4f98a2 | 946 | } |
6a884d7d | 947 | btrfs_clean_tree_block(buf); |
f0486c68 | 948 | *last_ref = 1; |
5d4f98a2 YZ |
949 | } |
950 | return 0; | |
951 | } | |
952 | ||
a6279470 FM |
953 | static struct extent_buffer *alloc_tree_block_no_bg_flush( |
954 | struct btrfs_trans_handle *trans, | |
955 | struct btrfs_root *root, | |
956 | u64 parent_start, | |
957 | const struct btrfs_disk_key *disk_key, | |
958 | int level, | |
959 | u64 hint, | |
960 | u64 empty_size) | |
961 | { | |
962 | struct btrfs_fs_info *fs_info = root->fs_info; | |
963 | struct extent_buffer *ret; | |
964 | ||
965 | /* | |
966 | * If we are COWing a node/leaf from the extent, chunk, device or free | |
967 | * space trees, make sure that we do not finish block group creation of | |
968 | * pending block groups. We do this to avoid a deadlock. | |
969 | * COWing can result in allocation of a new chunk, and flushing pending | |
970 | * block groups (btrfs_create_pending_block_groups()) can be triggered | |
971 | * when finishing allocation of a new chunk. Creation of a pending block | |
972 | * group modifies the extent, chunk, device and free space trees, | |
973 | * therefore we could deadlock with ourselves since we are holding a | |
974 | * lock on an extent buffer that btrfs_create_pending_block_groups() may | |
975 | * try to COW later. | |
976 | * For similar reasons, we also need to delay flushing pending block | |
977 | * groups when splitting a leaf or node, from one of those trees, since | |
978 | * we are holding a write lock on it and its parent or when inserting a | |
979 | * new root node for one of those trees. | |
980 | */ | |
981 | if (root == fs_info->extent_root || | |
982 | root == fs_info->chunk_root || | |
983 | root == fs_info->dev_root || | |
984 | root == fs_info->free_space_root) | |
985 | trans->can_flush_pending_bgs = false; | |
986 | ||
987 | ret = btrfs_alloc_tree_block(trans, root, parent_start, | |
988 | root->root_key.objectid, disk_key, level, | |
989 | hint, empty_size); | |
990 | trans->can_flush_pending_bgs = true; | |
991 | ||
992 | return ret; | |
993 | } | |
994 | ||
d352ac68 | 995 | /* |
d397712b CM |
996 | * does the dirty work in cow of a single block. The parent block (if |
997 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
998 | * dirty and returned locked. If you modify the block it needs to be marked | |
999 | * dirty again. | |
d352ac68 CM |
1000 | * |
1001 | * search_start -- an allocation hint for the new block | |
1002 | * | |
d397712b CM |
1003 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1004 | * bytes the allocator should try to find free next to the block it returns. | |
1005 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1006 | */ |
d397712b | 1007 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1008 | struct btrfs_root *root, |
1009 | struct extent_buffer *buf, | |
1010 | struct extent_buffer *parent, int parent_slot, | |
1011 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1012 | u64 search_start, u64 empty_size) |
02217ed2 | 1013 | { |
0b246afa | 1014 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1015 | struct btrfs_disk_key disk_key; |
5f39d397 | 1016 | struct extent_buffer *cow; |
be1a5564 | 1017 | int level, ret; |
f0486c68 | 1018 | int last_ref = 0; |
925baedd | 1019 | int unlock_orig = 0; |
0f5053eb | 1020 | u64 parent_start = 0; |
7bb86316 | 1021 | |
925baedd CM |
1022 | if (*cow_ret == buf) |
1023 | unlock_orig = 1; | |
1024 | ||
b9447ef8 | 1025 | btrfs_assert_tree_locked(buf); |
925baedd | 1026 | |
92a7cc42 | 1027 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 1028 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 1029 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 1030 | trans->transid != root->last_trans); |
5f39d397 | 1031 | |
7bb86316 | 1032 | level = btrfs_header_level(buf); |
31840ae1 | 1033 | |
5d4f98a2 YZ |
1034 | if (level == 0) |
1035 | btrfs_item_key(buf, &disk_key, 0); | |
1036 | else | |
1037 | btrfs_node_key(buf, &disk_key, 0); | |
1038 | ||
0f5053eb GR |
1039 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1040 | parent_start = parent->start; | |
5d4f98a2 | 1041 | |
a6279470 FM |
1042 | cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, |
1043 | level, search_start, empty_size); | |
54aa1f4d CM |
1044 | if (IS_ERR(cow)) |
1045 | return PTR_ERR(cow); | |
6702ed49 | 1046 | |
b4ce94de CM |
1047 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1048 | ||
58e8012c | 1049 | copy_extent_buffer_full(cow, buf); |
db94535d | 1050 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1051 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1052 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1053 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1054 | BTRFS_HEADER_FLAG_RELOC); | |
1055 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1056 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1057 | else | |
1058 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1059 | |
de37aa51 | 1060 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 1061 | |
be1a5564 | 1062 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1063 | if (ret) { |
66642832 | 1064 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1065 | return ret; |
1066 | } | |
1a40e23b | 1067 | |
92a7cc42 | 1068 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
83d4cfd4 | 1069 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1070 | if (ret) { |
66642832 | 1071 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1072 | return ret; |
93314e3b | 1073 | } |
83d4cfd4 | 1074 | } |
3fd0a558 | 1075 | |
02217ed2 | 1076 | if (buf == root->node) { |
925baedd | 1077 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1078 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1079 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1080 | parent_start = buf->start; | |
925baedd | 1081 | |
67439dad | 1082 | atomic_inc(&cow->refs); |
d9d19a01 DS |
1083 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1084 | BUG_ON(ret < 0); | |
240f62c8 | 1085 | rcu_assign_pointer(root->node, cow); |
925baedd | 1086 | |
f0486c68 | 1087 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1088 | last_ref); |
5f39d397 | 1089 | free_extent_buffer(buf); |
0b86a832 | 1090 | add_root_to_dirty_list(root); |
02217ed2 | 1091 | } else { |
5d4f98a2 | 1092 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1093 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1094 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1095 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1096 | cow->start); |
74493f7a CM |
1097 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1098 | trans->transid); | |
d6025579 | 1099 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1100 | if (last_ref) { |
db7279a2 | 1101 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1102 | if (ret) { |
66642832 | 1103 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1104 | return ret; |
1105 | } | |
1106 | } | |
f0486c68 | 1107 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1108 | last_ref); |
02217ed2 | 1109 | } |
925baedd CM |
1110 | if (unlock_orig) |
1111 | btrfs_tree_unlock(buf); | |
3083ee2e | 1112 | free_extent_buffer_stale(buf); |
ccd467d6 | 1113 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1114 | *cow_ret = cow; |
02217ed2 CM |
1115 | return 0; |
1116 | } | |
1117 | ||
5d9e75c4 JS |
1118 | /* |
1119 | * returns the logical address of the oldest predecessor of the given root. | |
1120 | * entries older than time_seq are ignored. | |
1121 | */ | |
bcd24dab DS |
1122 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1123 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1124 | { |
1125 | struct tree_mod_elem *tm; | |
1126 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1127 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1128 | int looped = 0; |
1129 | ||
1130 | if (!time_seq) | |
35a3621b | 1131 | return NULL; |
5d9e75c4 JS |
1132 | |
1133 | /* | |
298cfd36 CR |
1134 | * the very last operation that's logged for a root is the |
1135 | * replacement operation (if it is replaced at all). this has | |
1136 | * the logical address of the *new* root, making it the very | |
1137 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1138 | */ |
1139 | while (1) { | |
bcd24dab | 1140 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1141 | time_seq); |
1142 | if (!looped && !tm) | |
35a3621b | 1143 | return NULL; |
5d9e75c4 | 1144 | /* |
28da9fb4 JS |
1145 | * if there are no tree operation for the oldest root, we simply |
1146 | * return it. this should only happen if that (old) root is at | |
1147 | * level 0. | |
5d9e75c4 | 1148 | */ |
28da9fb4 JS |
1149 | if (!tm) |
1150 | break; | |
5d9e75c4 | 1151 | |
28da9fb4 JS |
1152 | /* |
1153 | * if there's an operation that's not a root replacement, we | |
1154 | * found the oldest version of our root. normally, we'll find a | |
1155 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1156 | */ | |
5d9e75c4 JS |
1157 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1158 | break; | |
1159 | ||
1160 | found = tm; | |
1161 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1162 | looped = 1; |
1163 | } | |
1164 | ||
a95236d9 JS |
1165 | /* if there's no old root to return, return what we found instead */ |
1166 | if (!found) | |
1167 | found = tm; | |
1168 | ||
5d9e75c4 JS |
1169 | return found; |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1174 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1175 | * time_seq). |
1176 | */ | |
1177 | static void | |
f1ca7e98 JB |
1178 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1179 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1180 | { |
1181 | u32 n; | |
1182 | struct rb_node *next; | |
1183 | struct tree_mod_elem *tm = first_tm; | |
1184 | unsigned long o_dst; | |
1185 | unsigned long o_src; | |
1186 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1187 | ||
1188 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1189 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1190 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1191 | /* |
1192 | * all the operations are recorded with the operator used for | |
1193 | * the modification. as we're going backwards, we do the | |
1194 | * opposite of each operation here. | |
1195 | */ | |
1196 | switch (tm->op) { | |
1197 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1198 | BUG_ON(tm->slot < n); | |
c730ae0c | 1199 | fallthrough; |
95c80bb1 | 1200 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1201 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1202 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1203 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1204 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1205 | tm->generation); | |
4c3e6969 | 1206 | n++; |
5d9e75c4 JS |
1207 | break; |
1208 | case MOD_LOG_KEY_REPLACE: | |
1209 | BUG_ON(tm->slot >= n); | |
1210 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1211 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1212 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1213 | tm->generation); | |
1214 | break; | |
1215 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1216 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1217 | n--; |
1218 | break; | |
1219 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1220 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1221 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1222 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1223 | tm->move.nr_items * p_size); |
1224 | break; | |
1225 | case MOD_LOG_ROOT_REPLACE: | |
1226 | /* | |
1227 | * this operation is special. for roots, this must be | |
1228 | * handled explicitly before rewinding. | |
1229 | * for non-roots, this operation may exist if the node | |
1230 | * was a root: root A -> child B; then A gets empty and | |
1231 | * B is promoted to the new root. in the mod log, we'll | |
1232 | * have a root-replace operation for B, a tree block | |
1233 | * that is no root. we simply ignore that operation. | |
1234 | */ | |
1235 | break; | |
1236 | } | |
1237 | next = rb_next(&tm->node); | |
1238 | if (!next) | |
1239 | break; | |
6b4df8b6 | 1240 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1241 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1242 | break; |
1243 | } | |
b1a09f1e | 1244 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1245 | btrfs_set_header_nritems(eb, n); |
1246 | } | |
1247 | ||
47fb091f | 1248 | /* |
01327610 | 1249 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1250 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1251 | * returned buffer is always read-locked. If the returned buffer is not the | |
1252 | * input buffer, the lock on the input buffer is released and the input buffer | |
1253 | * is freed (its refcount is decremented). | |
1254 | */ | |
5d9e75c4 | 1255 | static struct extent_buffer * |
9ec72677 JB |
1256 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1257 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1258 | { |
1259 | struct extent_buffer *eb_rewin; | |
1260 | struct tree_mod_elem *tm; | |
1261 | ||
1262 | if (!time_seq) | |
1263 | return eb; | |
1264 | ||
1265 | if (btrfs_header_level(eb) == 0) | |
1266 | return eb; | |
1267 | ||
1268 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1269 | if (!tm) | |
1270 | return eb; | |
1271 | ||
9ec72677 | 1272 | btrfs_set_path_blocking(path); |
300aa896 | 1273 | btrfs_set_lock_blocking_read(eb); |
9ec72677 | 1274 | |
5d9e75c4 JS |
1275 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1276 | BUG_ON(tm->slot != 0); | |
da17066c | 1277 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1278 | if (!eb_rewin) { |
9ec72677 | 1279 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1280 | free_extent_buffer(eb); |
1281 | return NULL; | |
1282 | } | |
5d9e75c4 JS |
1283 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1284 | btrfs_set_header_backref_rev(eb_rewin, | |
1285 | btrfs_header_backref_rev(eb)); | |
1286 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1287 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1288 | } else { |
1289 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1290 | if (!eb_rewin) { |
9ec72677 | 1291 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1292 | free_extent_buffer(eb); |
1293 | return NULL; | |
1294 | } | |
5d9e75c4 JS |
1295 | } |
1296 | ||
9ec72677 | 1297 | btrfs_tree_read_unlock_blocking(eb); |
5d9e75c4 JS |
1298 | free_extent_buffer(eb); |
1299 | ||
47fb091f | 1300 | btrfs_tree_read_lock(eb_rewin); |
f1ca7e98 | 1301 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1302 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1303 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1304 | |
1305 | return eb_rewin; | |
1306 | } | |
1307 | ||
8ba97a15 JS |
1308 | /* |
1309 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1310 | * value. If there are no changes, the current root->root_node is returned. If | |
1311 | * anything changed in between, there's a fresh buffer allocated on which the | |
1312 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1313 | * Returns NULL on error (with no locks held). | |
1314 | */ | |
5d9e75c4 JS |
1315 | static inline struct extent_buffer * |
1316 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1317 | { | |
0b246afa | 1318 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1319 | struct tree_mod_elem *tm; |
30b0463a JS |
1320 | struct extent_buffer *eb = NULL; |
1321 | struct extent_buffer *eb_root; | |
efad8a85 | 1322 | u64 eb_root_owner = 0; |
7bfdcf7f | 1323 | struct extent_buffer *old; |
a95236d9 | 1324 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1325 | u64 old_generation = 0; |
a95236d9 | 1326 | u64 logical; |
581c1760 | 1327 | int level; |
5d9e75c4 | 1328 | |
30b0463a | 1329 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1330 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1331 | if (!tm) |
30b0463a | 1332 | return eb_root; |
5d9e75c4 | 1333 | |
a95236d9 JS |
1334 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1335 | old_root = &tm->old_root; | |
1336 | old_generation = tm->generation; | |
1337 | logical = old_root->logical; | |
581c1760 | 1338 | level = old_root->level; |
a95236d9 | 1339 | } else { |
30b0463a | 1340 | logical = eb_root->start; |
581c1760 | 1341 | level = btrfs_header_level(eb_root); |
a95236d9 | 1342 | } |
5d9e75c4 | 1343 | |
0b246afa | 1344 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1345 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1346 | btrfs_tree_read_unlock(eb_root); |
1347 | free_extent_buffer(eb_root); | |
581c1760 | 1348 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1349 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1350 | if (!IS_ERR(old)) | |
1351 | free_extent_buffer(old); | |
0b246afa JM |
1352 | btrfs_warn(fs_info, |
1353 | "failed to read tree block %llu from get_old_root", | |
1354 | logical); | |
834328a8 | 1355 | } else { |
7bfdcf7f LB |
1356 | eb = btrfs_clone_extent_buffer(old); |
1357 | free_extent_buffer(old); | |
834328a8 JS |
1358 | } |
1359 | } else if (old_root) { | |
efad8a85 | 1360 | eb_root_owner = btrfs_header_owner(eb_root); |
30b0463a JS |
1361 | btrfs_tree_read_unlock(eb_root); |
1362 | free_extent_buffer(eb_root); | |
0b246afa | 1363 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1364 | } else { |
300aa896 | 1365 | btrfs_set_lock_blocking_read(eb_root); |
30b0463a | 1366 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1367 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1368 | free_extent_buffer(eb_root); |
834328a8 JS |
1369 | } |
1370 | ||
8ba97a15 JS |
1371 | if (!eb) |
1372 | return NULL; | |
1373 | btrfs_tree_read_lock(eb); | |
a95236d9 | 1374 | if (old_root) { |
5d9e75c4 JS |
1375 | btrfs_set_header_bytenr(eb, eb->start); |
1376 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
efad8a85 | 1377 | btrfs_set_header_owner(eb, eb_root_owner); |
a95236d9 JS |
1378 | btrfs_set_header_level(eb, old_root->level); |
1379 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1380 | } |
28da9fb4 | 1381 | if (tm) |
0b246afa | 1382 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1383 | else |
1384 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1385 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1386 | |
1387 | return eb; | |
1388 | } | |
1389 | ||
5b6602e7 JS |
1390 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1391 | { | |
1392 | struct tree_mod_elem *tm; | |
1393 | int level; | |
30b0463a | 1394 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1395 | |
bcd24dab | 1396 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1397 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1398 | level = tm->old_root.level; | |
1399 | } else { | |
30b0463a | 1400 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1401 | } |
30b0463a | 1402 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1403 | |
1404 | return level; | |
1405 | } | |
1406 | ||
5d4f98a2 YZ |
1407 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1408 | struct btrfs_root *root, | |
1409 | struct extent_buffer *buf) | |
1410 | { | |
f5ee5c9a | 1411 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1412 | return 0; |
fccb84c9 | 1413 | |
d1980131 DS |
1414 | /* Ensure we can see the FORCE_COW bit */ |
1415 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1416 | |
1417 | /* | |
1418 | * We do not need to cow a block if | |
1419 | * 1) this block is not created or changed in this transaction; | |
1420 | * 2) this block does not belong to TREE_RELOC tree; | |
1421 | * 3) the root is not forced COW. | |
1422 | * | |
1423 | * What is forced COW: | |
01327610 | 1424 | * when we create snapshot during committing the transaction, |
52042d8e | 1425 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
1426 | * block to ensure the metadata consistency. |
1427 | */ | |
5d4f98a2 YZ |
1428 | if (btrfs_header_generation(buf) == trans->transid && |
1429 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1430 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1431 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1432 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1433 | return 0; |
1434 | return 1; | |
1435 | } | |
1436 | ||
d352ac68 CM |
1437 | /* |
1438 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1439 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1440 | * once per transaction, as long as it hasn't been written yet |
1441 | */ | |
d397712b | 1442 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1443 | struct btrfs_root *root, struct extent_buffer *buf, |
1444 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1445 | struct extent_buffer **cow_ret) |
6702ed49 | 1446 | { |
0b246afa | 1447 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1448 | u64 search_start; |
f510cfec | 1449 | int ret; |
dc17ff8f | 1450 | |
83354f07 JB |
1451 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
1452 | btrfs_err(fs_info, | |
1453 | "COW'ing blocks on a fs root that's being dropped"); | |
1454 | ||
0b246afa | 1455 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1456 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1457 | trans->transid, |
0b246afa | 1458 | fs_info->running_transaction->transid); |
31b1a2bd | 1459 | |
0b246afa | 1460 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1461 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1462 | trans->transid, fs_info->generation); |
dc17ff8f | 1463 | |
5d4f98a2 | 1464 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1465 | trans->dirty = true; |
6702ed49 CM |
1466 | *cow_ret = buf; |
1467 | return 0; | |
1468 | } | |
c487685d | 1469 | |
