Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
5d4f98a2 YZ |
2 | /* |
3 | * Copyright (C) 2009 Oracle. All rights reserved. | |
5d4f98a2 YZ |
4 | */ |
5 | ||
6 | #include <linux/sched.h> | |
7 | #include <linux/pagemap.h> | |
8 | #include <linux/writeback.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/rbtree.h> | |
5a0e3ad6 | 11 | #include <linux/slab.h> |
726a3421 | 12 | #include <linux/error-injection.h> |
5d4f98a2 YZ |
13 | #include "ctree.h" |
14 | #include "disk-io.h" | |
15 | #include "transaction.h" | |
16 | #include "volumes.h" | |
17 | #include "locking.h" | |
18 | #include "btrfs_inode.h" | |
19 | #include "async-thread.h" | |
0af3d00b | 20 | #include "free-space-cache.h" |
62b99540 | 21 | #include "qgroup.h" |
cdccee99 | 22 | #include "print-tree.h" |
86736342 | 23 | #include "delalloc-space.h" |
aac0023c | 24 | #include "block-group.h" |
19b546d7 | 25 | #include "backref.h" |
e9a28dc5 | 26 | #include "misc.h" |
c2832898 | 27 | #include "subpage.h" |
7ae9bd18 | 28 | #include "zoned.h" |
26c2c454 | 29 | #include "inode-item.h" |
f1e5c618 | 30 | #include "space-info.h" |
c7f13d42 | 31 | #include "fs.h" |
07e81dc9 | 32 | #include "accessors.h" |
a0231804 | 33 | #include "extent-tree.h" |
45c40c8f | 34 | #include "root-tree.h" |
7c8ede16 | 35 | #include "file-item.h" |
67707479 | 36 | #include "relocation.h" |
7f0add25 | 37 | #include "super.h" |
103c1972 | 38 | #include "tree-checker.h" |
5d4f98a2 | 39 | |
0c891389 QW |
40 | /* |
41 | * Relocation overview | |
42 | * | |
43 | * [What does relocation do] | |
44 | * | |
45 | * The objective of relocation is to relocate all extents of the target block | |
46 | * group to other block groups. | |
47 | * This is utilized by resize (shrink only), profile converting, compacting | |
48 | * space, or balance routine to spread chunks over devices. | |
49 | * | |
50 | * Before | After | |
51 | * ------------------------------------------------------------------ | |
52 | * BG A: 10 data extents | BG A: deleted | |
53 | * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) | |
54 | * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) | |
55 | * | |
56 | * [How does relocation work] | |
57 | * | |
58 | * 1. Mark the target block group read-only | |
59 | * New extents won't be allocated from the target block group. | |
60 | * | |
61 | * 2.1 Record each extent in the target block group | |
62 | * To build a proper map of extents to be relocated. | |
63 | * | |
64 | * 2.2 Build data reloc tree and reloc trees | |
65 | * Data reloc tree will contain an inode, recording all newly relocated | |
66 | * data extents. | |
67 | * There will be only one data reloc tree for one data block group. | |
68 | * | |
69 | * Reloc tree will be a special snapshot of its source tree, containing | |
70 | * relocated tree blocks. | |
71 | * Each tree referring to a tree block in target block group will get its | |
72 | * reloc tree built. | |
73 | * | |
74 | * 2.3 Swap source tree with its corresponding reloc tree | |
75 | * Each involved tree only refers to new extents after swap. | |
76 | * | |
77 | * 3. Cleanup reloc trees and data reloc tree. | |
78 | * As old extents in the target block group are still referenced by reloc | |
79 | * trees, we need to clean them up before really freeing the target block | |
80 | * group. | |
81 | * | |
82 | * The main complexity is in steps 2.2 and 2.3. | |
83 | * | |
84 | * The entry point of relocation is relocate_block_group() function. | |
85 | */ | |
86 | ||
2a979612 | 87 | #define RELOCATION_RESERVED_NODES 256 |