ee22184b | 1470 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1471 | |
1472 | if (parent) | |
8bead258 DS |
1473 | btrfs_set_lock_blocking_write(parent); |
1474 | btrfs_set_lock_blocking_write(buf); | |
b4ce94de | 1475 | |
f616f5cd QW |
1476 | /* |
1477 | * Before CoWing this block for later modification, check if it's | |
1478 | * the subtree root and do the delayed subtree trace if needed. | |
1479 | * | |
1480 | * Also We don't care about the error, as it's handled internally. | |
1481 | */ | |
1482 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 1483 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1484 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1485 | |
1486 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1487 | ||
f510cfec | 1488 | return ret; |
6702ed49 CM |
1489 | } |
1490 | ||
d352ac68 CM |
1491 | /* |
1492 | * helper function for defrag to decide if two blocks pointed to by a | |
1493 | * node are actually close by | |
1494 | */ | |
6b80053d | 1495 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1496 | { |
6b80053d | 1497 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1498 | return 1; |
6b80053d | 1499 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1500 | return 1; |
1501 | return 0; | |
1502 | } | |
1503 | ||
081e9573 CM |
1504 | /* |
1505 | * compare two keys in a memcmp fashion | |
1506 | */ | |
310712b2 OS |
1507 | static int comp_keys(const struct btrfs_disk_key *disk, |
1508 | const struct btrfs_key *k2) | |
081e9573 CM |
1509 | { |
1510 | struct btrfs_key k1; | |
1511 | ||
1512 | btrfs_disk_key_to_cpu(&k1, disk); | |
1513 | ||
20736aba | 1514 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1515 | } |
1516 | ||
f3465ca4 JB |
1517 | /* |
1518 | * same as comp_keys only with two btrfs_key's | |
1519 | */ | |
e1f60a65 | 1520 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1521 | { |
1522 | if (k1->objectid > k2->objectid) | |
1523 | return 1; | |
1524 | if (k1->objectid < k2->objectid) | |
1525 | return -1; | |
1526 | if (k1->type > k2->type) | |
1527 | return 1; | |
1528 | if (k1->type < k2->type) | |
1529 | return -1; | |
1530 | if (k1->offset > k2->offset) | |
1531 | return 1; | |
1532 | if (k1->offset < k2->offset) | |
1533 | return -1; | |
1534 | return 0; | |
1535 | } | |
081e9573 | 1536 | |
d352ac68 CM |
1537 | /* |
1538 | * this is used by the defrag code to go through all the | |
1539 | * leaves pointed to by a node and reallocate them so that | |
1540 | * disk order is close to key order | |
1541 | */ | |
6702ed49 | 1542 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1543 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1544 | int start_slot, u64 *last_ret, |
a6b6e75e | 1545 | struct btrfs_key *progress) |
6702ed49 | 1546 | { |
0b246afa | 1547 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1548 | struct extent_buffer *cur; |
6702ed49 | 1549 | u64 blocknr; |
ca7a79ad | 1550 | u64 gen; |
e9d0b13b CM |
1551 | u64 search_start = *last_ret; |
1552 | u64 last_block = 0; | |
6702ed49 CM |
1553 | u64 other; |
1554 | u32 parent_nritems; | |
6702ed49 CM |
1555 | int end_slot; |
1556 | int i; | |
1557 | int err = 0; | |
f2183bde | 1558 | int parent_level; |
6b80053d CM |
1559 | int uptodate; |
1560 | u32 blocksize; | |
081e9573 CM |
1561 | int progress_passed = 0; |
1562 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1563 | |
5708b959 | 1564 | parent_level = btrfs_header_level(parent); |
5708b959 | 1565 | |
0b246afa JM |
1566 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1567 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1568 | |
6b80053d | 1569 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1570 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1571 | end_slot = parent_nritems - 1; |
6702ed49 | 1572 | |
5dfe2be7 | 1573 | if (parent_nritems <= 1) |
6702ed49 CM |
1574 | return 0; |
1575 | ||
8bead258 | 1576 | btrfs_set_lock_blocking_write(parent); |
b4ce94de | 1577 | |
5dfe2be7 | 1578 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1579 | struct btrfs_key first_key; |
6702ed49 | 1580 | int close = 1; |
a6b6e75e | 1581 | |
081e9573 CM |
1582 | btrfs_node_key(parent, &disk_key, i); |
1583 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1584 | continue; | |
1585 | ||
1586 | progress_passed = 1; | |
6b80053d | 1587 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1588 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1589 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1590 | if (last_block == 0) |
1591 | last_block = blocknr; | |
5708b959 | 1592 | |
6702ed49 | 1593 | if (i > 0) { |
6b80053d CM |
1594 | other = btrfs_node_blockptr(parent, i - 1); |
1595 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1596 | } |
5dfe2be7 | 1597 | if (!close && i < end_slot) { |
6b80053d CM |
1598 | other = btrfs_node_blockptr(parent, i + 1); |
1599 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1600 | } |
e9d0b13b CM |
1601 | if (close) { |
1602 | last_block = blocknr; | |
6702ed49 | 1603 | continue; |
e9d0b13b | 1604 | } |
6702ed49 | 1605 | |
0b246afa | 1606 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1607 | if (cur) |
b9fab919 | 1608 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1609 | else |
1610 | uptodate = 0; | |
5708b959 | 1611 | if (!cur || !uptodate) { |
6b80053d | 1612 | if (!cur) { |
581c1760 QW |
1613 | cur = read_tree_block(fs_info, blocknr, gen, |
1614 | parent_level - 1, | |
1615 | &first_key); | |
64c043de LB |
1616 | if (IS_ERR(cur)) { |
1617 | return PTR_ERR(cur); | |
1618 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1619 | free_extent_buffer(cur); |
97d9a8a4 | 1620 | return -EIO; |
416bc658 | 1621 | } |
6b80053d | 1622 | } else if (!uptodate) { |
581c1760 QW |
1623 | err = btrfs_read_buffer(cur, gen, |
1624 | parent_level - 1,&first_key); | |
018642a1 TI |
1625 | if (err) { |
1626 | free_extent_buffer(cur); | |
1627 | return err; | |
1628 | } | |
f2183bde | 1629 | } |
6702ed49 | 1630 | } |
e9d0b13b | 1631 | if (search_start == 0) |
6b80053d | 1632 | search_start = last_block; |
e9d0b13b | 1633 | |
e7a84565 | 1634 | btrfs_tree_lock(cur); |
8bead258 | 1635 | btrfs_set_lock_blocking_write(cur); |
6b80053d | 1636 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1637 | &cur, search_start, |
6b80053d | 1638 | min(16 * blocksize, |
9fa8cfe7 | 1639 | (end_slot - i) * blocksize)); |
252c38f0 | 1640 | if (err) { |
e7a84565 | 1641 | btrfs_tree_unlock(cur); |
6b80053d | 1642 | free_extent_buffer(cur); |
6702ed49 | 1643 | break; |
252c38f0 | 1644 | } |
e7a84565 CM |
1645 | search_start = cur->start; |
1646 | last_block = cur->start; | |
f2183bde | 1647 | *last_ret = search_start; |
e7a84565 CM |
1648 | btrfs_tree_unlock(cur); |
1649 | free_extent_buffer(cur); | |
6702ed49 CM |
1650 | } |
1651 | return err; | |
1652 | } | |
1653 | ||
74123bd7 | 1654 | /* |
5f39d397 CM |
1655 | * search for key in the extent_buffer. The items start at offset p, |
1656 | * and they are item_size apart. There are 'max' items in p. | |
1657 | * | |
74123bd7 CM |
1658 | * the slot in the array is returned via slot, and it points to |
1659 | * the place where you would insert key if it is not found in | |
1660 | * the array. | |
1661 | * | |
1662 | * slot may point to max if the key is bigger than all of the keys | |
1663 | */ | |
e02119d5 | 1664 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1665 | unsigned long p, int item_size, |
1666 | const struct btrfs_key *key, | |
e02119d5 | 1667 | int max, int *slot) |
be0e5c09 CM |
1668 | { |
1669 | int low = 0; | |
1670 | int high = max; | |
be0e5c09 | 1671 | int ret; |
5cd17f34 | 1672 | const int key_size = sizeof(struct btrfs_disk_key); |
be0e5c09 | 1673 | |
5e24e9af LB |
1674 | if (low > high) { |
1675 | btrfs_err(eb->fs_info, | |
1676 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1677 | __func__, low, high, eb->start, | |
1678 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1679 | return -EINVAL; | |
1680 | } | |
1681 | ||
d397712b | 1682 | while (low < high) { |
5cd17f34 DS |
1683 | unsigned long oip; |
1684 | unsigned long offset; | |
1685 | struct btrfs_disk_key *tmp; | |
1686 | struct btrfs_disk_key unaligned; | |
1687 | int mid; | |
1688 | ||
be0e5c09 | 1689 | mid = (low + high) / 2; |
5f39d397 | 1690 | offset = p + mid * item_size; |
5cd17f34 | 1691 | oip = offset_in_page(offset); |
5f39d397 | 1692 | |
5cd17f34 DS |
1693 | if (oip + key_size <= PAGE_SIZE) { |
1694 | const unsigned long idx = offset >> PAGE_SHIFT; | |
1695 | char *kaddr = page_address(eb->pages[idx]); | |
5f39d397 | 1696 | |
5cd17f34 | 1697 | tmp = (struct btrfs_disk_key *)(kaddr + oip); |
5f39d397 | 1698 | } else { |
5cd17f34 DS |
1699 | read_extent_buffer(eb, &unaligned, offset, key_size); |
1700 | tmp = &unaligned; | |
5f39d397 | 1701 | } |
5cd17f34 | 1702 | |
be0e5c09 CM |
1703 | ret = comp_keys(tmp, key); |
1704 | ||
1705 | if (ret < 0) | |
1706 | low = mid + 1; | |
1707 | else if (ret > 0) | |
1708 | high = mid; | |
1709 | else { | |
1710 | *slot = mid; | |
1711 | return 0; | |
1712 | } | |
1713 | } | |
1714 | *slot = low; | |
1715 | return 1; | |
1716 | } | |
1717 | ||
97571fd0 CM |
1718 | /* |
1719 | * simple bin_search frontend that does the right thing for | |
1720 | * leaves vs nodes | |
1721 | */ | |
a74b35ec | 1722 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
e3b83361 | 1723 | int *slot) |
be0e5c09 | 1724 | { |
e3b83361 | 1725 | if (btrfs_header_level(eb) == 0) |
5f39d397 CM |
1726 | return generic_bin_search(eb, |
1727 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1728 | sizeof(struct btrfs_item), |
5f39d397 | 1729 | key, btrfs_header_nritems(eb), |
7518a238 | 1730 | slot); |
f775738f | 1731 | else |
5f39d397 CM |
1732 | return generic_bin_search(eb, |
1733 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1734 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1735 | key, btrfs_header_nritems(eb), |
7518a238 | 1736 | slot); |
be0e5c09 CM |
1737 | } |
1738 | ||
f0486c68 YZ |
1739 | static void root_add_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 | ||
1747 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1748 | { | |
1749 | spin_lock(&root->accounting_lock); | |
1750 | btrfs_set_root_used(&root->root_item, | |
1751 | btrfs_root_used(&root->root_item) - size); | |
1752 | spin_unlock(&root->accounting_lock); | |
1753 | } | |
1754 | ||
d352ac68 CM |
1755 | /* given a node and slot number, this reads the blocks it points to. The |
1756 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1757 | */ |
4b231ae4 DS |
1758 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
1759 | int slot) | |
bb803951 | 1760 | { |
ca7a79ad | 1761 | int level = btrfs_header_level(parent); |
416bc658 | 1762 | struct extent_buffer *eb; |
581c1760 | 1763 | struct btrfs_key first_key; |
416bc658 | 1764 | |
fb770ae4 LB |
1765 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1766 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1767 | |
1768 | BUG_ON(level == 0); | |
1769 | ||
581c1760 | 1770 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 1771 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1772 | btrfs_node_ptr_generation(parent, slot), |
1773 | level - 1, &first_key); | |
fb770ae4 LB |
1774 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1775 | free_extent_buffer(eb); | |
1776 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1777 | } |
1778 | ||
1779 | return eb; | |
bb803951 CM |
1780 | } |
1781 | ||
d352ac68 CM |
1782 | /* |
1783 | * node level balancing, used to make sure nodes are in proper order for | |
1784 | * item deletion. We balance from the top down, so we have to make sure | |
1785 | * that a deletion won't leave an node completely empty later on. | |
1786 | */ | |
e02119d5 | 1787 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1788 | struct btrfs_root *root, |
1789 | struct btrfs_path *path, int level) | |
bb803951 | 1790 | { |
0b246afa | 1791 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1792 | struct extent_buffer *right = NULL; |
1793 | struct extent_buffer *mid; | |
1794 | struct extent_buffer *left = NULL; | |
1795 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1796 | int ret = 0; |
1797 | int wret; | |
1798 | int pslot; | |
bb803951 | 1799 | int orig_slot = path->slots[level]; |
79f95c82 | 1800 | u64 orig_ptr; |
bb803951 | 1801 | |
98e6b1eb | 1802 | ASSERT(level > 0); |
bb803951 | 1803 | |
5f39d397 | 1804 | mid = path->nodes[level]; |
b4ce94de | 1805 | |
bd681513 CM |
1806 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1807 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1808 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1809 | ||
1d4f8a0c | 1810 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1811 | |
a05a9bb1 | 1812 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1813 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1814 | pslot = path->slots[level + 1]; |
1815 | } | |
bb803951 | 1816 | |
40689478 CM |
1817 | /* |
1818 | * deal with the case where there is only one pointer in the root | |
1819 | * by promoting the node below to a root | |
1820 | */ | |
5f39d397 CM |
1821 | if (!parent) { |
1822 | struct extent_buffer *child; | |
bb803951 | 1823 | |
5f39d397 | 1824 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1825 | return 0; |
1826 | ||
1827 | /* promote the child to a root */ | |
4b231ae4 | 1828 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
1829 | if (IS_ERR(child)) { |
1830 | ret = PTR_ERR(child); | |
0b246afa | 1831 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1832 | goto enospc; |
1833 | } | |
1834 | ||
925baedd | 1835 | btrfs_tree_lock(child); |
8bead258 | 1836 | btrfs_set_lock_blocking_write(child); |
9fa8cfe7 | 1837 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1838 | if (ret) { |
1839 | btrfs_tree_unlock(child); | |
1840 | free_extent_buffer(child); | |
1841 | goto enospc; | |
1842 | } | |
2f375ab9 | 1843 | |
d9d19a01 DS |
1844 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1845 | BUG_ON(ret < 0); | |
240f62c8 | 1846 | rcu_assign_pointer(root->node, child); |
925baedd | 1847 | |
0b86a832 | 1848 | add_root_to_dirty_list(root); |
925baedd | 1849 | btrfs_tree_unlock(child); |
b4ce94de | 1850 | |
925baedd | 1851 | path->locks[level] = 0; |
bb803951 | 1852 | path->nodes[level] = NULL; |
6a884d7d | 1853 | btrfs_clean_tree_block(mid); |
925baedd | 1854 | btrfs_tree_unlock(mid); |
bb803951 | 1855 | /* once for the path */ |
5f39d397 | 1856 | free_extent_buffer(mid); |
f0486c68 YZ |
1857 | |
1858 | root_sub_used(root, mid->len); | |
5581a51a | 1859 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1860 | /* once for the root ptr */ |
3083ee2e | 1861 | free_extent_buffer_stale(mid); |
f0486c68 | 1862 | return 0; |
bb803951 | 1863 | } |
5f39d397 | 1864 | if (btrfs_header_nritems(mid) > |
0b246afa | 1865 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1866 | return 0; |
1867 | ||
4b231ae4 | 1868 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1869 | if (IS_ERR(left)) |
1870 | left = NULL; | |
1871 | ||
5f39d397 | 1872 | if (left) { |
925baedd | 1873 | btrfs_tree_lock(left); |
8bead258 | 1874 | btrfs_set_lock_blocking_write(left); |
5f39d397 | 1875 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1876 | parent, pslot - 1, &left); |
54aa1f4d CM |
1877 | if (wret) { |
1878 | ret = wret; | |
1879 | goto enospc; | |
1880 | } | |
2cc58cf2 | 1881 | } |
fb770ae4 | 1882 | |
4b231ae4 | 1883 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1884 | if (IS_ERR(right)) |
1885 | right = NULL; | |
1886 | ||
5f39d397 | 1887 | if (right) { |
925baedd | 1888 | btrfs_tree_lock(right); |
8bead258 | 1889 | btrfs_set_lock_blocking_write(right); |
5f39d397 | 1890 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1891 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1892 | if (wret) { |
1893 | ret = wret; | |
1894 | goto enospc; | |
1895 | } | |
1896 | } | |
1897 | ||
1898 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1899 | if (left) { |
1900 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 1901 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
1902 | if (wret < 0) |
1903 | ret = wret; | |
bb803951 | 1904 | } |
79f95c82 CM |
1905 | |
1906 | /* | |
1907 | * then try to empty the right most buffer into the middle | |
1908 | */ | |
5f39d397 | 1909 | if (right) { |
d30a668f | 1910 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1911 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1912 | ret = wret; |
5f39d397 | 1913 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 1914 | btrfs_clean_tree_block(right); |
925baedd | 1915 | btrfs_tree_unlock(right); |
afe5fea7 | 1916 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1917 | root_sub_used(root, right->len); |
5581a51a | 1918 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1919 | free_extent_buffer_stale(right); |
f0486c68 | 1920 | right = NULL; |
bb803951 | 1921 | } else { |
5f39d397 CM |
1922 | struct btrfs_disk_key right_key; |
1923 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1924 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1925 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1926 | BUG_ON(ret < 0); | |
5f39d397 CM |
1927 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1928 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1929 | } |
1930 | } | |
5f39d397 | 1931 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1932 | /* |
1933 | * we're not allowed to leave a node with one item in the | |
1934 | * tree during a delete. A deletion from lower in the tree | |
1935 | * could try to delete the only pointer in this node. | |
1936 | * So, pull some keys from the left. | |
1937 | * There has to be a left pointer at this point because | |
1938 | * otherwise we would have pulled some pointers from the | |
1939 | * right | |
1940 | */ | |
305a26af MF |
1941 | if (!left) { |
1942 | ret = -EROFS; | |
0b246afa | 1943 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1944 | goto enospc; |
1945 | } | |
55d32ed8 | 1946 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1947 | if (wret < 0) { |
79f95c82 | 1948 | ret = wret; |
54aa1f4d CM |
1949 | goto enospc; |
1950 | } | |
bce4eae9 | 1951 | if (wret == 1) { |
d30a668f | 1952 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1953 | if (wret < 0) |
1954 | ret = wret; | |
1955 | } | |
79f95c82 CM |
1956 | BUG_ON(wret == 1); |
1957 | } | |
5f39d397 | 1958 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 1959 | btrfs_clean_tree_block(mid); |
925baedd | 1960 | btrfs_tree_unlock(mid); |
afe5fea7 | 1961 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1962 | root_sub_used(root, mid->len); |
5581a51a | 1963 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1964 | free_extent_buffer_stale(mid); |
f0486c68 | 1965 | mid = NULL; |
79f95c82 CM |
1966 | } else { |
1967 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1968 | struct btrfs_disk_key mid_key; |
1969 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
1970 | ret = tree_mod_log_insert_key(parent, pslot, |
1971 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1972 | BUG_ON(ret < 0); | |
5f39d397 CM |
1973 | btrfs_set_node_key(parent, &mid_key, pslot); |
1974 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1975 | } |
bb803951 | 1976 | |
79f95c82 | 1977 | /* update the path */ |
5f39d397 CM |
1978 | if (left) { |
1979 | if (btrfs_header_nritems(left) > orig_slot) { | |
67439dad | 1980 | atomic_inc(&left->refs); |
925baedd | 1981 | /* left was locked after cow */ |
5f39d397 | 1982 | path->nodes[level] = left; |
bb803951 CM |
1983 | path->slots[level + 1] -= 1; |
1984 | path->slots[level] = orig_slot; | |
925baedd CM |
1985 | if (mid) { |
1986 | btrfs_tree_unlock(mid); | |
5f39d397 | 1987 | free_extent_buffer(mid); |
925baedd | 1988 | } |
bb803951 | 1989 | } else { |
5f39d397 | 1990 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1991 | path->slots[level] = orig_slot; |
1992 | } | |
1993 | } | |
79f95c82 | 1994 | /* double check we haven't messed things up */ |
e20d96d6 | 1995 | if (orig_ptr != |
5f39d397 | 1996 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1997 | BUG(); |
54aa1f4d | 1998 | enospc: |
925baedd CM |
1999 | if (right) { |
2000 | btrfs_tree_unlock(right); | |
5f39d397 | 2001 | free_extent_buffer(right); |
925baedd CM |
2002 | } |
2003 | if (left) { | |
2004 | if (path->nodes[level] != left) | |
2005 | btrfs_tree_unlock(left); | |
5f39d397 | 2006 | free_extent_buffer(left); |
925baedd | 2007 | } |
bb803951 CM |
2008 | return ret; |
2009 | } | |
2010 | ||
d352ac68 CM |
2011 | /* Node balancing for insertion. Here we only split or push nodes around |
2012 | * when they are completely full. This is also done top down, so we | |
2013 | * have to be pessimistic. | |
2014 | */ | |
d397712b | 2015 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2016 | struct btrfs_root *root, |
2017 | struct btrfs_path *path, int level) | |
e66f709b | 2018 | { |
0b246afa | 2019 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2020 | struct extent_buffer *right = NULL; |
2021 | struct extent_buffer *mid; | |
2022 | struct extent_buffer *left = NULL; | |
2023 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2024 | int ret = 0; |
2025 | int wret; | |
2026 | int pslot; | |
2027 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2028 | |
2029 | if (level == 0) | |
2030 | return 1; | |
2031 | ||
5f39d397 | 2032 | mid = path->nodes[level]; |
7bb86316 | 2033 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2034 | |
a05a9bb1 | 2035 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2036 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2037 | pslot = path->slots[level + 1]; |
2038 | } | |
e66f709b | 2039 | |
5f39d397 | 2040 | if (!parent) |
e66f709b | 2041 | return 1; |
e66f709b | 2042 | |
4b231ae4 | 2043 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
2044 | if (IS_ERR(left)) |
2045 | left = NULL; | |
e66f709b CM |
2046 | |
2047 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2048 | if (left) { |
e66f709b | 2049 | u32 left_nr; |
925baedd CM |
2050 | |
2051 | btrfs_tree_lock(left); | |
8bead258 | 2052 | btrfs_set_lock_blocking_write(left); |
b4ce94de | 2053 | |
5f39d397 | 2054 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2055 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2056 | wret = 1; |
2057 | } else { | |
5f39d397 | 2058 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2059 | pslot - 1, &left); |
54aa1f4d CM |
2060 | if (ret) |
2061 | wret = 1; | |
2062 | else { | |
d30a668f | 2063 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 2064 | } |
33ade1f8 | 2065 | } |
e66f709b CM |
2066 | if (wret < 0) |
2067 | ret = wret; | |
2068 | if (wret == 0) { | |
5f39d397 | 2069 | struct btrfs_disk_key disk_key; |
e66f709b | 2070 | orig_slot += left_nr; |
5f39d397 | 2071 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2072 | ret = tree_mod_log_insert_key(parent, pslot, |
2073 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2074 | BUG_ON(ret < 0); | |
5f39d397 CM |
2075 | btrfs_set_node_key(parent, &disk_key, pslot); |
2076 | btrfs_mark_buffer_dirty(parent); | |
2077 | if (btrfs_header_nritems(left) > orig_slot) { | |
2078 | path->nodes[level] = left; | |
e66f709b CM |
2079 | path->slots[level + 1] -= 1; |
2080 | path->slots[level] = orig_slot; | |
925baedd | 2081 | btrfs_tree_unlock(mid); |
5f39d397 | 2082 | free_extent_buffer(mid); |
e66f709b CM |
2083 | } else { |
2084 | orig_slot -= | |
5f39d397 | 2085 | btrfs_header_nritems(left); |
e66f709b | 2086 | path->slots[level] = orig_slot; |
925baedd | 2087 | btrfs_tree_unlock(left); |
5f39d397 | 2088 | free_extent_buffer(left); |
e66f709b | 2089 | } |
e66f709b CM |
2090 | return 0; |
2091 | } | |
925baedd | 2092 | btrfs_tree_unlock(left); |
5f39d397 | 2093 | free_extent_buffer(left); |
e66f709b | 2094 | } |
4b231ae4 | 2095 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
2096 | if (IS_ERR(right)) |
2097 | right = NULL; | |
e66f709b CM |
2098 | |
2099 | /* | |
2100 | * then try to empty the right most buffer into the middle | |
2101 | */ | |
5f39d397 | 2102 | if (right) { |
33ade1f8 | 2103 | u32 right_nr; |
b4ce94de | 2104 | |
925baedd | 2105 | btrfs_tree_lock(right); |
8bead258 | 2106 | btrfs_set_lock_blocking_write(right); |
b4ce94de | 2107 | |
5f39d397 | 2108 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2109 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2110 | wret = 1; |
2111 | } else { | |
5f39d397 CM |
2112 | ret = btrfs_cow_block(trans, root, right, |
2113 | parent, pslot + 1, | |
9fa8cfe7 | 2114 | &right); |
54aa1f4d CM |
2115 | if (ret) |
2116 | wret = 1; | |
2117 | else { | |
55d32ed8 | 2118 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 2119 | } |
33ade1f8 | 2120 | } |
e66f709b CM |
2121 | if (wret < 0) |
2122 | ret = wret; | |
2123 | if (wret == 0) { | |
5f39d397 CM |
2124 | struct btrfs_disk_key disk_key; |
2125 | ||
2126 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2127 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2128 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2129 | BUG_ON(ret < 0); | |
5f39d397 CM |
2130 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2131 | btrfs_mark_buffer_dirty(parent); | |
2132 | ||
2133 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2134 | path->nodes[level] = right; | |
e66f709b CM |
2135 | path->slots[level + 1] += 1; |
2136 | path->slots[level] = orig_slot - | |
5f39d397 | 2137 | btrfs_header_nritems(mid); |
925baedd | 2138 | btrfs_tree_unlock(mid); |
5f39d397 | 2139 | free_extent_buffer(mid); |
e66f709b | 2140 | } else { |
925baedd | 2141 | btrfs_tree_unlock(right); |
5f39d397 | 2142 | free_extent_buffer(right); |
e66f709b | 2143 | } |
e66f709b CM |
2144 | return 0; |
2145 | } | |
925baedd | 2146 | btrfs_tree_unlock(right); |
5f39d397 | 2147 | free_extent_buffer(right); |
e66f709b | 2148 | } |
e66f709b CM |
2149 | return 1; |
2150 | } | |
2151 | ||
3c69faec | 2152 | /* |
d352ac68 CM |
2153 | * readahead one full node of leaves, finding things that are close |
2154 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2155 | */ |
2ff7e61e | 2156 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2157 | struct btrfs_path *path, |
2158 | int level, int slot, u64 objectid) | |
3c69faec | 2159 | { |
5f39d397 | 2160 | struct extent_buffer *node; |
01f46658 | 2161 | struct btrfs_disk_key disk_key; |
3c69faec | 2162 | u32 nritems; |
3c69faec | 2163 | u64 search; |
a7175319 | 2164 | u64 target; |
6b80053d | 2165 | u64 nread = 0; |
5f39d397 | 2166 | struct extent_buffer *eb; |
6b80053d CM |
2167 | u32 nr; |
2168 | u32 blocksize; | |
2169 | u32 nscan = 0; | |
db94535d | 2170 | |
a6b6e75e | 2171 | if (level != 1) |
6702ed49 CM |
2172 | return; |
2173 | ||
2174 | if (!path->nodes[level]) | |
3c69faec CM |
2175 | return; |
2176 | ||
5f39d397 | 2177 | node = path->nodes[level]; |
925baedd | 2178 | |
3c69faec | 2179 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2180 | blocksize = fs_info->nodesize; |
2181 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2182 | if (eb) { |
2183 | free_extent_buffer(eb); | |
3c69faec CM |
2184 | return; |
2185 | } | |
2186 | ||
a7175319 | 2187 | target = search; |
6b80053d | 2188 | |
5f39d397 | 2189 | nritems = btrfs_header_nritems(node); |
6b80053d | 2190 | nr = slot; |
25b8b936 | 2191 | |
d397712b | 2192 | while (1) { |
e4058b54 | 2193 | if (path->reada == READA_BACK) { |
6b80053d CM |
2194 | if (nr == 0) |
2195 | break; | |
2196 | nr--; | |
e4058b54 | 2197 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2198 | nr++; |
2199 | if (nr >= nritems) | |
2200 | break; | |
3c69faec | 2201 | } |
e4058b54 | 2202 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2203 | btrfs_node_key(node, &disk_key, nr); |
2204 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2205 | break; | |
2206 | } | |
6b80053d | 2207 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2208 | if ((search <= target && target - search <= 65536) || |
2209 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2210 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2211 | nread += blocksize; |
2212 | } | |
2213 | nscan++; | |
a7175319 | 2214 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2215 | break; |
3c69faec CM |
2216 | } |
2217 | } | |
925baedd | 2218 | |
2ff7e61e | 2219 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2220 | struct btrfs_path *path, int level) |
b4ce94de CM |
2221 | { |
2222 | int slot; | |
2223 | int nritems; | |
2224 | struct extent_buffer *parent; | |
2225 | struct extent_buffer *eb; | |
2226 | u64 gen; | |
2227 | u64 block1 = 0; | |
2228 | u64 block2 = 0; | |
b4ce94de | 2229 | |
8c594ea8 | 2230 | parent = path->nodes[level + 1]; |
b4ce94de | 2231 | if (!parent) |
0b08851f | 2232 | return; |
b4ce94de CM |
2233 | |
2234 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2235 | slot = path->slots[level + 1]; |
b4ce94de CM |
2236 | |
2237 | if (slot > 0) { | |
2238 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2239 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2240 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2241 | /* |
2242 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2243 | * don't want to return eagain here. That will loop | |
2244 | * forever | |
2245 | */ | |
2246 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2247 | block1 = 0; |
2248 | free_extent_buffer(eb); | |
2249 | } | |
8c594ea8 | 2250 | if (slot + 1 < nritems) { |
b4ce94de CM |
2251 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2252 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2253 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2254 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2255 | block2 = 0; |
2256 | free_extent_buffer(eb); | |
2257 | } | |
8c594ea8 | 2258 | |
0b08851f | 2259 | if (block1) |
2ff7e61e | 2260 | readahead_tree_block(fs_info, block1); |
0b08851f | 2261 | if (block2) |
2ff7e61e | 2262 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2263 | } |
2264 | ||
2265 | ||
d352ac68 | 2266 | /* |
d397712b CM |
2267 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2268 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2269 | * operations on the tree might require changing key pointers higher up in the | |
2270 | * tree. | |
d352ac68 | 2271 | * |
d397712b CM |
2272 | * callers might also have set path->keep_locks, which tells this code to keep |
2273 | * the lock if the path points to the last slot in the block. This is part of | |
2274 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2275 | * |
d397712b CM |
2276 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2277 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2278 | */ |
e02119d5 | 2279 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2280 | int lowest_unlock, int min_write_lock_level, |
2281 | int *write_lock_level) | |
925baedd CM |
2282 | { |
2283 | int i; | |
2284 | int skip_level = level; | |
051e1b9f | 2285 | int no_skips = 0; |
925baedd CM |
2286 | struct extent_buffer *t; |
2287 | ||
2288 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2289 | if (!path->nodes[i]) | |
2290 | break; | |
2291 | if (!path->locks[i]) | |
2292 | break; | |
051e1b9f | 2293 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2294 | skip_level = i + 1; |
2295 | continue; | |
2296 | } | |
051e1b9f | 2297 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2298 | u32 nritems; |
2299 | t = path->nodes[i]; | |
2300 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2301 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2302 | skip_level = i + 1; |
2303 | continue; | |
2304 | } | |
2305 | } | |
051e1b9f CM |
2306 | if (skip_level < i && i >= lowest_unlock) |
2307 | no_skips = 1; | |
2308 | ||
925baedd | 2309 | t = path->nodes[i]; |
d80bb3f9 | 2310 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2311 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2312 | path->locks[i] = 0; |
f7c79f30 CM |
2313 | if (write_lock_level && |
2314 | i > min_write_lock_level && | |
2315 | i <= *write_lock_level) { | |
2316 | *write_lock_level = i - 1; | |
2317 | } | |
925baedd CM |
2318 | } |
2319 | } | |
2320 | } | |
2321 | ||
c8c42864 CM |
2322 | /* |
2323 | * helper function for btrfs_search_slot. The goal is to find a block | |
2324 | * in cache without setting the path to blocking. If we find the block | |
2325 | * we return zero and the path is unchanged. | |
2326 | * | |
2327 | * If we can't find the block, we set the path blocking and do some | |
2328 | * reada. -EAGAIN is returned and the search must be repeated. | |
2329 | */ | |
2330 | static int | |
d07b8528 LB |
2331 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2332 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2333 | const struct btrfs_key *key) |
c8c42864 | 2334 | { |
0b246afa | 2335 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2336 | u64 blocknr; |
2337 | u64 gen; | |
c8c42864 | 2338 | struct extent_buffer *tmp; |
581c1760 | 2339 | struct btrfs_key first_key; |
76a05b35 | 2340 | int ret; |
581c1760 | 2341 | int parent_level; |
c8c42864 | 2342 | |
213ff4b7 NB |
2343 | blocknr = btrfs_node_blockptr(*eb_ret, slot); |
2344 | gen = btrfs_node_ptr_generation(*eb_ret, slot); | |
2345 | parent_level = btrfs_header_level(*eb_ret); | |
2346 | btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); | |
c8c42864 | 2347 | |
0b246afa | 2348 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2349 | if (tmp) { |
b9fab919 | 2350 | /* first we do an atomic uptodate check */ |
bdf7c00e | 2351 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
2352 | /* |
2353 | * Do extra check for first_key, eb can be stale due to | |
2354 | * being cached, read from scrub, or have multiple | |
2355 | * parents (shared tree blocks). | |
2356 | */ | |
e064d5e9 | 2357 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
2358 | parent_level - 1, &first_key, gen)) { |
2359 | free_extent_buffer(tmp); | |
2360 | return -EUCLEAN; | |
2361 | } | |
bdf7c00e JB |
2362 | *eb_ret = tmp; |
2363 | return 0; | |
2364 | } | |
2365 | ||
2366 | /* the pages were up to date, but we failed | |
2367 | * the generation number check. Do a full | |
2368 | * read for the generation number that is correct. | |
2369 | * We must do this without dropping locks so | |
2370 | * we can trust our generation number | |
2371 | */ | |
2372 | btrfs_set_path_blocking(p); | |
2373 | ||
2374 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2375 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2376 | if (!ret) { |
2377 | *eb_ret = tmp; | |
2378 | return 0; | |
cb44921a | 2379 | } |
bdf7c00e JB |
2380 | free_extent_buffer(tmp); |
2381 | btrfs_release_path(p); | |
2382 | return -EIO; | |
c8c42864 CM |
2383 | } |
2384 | ||
2385 | /* | |
2386 | * reduce lock contention at high levels | |
2387 | * of the btree by dropping locks before | |
76a05b35 CM |
2388 | * we read. Don't release the lock on the current |
2389 | * level because we need to walk this node to figure | |
2390 | * out which blocks to read. | |
c8c42864 | 2391 | */ |
8c594ea8 CM |
2392 | btrfs_unlock_up_safe(p, level + 1); |
2393 | btrfs_set_path_blocking(p); | |
2394 | ||
e4058b54 | 2395 | if (p->reada != READA_NONE) |
2ff7e61e | 2396 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2397 | |
76a05b35 | 2398 | ret = -EAGAIN; |
02a3307a | 2399 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2400 | &first_key); |
64c043de | 2401 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2402 | /* |
2403 | * If the read above didn't mark this buffer up to date, | |
2404 | * it will never end up being up to date. Set ret to EIO now | |
2405 | * and give up so that our caller doesn't loop forever | |
2406 | * on our EAGAINs. | |
2407 | */ | |
e6a1d6fd | 2408 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2409 | ret = -EIO; |
c8c42864 | 2410 | free_extent_buffer(tmp); |
c871b0f2 LB |
2411 | } else { |
2412 | ret = PTR_ERR(tmp); | |
76a05b35 | 2413 | } |
02a3307a LB |
2414 | |
2415 | btrfs_release_path(p); | |
76a05b35 | 2416 | return ret; |
c8c42864 CM |
2417 | } |
2418 | ||
2419 | /* | |
2420 | * helper function for btrfs_search_slot. This does all of the checks | |
2421 | * for node-level blocks and does any balancing required based on | |
2422 | * the ins_len. | |
2423 | * | |
2424 | * If no extra work was required, zero is returned. If we had to | |
2425 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2426 | * start over | |
2427 | */ | |
2428 | static int | |
2429 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2430 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2431 | struct extent_buffer *b, int level, int ins_len, |
2432 | int *write_lock_level) | |
c8c42864 | 2433 | { |
0b246afa | 2434 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2435 | int ret; |
0b246afa | 2436 | |
c8c42864 | 2437 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2438 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2439 | int sret; |
2440 | ||
bd681513 CM |
2441 | if (*write_lock_level < level + 1) { |
2442 | *write_lock_level = level + 1; | |
2443 | btrfs_release_path(p); | |
2444 | goto again; | |
2445 | } | |
2446 | ||
c8c42864 | 2447 | btrfs_set_path_blocking(p); |
2ff7e61e | 2448 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2449 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2450 | |
2451 | BUG_ON(sret > 0); | |
2452 | if (sret) { | |
2453 | ret = sret; | |
2454 | goto done; | |
2455 | } | |
2456 | b = p->nodes[level]; | |
2457 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2458 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2459 | int sret; |
2460 | ||
bd681513 CM |
2461 | if (*write_lock_level < level + 1) { |
2462 | *write_lock_level = level + 1; | |
2463 | btrfs_release_path(p); | |
2464 | goto again; | |
2465 | } | |
2466 | ||
c8c42864 | 2467 | btrfs_set_path_blocking(p); |
2ff7e61e | 2468 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2469 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2470 | |
2471 | if (sret) { | |
2472 | ret = sret; | |
2473 | goto done; | |
2474 | } | |
2475 | b = p->nodes[level]; | |
2476 | if (!b) { | |
b3b4aa74 | 2477 | btrfs_release_path(p); |
c8c42864 CM |
2478 | goto again; |
2479 | } | |
2480 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2481 | } | |
2482 | return 0; | |
2483 | ||
2484 | again: | |
2485 | ret = -EAGAIN; | |
2486 | done: | |
2487 | return ret; | |
2488 | } | |
2489 | ||
381cf658 | 2490 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2491 | u64 iobjectid, u64 ioff, u8 key_type, |
2492 | struct btrfs_key *found_key) | |
2493 | { | |
2494 | int ret; | |
2495 | struct btrfs_key key; | |
2496 | struct extent_buffer *eb; | |
381cf658 DS |
2497 | |
2498 | ASSERT(path); | |
1d4c08e0 | 2499 | ASSERT(found_key); |
e33d5c3d KN |
2500 | |
2501 | key.type = key_type; | |
2502 | key.objectid = iobjectid; | |
2503 | key.offset = ioff; | |
2504 | ||
2505 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2506 | if (ret < 0) |
e33d5c3d KN |
2507 | return ret; |
2508 | ||
2509 | eb = path->nodes[0]; | |
2510 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2511 | ret = btrfs_next_leaf(fs_root, path); | |
2512 | if (ret) | |
2513 | return ret; | |
2514 | eb = path->nodes[0]; | |
2515 | } | |
2516 | ||
2517 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2518 | if (found_key->type != key.type || | |
2519 | found_key->objectid != key.objectid) | |
2520 | return 1; | |
2521 | ||
2522 | return 0; | |
2523 | } | |
2524 | ||
1fc28d8e LB |
2525 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2526 | struct btrfs_path *p, | |
2527 | int write_lock_level) | |
2528 | { | |
2529 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2530 | struct extent_buffer *b; | |
2531 | int root_lock; | |
2532 | int level = 0; | |
2533 | ||
2534 | /* We try very hard to do read locks on the root */ | |
2535 | root_lock = BTRFS_READ_LOCK; | |
2536 | ||
2537 | if (p->search_commit_root) { | |
be6821f8 FM |
2538 | /* |
2539 | * The commit roots are read only so we always do read locks, | |
2540 | * and we always must hold the commit_root_sem when doing | |
2541 | * searches on them, the only exception is send where we don't | |
2542 | * want to block transaction commits for a long time, so | |
2543 | * we need to clone the commit root in order to avoid races | |
2544 | * with transaction commits that create a snapshot of one of | |
2545 | * the roots used by a send operation. | |
2546 | */ | |
2547 | if (p->need_commit_sem) { | |
1fc28d8e | 2548 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 2549 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 2550 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
2551 | if (!b) |
2552 | return ERR_PTR(-ENOMEM); | |
2553 | ||
2554 | } else { | |
2555 | b = root->commit_root; | |
67439dad | 2556 | atomic_inc(&b->refs); |
be6821f8 FM |
2557 | } |
2558 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
2559 | /* |
2560 | * Ensure that all callers have set skip_locking when | |
2561 | * p->search_commit_root = 1. | |
2562 | */ | |
2563 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2564 | |
2565 | goto out; | |
2566 | } | |
2567 | ||
2568 | if (p->skip_locking) { | |
2569 | b = btrfs_root_node(root); | |
2570 | level = btrfs_header_level(b); | |
2571 | goto out; | |
2572 | } | |
2573 | ||
2574 | /* | |
662c653b LB |
2575 | * If the level is set to maximum, we can skip trying to get the read |
2576 | * lock. | |
1fc28d8e | 2577 | */ |
662c653b LB |
2578 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2579 | /* | |
2580 | * We don't know the level of the root node until we actually | |
2581 | * have it read locked | |
2582 | */ | |
2583 | b = btrfs_read_lock_root_node(root); | |
2584 | level = btrfs_header_level(b); | |
2585 | if (level > write_lock_level) | |
2586 | goto out; | |
2587 | ||
2588 | /* Whoops, must trade for write lock */ | |
2589 | btrfs_tree_read_unlock(b); | |
2590 | free_extent_buffer(b); | |
2591 | } | |
1fc28d8e | 2592 | |
1fc28d8e LB |
2593 | b = btrfs_lock_root_node(root); |
2594 | root_lock = BTRFS_WRITE_LOCK; | |
2595 | ||
2596 | /* The level might have changed, check again */ | |
2597 | level = btrfs_header_level(b); | |
2598 | ||
2599 | out: | |
2600 | p->nodes[level] = b; | |
2601 | if (!p->skip_locking) | |
2602 | p->locks[level] = root_lock; | |
2603 | /* | |
2604 | * Callers are responsible for dropping b's references. | |
2605 | */ | |
2606 | return b; | |
2607 | } | |
2608 | ||
2609 | ||
74123bd7 | 2610 | /* |
4271ecea NB |
2611 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2612 | * modifications to preserve tree invariants. | |
74123bd7 | 2613 | * |
4271ecea NB |
2614 | * @trans: Handle of transaction, used when modifying the tree |
2615 | * @p: Holds all btree nodes along the search path | |
2616 | * @root: The root node of the tree | |
2617 | * @key: The key we are looking for | |
2618 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2619 | * deletions it's -1. 0 for plain searches | |
2620 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2621 | * when modifying the tree. | |
97571fd0 | 2622 | * |
4271ecea NB |
2623 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2624 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2625 | * | |
2626 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2627 | * of the path (level 0) | |
2628 | * | |
2629 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2630 | * points to the slot where it should be inserted | |
2631 | * | |