5d4f98a2 YZ |
88 | /* |
89 | * map address of tree root to tree | |
90 | */ | |
91 | struct mapping_node { | |
e9a28dc5 QW |
92 | struct { |
93 | struct rb_node rb_node; | |
94 | u64 bytenr; | |
95 | }; /* Use rb_simle_node for search/insert */ | |
5d4f98a2 YZ |
96 | void *data; |
97 | }; | |
98 | ||
99 | struct mapping_tree { | |
100 | struct rb_root rb_root; | |
101 | spinlock_t lock; | |
102 | }; | |
103 | ||
104 | /* | |
105 | * present a tree block to process | |
106 | */ | |
107 | struct tree_block { | |
e9a28dc5 QW |
108 | struct { |
109 | struct rb_node rb_node; | |
110 | u64 bytenr; | |
111 | }; /* Use rb_simple_node for search/insert */ | |
f7ba2d37 | 112 | u64 owner; |
5d4f98a2 | 113 | struct btrfs_key key; |
a3bb700f DS |
114 | u8 level; |
115 | bool key_ready; | |
5d4f98a2 YZ |
116 | }; |
117 | ||
0257bb82 YZ |
118 | #define MAX_EXTENTS 128 |
119 | ||
120 | struct file_extent_cluster { | |
121 | u64 start; | |
122 | u64 end; | |
123 | u64 boundary[MAX_EXTENTS]; | |
124 | unsigned int nr; | |
2672a051 | 125 | u64 owning_root; |
0257bb82 YZ |
126 | }; |
127 | ||
8daf07cf DS |
128 | /* Stages of data relocation. */ |
129 | enum reloc_stage { | |
130 | MOVE_DATA_EXTENTS, | |
131 | UPDATE_DATA_PTRS | |
0257bb82 YZ |
132 | }; |
133 | ||
5d4f98a2 YZ |
134 | struct reloc_control { |
135 | /* block group to relocate */ | |
32da5386 | 136 | struct btrfs_block_group *block_group; |
5d4f98a2 YZ |
137 | /* extent tree */ |
138 | struct btrfs_root *extent_root; | |
139 | /* inode for moving data */ | |
140 | struct inode *data_inode; | |
3fd0a558 YZ |
141 | |
142 | struct btrfs_block_rsv *block_rsv; | |
143 | ||
a26195a5 | 144 | struct btrfs_backref_cache backref_cache; |
3fd0a558 YZ |
145 | |
146 | struct file_extent_cluster cluster; | |
5d4f98a2 YZ |
147 | /* tree blocks have been processed */ |
148 | struct extent_io_tree processed_blocks; | |
149 | /* map start of tree root to corresponding reloc tree */ | |
150 | struct mapping_tree reloc_root_tree; | |
151 | /* list of reloc trees */ | |
152 | struct list_head reloc_roots; | |
d2311e69 QW |
153 | /* list of subvolume trees that get relocated */ |
154 | struct list_head dirty_subvol_roots; | |
3fd0a558 YZ |
155 | /* size of metadata reservation for merging reloc trees */ |
156 | u64 merging_rsv_size; | |
157 | /* size of relocated tree nodes */ | |
158 | u64 nodes_relocated; | |
0647bf56 WS |
159 | /* reserved size for block group relocation*/ |
160 | u64 reserved_bytes; | |
3fd0a558 | 161 | |
5d4f98a2 YZ |
162 | u64 search_start; |
163 | u64 extents_found; | |
3fd0a558 | 164 | |
8daf07cf | 165 | enum reloc_stage stage; |
d23d42e3 DS |
166 | bool create_reloc_tree; |
167 | bool merge_reloc_tree; | |
168 | bool found_file_extent; | |
5d4f98a2 YZ |
169 | }; |
170 | ||
9569cc20 | 171 | static void mark_block_processed(struct reloc_control *rc, |
a26195a5 | 172 | struct btrfs_backref_node *node) |
9569cc20 QW |
173 | { |
174 | u32 blocksize; | |
175 | ||
176 | if (node->level == 0 || | |
177 | in_range(node->bytenr, rc->block_group->start, | |
178 | rc->block_group->length)) { | |
179 | blocksize = rc->extent_root->fs_info->nodesize; | |
0acd32c2 | 180 | set_extent_bit(&rc->processed_blocks, node->bytenr, |
1d126800 | 181 | node->bytenr + blocksize - 1, EXTENT_DIRTY, NULL); |
9569cc20 QW |
182 | } |
183 | node->processed = 1; | |
184 | } | |
185 | ||
5d4f98a2 YZ |
186 | /* |
187 | * walk up backref nodes until reach node presents tree root | |
188 | */ | |
a26195a5 QW |
189 | static struct btrfs_backref_node *walk_up_backref( |