2632 | * If an error is encountered while searching the tree a negative error number | |
2633 | * is returned | |
74123bd7 | 2634 | */ |
310712b2 OS |
2635 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2636 | const struct btrfs_key *key, struct btrfs_path *p, | |
2637 | int ins_len, int cow) | |
be0e5c09 | 2638 | { |
5f39d397 | 2639 | struct extent_buffer *b; |
be0e5c09 CM |
2640 | int slot; |
2641 | int ret; | |
33c66f43 | 2642 | int err; |
be0e5c09 | 2643 | int level; |
925baedd | 2644 | int lowest_unlock = 1; |
bd681513 CM |
2645 | /* everything at write_lock_level or lower must be write locked */ |
2646 | int write_lock_level = 0; | |
9f3a7427 | 2647 | u8 lowest_level = 0; |
f7c79f30 | 2648 | int min_write_lock_level; |
d7396f07 | 2649 | int prev_cmp; |
9f3a7427 | 2650 | |
6702ed49 | 2651 | lowest_level = p->lowest_level; |
323ac95b | 2652 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2653 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2654 | BUG_ON(!cow && ins_len); |
25179201 | 2655 | |
bd681513 | 2656 | if (ins_len < 0) { |
925baedd | 2657 | lowest_unlock = 2; |
65b51a00 | 2658 | |
bd681513 CM |
2659 | /* when we are removing items, we might have to go up to level |
2660 | * two as we update tree pointers Make sure we keep write | |
2661 | * for those levels as well | |
2662 | */ | |
2663 | write_lock_level = 2; | |
2664 | } else if (ins_len > 0) { | |
2665 | /* | |
2666 | * for inserting items, make sure we have a write lock on | |
2667 | * level 1 so we can update keys | |
2668 | */ | |
2669 | write_lock_level = 1; | |
2670 | } | |
2671 | ||
2672 | if (!cow) | |
2673 | write_lock_level = -1; | |
2674 | ||
09a2a8f9 | 2675 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2676 | write_lock_level = BTRFS_MAX_LEVEL; |
2677 | ||
f7c79f30 CM |
2678 | min_write_lock_level = write_lock_level; |
2679 | ||
bb803951 | 2680 | again: |
d7396f07 | 2681 | prev_cmp = -1; |
1fc28d8e | 2682 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2683 | if (IS_ERR(b)) { |
2684 | ret = PTR_ERR(b); | |
2685 | goto done; | |
2686 | } | |
925baedd | 2687 | |
eb60ceac | 2688 | while (b) { |
f624d976 QW |
2689 | int dec = 0; |
2690 | ||
5f39d397 | 2691 | level = btrfs_header_level(b); |
65b51a00 | 2692 | |
02217ed2 | 2693 | if (cow) { |
9ea2c7c9 NB |
2694 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2695 | ||
c8c42864 CM |
2696 | /* |
2697 | * if we don't really need to cow this block | |
2698 | * then we don't want to set the path blocking, | |
2699 | * so we test it here | |
2700 | */ | |
64c12921 JM |
2701 | if (!should_cow_block(trans, root, b)) { |
2702 | trans->dirty = true; | |
65b51a00 | 2703 | goto cow_done; |
64c12921 | 2704 | } |
5d4f98a2 | 2705 | |
bd681513 CM |
2706 | /* |
2707 | * must have write locks on this node and the | |
2708 | * parent | |
2709 | */ | |
5124e00e JB |
2710 | if (level > write_lock_level || |
2711 | (level + 1 > write_lock_level && | |
2712 | level + 1 < BTRFS_MAX_LEVEL && | |
2713 | p->nodes[level + 1])) { | |
bd681513 CM |
2714 | write_lock_level = level + 1; |
2715 | btrfs_release_path(p); | |
2716 | goto again; | |
2717 | } | |
2718 | ||
160f4089 | 2719 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2720 | if (last_level) |
2721 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2722 | &b); | |
2723 | else | |
2724 | err = btrfs_cow_block(trans, root, b, | |
2725 | p->nodes[level + 1], | |
2726 | p->slots[level + 1], &b); | |
33c66f43 | 2727 | if (err) { |
33c66f43 | 2728 | ret = err; |
65b51a00 | 2729 | goto done; |
54aa1f4d | 2730 | } |
02217ed2 | 2731 | } |
65b51a00 | 2732 | cow_done: |
eb60ceac | 2733 | p->nodes[level] = b; |
52398340 LB |
2734 | /* |
2735 | * Leave path with blocking locks to avoid massive | |
2736 | * lock context switch, this is made on purpose. | |
2737 | */ | |
b4ce94de CM |
2738 | |
2739 | /* | |
2740 | * we have a lock on b and as long as we aren't changing | |
2741 | * the tree, there is no way to for the items in b to change. | |
2742 | * It is safe to drop the lock on our parent before we | |
2743 | * go through the expensive btree search on b. | |
2744 | * | |
eb653de1 FDBM |
2745 | * If we're inserting or deleting (ins_len != 0), then we might |
2746 | * be changing slot zero, which may require changing the parent. | |
2747 | * So, we can't drop the lock until after we know which slot | |
2748 | * we're operating on. | |
b4ce94de | 2749 | */ |
eb653de1 FDBM |
2750 | if (!ins_len && !p->keep_locks) { |
2751 | int u = level + 1; | |
2752 | ||
2753 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2754 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2755 | p->locks[u] = 0; | |
2756 | } | |
2757 | } | |
b4ce94de | 2758 | |
995e9a16 NB |
2759 | /* |
2760 | * If btrfs_bin_search returns an exact match (prev_cmp == 0) | |
2761 | * we can safely assume the target key will always be in slot 0 | |
2762 | * on lower levels due to the invariants BTRFS' btree provides, | |
2763 | * namely that a btrfs_key_ptr entry always points to the | |
2764 | * lowest key in the child node, thus we can skip searching | |
2765 | * lower levels | |
2766 | */ | |
2767 | if (prev_cmp == 0) { | |
2768 | slot = 0; | |
2769 | ret = 0; | |
2770 | } else { | |
2771 | ret = btrfs_bin_search(b, key, &slot); | |
2772 | prev_cmp = ret; | |
2773 | if (ret < 0) | |
2774 | goto done; | |
2775 | } | |
b4ce94de | 2776 | |
f624d976 | 2777 | if (level == 0) { |
be0e5c09 | 2778 | p->slots[level] = slot; |
87b29b20 | 2779 | if (ins_len > 0 && |
e902baac | 2780 | btrfs_leaf_free_space(b) < ins_len) { |
bd681513 CM |
2781 | if (write_lock_level < 1) { |
2782 | write_lock_level = 1; | |
2783 | btrfs_release_path(p); | |
2784 | goto again; | |
2785 | } | |
2786 | ||
b4ce94de | 2787 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2788 | err = split_leaf(trans, root, key, |
2789 | p, ins_len, ret == 0); | |
b4ce94de | 2790 | |
33c66f43 YZ |
2791 | BUG_ON(err > 0); |
2792 | if (err) { | |
2793 | ret = err; | |
65b51a00 CM |
2794 | goto done; |
2795 | } | |
5c680ed6 | 2796 | } |
459931ec | 2797 | if (!p->search_for_split) |
f7c79f30 | 2798 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2799 | min_write_lock_level, NULL); |
65b51a00 | 2800 | goto done; |
be0e5c09 | 2801 | } |
f624d976 QW |
2802 | if (ret && slot > 0) { |
2803 | dec = 1; | |
2804 | slot--; | |
2805 | } | |
2806 | p->slots[level] = slot; | |
2807 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
2808 | &write_lock_level); | |
2809 | if (err == -EAGAIN) | |
2810 | goto again; | |
2811 | if (err) { | |
2812 | ret = err; | |
2813 | goto done; | |
2814 | } | |
2815 | b = p->nodes[level]; | |
2816 | slot = p->slots[level]; | |
2817 | ||
2818 | /* | |
2819 | * Slot 0 is special, if we change the key we have to update | |
2820 | * the parent pointer which means we must have a write lock on | |
2821 | * the parent | |
2822 | */ | |
2823 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
2824 | write_lock_level = level + 1; | |
2825 | btrfs_release_path(p); | |
2826 | goto again; | |
2827 | } | |
2828 | ||
2829 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
2830 | &write_lock_level); | |
2831 | ||
2832 | if (level == lowest_level) { | |
2833 | if (dec) | |
2834 | p->slots[level]++; | |
2835 | goto done; | |
2836 | } | |
2837 | ||
2838 | err = read_block_for_search(root, p, &b, level, slot, key); | |
2839 | if (err == -EAGAIN) | |
2840 | goto again; | |
2841 | if (err) { | |
2842 | ret = err; | |
2843 | goto done; | |
2844 | } | |
2845 | ||
2846 | if (!p->skip_locking) { | |
2847 | level = btrfs_header_level(b); | |
2848 | if (level <= write_lock_level) { | |
2849 | if (!btrfs_try_tree_write_lock(b)) { | |
2850 | btrfs_set_path_blocking(p); | |
2851 | btrfs_tree_lock(b); | |
2852 | } | |
2853 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2854 | } else { | |
2855 | if (!btrfs_tree_read_lock_atomic(b)) { | |
2856 | btrfs_set_path_blocking(p); | |
2857 | btrfs_tree_read_lock(b); | |
2858 | } | |
2859 | p->locks[level] = BTRFS_READ_LOCK; | |
2860 | } | |
2861 | p->nodes[level] = b; | |
2862 | } | |
be0e5c09 | 2863 | } |
65b51a00 CM |
2864 | ret = 1; |
2865 | done: | |
b4ce94de CM |
2866 | /* |
2867 | * we don't really know what they plan on doing with the path | |
2868 | * from here on, so for now just mark it as blocking | |
2869 | */ | |
b9473439 CM |
2870 | if (!p->leave_spinning) |
2871 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2872 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2873 | btrfs_release_path(p); |
65b51a00 | 2874 | return ret; |
be0e5c09 CM |
2875 | } |
2876 | ||
5d9e75c4 JS |
2877 | /* |
2878 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2879 | * current state of the tree together with the operations recorded in the tree | |
2880 | * modification log to search for the key in a previous version of this tree, as | |
2881 | * denoted by the time_seq parameter. | |
2882 | * | |
2883 | * Naturally, there is no support for insert, delete or cow operations. | |
2884 | * | |
2885 | * The resulting path and return value will be set up as if we called | |
2886 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2887 | */ | |
310712b2 | 2888 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2889 | struct btrfs_path *p, u64 time_seq) |
2890 | { | |
0b246afa | 2891 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2892 | struct extent_buffer *b; |
2893 | int slot; | |
2894 | int ret; | |
2895 | int err; | |
2896 | int level; | |
2897 | int lowest_unlock = 1; | |
2898 | u8 lowest_level = 0; | |
2899 | ||
2900 | lowest_level = p->lowest_level; | |
2901 | WARN_ON(p->nodes[0] != NULL); | |
2902 | ||
2903 | if (p->search_commit_root) { | |
2904 | BUG_ON(time_seq); | |
2905 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2906 | } | |
2907 | ||
2908 | again: | |
5d9e75c4 | 2909 | b = get_old_root(root, time_seq); |
315bed43 NB |
2910 | if (!b) { |
2911 | ret = -EIO; | |
2912 | goto done; | |
2913 | } | |
5d9e75c4 | 2914 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2915 | p->locks[level] = BTRFS_READ_LOCK; |
2916 | ||
2917 | while (b) { | |
abe9339d QW |
2918 | int dec = 0; |
2919 | ||
5d9e75c4 JS |
2920 | level = btrfs_header_level(b); |
2921 | p->nodes[level] = b; | |
5d9e75c4 JS |
2922 | |
2923 | /* | |
2924 | * we have a lock on b and as long as we aren't changing | |
2925 | * the tree, there is no way to for the items in b to change. | |
2926 | * It is safe to drop the lock on our parent before we | |
2927 | * go through the expensive btree search on b. | |
2928 | */ | |
2929 | btrfs_unlock_up_safe(p, level + 1); | |
2930 | ||
995e9a16 | 2931 | ret = btrfs_bin_search(b, key, &slot); |
cbca7d59 FM |
2932 | if (ret < 0) |
2933 | goto done; | |
5d9e75c4 | 2934 | |
abe9339d | 2935 | if (level == 0) { |
5d9e75c4 JS |
2936 | p->slots[level] = slot; |
2937 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
2938 | goto done; |
2939 | } | |
5d9e75c4 | 2940 | |
abe9339d QW |
2941 | if (ret && slot > 0) { |
2942 | dec = 1; | |
2943 | slot--; | |
2944 | } | |
2945 | p->slots[level] = slot; | |
2946 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 2947 | |
abe9339d QW |
2948 | if (level == lowest_level) { |
2949 | if (dec) | |
2950 | p->slots[level]++; | |
2951 | goto done; | |
2952 | } | |
5d9e75c4 | 2953 | |
abe9339d QW |
2954 | err = read_block_for_search(root, p, &b, level, slot, key); |
2955 | if (err == -EAGAIN) | |
2956 | goto again; | |
2957 | if (err) { | |
2958 | ret = err; | |
5d9e75c4 JS |
2959 | goto done; |
2960 | } | |
abe9339d QW |
2961 | |
2962 | level = btrfs_header_level(b); | |
2963 | if (!btrfs_tree_read_lock_atomic(b)) { | |
2964 | btrfs_set_path_blocking(p); | |
2965 | btrfs_tree_read_lock(b); | |
2966 | } | |
2967 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); | |
2968 | if (!b) { | |
2969 | ret = -ENOMEM; | |
2970 | goto done; | |
2971 | } | |
2972 | p->locks[level] = BTRFS_READ_LOCK; | |
2973 | p->nodes[level] = b; | |
5d9e75c4 JS |
2974 | } |
2975 | ret = 1; | |
2976 | done: | |
2977 | if (!p->leave_spinning) | |
2978 | btrfs_set_path_blocking(p); | |
2979 | if (ret < 0) | |
2980 | btrfs_release_path(p); | |
2981 | ||
2982 | return ret; | |
2983 | } | |
2984 | ||
2f38b3e1 AJ |
2985 | /* |
2986 | * helper to use instead of search slot if no exact match is needed but | |
2987 | * instead the next or previous item should be returned. | |
2988 | * When find_higher is true, the next higher item is returned, the next lower | |
2989 | * otherwise. | |
2990 | * When return_any and find_higher are both true, and no higher item is found, | |
2991 | * return the next lower instead. | |
2992 | * When return_any is true and find_higher is false, and no lower item is found, | |
2993 | * return the next higher instead. | |
2994 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2995 | * < 0 on error | |
2996 | */ | |
2997 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
2998 | const struct btrfs_key *key, |
2999 | struct btrfs_path *p, int find_higher, | |
3000 | int return_any) | |
2f38b3e1 AJ |
3001 | { |
3002 | int ret; | |
3003 | struct extent_buffer *leaf; | |
3004 | ||
3005 | again: | |
3006 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3007 | if (ret <= 0) | |
3008 | return ret; | |
3009 | /* | |
3010 | * a return value of 1 means the path is at the position where the | |
3011 | * item should be inserted. Normally this is the next bigger item, | |
3012 | * but in case the previous item is the last in a leaf, path points | |
3013 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3014 | * item. | |
3015 | */ | |
3016 | leaf = p->nodes[0]; | |
3017 | ||
3018 | if (find_higher) { | |
3019 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3020 | ret = btrfs_next_leaf(root, p); | |
3021 | if (ret <= 0) | |
3022 | return ret; | |
3023 | if (!return_any) | |
3024 | return 1; | |
3025 | /* | |
3026 | * no higher item found, return the next | |
3027 | * lower instead | |
3028 | */ | |
3029 | return_any = 0; | |
3030 | find_higher = 0; | |
3031 | btrfs_release_path(p); | |
3032 | goto again; | |
3033 | } | |
3034 | } else { | |
e6793769 AJ |
3035 | if (p->slots[0] == 0) { |
3036 | ret = btrfs_prev_leaf(root, p); | |
3037 | if (ret < 0) | |
3038 | return ret; | |
3039 | if (!ret) { | |
23c6bf6a FDBM |
3040 | leaf = p->nodes[0]; |
3041 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3042 | p->slots[0]--; | |
e6793769 | 3043 | return 0; |
2f38b3e1 | 3044 | } |
e6793769 AJ |
3045 | if (!return_any) |
3046 | return 1; | |
3047 | /* | |
3048 | * no lower item found, return the next | |
3049 | * higher instead | |
3050 | */ | |
3051 | return_any = 0; | |
3052 | find_higher = 1; | |
3053 | btrfs_release_path(p); | |
3054 | goto again; | |
3055 | } else { | |
2f38b3e1 AJ |
3056 | --p->slots[0]; |
3057 | } | |
3058 | } | |
3059 | return 0; | |
3060 | } | |
3061 | ||
74123bd7 CM |
3062 | /* |
3063 | * adjust the pointers going up the tree, starting at level | |
3064 | * making sure the right key of each node is points to 'key'. | |
3065 | * This is used after shifting pointers to the left, so it stops | |
3066 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3067 | * higher levels | |
aa5d6bed | 3068 | * |
74123bd7 | 3069 | */ |
b167fa91 | 3070 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3071 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3072 | { |
3073 | int i; | |
5f39d397 | 3074 | struct extent_buffer *t; |
0e82bcfe | 3075 | int ret; |
5f39d397 | 3076 | |
234b63a0 | 3077 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3078 | int tslot = path->slots[i]; |
0e82bcfe | 3079 | |
eb60ceac | 3080 | if (!path->nodes[i]) |
be0e5c09 | 3081 | break; |
5f39d397 | 3082 | t = path->nodes[i]; |
0e82bcfe DS |
3083 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3084 | GFP_ATOMIC); | |
3085 | BUG_ON(ret < 0); | |
5f39d397 | 3086 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3087 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3088 | if (tslot != 0) |
3089 | break; | |
3090 | } | |
3091 | } | |
3092 | ||
31840ae1 ZY |
3093 | /* |
3094 | * update item key. | |
3095 | * | |
3096 | * This function isn't completely safe. It's the caller's responsibility | |
3097 | * that the new key won't break the order | |
3098 | */ | |
b7a0365e DD |
3099 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3100 | struct btrfs_path *path, | |
310712b2 | 3101 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3102 | { |
3103 | struct btrfs_disk_key disk_key; | |
3104 | struct extent_buffer *eb; | |
3105 | int slot; | |
3106 | ||
3107 | eb = path->nodes[0]; | |
3108 | slot = path->slots[0]; | |
3109 | if (slot > 0) { | |
3110 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
3111 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
3112 | btrfs_crit(fs_info, | |
3113 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3114 | slot, btrfs_disk_key_objectid(&disk_key), | |
3115 | btrfs_disk_key_type(&disk_key), | |
3116 | btrfs_disk_key_offset(&disk_key), | |
3117 | new_key->objectid, new_key->type, | |
3118 | new_key->offset); | |
3119 | btrfs_print_leaf(eb); | |
3120 | BUG(); | |
3121 | } | |
31840ae1 ZY |
3122 | } |
3123 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3124 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
3125 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
3126 | btrfs_crit(fs_info, | |
3127 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3128 | slot, btrfs_disk_key_objectid(&disk_key), | |
3129 | btrfs_disk_key_type(&disk_key), | |
3130 | btrfs_disk_key_offset(&disk_key), | |
3131 | new_key->objectid, new_key->type, | |
3132 | new_key->offset); | |
3133 | btrfs_print_leaf(eb); | |
3134 | BUG(); | |
3135 | } | |
31840ae1 ZY |
3136 | } |
3137 | ||
3138 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3139 | btrfs_set_item_key(eb, &disk_key, slot); | |
3140 | btrfs_mark_buffer_dirty(eb); | |
3141 | if (slot == 0) | |
b167fa91 | 3142 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3143 | } |
3144 | ||
74123bd7 CM |
3145 | /* |
3146 | * try to push data from one node into the next node left in the | |
79f95c82 | 3147 | * tree. |
aa5d6bed CM |
3148 | * |
3149 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3150 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3151 | */ |
98ed5174 | 3152 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 3153 | struct extent_buffer *dst, |
971a1f66 | 3154 | struct extent_buffer *src, int empty) |
be0e5c09 | 3155 | { |
d30a668f | 3156 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 3157 | int push_items = 0; |
bb803951 CM |
3158 | int src_nritems; |
3159 | int dst_nritems; | |
aa5d6bed | 3160 | int ret = 0; |
be0e5c09 | 3161 | |
5f39d397 CM |
3162 | src_nritems = btrfs_header_nritems(src); |
3163 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3164 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3165 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3166 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3167 | |
bce4eae9 | 3168 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3169 | return 1; |
3170 | ||
d397712b | 3171 | if (push_items <= 0) |
be0e5c09 CM |
3172 | return 1; |
3173 | ||
bce4eae9 | 3174 | if (empty) { |
971a1f66 | 3175 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3176 | if (push_items < src_nritems) { |
3177 | /* leave at least 8 pointers in the node if | |
3178 | * we aren't going to empty it | |
3179 | */ | |
3180 | if (src_nritems - push_items < 8) { | |
3181 | if (push_items <= 8) | |
3182 | return 1; | |
3183 | push_items -= 8; | |
3184 | } | |
3185 | } | |
3186 | } else | |
3187 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3188 | |
ed874f0d | 3189 | ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 3190 | if (ret) { |
66642832 | 3191 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3192 | return ret; |
3193 | } | |
5f39d397 CM |
3194 | copy_extent_buffer(dst, src, |
3195 | btrfs_node_key_ptr_offset(dst_nritems), | |
3196 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3197 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3198 | |
bb803951 | 3199 | if (push_items < src_nritems) { |
57911b8b | 3200 | /* |
bf1d3425 DS |
3201 | * Don't call tree_mod_log_insert_move here, key removal was |
3202 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3203 | */ |
5f39d397 CM |
3204 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3205 | btrfs_node_key_ptr_offset(push_items), | |
3206 | (src_nritems - push_items) * | |
3207 | sizeof(struct btrfs_key_ptr)); | |
3208 | } | |
3209 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3210 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3211 | btrfs_mark_buffer_dirty(src); | |
3212 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3213 | |
79f95c82 CM |
3214 | return ret; |
3215 | } | |
3216 | ||
3217 | /* | |
3218 | * try to push data from one node into the next node right in the | |
3219 | * tree. | |
3220 | * | |
3221 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3222 | * error, and > 0 if there was no room in the right hand block. | |
3223 | * | |
3224 | * this will only push up to 1/2 the contents of the left node over | |
3225 | */ | |
5f39d397 | 3226 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3227 | struct extent_buffer *dst, |
3228 | struct extent_buffer *src) | |
79f95c82 | 3229 | { |
55d32ed8 | 3230 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
3231 | int push_items = 0; |
3232 | int max_push; | |
3233 | int src_nritems; | |
3234 | int dst_nritems; | |
3235 | int ret = 0; | |
79f95c82 | 3236 | |
7bb86316 CM |
3237 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3238 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3239 | ||
5f39d397 CM |
3240 | src_nritems = btrfs_header_nritems(src); |
3241 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3242 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3243 | if (push_items <= 0) |
79f95c82 | 3244 | return 1; |
bce4eae9 | 3245 | |
d397712b | 3246 | if (src_nritems < 4) |
bce4eae9 | 3247 | return 1; |
79f95c82 CM |
3248 | |
3249 | max_push = src_nritems / 2 + 1; | |
3250 | /* don't try to empty the node */ | |
d397712b | 3251 | if (max_push >= src_nritems) |
79f95c82 | 3252 | return 1; |
252c38f0 | 3253 | |
79f95c82 CM |
3254 | if (max_push < push_items) |
3255 | push_items = max_push; | |
3256 | ||
bf1d3425 DS |
3257 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3258 | BUG_ON(ret < 0); | |
5f39d397 CM |
3259 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3260 | btrfs_node_key_ptr_offset(0), | |
3261 | (dst_nritems) * | |
3262 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3263 | |
ed874f0d DS |
3264 | ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
3265 | push_items); | |
5de865ee | 3266 | if (ret) { |
66642832 | 3267 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3268 | return ret; |
3269 | } | |
5f39d397 CM |
3270 | copy_extent_buffer(dst, src, |
3271 | btrfs_node_key_ptr_offset(0), | |
3272 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3273 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3274 | |
5f39d397 CM |
3275 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3276 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3277 | |
5f39d397 CM |
3278 | btrfs_mark_buffer_dirty(src); |
3279 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3280 | |
aa5d6bed | 3281 | return ret; |
be0e5c09 CM |
3282 | } |
3283 | ||
97571fd0 CM |
3284 | /* |
3285 | * helper function to insert a new root level in the tree. | |
3286 | * A new node is allocated, and a single item is inserted to | |
3287 | * point to the existing root | |
aa5d6bed CM |
3288 | * |
3289 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3290 | */ |
d397712b | 3291 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3292 | struct btrfs_root *root, |
fdd99c72 | 3293 | struct btrfs_path *path, int level) |
5c680ed6 | 3294 | { |
0b246afa | 3295 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3296 | u64 lower_gen; |
5f39d397 CM |
3297 | struct extent_buffer *lower; |
3298 | struct extent_buffer *c; | |
925baedd | 3299 | struct extent_buffer *old; |
5f39d397 | 3300 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3301 | int ret; |
5c680ed6 CM |
3302 | |
3303 | BUG_ON(path->nodes[level]); | |
3304 | BUG_ON(path->nodes[level-1] != root->node); | |
3305 | ||
7bb86316 CM |
3306 | lower = path->nodes[level-1]; |
3307 | if (level == 1) | |
3308 | btrfs_item_key(lower, &lower_key, 0); | |
3309 | else | |
3310 | btrfs_node_key(lower, &lower_key, 0); | |
3311 | ||
a6279470 FM |
3312 | c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, |
3313 | root->node->start, 0); | |
5f39d397 CM |
3314 | if (IS_ERR(c)) |
3315 | return PTR_ERR(c); | |
925baedd | 3316 | |
0b246afa | 3317 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3318 | |
5f39d397 | 3319 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3320 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3321 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3322 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3323 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3324 | |
3325 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3326 | |
5f39d397 | 3327 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3328 | |
925baedd | 3329 | old = root->node; |
d9d19a01 DS |
3330 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3331 | BUG_ON(ret < 0); | |
240f62c8 | 3332 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3333 | |
3334 | /* the super has an extra ref to root->node */ | |
3335 | free_extent_buffer(old); | |
3336 | ||
0b86a832 | 3337 | add_root_to_dirty_list(root); |
67439dad | 3338 | atomic_inc(&c->refs); |
5f39d397 | 3339 | path->nodes[level] = c; |
95449a16 | 3340 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3341 | path->slots[level] = 0; |
3342 | return 0; | |
3343 | } | |
3344 | ||
74123bd7 CM |
3345 | /* |
3346 | * worker function to insert a single pointer in a node. | |
3347 | * the node should have enough room for the pointer already | |
97571fd0 | 3348 | * |
74123bd7 CM |
3349 | * slot and level indicate where you want the key to go, and |
3350 | * blocknr is the block the key points to. | |
3351 | */ | |
143bede5 | 3352 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 3353 | struct btrfs_path *path, |
143bede5 | 3354 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3355 | int slot, int level) |
74123bd7 | 3356 | { |
5f39d397 | 3357 | struct extent_buffer *lower; |
74123bd7 | 3358 | int nritems; |
f3ea38da | 3359 | int ret; |
5c680ed6 CM |
3360 | |
3361 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3362 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3363 | lower = path->nodes[level]; |
3364 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3365 | BUG_ON(slot > nritems); |
6ad3cf6d | 3366 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 3367 | if (slot != nritems) { |
bf1d3425 DS |
3368 | if (level) { |
3369 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3370 | nritems - slot); |
bf1d3425 DS |
3371 | BUG_ON(ret < 0); |
3372 | } | |
5f39d397 CM |
3373 | memmove_extent_buffer(lower, |
3374 | btrfs_node_key_ptr_offset(slot + 1), | |
3375 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3376 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3377 | } |
c3e06965 | 3378 | if (level) { |
e09c2efe DS |
3379 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3380 | GFP_NOFS); | |
f3ea38da JS |
3381 | BUG_ON(ret < 0); |
3382 | } | |
5f39d397 | 3383 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3384 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3385 | WARN_ON(trans->transid == 0); |
3386 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3387 | btrfs_set_header_nritems(lower, nritems + 1); |
3388 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3389 | } |
3390 | ||
97571fd0 CM |
3391 | /* |
3392 | * split the node at the specified level in path in two. | |
3393 | * The path is corrected to point to the appropriate node after the split | |
3394 | * | |
3395 | * Before splitting this tries to make some room in the node by pushing | |
3396 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3397 | * |
3398 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3399 | */ |
e02119d5 CM |
3400 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3401 | struct btrfs_root *root, | |
3402 | struct btrfs_path *path, int level) | |
be0e5c09 | 3403 | { |
0b246afa | 3404 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3405 | struct extent_buffer *c; |
3406 | struct extent_buffer *split; | |
3407 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3408 | int mid; |
5c680ed6 | 3409 | int ret; |
7518a238 | 3410 | u32 c_nritems; |
eb60ceac | 3411 | |
5f39d397 | 3412 | c = path->nodes[level]; |
7bb86316 | 3413 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3414 | if (c == root->node) { |
d9abbf1c | 3415 | /* |
90f8d62e JS |
3416 | * trying to split the root, lets make a new one |
3417 | * | |
fdd99c72 | 3418 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3419 | * insert_new_root, because that root buffer will be kept as a |
3420 | * normal node. We are going to log removal of half of the | |
3421 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3422 | * tree lock on the buffer, which is why we cannot race with | |
3423 | * other tree_mod_log users. | |
d9abbf1c | 3424 | */ |
fdd99c72 | 3425 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3426 | if (ret) |
3427 | return ret; | |
b3612421 | 3428 | } else { |
e66f709b | 3429 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3430 | c = path->nodes[level]; |
3431 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3432 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3433 | return 0; |
54aa1f4d CM |
3434 | if (ret < 0) |
3435 | return ret; | |
be0e5c09 | 3436 | } |
e66f709b | 3437 | |
5f39d397 | 3438 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3439 | mid = (c_nritems + 1) / 2; |
3440 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3441 | |
a6279470 FM |
3442 | split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, |
3443 | c->start, 0); | |
5f39d397 CM |
3444 | if (IS_ERR(split)) |
3445 | return PTR_ERR(split); | |
3446 | ||
0b246afa | 3447 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3448 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3449 | |
ed874f0d | 3450 | ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 3451 | if (ret) { |
66642832 | 3452 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3453 | return ret; |
3454 | } | |
5f39d397 CM |
3455 | copy_extent_buffer(split, c, |
3456 | btrfs_node_key_ptr_offset(0), | |
3457 | btrfs_node_key_ptr_offset(mid), | |
3458 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3459 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3460 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3461 | ret = 0; |
3462 | ||
5f39d397 CM |
3463 | btrfs_mark_buffer_dirty(c); |
3464 | btrfs_mark_buffer_dirty(split); | |
3465 | ||
6ad3cf6d | 3466 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 3467 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3468 | |
5de08d7d | 3469 | if (path->slots[level] >= mid) { |
5c680ed6 | 3470 | path->slots[level] -= mid; |
925baedd | 3471 | btrfs_tree_unlock(c); |
5f39d397 CM |
3472 | free_extent_buffer(c); |
3473 | path->nodes[level] = split; | |
5c680ed6 CM |
3474 | path->slots[level + 1] += 1; |
3475 | } else { | |
925baedd | 3476 | btrfs_tree_unlock(split); |
5f39d397 | 3477 | free_extent_buffer(split); |
be0e5c09 | 3478 | } |
aa5d6bed | 3479 | return ret; |
be0e5c09 CM |
3480 | } |
3481 | ||
74123bd7 CM |
3482 | /* |
3483 | * how many bytes are required to store the items in a leaf. start | |
3484 | * and nr indicate which items in the leaf to check. This totals up the | |
3485 | * space used both by the item structs and the item data | |
3486 | */ | |
5f39d397 | 3487 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3488 | { |
41be1f3b JB |
3489 | struct btrfs_item *start_item; |
3490 | struct btrfs_item *end_item; | |
be0e5c09 | 3491 | int data_len; |
5f39d397 | 3492 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3493 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3494 | |
3495 | if (!nr) | |
3496 | return 0; | |
dd3cc16b RK |
3497 | start_item = btrfs_item_nr(start); |
3498 | end_item = btrfs_item_nr(end); | |
a31356b9 DS |
3499 | data_len = btrfs_item_offset(l, start_item) + |
3500 | btrfs_item_size(l, start_item); | |
3501 | data_len = data_len - btrfs_item_offset(l, end_item); | |
0783fcfc | 3502 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3503 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3504 | return data_len; |
3505 | } | |
3506 | ||
d4dbff95 CM |
3507 | /* |
3508 | * The space between the end of the leaf items and | |
3509 | * the start of the leaf data. IOW, how much room | |
3510 | * the leaf has left for both items and data | |
3511 | */ | |
e902baac | 3512 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 3513 | { |
e902baac | 3514 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
3515 | int nritems = btrfs_header_nritems(leaf); |
3516 | int ret; | |
0b246afa JM |
3517 | |
3518 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3519 | if (ret < 0) { |
0b246afa JM |
3520 | btrfs_crit(fs_info, |
3521 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3522 | ret, | |
3523 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3524 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3525 | } |
3526 | return ret; | |
d4dbff95 CM |
3527 | } |
3528 | ||
99d8f83c CM |
3529 | /* |
3530 | * min slot controls the lowest index we're willing to push to the | |
3531 | * right. We'll push up to and including min_slot, but no lower | |
3532 | */ | |
f72f0010 | 3533 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
3534 | int data_size, int empty, |
3535 | struct extent_buffer *right, | |
99d8f83c CM |
3536 | int free_space, u32 left_nritems, |
3537 | u32 min_slot) | |
00ec4c51 | 3538 | { |
f72f0010 | 3539 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 3540 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3541 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3542 | struct btrfs_map_token token; |
5f39d397 | 3543 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3544 | int slot; |
34a38218 | 3545 | u32 i; |
00ec4c51 CM |
3546 | int push_space = 0; |
3547 | int push_items = 0; | |
0783fcfc | 3548 | struct btrfs_item *item; |
34a38218 | 3549 | u32 nr; |
7518a238 | 3550 | u32 right_nritems; |
5f39d397 | 3551 | u32 data_end; |
db94535d | 3552 | u32 this_item_size; |
00ec4c51 | 3553 | |
34a38218 CM |
3554 | if (empty) |
3555 | nr = 0; | |
3556 | else | |
99d8f83c | 3557 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3558 | |
31840ae1 | 3559 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3560 | push_space += data_size; |
31840ae1 | 3561 | |
44871b1b | 3562 | slot = path->slots[1]; |
34a38218 CM |
3563 | i = left_nritems - 1; |
3564 | while (i >= nr) { | |
dd3cc16b | 3565 | item = btrfs_item_nr(i); |
db94535d | 3566 | |
31840ae1 ZY |
3567 | if (!empty && push_items > 0) { |
3568 | if (path->slots[0] > i) | |
3569 | break; | |
3570 | if (path->slots[0] == i) { | |
e902baac DS |
3571 | int space = btrfs_leaf_free_space(left); |
3572 | ||
31840ae1 ZY |
3573 | if (space + push_space * 2 > free_space) |
3574 | break; | |
3575 | } | |
3576 | } | |
3577 | ||
00ec4c51 | 3578 | if (path->slots[0] == i) |
87b29b20 | 3579 | push_space += data_size; |
db94535d | 3580 | |
db94535d CM |
3581 | this_item_size = btrfs_item_size(left, item); |
3582 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3583 | break; |
31840ae1 | 3584 | |
00ec4c51 | 3585 | push_items++; |
db94535d | 3586 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3587 | if (i == 0) |
3588 | break; | |
3589 | i--; | |
db94535d | 3590 | } |
5f39d397 | 3591 | |
925baedd CM |
3592 | if (push_items == 0) |
3593 | goto out_unlock; | |
5f39d397 | 3594 | |
6c1500f2 | 3595 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3596 | |
00ec4c51 | 3597 | /* push left to right */ |
5f39d397 | 3598 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3599 | |
5f39d397 | 3600 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
8f881e8c | 3601 | push_space -= leaf_data_end(left); |
5f39d397 | 3602 | |
00ec4c51 | 3603 | /* make room in the right data area */ |
8f881e8c | 3604 | data_end = leaf_data_end(right); |
5f39d397 | 3605 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3606 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3607 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3608 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3609 | |
00ec4c51 | 3610 | /* copy from the left data area */ |
3d9ec8c4 | 3611 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3612 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 3613 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 3614 | push_space); |
5f39d397 CM |
3615 | |
3616 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3617 | btrfs_item_nr_offset(0), | |
3618 | right_nritems * sizeof(struct btrfs_item)); | |
3619 | ||
00ec4c51 | 3620 | /* copy the items from left to right */ |
5f39d397 CM |
3621 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3622 | btrfs_item_nr_offset(left_nritems - push_items), | |
3623 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3624 | |
3625 | /* update the item pointers */ | |
c82f823c | 3626 | btrfs_init_map_token(&token, right); |
7518a238 | 3627 | right_nritems += push_items; |
5f39d397 | 3628 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3629 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3630 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3631 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
3632 | push_space -= btrfs_token_item_size(&token, item); |
3633 | btrfs_set_token_item_offset(&token, item, push_space); | |
db94535d CM |
3634 | } |
3635 | ||
7518a238 | 3636 | left_nritems -= push_items; |
5f39d397 | 3637 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3638 | |
34a38218 CM |
3639 | if (left_nritems) |
3640 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3641 | else |
6a884d7d | 3642 | btrfs_clean_tree_block(left); |
f0486c68 | 3643 | |
5f39d397 | 3644 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3645 | |
5f39d397 CM |
3646 | btrfs_item_key(right, &disk_key, 0); |
3647 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3648 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3649 | |
00ec4c51 | 3650 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3651 | if (path->slots[0] >= left_nritems) { |
3652 | path->slots[0] -= left_nritems; | |
925baedd | 3653 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 3654 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 3655 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3656 | free_extent_buffer(path->nodes[0]); |
3657 | path->nodes[0] = right; | |
00ec4c51 CM |
3658 | path->slots[1] += 1; |
3659 | } else { | |
925baedd | 3660 | btrfs_tree_unlock(right); |
5f39d397 | 3661 | free_extent_buffer(right); |
00ec4c51 CM |
3662 | } |
3663 | return 0; | |
925baedd CM |
3664 | |
3665 | out_unlock: | |
3666 | btrfs_tree_unlock(right); | |
3667 | free_extent_buffer(right); | |
3668 | return 1; | |
00ec4c51 | 3669 | } |
925baedd | 3670 | |
44871b1b CM |
3671 | /* |
3672 | * push some data in the path leaf to the right, trying to free up at | |
3673 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3674 | * | |
3675 | * returns 1 if the push failed because the other node didn't have enough | |
3676 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3677 | * |
3678 | * this will push starting from min_slot to the end of the leaf. It won't | |
3679 | * push any slot lower than min_slot | |
44871b1b CM |
3680 | */ |
3681 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3682 | *root, struct btrfs_path *path, |
3683 | int min_data_size, int data_size, | |
3684 | int empty, u32 min_slot) | |
44871b1b CM |
3685 | { |
3686 | struct extent_buffer *left = path->nodes[0]; | |
3687 | struct extent_buffer *right; | |
3688 | struct extent_buffer *upper; | |
3689 | int slot; | |
3690 | int free_space; | |
3691 | u32 left_nritems; | |
3692 | int ret; | |
3693 | ||
3694 | if (!path->nodes[1]) | |
3695 | return 1; | |
3696 | ||
3697 | slot = path->slots[1]; | |
3698 | upper = path->nodes[1]; | |
3699 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3700 | return 1; | |
3701 | ||
3702 | btrfs_assert_tree_locked(path->nodes[1]); | |
3703 | ||
4b231ae4 | 3704 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3705 | /* |
3706 | * slot + 1 is not valid or we fail to read the right node, | |
3707 | * no big deal, just return. | |
3708 | */ | |
3709 | if (IS_ERR(right)) | |
91ca338d TI |
3710 | return 1; |
3711 | ||
44871b1b | 3712 | btrfs_tree_lock(right); |
8bead258 | 3713 | btrfs_set_lock_blocking_write(right); |
44871b1b | 3714 | |
e902baac | 3715 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3716 | if (free_space < data_size) |
3717 | goto out_unlock; | |
3718 | ||
3719 | /* cow and double check */ | |
3720 | ret = btrfs_cow_block(trans, root, right, upper, | |
3721 | slot + 1, &right); | |
3722 | if (ret) | |
3723 | goto out_unlock; | |
3724 | ||
e902baac | 3725 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3726 | if (free_space < data_size) |
3727 | goto out_unlock; | |
3728 | ||
3729 | left_nritems = btrfs_header_nritems(left); | |
3730 | if (left_nritems == 0) | |
3731 | goto out_unlock; | |
3732 | ||
2ef1fed2 FDBM |
3733 | if (path->slots[0] == left_nritems && !empty) { |
3734 | /* Key greater than all keys in the leaf, right neighbor has | |
3735 | * enough room for it and we're not emptying our leaf to delete | |
3736 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3737 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3738 | btrfs_tree_unlock(left); |
3739 | free_extent_buffer(left); | |
3740 | path->nodes[0] = right; | |
3741 | path->slots[0] = 0; | |
3742 | path->slots[1]++; | |
3743 | return 0; | |
3744 | } | |
3745 | ||
f72f0010 | 3746 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3747 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3748 | out_unlock: |
3749 | btrfs_tree_unlock(right); | |
3750 | free_extent_buffer(right); | |
3751 | return 1; | |
3752 | } | |
3753 | ||
74123bd7 CM |
3754 | /* |
3755 | * push some data in the path leaf to the left, trying to free up at | |
3756 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3757 | * |
3758 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3759 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3760 | * items | |
74123bd7 | 3761 | */ |
8087c193 | 3762 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3763 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3764 | int free_space, u32 right_nritems, |
3765 | u32 max_slot) | |
be0e5c09 | 3766 | { |
8087c193 | 3767 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3768 | struct btrfs_disk_key disk_key; |
3769 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3770 | int i; |
be0e5c09 CM |
3771 | int push_space = 0; |
3772 | int push_items = 0; | |
0783fcfc | 3773 | struct btrfs_item *item; |
7518a238 | 3774 | u32 old_left_nritems; |
34a38218 | 3775 | u32 nr; |
aa5d6bed | 3776 | int ret = 0; |
db94535d CM |
3777 | u32 this_item_size; |
3778 | u32 old_left_item_size; | |
cfed81a0 CM |
3779 | struct btrfs_map_token token; |
3780 | ||