190 | struct btrfs_backref_node *node, | |
191 | struct btrfs_backref_edge *edges[], int *index) | |
5d4f98a2 | 192 | { |
a26195a5 | 193 | struct btrfs_backref_edge *edge; |
5d4f98a2 YZ |
194 | int idx = *index; |
195 | ||
196 | while (!list_empty(&node->upper)) { | |
197 | edge = list_entry(node->upper.next, | |
a26195a5 | 198 | struct btrfs_backref_edge, list[LOWER]); |
5d4f98a2 YZ |
199 | edges[idx++] = edge; |
200 | node = edge->node[UPPER]; | |
201 | } | |
3fd0a558 | 202 | BUG_ON(node->detached); |
5d4f98a2 YZ |
203 | *index = idx; |
204 | return node; | |
205 | } | |
206 | ||
207 | /* | |
208 | * walk down backref nodes to find start of next reference path | |
209 | */ | |
a26195a5 QW |
210 | static struct btrfs_backref_node *walk_down_backref( |
211 | struct btrfs_backref_edge *edges[], int *index) | |
5d4f98a2 | 212 | { |
a26195a5 QW |
213 | struct btrfs_backref_edge *edge; |
214 | struct btrfs_backref_node *lower; | |
5d4f98a2 YZ |
215 | int idx = *index; |
216 | ||
217 | while (idx > 0) { | |
218 | edge = edges[idx - 1]; | |
219 | lower = edge->node[LOWER]; | |
220 | if (list_is_last(&edge->list[LOWER], &lower->upper)) { | |
221 | idx--; | |
222 | continue; | |
223 | } | |
224 | edge = list_entry(edge->list[LOWER].next, | |
a26195a5 | 225 | struct btrfs_backref_edge, list[LOWER]); |
5d4f98a2 YZ |
226 | edges[idx - 1] = edge; |
227 | *index = idx; | |
228 | return edge->node[UPPER]; | |
229 | } | |
230 | *index = 0; | |
231 | return NULL; | |
232 | } | |
5d4f98a2 | 233 | |
a26195a5 QW |
234 | static void update_backref_node(struct btrfs_backref_cache *cache, |
235 | struct btrfs_backref_node *node, u64 bytenr) | |
3fd0a558 YZ |
236 | { |
237 | struct rb_node *rb_node; | |
238 | rb_erase(&node->rb_node, &cache->rb_root); | |
239 | node->bytenr = bytenr; | |
e9a28dc5 | 240 | rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); |
43c04fb1 | 241 | if (rb_node) |
982c92cb | 242 | btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); |
3fd0a558 YZ |
243 | } |
244 | ||
245 | /* | |
246 | * update backref cache after a transaction commit | |
247 | */ | |
248 | static int update_backref_cache(struct btrfs_trans_handle *trans, | |
a26195a5 | 249 | struct btrfs_backref_cache *cache) |
3fd0a558 | 250 | { |
a26195a5 | 251 | struct btrfs_backref_node *node; |
3fd0a558 YZ |
252 | int level = 0; |
253 | ||
254 | if (cache->last_trans == 0) { | |
255 | cache->last_trans = trans->transid; | |
256 | return 0; | |
257 | } | |
258 | ||
259 | if (cache->last_trans == trans->transid) | |
260 | return 0; | |
261 | ||
262 | /* | |
263 | * detached nodes are used to avoid unnecessary backref | |
264 | * lookup. transaction commit changes the extent tree. | |
265 | * so the detached nodes are no longer useful. | |
266 | */ | |
267 | while (!list_empty(&cache->detached)) { | |
268 | node = list_entry(cache->detached.next, | |
a26195a5 | 269 | struct btrfs_backref_node, list); |
023acb07 | 270 | btrfs_backref_cleanup_node(cache, node); |
3fd0a558 YZ |
271 | } |
272 | ||
273 | while (!list_empty(&cache->changed)) { | |
274 | node = list_entry(cache->changed.next, | |
a26195a5 | 275 | struct btrfs_backref_node, list); |
3fd0a558 YZ |
276 | list_del_init(&node->list); |
277 | BUG_ON(node->pending); | |
278 | update_backref_node(cache, node, node->new_bytenr); | |
279 | } | |
280 | ||
281 | /* | |
282 | * some nodes can be left in the pending list if there were | |
283 | * errors during processing the pending nodes. | |
284 | */ | |
285 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