34a38218 | 3781 | if (empty) |
99d8f83c | 3782 | nr = min(right_nritems, max_slot); |
34a38218 | 3783 | else |
99d8f83c | 3784 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3785 | |
3786 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3787 | item = btrfs_item_nr(i); |
db94535d | 3788 | |
31840ae1 ZY |
3789 | if (!empty && push_items > 0) { |
3790 | if (path->slots[0] < i) | |
3791 | break; | |
3792 | if (path->slots[0] == i) { | |
e902baac DS |
3793 | int space = btrfs_leaf_free_space(right); |
3794 | ||
31840ae1 ZY |
3795 | if (space + push_space * 2 > free_space) |
3796 | break; | |
3797 | } | |
3798 | } | |
3799 | ||
be0e5c09 | 3800 | if (path->slots[0] == i) |
87b29b20 | 3801 | push_space += data_size; |
db94535d CM |
3802 | |
3803 | this_item_size = btrfs_item_size(right, item); | |
3804 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3805 | break; |
db94535d | 3806 | |
be0e5c09 | 3807 | push_items++; |
db94535d CM |
3808 | push_space += this_item_size + sizeof(*item); |
3809 | } | |
3810 | ||
be0e5c09 | 3811 | if (push_items == 0) { |
925baedd CM |
3812 | ret = 1; |
3813 | goto out; | |
be0e5c09 | 3814 | } |
fae7f21c | 3815 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3816 | |
be0e5c09 | 3817 | /* push data from right to left */ |
5f39d397 CM |
3818 | copy_extent_buffer(left, right, |
3819 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3820 | btrfs_item_nr_offset(0), | |
3821 | push_items * sizeof(struct btrfs_item)); | |
3822 | ||
0b246afa | 3823 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3824 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3825 | |
3d9ec8c4 | 3826 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3827 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3828 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3829 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3830 | push_space); |
5f39d397 | 3831 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3832 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3833 | |
c82f823c | 3834 | btrfs_init_map_token(&token, left); |
db94535d | 3835 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3836 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3837 | u32 ioff; |
db94535d | 3838 | |
dd3cc16b | 3839 | item = btrfs_item_nr(i); |
db94535d | 3840 | |
cc4c13d5 DS |
3841 | ioff = btrfs_token_item_offset(&token, item); |
3842 | btrfs_set_token_item_offset(&token, item, | |
3843 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); | |
be0e5c09 | 3844 | } |
5f39d397 | 3845 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3846 | |
3847 | /* fixup right node */ | |
31b1a2bd JL |
3848 | if (push_items > right_nritems) |
3849 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3850 | right_nritems); |
34a38218 CM |
3851 | |
3852 | if (push_items < right_nritems) { | |
3853 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
8f881e8c | 3854 | leaf_data_end(right); |
3d9ec8c4 | 3855 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3856 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3857 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3858 | leaf_data_end(right), push_space); |
34a38218 CM |
3859 | |
3860 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3861 | btrfs_item_nr_offset(push_items), |
3862 | (btrfs_header_nritems(right) - push_items) * | |
3863 | sizeof(struct btrfs_item)); | |
34a38218 | 3864 | } |
c82f823c DS |
3865 | |
3866 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3867 | right_nritems -= push_items; |
3868 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3869 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3870 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3871 | item = btrfs_item_nr(i); |
db94535d | 3872 | |
cc4c13d5 DS |
3873 | push_space = push_space - btrfs_token_item_size(&token, item); |
3874 | btrfs_set_token_item_offset(&token, item, push_space); | |
db94535d | 3875 | } |
eb60ceac | 3876 | |
5f39d397 | 3877 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3878 | if (right_nritems) |
3879 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3880 | else |
6a884d7d | 3881 | btrfs_clean_tree_block(right); |
098f59c2 | 3882 | |
5f39d397 | 3883 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3884 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3885 | |
3886 | /* then fixup the leaf pointer in the path */ | |
3887 | if (path->slots[0] < push_items) { | |
3888 | path->slots[0] += old_left_nritems; | |
925baedd | 3889 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3890 | free_extent_buffer(path->nodes[0]); |
3891 | path->nodes[0] = left; | |
be0e5c09 CM |
3892 | path->slots[1] -= 1; |
3893 | } else { | |
925baedd | 3894 | btrfs_tree_unlock(left); |
5f39d397 | 3895 | free_extent_buffer(left); |
be0e5c09 CM |
3896 | path->slots[0] -= push_items; |
3897 | } | |
eb60ceac | 3898 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3899 | return ret; |
925baedd CM |
3900 | out: |
3901 | btrfs_tree_unlock(left); | |
3902 | free_extent_buffer(left); | |
3903 | return ret; | |
be0e5c09 CM |
3904 | } |
3905 | ||
44871b1b CM |
3906 | /* |
3907 | * push some data in the path leaf to the left, trying to free up at | |
3908 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3909 | * |
3910 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3911 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3912 | * items | |
44871b1b CM |
3913 | */ |
3914 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3915 | *root, struct btrfs_path *path, int min_data_size, |
3916 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3917 | { |
3918 | struct extent_buffer *right = path->nodes[0]; | |
3919 | struct extent_buffer *left; | |
3920 | int slot; | |
3921 | int free_space; | |
3922 | u32 right_nritems; | |
3923 | int ret = 0; | |
3924 | ||
3925 | slot = path->slots[1]; | |
3926 | if (slot == 0) | |
3927 | return 1; | |
3928 | if (!path->nodes[1]) | |
3929 | return 1; | |
3930 | ||
3931 | right_nritems = btrfs_header_nritems(right); | |
3932 | if (right_nritems == 0) | |
3933 | return 1; | |
3934 | ||
3935 | btrfs_assert_tree_locked(path->nodes[1]); | |
3936 | ||
4b231ae4 | 3937 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
3938 | /* |
3939 | * slot - 1 is not valid or we fail to read the left node, | |
3940 | * no big deal, just return. | |
3941 | */ | |
3942 | if (IS_ERR(left)) | |
91ca338d TI |
3943 | return 1; |
3944 | ||
44871b1b | 3945 | btrfs_tree_lock(left); |
8bead258 | 3946 | btrfs_set_lock_blocking_write(left); |
44871b1b | 3947 | |
e902baac | 3948 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3949 | if (free_space < data_size) { |
3950 | ret = 1; | |
3951 | goto out; | |
3952 | } | |
3953 | ||
3954 | /* cow and double check */ | |
3955 | ret = btrfs_cow_block(trans, root, left, | |
3956 | path->nodes[1], slot - 1, &left); | |
3957 | if (ret) { | |
3958 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3959 | if (ret == -ENOSPC) |
3960 | ret = 1; | |
44871b1b CM |
3961 | goto out; |
3962 | } | |
3963 | ||
e902baac | 3964 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3965 | if (free_space < data_size) { |
3966 | ret = 1; | |
3967 | goto out; | |
3968 | } | |
3969 | ||
8087c193 | 3970 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
3971 | empty, left, free_space, right_nritems, |
3972 | max_slot); | |
44871b1b CM |
3973 | out: |
3974 | btrfs_tree_unlock(left); | |
3975 | free_extent_buffer(left); | |
3976 | return ret; | |
3977 | } | |
3978 | ||
3979 | /* | |
3980 | * split the path's leaf in two, making sure there is at least data_size | |
3981 | * available for the resulting leaf level of the path. | |
44871b1b | 3982 | */ |
143bede5 | 3983 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
3984 | struct btrfs_path *path, |
3985 | struct extent_buffer *l, | |
3986 | struct extent_buffer *right, | |
3987 | int slot, int mid, int nritems) | |
44871b1b | 3988 | { |
94f94ad9 | 3989 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
3990 | int data_copy_size; |
3991 | int rt_data_off; | |
3992 | int i; | |
44871b1b | 3993 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3994 | struct btrfs_map_token token; |
3995 | ||
44871b1b CM |
3996 | nritems = nritems - mid; |
3997 | btrfs_set_header_nritems(right, nritems); | |
8f881e8c | 3998 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); |
44871b1b CM |
3999 | |
4000 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4001 | btrfs_item_nr_offset(mid), | |
4002 | nritems * sizeof(struct btrfs_item)); | |
4003 | ||
4004 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4005 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4006 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 4007 | leaf_data_end(l), data_copy_size); |
44871b1b | 4008 | |
0b246afa | 4009 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b | 4010 | |
c82f823c | 4011 | btrfs_init_map_token(&token, right); |
44871b1b | 4012 | for (i = 0; i < nritems; i++) { |
dd3cc16b | 4013 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4014 | u32 ioff; |
4015 | ||
cc4c13d5 DS |
4016 | ioff = btrfs_token_item_offset(&token, item); |
4017 | btrfs_set_token_item_offset(&token, item, ioff + rt_data_off); | |
44871b1b CM |
4018 | } |
4019 | ||
44871b1b | 4020 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4021 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 4022 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
4023 | |
4024 | btrfs_mark_buffer_dirty(right); | |
4025 | btrfs_mark_buffer_dirty(l); | |
4026 | BUG_ON(path->slots[0] != slot); | |
4027 | ||
44871b1b CM |
4028 | if (mid <= slot) { |
4029 | btrfs_tree_unlock(path->nodes[0]); | |
4030 | free_extent_buffer(path->nodes[0]); | |
4031 | path->nodes[0] = right; | |
4032 | path->slots[0] -= mid; | |
4033 | path->slots[1] += 1; | |
4034 | } else { | |
4035 | btrfs_tree_unlock(right); | |
4036 | free_extent_buffer(right); | |
4037 | } | |
4038 | ||
4039 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4040 | } |
4041 | ||
99d8f83c CM |
4042 | /* |
4043 | * double splits happen when we need to insert a big item in the middle | |
4044 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4045 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4046 | * A B C | |
4047 | * | |
4048 | * We avoid this by trying to push the items on either side of our target | |
4049 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4050 | * completely. | |
4051 | */ | |
4052 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4053 | struct btrfs_root *root, | |
4054 | struct btrfs_path *path, | |
4055 | int data_size) | |
4056 | { | |
4057 | int ret; | |
4058 | int progress = 0; | |
4059 | int slot; | |
4060 | u32 nritems; | |
5a4267ca | 4061 | int space_needed = data_size; |
99d8f83c CM |
4062 | |
4063 | slot = path->slots[0]; | |
5a4267ca | 4064 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 4065 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
4066 | |
4067 | /* | |
4068 | * try to push all the items after our slot into the | |
4069 | * right leaf | |
4070 | */ | |
5a4267ca | 4071 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4072 | if (ret < 0) |
4073 | return ret; | |
4074 | ||
4075 | if (ret == 0) | |
4076 | progress++; | |
4077 | ||
4078 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4079 | /* | |
4080 | * our goal is to get our slot at the start or end of a leaf. If | |
4081 | * we've done so we're done | |
4082 | */ | |
4083 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4084 | return 0; | |
4085 | ||
e902baac | 4086 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4087 | return 0; |
4088 | ||
4089 | /* try to push all the items before our slot into the next leaf */ | |
4090 | slot = path->slots[0]; | |
263d3995 FM |
4091 | space_needed = data_size; |
4092 | if (slot > 0) | |
e902baac | 4093 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 4094 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4095 | if (ret < 0) |
4096 | return ret; | |
4097 | ||
4098 | if (ret == 0) | |
4099 | progress++; | |
4100 | ||
4101 | if (progress) | |
4102 | return 0; | |
4103 | return 1; | |
4104 | } | |
4105 | ||
74123bd7 CM |
4106 | /* |
4107 | * split the path's leaf in two, making sure there is at least data_size | |
4108 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4109 | * |
4110 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4111 | */ |
e02119d5 CM |
4112 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4113 | struct btrfs_root *root, | |
310712b2 | 4114 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4115 | struct btrfs_path *path, int data_size, |
4116 | int extend) | |
be0e5c09 | 4117 | { |
5d4f98a2 | 4118 | struct btrfs_disk_key disk_key; |
5f39d397 | 4119 | struct extent_buffer *l; |
7518a238 | 4120 | u32 nritems; |
eb60ceac CM |
4121 | int mid; |
4122 | int slot; | |
5f39d397 | 4123 | struct extent_buffer *right; |
b7a0365e | 4124 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4125 | int ret = 0; |
aa5d6bed | 4126 | int wret; |
5d4f98a2 | 4127 | int split; |
cc0c5538 | 4128 | int num_doubles = 0; |
99d8f83c | 4129 | int tried_avoid_double = 0; |
aa5d6bed | 4130 | |
a5719521 YZ |
4131 | l = path->nodes[0]; |
4132 | slot = path->slots[0]; | |
4133 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4134 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4135 | return -EOVERFLOW; |
4136 | ||
40689478 | 4137 | /* first try to make some room by pushing left and right */ |
33157e05 | 4138 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4139 | int space_needed = data_size; |
4140 | ||
4141 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 4142 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4143 | |
4144 | wret = push_leaf_right(trans, root, path, space_needed, | |
4145 | space_needed, 0, 0); | |
d397712b | 4146 | if (wret < 0) |
eaee50e8 | 4147 | return wret; |
3685f791 | 4148 | if (wret) { |
263d3995 FM |
4149 | space_needed = data_size; |
4150 | if (slot > 0) | |
e902baac | 4151 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4152 | wret = push_leaf_left(trans, root, path, space_needed, |
4153 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4154 | if (wret < 0) |
4155 | return wret; | |
4156 | } | |
4157 | l = path->nodes[0]; | |
aa5d6bed | 4158 | |
3685f791 | 4159 | /* did the pushes work? */ |
e902baac | 4160 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 4161 | return 0; |
3326d1b0 | 4162 | } |
aa5d6bed | 4163 | |
5c680ed6 | 4164 | if (!path->nodes[1]) { |
fdd99c72 | 4165 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4166 | if (ret) |
4167 | return ret; | |
4168 | } | |
cc0c5538 | 4169 | again: |
5d4f98a2 | 4170 | split = 1; |
cc0c5538 | 4171 | l = path->nodes[0]; |
eb60ceac | 4172 | slot = path->slots[0]; |
5f39d397 | 4173 | nritems = btrfs_header_nritems(l); |
d397712b | 4174 | mid = (nritems + 1) / 2; |
54aa1f4d | 4175 | |
5d4f98a2 YZ |
4176 | if (mid <= slot) { |
4177 | if (nritems == 1 || | |
4178 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4179 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4180 | if (slot >= nritems) { |
4181 | split = 0; | |
4182 | } else { | |
4183 | mid = slot; | |
4184 | if (mid != nritems && | |
4185 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4186 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4187 | if (data_size && !tried_avoid_double) |
4188 | goto push_for_double; | |
5d4f98a2 YZ |
4189 | split = 2; |
4190 | } | |
4191 | } | |
4192 | } | |
4193 | } else { | |
4194 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4195 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4196 | if (!extend && data_size && slot == 0) { |
4197 | split = 0; | |
4198 | } else if ((extend || !data_size) && slot == 0) { | |
4199 | mid = 1; | |
4200 | } else { | |
4201 | mid = slot; | |
4202 | if (mid != nritems && | |
4203 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4204 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4205 | if (data_size && !tried_avoid_double) |
4206 | goto push_for_double; | |
67871254 | 4207 | split = 2; |
5d4f98a2 YZ |
4208 | } |
4209 | } | |
4210 | } | |
4211 | } | |
4212 | ||
4213 | if (split == 0) | |
4214 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4215 | else | |
4216 | btrfs_item_key(l, &disk_key, mid); | |
4217 | ||
a6279470 FM |
4218 | right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, |
4219 | l->start, 0); | |
f0486c68 | 4220 | if (IS_ERR(right)) |
5f39d397 | 4221 | return PTR_ERR(right); |
f0486c68 | 4222 | |
0b246afa | 4223 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4224 | |
5d4f98a2 YZ |
4225 | if (split == 0) { |
4226 | if (mid <= slot) { | |
4227 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4228 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4229 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4230 | btrfs_tree_unlock(path->nodes[0]); |
4231 | free_extent_buffer(path->nodes[0]); | |
4232 | path->nodes[0] = right; | |
4233 | path->slots[0] = 0; | |
4234 | path->slots[1] += 1; | |
4235 | } else { | |
4236 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4237 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4238 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
4239 | btrfs_tree_unlock(path->nodes[0]); |
4240 | free_extent_buffer(path->nodes[0]); | |
4241 | path->nodes[0] = right; | |
4242 | path->slots[0] = 0; | |
143bede5 | 4243 | if (path->slots[1] == 0) |
b167fa91 | 4244 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4245 | } |
196e0249 LB |
4246 | /* |
4247 | * We create a new leaf 'right' for the required ins_len and | |
4248 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4249 | * the content of ins_len to 'right'. | |
4250 | */ | |
5d4f98a2 | 4251 | return ret; |
d4dbff95 | 4252 | } |
74123bd7 | 4253 | |
94f94ad9 | 4254 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 4255 | |
5d4f98a2 | 4256 | if (split == 2) { |
cc0c5538 CM |
4257 | BUG_ON(num_doubles != 0); |
4258 | num_doubles++; | |
4259 | goto again; | |
a429e513 | 4260 | } |
44871b1b | 4261 | |
143bede5 | 4262 | return 0; |
99d8f83c CM |
4263 | |
4264 | push_for_double: | |
4265 | push_for_double_split(trans, root, path, data_size); | |
4266 | tried_avoid_double = 1; | |
e902baac | 4267 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4268 | return 0; |
4269 | goto again; | |
be0e5c09 CM |
4270 | } |
4271 | ||
ad48fd75 YZ |
4272 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4273 | struct btrfs_root *root, | |
4274 | struct btrfs_path *path, int ins_len) | |
459931ec | 4275 | { |
ad48fd75 | 4276 | struct btrfs_key key; |
459931ec | 4277 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4278 | struct btrfs_file_extent_item *fi; |
4279 | u64 extent_len = 0; | |
4280 | u32 item_size; | |
4281 | int ret; | |
459931ec CM |
4282 | |
4283 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4284 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4285 | ||
4286 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4287 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4288 | ||
e902baac | 4289 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 4290 | return 0; |
459931ec CM |
4291 | |
4292 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4293 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4294 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4295 | struct btrfs_file_extent_item); | |
4296 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4297 | } | |
b3b4aa74 | 4298 | btrfs_release_path(path); |
459931ec | 4299 | |
459931ec | 4300 | path->keep_locks = 1; |
ad48fd75 YZ |
4301 | path->search_for_split = 1; |
4302 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4303 | path->search_for_split = 0; |
a8df6fe6 FM |
4304 | if (ret > 0) |
4305 | ret = -EAGAIN; | |
ad48fd75 YZ |
4306 | if (ret < 0) |
4307 | goto err; | |
459931ec | 4308 | |
ad48fd75 YZ |
4309 | ret = -EAGAIN; |
4310 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4311 | /* if our item isn't there, return now */ |
4312 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4313 | goto err; |
4314 | ||
109f6aef | 4315 | /* the leaf has changed, it now has room. return now */ |
e902baac | 4316 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
4317 | goto err; |
4318 | ||
ad48fd75 YZ |
4319 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4320 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4321 | struct btrfs_file_extent_item); | |
4322 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4323 | goto err; | |
459931ec CM |
4324 | } |
4325 | ||
b9473439 | 4326 | btrfs_set_path_blocking(path); |
ad48fd75 | 4327 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4328 | if (ret) |
4329 | goto err; | |
459931ec | 4330 | |
ad48fd75 | 4331 | path->keep_locks = 0; |
b9473439 | 4332 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4333 | return 0; |
4334 | err: | |
4335 | path->keep_locks = 0; | |
4336 | return ret; | |
4337 | } | |
4338 | ||
25263cd7 | 4339 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 4340 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4341 | unsigned long split_offset) |
4342 | { | |
4343 | struct extent_buffer *leaf; | |
4344 | struct btrfs_item *item; | |
4345 | struct btrfs_item *new_item; | |
4346 | int slot; | |
4347 | char *buf; | |
4348 | u32 nritems; | |
4349 | u32 item_size; | |
4350 | u32 orig_offset; | |
4351 | struct btrfs_disk_key disk_key; | |
4352 | ||
b9473439 | 4353 | leaf = path->nodes[0]; |
e902baac | 4354 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4355 | |
b4ce94de CM |
4356 | btrfs_set_path_blocking(path); |
4357 | ||
dd3cc16b | 4358 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4359 | orig_offset = btrfs_item_offset(leaf, item); |
4360 | item_size = btrfs_item_size(leaf, item); | |
4361 | ||
459931ec | 4362 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4363 | if (!buf) |
4364 | return -ENOMEM; | |
4365 | ||
459931ec CM |
4366 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4367 | path->slots[0]), item_size); | |
459931ec | 4368 | |
ad48fd75 | 4369 | slot = path->slots[0] + 1; |
459931ec | 4370 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4371 | if (slot != nritems) { |
4372 | /* shift the items */ | |
4373 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4374 | btrfs_item_nr_offset(slot), |
4375 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4376 | } |
4377 | ||
4378 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4379 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4380 | ||
dd3cc16b | 4381 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4382 | |
4383 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4384 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4385 | ||
4386 | btrfs_set_item_offset(leaf, item, | |
4387 | orig_offset + item_size - split_offset); | |
4388 | btrfs_set_item_size(leaf, item, split_offset); | |
4389 | ||
4390 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4391 | ||
4392 | /* write the data for the start of the original item */ | |
4393 | write_extent_buffer(leaf, buf, | |
4394 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4395 | split_offset); | |
4396 | ||
4397 | /* write the data for the new item */ | |
4398 | write_extent_buffer(leaf, buf + split_offset, | |
4399 | btrfs_item_ptr_offset(leaf, slot), | |
4400 | item_size - split_offset); | |
4401 | btrfs_mark_buffer_dirty(leaf); | |
4402 | ||
e902baac | 4403 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 4404 | kfree(buf); |
ad48fd75 YZ |
4405 | return 0; |
4406 | } | |
4407 | ||
4408 | /* | |
4409 | * This function splits a single item into two items, | |
4410 | * giving 'new_key' to the new item and splitting the | |
4411 | * old one at split_offset (from the start of the item). | |
4412 | * | |
4413 | * The path may be released by this operation. After | |
4414 | * the split, the path is pointing to the old item. The | |
4415 | * new item is going to be in the same node as the old one. | |
4416 | * | |
4417 | * Note, the item being split must be smaller enough to live alone on | |
4418 | * a tree block with room for one extra struct btrfs_item | |
4419 | * | |
4420 | * This allows us to split the item in place, keeping a lock on the | |
4421 | * leaf the entire time. | |
4422 | */ | |
4423 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4424 | struct btrfs_root *root, | |
4425 | struct btrfs_path *path, | |
310712b2 | 4426 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4427 | unsigned long split_offset) |
4428 | { | |
4429 | int ret; | |
4430 | ret = setup_leaf_for_split(trans, root, path, | |
4431 | sizeof(struct btrfs_item)); | |
4432 | if (ret) | |
4433 | return ret; | |
4434 | ||
25263cd7 | 4435 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
4436 | return ret; |
4437 | } | |
4438 | ||
ad48fd75 YZ |
4439 | /* |
4440 | * This function duplicate a item, giving 'new_key' to the new item. | |
4441 | * It guarantees both items live in the same tree leaf and the new item | |
4442 | * is contiguous with the original item. | |
4443 | * | |
4444 | * This allows us to split file extent in place, keeping a lock on the | |
4445 | * leaf the entire time. | |
4446 | */ | |
4447 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4448 | struct btrfs_root *root, | |
4449 | struct btrfs_path *path, | |
310712b2 | 4450 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4451 | { |
4452 | struct extent_buffer *leaf; | |
4453 | int ret; | |
4454 | u32 item_size; | |
4455 | ||
4456 | leaf = path->nodes[0]; | |
4457 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4458 | ret = setup_leaf_for_split(trans, root, path, | |
4459 | item_size + sizeof(struct btrfs_item)); | |
4460 | if (ret) | |
4461 | return ret; | |
4462 | ||
4463 | path->slots[0]++; | |
afe5fea7 | 4464 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4465 | item_size, item_size + |
4466 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4467 | leaf = path->nodes[0]; |
4468 | memcpy_extent_buffer(leaf, | |
4469 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4470 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4471 | item_size); | |
4472 | return 0; | |
4473 | } | |
4474 | ||
d352ac68 CM |
4475 | /* |
4476 | * make the item pointed to by the path smaller. new_size indicates | |
4477 | * how small to make it, and from_end tells us if we just chop bytes | |
4478 | * off the end of the item or if we shift the item to chop bytes off | |
4479 | * the front. | |
4480 | */ | |
78ac4f9e | 4481 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 4482 | { |
b18c6685 | 4483 | int slot; |
5f39d397 CM |
4484 | struct extent_buffer *leaf; |
4485 | struct btrfs_item *item; | |
b18c6685 CM |
4486 | u32 nritems; |
4487 | unsigned int data_end; | |
4488 | unsigned int old_data_start; | |
4489 | unsigned int old_size; | |
4490 | unsigned int size_diff; | |
4491 | int i; | |
cfed81a0 CM |
4492 | struct btrfs_map_token token; |
4493 | ||
5f39d397 | 4494 | leaf = path->nodes[0]; |
179e29e4 CM |
4495 | slot = path->slots[0]; |
4496 | ||
4497 | old_size = btrfs_item_size_nr(leaf, slot); | |
4498 | if (old_size == new_size) | |
143bede5 | 4499 | return; |
b18c6685 | 4500 | |
5f39d397 | 4501 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4502 | data_end = leaf_data_end(leaf); |
b18c6685 | 4503 | |
5f39d397 | 4504 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4505 | |
b18c6685 CM |
4506 | size_diff = old_size - new_size; |
4507 | ||
4508 | BUG_ON(slot < 0); | |
4509 | BUG_ON(slot >= nritems); | |
4510 | ||
4511 | /* | |
4512 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4513 | */ | |
4514 | /* first correct the data pointers */ | |
c82f823c | 4515 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 4516 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4517 | u32 ioff; |
dd3cc16b | 4518 | item = btrfs_item_nr(i); |
db94535d | 4519 | |
cc4c13d5 DS |
4520 | ioff = btrfs_token_item_offset(&token, item); |
4521 | btrfs_set_token_item_offset(&token, item, ioff + size_diff); | |
b18c6685 | 4522 | } |
db94535d | 4523 | |
b18c6685 | 4524 | /* shift the data */ |
179e29e4 | 4525 | if (from_end) { |
3d9ec8c4 NB |
4526 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4527 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4528 | data_end, old_data_start + new_size - data_end); |
4529 | } else { | |
4530 | struct btrfs_disk_key disk_key; | |
4531 | u64 offset; | |
4532 | ||
4533 | btrfs_item_key(leaf, &disk_key, slot); | |
4534 | ||
4535 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4536 | unsigned long ptr; | |
4537 | struct btrfs_file_extent_item *fi; | |
4538 | ||
4539 | fi = btrfs_item_ptr(leaf, slot, | |
4540 | struct btrfs_file_extent_item); | |
4541 | fi = (struct btrfs_file_extent_item *)( | |
4542 | (unsigned long)fi - size_diff); | |
4543 | ||
4544 | if (btrfs_file_extent_type(leaf, fi) == | |
4545 | BTRFS_FILE_EXTENT_INLINE) { | |
4546 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4547 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4548 | (unsigned long)fi, |
7ec20afb | 4549 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4550 | } |
4551 | } | |
4552 | ||
3d9ec8c4 NB |
4553 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4554 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4555 | data_end, old_data_start - data_end); |
4556 | ||
4557 | offset = btrfs_disk_key_offset(&disk_key); | |
4558 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4559 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4560 | if (slot == 0) | |
b167fa91 | 4561 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4562 | } |
5f39d397 | 4563 | |
dd3cc16b | 4564 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4565 | btrfs_set_item_size(leaf, item, new_size); |
4566 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4567 | |
e902baac | 4568 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4569 | btrfs_print_leaf(leaf); |
b18c6685 | 4570 | BUG(); |
5f39d397 | 4571 | } |
b18c6685 CM |
4572 | } |
4573 | ||
d352ac68 | 4574 | /* |
8f69dbd2 | 4575 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4576 | */ |
c71dd880 | 4577 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 4578 | { |
6567e837 | 4579 | int slot; |
5f39d397 CM |
4580 | struct extent_buffer *leaf; |
4581 | struct btrfs_item *item; | |
6567e837 CM |
4582 | u32 nritems; |
4583 | unsigned int data_end; | |
4584 | unsigned int old_data; | |
4585 | unsigned int old_size; | |
4586 | int i; | |
cfed81a0 CM |
4587 | struct btrfs_map_token token; |
4588 | ||
5f39d397 | 4589 | leaf = path->nodes[0]; |
6567e837 | 4590 | |
5f39d397 | 4591 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4592 | data_end = leaf_data_end(leaf); |
6567e837 | 4593 | |
e902baac | 4594 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 4595 | btrfs_print_leaf(leaf); |
6567e837 | 4596 | BUG(); |
5f39d397 | 4597 | } |
6567e837 | 4598 | slot = path->slots[0]; |
5f39d397 | 4599 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4600 | |
4601 | BUG_ON(slot < 0); | |
3326d1b0 | 4602 | if (slot >= nritems) { |
a4f78750 | 4603 | btrfs_print_leaf(leaf); |
c71dd880 | 4604 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 4605 | slot, nritems); |
290342f6 | 4606 | BUG(); |
3326d1b0 | 4607 | } |
6567e837 CM |
4608 | |
4609 | /* | |
4610 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4611 | */ | |
4612 | /* first correct the data pointers */ | |
c82f823c | 4613 | btrfs_init_map_token(&token, leaf); |
6567e837 | 4614 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4615 | u32 ioff; |
dd3cc16b | 4616 | item = btrfs_item_nr(i); |
db94535d | 4617 | |
cc4c13d5 DS |
4618 | ioff = btrfs_token_item_offset(&token, item); |
4619 | btrfs_set_token_item_offset(&token, item, ioff - data_size); | |
6567e837 | 4620 | } |
5f39d397 | 4621 | |
6567e837 | 4622 | /* shift the data */ |
3d9ec8c4 NB |
4623 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4624 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4625 | data_end, old_data - data_end); |
5f39d397 | 4626 | |
6567e837 | 4627 | data_end = old_data; |
5f39d397 | 4628 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4629 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4630 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4631 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4632 | |
e902baac | 4633 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4634 | btrfs_print_leaf(leaf); |
6567e837 | 4635 | BUG(); |
5f39d397 | 4636 | } |
6567e837 CM |
4637 | } |
4638 | ||
74123bd7 | 4639 | /* |
44871b1b CM |
4640 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4641 | * to save stack depth by doing the bulk of the work in a function | |
4642 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4643 | */ |
afe5fea7 | 4644 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4645 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4646 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4647 | { |
0b246afa | 4648 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4649 | struct btrfs_item *item; |
9c58309d | 4650 | int i; |
7518a238 | 4651 | u32 nritems; |
be0e5c09 | 4652 | unsigned int data_end; |
e2fa7227 | 4653 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4654 | struct extent_buffer *leaf; |
4655 | int slot; | |
cfed81a0 CM |
4656 | struct btrfs_map_token token; |
4657 | ||
24cdc847 FM |
4658 | if (path->slots[0] == 0) { |
4659 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4660 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4661 | } |
4662 | btrfs_unlock_up_safe(path, 1); | |
4663 | ||
5f39d397 | 4664 | leaf = path->nodes[0]; |
44871b1b | 4665 | slot = path->slots[0]; |
74123bd7 | 4666 | |
5f39d397 | 4667 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4668 | data_end = leaf_data_end(leaf); |
eb60ceac | 4669 | |
e902baac | 4670 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4671 | btrfs_print_leaf(leaf); |
0b246afa | 4672 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4673 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4674 | BUG(); |
d4dbff95 | 4675 | } |
5f39d397 | 4676 | |
c82f823c | 4677 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4678 | if (slot != nritems) { |
5f39d397 | 4679 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4680 | |
5f39d397 | 4681 | if (old_data < data_end) { |
a4f78750 | 4682 | btrfs_print_leaf(leaf); |
0b246afa | 4683 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4684 | slot, old_data, data_end); |
290342f6 | 4685 | BUG(); |
5f39d397 | 4686 | } |
be0e5c09 CM |
4687 | /* |
4688 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4689 | */ | |
4690 | /* first correct the data pointers */ | |
0783fcfc | 4691 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4692 | u32 ioff; |
db94535d | 4693 | |
62e85577 | 4694 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
4695 | ioff = btrfs_token_item_offset(&token, item); |
4696 | btrfs_set_token_item_offset(&token, item, | |
4697 | ioff - total_data); | |
0783fcfc | 4698 | } |
be0e5c09 | 4699 | /* shift the items */ |
9c58309d | 4700 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4701 | btrfs_item_nr_offset(slot), |
d6025579 | 4702 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4703 | |
4704 | /* shift the data */ | |
3d9ec8c4 NB |
4705 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4706 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4707 | data_end, old_data - data_end); |
be0e5c09 CM |
4708 | data_end = old_data; |
4709 | } | |
5f39d397 | 4710 | |
62e2749e | 4711 | /* setup the item for the new data */ |
9c58309d CM |
4712 | for (i = 0; i < nr; i++) { |
4713 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4714 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4715 | item = btrfs_item_nr(slot + i); |
cc4c13d5 | 4716 | btrfs_set_token_item_offset(&token, item, data_end - data_size[i]); |
9c58309d | 4717 | data_end -= data_size[i]; |
cc4c13d5 | 4718 | btrfs_set_token_item_size(&token, item, data_size[i]); |
9c58309d | 4719 | } |
44871b1b | 4720 | |
9c58309d | 4721 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4722 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4723 | |
e902baac | 4724 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4725 | btrfs_print_leaf(leaf); |
be0e5c09 | 4726 | BUG(); |
5f39d397 | 4727 | } |
44871b1b CM |
4728 | } |
4729 | ||
4730 | /* | |
4731 | * Given a key and some data, insert items into the tree. | |
4732 | * This does all the path init required, making room in the tree if needed. | |
4733 | */ | |
4734 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4735 | struct btrfs_root *root, | |
4736 | struct btrfs_path *path, | |
310712b2 | 4737 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4738 | int nr) |
4739 | { | |
44871b1b CM |
4740 | int ret = 0; |
4741 | int slot; | |
4742 | int i; | |
4743 | u32 total_size = 0; | |
4744 | u32 total_data = 0; | |
4745 | ||
4746 | for (i = 0; i < nr; i++) | |
4747 | total_data += data_size[i]; | |
4748 | ||
4749 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4750 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4751 | if (ret == 0) | |
4752 | return -EEXIST; | |
4753 | if (ret < 0) | |
143bede5 | 4754 | return ret; |
44871b1b | 4755 | |
44871b1b CM |
4756 | slot = path->slots[0]; |
4757 | BUG_ON(slot < 0); | |
4758 | ||
afe5fea7 | 4759 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4760 | total_data, total_size, nr); |
143bede5 | 4761 | return 0; |
62e2749e CM |
4762 | } |
4763 | ||
4764 | /* | |
4765 | * Given a key and some data, insert an item into the tree. | |
4766 | * This does all the path init required, making room in the tree if needed. | |
4767 | */ | |
310712b2 OS |
4768 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4769 | const struct btrfs_key *cpu_key, void *data, | |
4770 | u32 data_size) | |
62e2749e CM |
4771 | { |
4772 | int ret = 0; | |
2c90e5d6 | 4773 | struct btrfs_path *path; |
5f39d397 CM |
4774 | struct extent_buffer *leaf; |
4775 | unsigned long ptr; | |
62e2749e | 4776 | |
2c90e5d6 | 4777 | path = btrfs_alloc_path(); |
db5b493a TI |
4778 | if (!path) |
4779 | return -ENOMEM; | |
2c90e5d6 | 4780 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4781 | if (!ret) { |
5f39d397 CM |
4782 | leaf = path->nodes[0]; |
4783 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4784 | write_extent_buffer(leaf, data, ptr, data_size); | |
4785 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4786 | } |
2c90e5d6 | 4787 | btrfs_free_path(path); |
aa5d6bed | 4788 | return ret; |
be0e5c09 CM |
4789 | } |
4790 | ||
74123bd7 | 4791 | /* |
5de08d7d | 4792 | * delete the pointer from a given node. |
74123bd7 | 4793 | * |
d352ac68 CM |
4794 | * the tree should have been previously balanced so the deletion does not |
4795 | * empty a node. | |
74123bd7 | 4796 | */ |
afe5fea7 TI |
4797 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4798 | int level, int slot) | |
be0e5c09 | 4799 | { |
5f39d397 | 4800 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4801 | u32 nritems; |
f3ea38da | 4802 | int ret; |
be0e5c09 | 4803 | |
5f39d397 | 4804 | nritems = btrfs_header_nritems(parent); |
d397712b | 4805 | if (slot != nritems - 1) { |
bf1d3425 DS |
4806 | if (level) { |
4807 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4808 | nritems - slot - 1); |
bf1d3425 DS |
4809 | BUG_ON(ret < 0); |
4810 | } | |
5f39d397 CM |
4811 | memmove_extent_buffer(parent, |
4812 | btrfs_node_key_ptr_offset(slot), | |
4813 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4814 | sizeof(struct btrfs_key_ptr) * |
4815 | (nritems - slot - 1)); | |
57ba86c0 | 4816 | } else if (level) { |
e09c2efe DS |
4817 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4818 | GFP_NOFS); | |
57ba86c0 | 4819 | BUG_ON(ret < 0); |
bb803951 | 4820 | } |
f3ea38da | 4821 | |
7518a238 | 4822 | nritems--; |
5f39d397 | 4823 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4824 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4825 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4826 | /* just turn the root into a leaf and break */ |
5f39d397 | 4827 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4828 | } else if (slot == 0) { |
5f39d397 CM |
4829 | struct btrfs_disk_key disk_key; |
4830 | ||
4831 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4832 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4833 | } |
d6025579 | 4834 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4835 | } |
4836 | ||
323ac95b CM |
4837 | /* |
4838 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4839 | * path->nodes[1]. |
323ac95b CM |
4840 | * |
4841 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4842 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4843 | * | |
4844 | * The path must have already been setup for deleting the leaf, including | |
4845 | * all the proper balancing. path->nodes[1] must be locked. | |
4846 | */ | |
143bede5 JM |
4847 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4848 | struct btrfs_root *root, | |
4849 | struct btrfs_path *path, | |
4850 | struct extent_buffer *leaf) | |
323ac95b | 4851 | { |
5d4f98a2 | 4852 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4853 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4854 | |
4d081c41 CM |
4855 | /* |
4856 | * btrfs_free_extent is expensive, we want to make sure we | |
4857 | * aren't holding any locks when we call it | |
4858 | */ | |
4859 | btrfs_unlock_up_safe(path, 0); | |
4860 | ||
f0486c68 YZ |
4861 | root_sub_used(root, leaf->len); |
4862 | ||
67439dad | 4863 | atomic_inc(&leaf->refs); |
5581a51a | 4864 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4865 | free_extent_buffer_stale(leaf); |
323ac95b | 4866 | } |
74123bd7 CM |
4867 | /* |
4868 | * delete the item at the leaf level in path. If that empties | |
4869 | * the leaf, remove it from the tree | |
4870 | */ | |
85e21bac CM |
4871 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4872 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4873 | { |
0b246afa | 4874 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4875 | struct extent_buffer *leaf; |
4876 | struct btrfs_item *item; | |
ce0eac2a AM |
4877 | u32 last_off; |
4878 | u32 dsize = 0; | |
aa5d6bed CM |
4879 | int ret = 0; |
4880 | int wret; | |
85e21bac | 4881 | int i; |
7518a238 | 4882 | u32 nritems; |
be0e5c09 | 4883 | |
5f39d397 | 4884 | leaf = path->nodes[0]; |
85e21bac CM |
4885 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4886 | ||
4887 | for (i = 0; i < nr; i++) | |
4888 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4889 | ||
5f39d397 | 4890 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4891 | |
85e21bac | 4892 | if (slot + nr != nritems) { |
8f881e8c | 4893 | int data_end = leaf_data_end(leaf); |
c82f823c | 4894 | struct btrfs_map_token token; |
5f39d397 | 4895 | |
3d9ec8c4 | 4896 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4897 | data_end + dsize, |
3d9ec8c4 | 4898 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4899 | last_off - data_end); |
5f39d397 | 4900 | |
c82f823c | 4901 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4902 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4903 | u32 ioff; |
db94535d | 4904 | |
dd3cc16b | 4905 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
4906 | ioff = btrfs_token_item_offset(&token, item); |
4907 | btrfs_set_token_item_offset(&token, item, ioff + dsize); | |
0783fcfc | 4908 | } |
db94535d | 4909 | |