286 | list_for_each_entry(node, &cache->pending[level], list) { | |
287 | BUG_ON(!node->pending); | |
288 | if (node->bytenr == node->new_bytenr) | |
289 | continue; | |
290 | update_backref_node(cache, node, node->new_bytenr); | |
291 | } | |
292 | } | |
293 | ||
294 | cache->last_trans = 0; | |
295 | return 1; | |
296 | } | |
297 | ||
ab7c8bbf | 298 | static bool reloc_root_is_dead(const struct btrfs_root *root) |
6282675e QW |
299 | { |
300 | /* | |
301 | * Pair with set_bit/clear_bit in clean_dirty_subvols and | |
302 | * btrfs_update_reloc_root. We need to see the updated bit before | |
303 | * trying to access reloc_root | |
304 | */ | |
305 | smp_rmb(); | |
306 | if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) | |
307 | return true; | |
308 | return false; | |
309 | } | |
310 | ||
311 | /* | |
312 | * Check if this subvolume tree has valid reloc tree. | |
313 | * | |
314 | * Reloc tree after swap is considered dead, thus not considered as valid. | |
315 | * This is enough for most callers, as they don't distinguish dead reloc root | |
55465730 QW |
316 | * from no reloc root. But btrfs_should_ignore_reloc_root() below is a |
317 | * special case. | |
6282675e | 318 | */ |
ab7c8bbf | 319 | static bool have_reloc_root(const struct btrfs_root *root) |
6282675e QW |
320 | { |
321 | if (reloc_root_is_dead(root)) | |
322 | return false; | |
323 | if (!root->reloc_root) | |
324 | return false; | |
325 | return true; | |
326 | } | |
f2a97a9d | 327 | |
ab7c8bbf | 328 | bool btrfs_should_ignore_reloc_root(const struct btrfs_root *root) |
3fd0a558 YZ |
329 | { |
330 | struct btrfs_root *reloc_root; | |
331 | ||
92a7cc42 | 332 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
32f2abca | 333 | return false; |
3fd0a558 | 334 | |
6282675e QW |
335 | /* This root has been merged with its reloc tree, we can ignore it */ |
336 | if (reloc_root_is_dead(root)) | |
32f2abca | 337 | return true; |
6282675e | 338 | |
3fd0a558 YZ |
339 | reloc_root = root->reloc_root; |
340 | if (!reloc_root) | |
32f2abca | 341 | return false; |
3fd0a558 | 342 | |
4d4225fc JB |
343 | if (btrfs_header_generation(reloc_root->commit_root) == |
344 | root->fs_info->running_transaction->transid) | |
32f2abca | 345 | return false; |
3fd0a558 | 346 | /* |
32f2abca DS |
347 | * If there is reloc tree and it was created in previous transaction |
348 | * backref lookup can find the reloc tree, so backref node for the fs | |
349 | * tree root is useless for relocation. | |
3fd0a558 | 350 | */ |
32f2abca | 351 | return true; |
3fd0a558 | 352 | } |
55465730 | 353 | |
5d4f98a2 YZ |
354 | /* |
355 | * find reloc tree by address of tree root | |
356 | */ | |
2433bea5 | 357 | struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) |
5d4f98a2 | 358 | { |
2433bea5 | 359 | struct reloc_control *rc = fs_info->reloc_ctl; |
5d4f98a2 YZ |
360 | struct rb_node *rb_node; |
361 | struct mapping_node *node; | |
362 | struct btrfs_root *root = NULL; | |
363 | ||
2433bea5 | 364 | ASSERT(rc); |
5d4f98a2 | 365 | spin_lock(&rc->reloc_root_tree.lock); |
e9a28dc5 | 366 | rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); |
5d4f98a2 YZ |
367 | if (rb_node) { |
368 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
0d031dc4 | 369 | root = node->data; |
5d4f98a2 YZ |
370 | } |
371 | spin_unlock(&rc->reloc_root_tree.lock); | |
00246528 | 372 | return btrfs_grab_root(root); |
5d4f98a2 YZ |
373 | } |
374 | ||
29db137b QW |
375 | /* |
376 | * For useless nodes, do two major clean ups: | |
377 | * | |
378 | * - Cleanup the children edges and nodes | |