5f39d397 | 4910 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4911 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4912 | sizeof(struct btrfs_item) * |
85e21bac | 4913 | (nritems - slot - nr)); |
be0e5c09 | 4914 | } |
85e21bac CM |
4915 | btrfs_set_header_nritems(leaf, nritems - nr); |
4916 | nritems -= nr; | |
5f39d397 | 4917 | |
74123bd7 | 4918 | /* delete the leaf if we've emptied it */ |
7518a238 | 4919 | if (nritems == 0) { |
5f39d397 CM |
4920 | if (leaf == root->node) { |
4921 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4922 | } else { |
f0486c68 | 4923 | btrfs_set_path_blocking(path); |
6a884d7d | 4924 | btrfs_clean_tree_block(leaf); |
143bede5 | 4925 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4926 | } |
be0e5c09 | 4927 | } else { |
7518a238 | 4928 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4929 | if (slot == 0) { |
5f39d397 CM |
4930 | struct btrfs_disk_key disk_key; |
4931 | ||
4932 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4933 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 4934 | } |
aa5d6bed | 4935 | |
74123bd7 | 4936 | /* delete the leaf if it is mostly empty */ |
0b246afa | 4937 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
4938 | /* push_leaf_left fixes the path. |
4939 | * make sure the path still points to our leaf | |
4940 | * for possible call to del_ptr below | |
4941 | */ | |
4920c9ac | 4942 | slot = path->slots[1]; |
67439dad | 4943 | atomic_inc(&leaf->refs); |
5f39d397 | 4944 | |
b9473439 | 4945 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4946 | wret = push_leaf_left(trans, root, path, 1, 1, |
4947 | 1, (u32)-1); | |
54aa1f4d | 4948 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4949 | ret = wret; |
5f39d397 CM |
4950 | |
4951 | if (path->nodes[0] == leaf && | |
4952 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4953 | wret = push_leaf_right(trans, root, path, 1, |
4954 | 1, 1, 0); | |
54aa1f4d | 4955 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4956 | ret = wret; |
4957 | } | |
5f39d397 CM |
4958 | |
4959 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4960 | path->slots[1] = slot; |
143bede5 | 4961 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4962 | free_extent_buffer(leaf); |
143bede5 | 4963 | ret = 0; |
5de08d7d | 4964 | } else { |
925baedd CM |
4965 | /* if we're still in the path, make sure |
4966 | * we're dirty. Otherwise, one of the | |
4967 | * push_leaf functions must have already | |
4968 | * dirtied this buffer | |
4969 | */ | |
4970 | if (path->nodes[0] == leaf) | |
4971 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4972 | free_extent_buffer(leaf); |
be0e5c09 | 4973 | } |
d5719762 | 4974 | } else { |
5f39d397 | 4975 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4976 | } |
4977 | } | |
aa5d6bed | 4978 | return ret; |
be0e5c09 CM |
4979 | } |
4980 | ||
7bb86316 | 4981 | /* |
925baedd | 4982 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4983 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4984 | * returns < 0 on io errors. | |
d352ac68 CM |
4985 | * |
4986 | * This may release the path, and so you may lose any locks held at the | |
4987 | * time you call it. | |
7bb86316 | 4988 | */ |
16e7549f | 4989 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 4990 | { |
925baedd CM |
4991 | struct btrfs_key key; |
4992 | struct btrfs_disk_key found_key; | |
4993 | int ret; | |
7bb86316 | 4994 | |
925baedd | 4995 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4996 | |
e8b0d724 | 4997 | if (key.offset > 0) { |
925baedd | 4998 | key.offset--; |
e8b0d724 | 4999 | } else if (key.type > 0) { |
925baedd | 5000 | key.type--; |
e8b0d724 FDBM |
5001 | key.offset = (u64)-1; |
5002 | } else if (key.objectid > 0) { | |
925baedd | 5003 | key.objectid--; |
e8b0d724 FDBM |
5004 | key.type = (u8)-1; |
5005 | key.offset = (u64)-1; | |
5006 | } else { | |
925baedd | 5007 | return 1; |
e8b0d724 | 5008 | } |
7bb86316 | 5009 | |
b3b4aa74 | 5010 | btrfs_release_path(path); |
925baedd CM |
5011 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5012 | if (ret < 0) | |
5013 | return ret; | |
5014 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5015 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5016 | /* |
5017 | * We might have had an item with the previous key in the tree right | |
5018 | * before we released our path. And after we released our path, that | |
5019 | * item might have been pushed to the first slot (0) of the leaf we | |
5020 | * were holding due to a tree balance. Alternatively, an item with the | |
5021 | * previous key can exist as the only element of a leaf (big fat item). | |
5022 | * Therefore account for these 2 cases, so that our callers (like | |
5023 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5024 | * the previous key we computed above. | |
5025 | */ | |
5026 | if (ret <= 0) | |
925baedd CM |
5027 | return 0; |
5028 | return 1; | |
7bb86316 CM |
5029 | } |
5030 | ||
3f157a2f CM |
5031 | /* |
5032 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5033 | * for nodes or leaves that are have a minimum transaction id. |
5034 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5035 | * |
5036 | * This does not cow, but it does stuff the starting key it finds back | |
5037 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5038 | * key and get a writable path. | |
5039 | * | |
3f157a2f CM |
5040 | * This honors path->lowest_level to prevent descent past a given level |
5041 | * of the tree. | |
5042 | * | |
d352ac68 CM |
5043 | * min_trans indicates the oldest transaction that you are interested |
5044 | * in walking through. Any nodes or leaves older than min_trans are | |
5045 | * skipped over (without reading them). | |
5046 | * | |
3f157a2f CM |
5047 | * returns zero if something useful was found, < 0 on error and 1 if there |
5048 | * was nothing in the tree that matched the search criteria. | |
5049 | */ | |
5050 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5051 | struct btrfs_path *path, |
3f157a2f CM |
5052 | u64 min_trans) |
5053 | { | |
5054 | struct extent_buffer *cur; | |
5055 | struct btrfs_key found_key; | |
5056 | int slot; | |
9652480b | 5057 | int sret; |
3f157a2f CM |
5058 | u32 nritems; |
5059 | int level; | |
5060 | int ret = 1; | |
f98de9b9 | 5061 | int keep_locks = path->keep_locks; |
3f157a2f | 5062 | |
f98de9b9 | 5063 | path->keep_locks = 1; |
3f157a2f | 5064 | again: |
bd681513 | 5065 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5066 | level = btrfs_header_level(cur); |
e02119d5 | 5067 | WARN_ON(path->nodes[level]); |
3f157a2f | 5068 | path->nodes[level] = cur; |
bd681513 | 5069 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5070 | |
5071 | if (btrfs_header_generation(cur) < min_trans) { | |
5072 | ret = 1; | |
5073 | goto out; | |
5074 | } | |
d397712b | 5075 | while (1) { |
3f157a2f CM |
5076 | nritems = btrfs_header_nritems(cur); |
5077 | level = btrfs_header_level(cur); | |
e3b83361 | 5078 | sret = btrfs_bin_search(cur, min_key, &slot); |
cbca7d59 FM |
5079 | if (sret < 0) { |
5080 | ret = sret; | |
5081 | goto out; | |
5082 | } | |
3f157a2f | 5083 | |
323ac95b CM |
5084 | /* at the lowest level, we're done, setup the path and exit */ |
5085 | if (level == path->lowest_level) { | |
e02119d5 CM |
5086 | if (slot >= nritems) |
5087 | goto find_next_key; | |
3f157a2f CM |
5088 | ret = 0; |
5089 | path->slots[level] = slot; | |
5090 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5091 | goto out; | |
5092 | } | |
9652480b Y |
5093 | if (sret && slot > 0) |
5094 | slot--; | |
3f157a2f | 5095 | /* |
de78b51a ES |
5096 | * check this node pointer against the min_trans parameters. |
5097 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5098 | */ |
d397712b | 5099 | while (slot < nritems) { |
3f157a2f | 5100 | u64 gen; |
e02119d5 | 5101 | |
3f157a2f CM |
5102 | gen = btrfs_node_ptr_generation(cur, slot); |
5103 | if (gen < min_trans) { | |
5104 | slot++; | |
5105 | continue; | |
5106 | } | |
de78b51a | 5107 | break; |
3f157a2f | 5108 | } |
e02119d5 | 5109 | find_next_key: |
3f157a2f CM |
5110 | /* |
5111 | * we didn't find a candidate key in this node, walk forward | |
5112 | * and find another one | |
5113 | */ | |
5114 | if (slot >= nritems) { | |
e02119d5 | 5115 | path->slots[level] = slot; |
b4ce94de | 5116 | btrfs_set_path_blocking(path); |
e02119d5 | 5117 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5118 | min_trans); |
e02119d5 | 5119 | if (sret == 0) { |
b3b4aa74 | 5120 | btrfs_release_path(path); |
3f157a2f CM |
5121 | goto again; |
5122 | } else { | |
5123 | goto out; | |
5124 | } | |
5125 | } | |
5126 | /* save our key for returning back */ | |
5127 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5128 | path->slots[level] = slot; | |
5129 | if (level == path->lowest_level) { | |
5130 | ret = 0; | |
3f157a2f CM |
5131 | goto out; |
5132 | } | |
b4ce94de | 5133 | btrfs_set_path_blocking(path); |
4b231ae4 | 5134 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
5135 | if (IS_ERR(cur)) { |
5136 | ret = PTR_ERR(cur); | |
5137 | goto out; | |
5138 | } | |
3f157a2f | 5139 | |
bd681513 | 5140 | btrfs_tree_read_lock(cur); |
b4ce94de | 5141 | |
bd681513 | 5142 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5143 | path->nodes[level - 1] = cur; |
f7c79f30 | 5144 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5145 | } |
5146 | out: | |
f98de9b9 FM |
5147 | path->keep_locks = keep_locks; |
5148 | if (ret == 0) { | |
5149 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5150 | btrfs_set_path_blocking(path); | |
3f157a2f | 5151 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5152 | } |
3f157a2f CM |
5153 | return ret; |
5154 | } | |
5155 | ||
5156 | /* | |
5157 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5158 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5159 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5160 | * |
5161 | * 0 is returned if another key is found, < 0 if there are any errors | |
5162 | * and 1 is returned if there are no higher keys in the tree | |
5163 | * | |
5164 | * path->keep_locks should be set to 1 on the search made before | |
5165 | * calling this function. | |
5166 | */ | |
e7a84565 | 5167 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5168 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5169 | { |
e7a84565 CM |
5170 | int slot; |
5171 | struct extent_buffer *c; | |
5172 | ||
6a9fb468 | 5173 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 5174 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5175 | if (!path->nodes[level]) |
5176 | return 1; | |
5177 | ||
5178 | slot = path->slots[level] + 1; | |
5179 | c = path->nodes[level]; | |
3f157a2f | 5180 | next: |
e7a84565 | 5181 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5182 | int ret; |
5183 | int orig_lowest; | |
5184 | struct btrfs_key cur_key; | |
5185 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5186 | !path->nodes[level + 1]) | |
e7a84565 | 5187 | return 1; |
33c66f43 | 5188 | |
6a9fb468 | 5189 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
5190 | level++; |
5191 | continue; | |
5192 | } | |
5193 | ||
5194 | slot = btrfs_header_nritems(c) - 1; | |
5195 | if (level == 0) | |
5196 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5197 | else | |
5198 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5199 | ||
5200 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5201 | btrfs_release_path(path); |
33c66f43 YZ |
5202 | path->lowest_level = level; |
5203 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5204 | 0, 0); | |
5205 | path->lowest_level = orig_lowest; | |
5206 | if (ret < 0) | |
5207 | return ret; | |
5208 | ||
5209 | c = path->nodes[level]; | |
5210 | slot = path->slots[level]; | |
5211 | if (ret == 0) | |
5212 | slot++; | |
5213 | goto next; | |
e7a84565 | 5214 | } |
33c66f43 | 5215 | |
e7a84565 CM |
5216 | if (level == 0) |
5217 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5218 | else { |
3f157a2f CM |
5219 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5220 | ||
3f157a2f CM |
5221 | if (gen < min_trans) { |
5222 | slot++; | |
5223 | goto next; | |
5224 | } | |
e7a84565 | 5225 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5226 | } |
e7a84565 CM |
5227 | return 0; |
5228 | } | |
5229 | return 1; | |
5230 | } | |
5231 | ||
97571fd0 | 5232 | /* |
925baedd | 5233 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5234 | * returns 0 if it found something or 1 if there are no greater leaves. |
5235 | * returns < 0 on io errors. | |
97571fd0 | 5236 | */ |
234b63a0 | 5237 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5238 | { |
5239 | return btrfs_next_old_leaf(root, path, 0); | |
5240 | } | |
5241 | ||
5242 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5243 | u64 time_seq) | |
d97e63b6 CM |
5244 | { |
5245 | int slot; | |
8e73f275 | 5246 | int level; |
5f39d397 | 5247 | struct extent_buffer *c; |
8e73f275 | 5248 | struct extent_buffer *next; |
925baedd CM |
5249 | struct btrfs_key key; |
5250 | u32 nritems; | |
5251 | int ret; | |
8e73f275 | 5252 | int old_spinning = path->leave_spinning; |
bd681513 | 5253 | int next_rw_lock = 0; |
925baedd CM |
5254 | |
5255 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5256 | if (nritems == 0) |
925baedd | 5257 | return 1; |
925baedd | 5258 | |
8e73f275 CM |
5259 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5260 | again: | |
5261 | level = 1; | |
5262 | next = NULL; | |
bd681513 | 5263 | next_rw_lock = 0; |
b3b4aa74 | 5264 | btrfs_release_path(path); |
8e73f275 | 5265 | |
a2135011 | 5266 | path->keep_locks = 1; |
31533fb2 | 5267 | path->leave_spinning = 1; |
8e73f275 | 5268 | |
3d7806ec JS |
5269 | if (time_seq) |
5270 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5271 | else | |
5272 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5273 | path->keep_locks = 0; |
5274 | ||
5275 | if (ret < 0) | |
5276 | return ret; | |
5277 | ||
a2135011 | 5278 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5279 | /* |
5280 | * by releasing the path above we dropped all our locks. A balance | |
5281 | * could have added more items next to the key that used to be | |
5282 | * at the very end of the block. So, check again here and | |
5283 | * advance the path if there are now more items available. | |
5284 | */ | |
a2135011 | 5285 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5286 | if (ret == 0) |
5287 | path->slots[0]++; | |
8e73f275 | 5288 | ret = 0; |
925baedd CM |
5289 | goto done; |
5290 | } | |
0b43e04f LB |
5291 | /* |
5292 | * So the above check misses one case: | |
5293 | * - after releasing the path above, someone has removed the item that | |
5294 | * used to be at the very end of the block, and balance between leafs | |
5295 | * gets another one with bigger key.offset to replace it. | |
5296 | * | |
5297 | * This one should be returned as well, or we can get leaf corruption | |
5298 | * later(esp. in __btrfs_drop_extents()). | |
5299 | * | |
5300 | * And a bit more explanation about this check, | |
5301 | * with ret > 0, the key isn't found, the path points to the slot | |
5302 | * where it should be inserted, so the path->slots[0] item must be the | |
5303 | * bigger one. | |
5304 | */ | |
5305 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5306 | ret = 0; | |
5307 | goto done; | |
5308 | } | |
d97e63b6 | 5309 | |
d397712b | 5310 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5311 | if (!path->nodes[level]) { |
5312 | ret = 1; | |
5313 | goto done; | |
5314 | } | |
5f39d397 | 5315 | |
d97e63b6 CM |
5316 | slot = path->slots[level] + 1; |
5317 | c = path->nodes[level]; | |
5f39d397 | 5318 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5319 | level++; |
8e73f275 CM |
5320 | if (level == BTRFS_MAX_LEVEL) { |
5321 | ret = 1; | |
5322 | goto done; | |
5323 | } | |
d97e63b6 CM |
5324 | continue; |
5325 | } | |
5f39d397 | 5326 | |
925baedd | 5327 | if (next) { |
bd681513 | 5328 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5329 | free_extent_buffer(next); |
925baedd | 5330 | } |
5f39d397 | 5331 | |
8e73f275 | 5332 | next = c; |
bd681513 | 5333 | next_rw_lock = path->locks[level]; |
d07b8528 | 5334 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5335 | slot, &key); |
8e73f275 CM |
5336 | if (ret == -EAGAIN) |
5337 | goto again; | |
5f39d397 | 5338 | |
76a05b35 | 5339 | if (ret < 0) { |
b3b4aa74 | 5340 | btrfs_release_path(path); |
76a05b35 CM |
5341 | goto done; |
5342 | } | |
5343 | ||
5cd57b2c | 5344 | if (!path->skip_locking) { |
bd681513 | 5345 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5346 | if (!ret && time_seq) { |
5347 | /* | |
5348 | * If we don't get the lock, we may be racing | |
5349 | * with push_leaf_left, holding that lock while | |
5350 | * itself waiting for the leaf we've currently | |
5351 | * locked. To solve this situation, we give up | |
5352 | * on our lock and cycle. | |
5353 | */ | |
cf538830 | 5354 | free_extent_buffer(next); |
d42244a0 JS |
5355 | btrfs_release_path(path); |
5356 | cond_resched(); | |
5357 | goto again; | |
5358 | } | |
8e73f275 CM |
5359 | if (!ret) { |
5360 | btrfs_set_path_blocking(path); | |
bd681513 | 5361 | btrfs_tree_read_lock(next); |
8e73f275 | 5362 | } |
31533fb2 | 5363 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5364 | } |
d97e63b6 CM |
5365 | break; |
5366 | } | |
5367 | path->slots[level] = slot; | |
d397712b | 5368 | while (1) { |
d97e63b6 CM |
5369 | level--; |
5370 | c = path->nodes[level]; | |
925baedd | 5371 | if (path->locks[level]) |
bd681513 | 5372 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5373 | |
5f39d397 | 5374 | free_extent_buffer(c); |
d97e63b6 CM |
5375 | path->nodes[level] = next; |
5376 | path->slots[level] = 0; | |
a74a4b97 | 5377 | if (!path->skip_locking) |
bd681513 | 5378 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5379 | if (!level) |
5380 | break; | |
b4ce94de | 5381 | |
d07b8528 | 5382 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5383 | 0, &key); |
8e73f275 CM |
5384 | if (ret == -EAGAIN) |
5385 | goto again; | |
5386 | ||
76a05b35 | 5387 | if (ret < 0) { |
b3b4aa74 | 5388 | btrfs_release_path(path); |
76a05b35 CM |
5389 | goto done; |
5390 | } | |
5391 | ||
5cd57b2c | 5392 | if (!path->skip_locking) { |
bd681513 | 5393 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5394 | if (!ret) { |
5395 | btrfs_set_path_blocking(path); | |
bd681513 | 5396 | btrfs_tree_read_lock(next); |
bd681513 | 5397 | } |
31533fb2 | 5398 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5399 | } |
d97e63b6 | 5400 | } |
8e73f275 | 5401 | ret = 0; |
925baedd | 5402 | done: |
f7c79f30 | 5403 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5404 | path->leave_spinning = old_spinning; |
5405 | if (!old_spinning) | |
5406 | btrfs_set_path_blocking(path); | |
5407 | ||
5408 | return ret; | |
d97e63b6 | 5409 | } |
0b86a832 | 5410 | |
3f157a2f CM |
5411 | /* |
5412 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5413 | * searching until it gets past min_objectid or finds an item of 'type' | |
5414 | * | |
5415 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5416 | */ | |
0b86a832 CM |
5417 | int btrfs_previous_item(struct btrfs_root *root, |
5418 | struct btrfs_path *path, u64 min_objectid, | |
5419 | int type) | |
5420 | { | |
5421 | struct btrfs_key found_key; | |
5422 | struct extent_buffer *leaf; | |
e02119d5 | 5423 | u32 nritems; |
0b86a832 CM |
5424 | int ret; |
5425 | ||
d397712b | 5426 | while (1) { |
0b86a832 | 5427 | if (path->slots[0] == 0) { |
b4ce94de | 5428 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5429 | ret = btrfs_prev_leaf(root, path); |
5430 | if (ret != 0) | |
5431 | return ret; | |
5432 | } else { | |
5433 | path->slots[0]--; | |
5434 | } | |
5435 | leaf = path->nodes[0]; | |
e02119d5 CM |
5436 | nritems = btrfs_header_nritems(leaf); |
5437 | if (nritems == 0) | |
5438 | return 1; | |
5439 | if (path->slots[0] == nritems) | |
5440 | path->slots[0]--; | |
5441 | ||
0b86a832 | 5442 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5443 | if (found_key.objectid < min_objectid) |
5444 | break; | |
0a4eefbb YZ |
5445 | if (found_key.type == type) |
5446 | return 0; | |
e02119d5 CM |
5447 | if (found_key.objectid == min_objectid && |
5448 | found_key.type < type) | |
5449 | break; | |
0b86a832 CM |
5450 | } |
5451 | return 1; | |
5452 | } | |
ade2e0b3 WS |
5453 | |
5454 | /* | |
5455 | * search in extent tree to find a previous Metadata/Data extent item with | |
5456 | * min objecitd. | |
5457 | * | |
5458 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5459 | */ | |
5460 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5461 | struct btrfs_path *path, u64 min_objectid) | |
5462 | { | |
5463 | struct btrfs_key found_key; | |
5464 | struct extent_buffer *leaf; | |
5465 | u32 nritems; | |
5466 | int ret; | |
5467 | ||
5468 | while (1) { | |
5469 | if (path->slots[0] == 0) { | |
5470 | btrfs_set_path_blocking(path); | |
5471 | ret = btrfs_prev_leaf(root, path); | |
5472 | if (ret != 0) | |
5473 | return ret; | |
5474 | } else { | |
5475 | path->slots[0]--; | |
5476 | } | |
5477 | leaf = path->nodes[0]; | |
5478 | nritems = btrfs_header_nritems(leaf); | |
5479 | if (nritems == 0) | |
5480 | return 1; | |
5481 | if (path->slots[0] == nritems) | |
5482 | path->slots[0]--; | |
5483 | ||
5484 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5485 | if (found_key.objectid < min_objectid) | |
5486 | break; | |
5487 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5488 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5489 | return 0; | |
5490 | if (found_key.objectid == min_objectid && | |
5491 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5492 | break; | |
5493 | } | |
5494 | return 1; | |
5495 | } |