379 | * If child node is also orphan (no parent) during cleanup, then the child | |
380 | * node will also be cleaned up. | |
381 | * | |
382 | * - Freeing up leaves (level 0), keeps nodes detached | |
383 | * For nodes, the node is still cached as "detached" | |
384 | * | |
385 | * Return false if @node is not in the @useless_nodes list. | |
386 | * Return true if @node is in the @useless_nodes list. | |
387 | */ | |
388 | static bool handle_useless_nodes(struct reloc_control *rc, | |
a26195a5 | 389 | struct btrfs_backref_node *node) |
29db137b | 390 | { |
a26195a5 | 391 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
29db137b QW |
392 | struct list_head *useless_node = &cache->useless_node; |
393 | bool ret = false; | |
394 | ||
395 | while (!list_empty(useless_node)) { | |
a26195a5 | 396 | struct btrfs_backref_node *cur; |
29db137b | 397 | |
a26195a5 | 398 | cur = list_first_entry(useless_node, struct btrfs_backref_node, |
29db137b QW |
399 | list); |
400 | list_del_init(&cur->list); | |
401 | ||
402 | /* Only tree root nodes can be added to @useless_nodes */ | |
403 | ASSERT(list_empty(&cur->upper)); | |
404 | ||
405 | if (cur == node) | |
406 | ret = true; | |
407 | ||
408 | /* The node is the lowest node */ | |
409 | if (cur->lowest) { | |
410 | list_del_init(&cur->lower); | |
411 | cur->lowest = 0; | |
412 | } | |
413 | ||
414 | /* Cleanup the lower edges */ | |
415 | while (!list_empty(&cur->lower)) { | |
a26195a5 QW |
416 | struct btrfs_backref_edge *edge; |
417 | struct btrfs_backref_node *lower; | |
29db137b QW |
418 | |
419 | edge = list_entry(cur->lower.next, | |
a26195a5 | 420 | struct btrfs_backref_edge, list[UPPER]); |
29db137b QW |
421 | list_del(&edge->list[UPPER]); |
422 | list_del(&edge->list[LOWER]); | |
423 | lower = edge->node[LOWER]; | |
741188d3 | 424 | btrfs_backref_free_edge(cache, edge); |
29db137b QW |
425 | |
426 | /* Child node is also orphan, queue for cleanup */ | |
427 | if (list_empty(&lower->upper)) | |
428 | list_add(&lower->list, useless_node); | |
429 | } | |
430 | /* Mark this block processed for relocation */ | |
431 | mark_block_processed(rc, cur); | |
432 | ||
433 | /* | |
434 | * Backref nodes for tree leaves are deleted from the cache. | |
435 | * Backref nodes for upper level tree blocks are left in the | |
436 | * cache to avoid unnecessary backref lookup. | |
437 | */ | |
438 | if (cur->level > 0) { | |
439 | list_add(&cur->list, &cache->detached); | |
440 | cur->detached = 1; | |
441 | } else { | |
442 | rb_erase(&cur->rb_node, &cache->rb_root); | |
741188d3 | 443 | btrfs_backref_free_node(cache, cur); |
29db137b QW |
444 | } |
445 | } | |
446 | return ret; | |
447 | } | |
448 | ||
e7d571c7 QW |
449 | /* |
450 | * Build backref tree for a given tree block. Root of the backref tree | |
451 | * corresponds the tree block, leaves of the backref tree correspond roots of | |
452 | * b-trees that reference the tree block. | |
453 | * | |
454 | * The basic idea of this function is check backrefs of a given block to find | |
455 | * upper level blocks that reference the block, and then check backrefs of | |
456 | * these upper level blocks recursively. The recursion stops when tree root is | |
457 | * reached or backrefs for the block is cached. | |
458 | * | |
459 | * NOTE: if we find that backrefs for a block are cached, we know backrefs for | |
460 | * all upper level blocks that directly/indirectly reference the block are also | |
461 | * cached. | |
462 | */ | |
a26195a5 | 463 | static noinline_for_stack struct btrfs_backref_node *build_backref_tree( |
eb96e221 | 464 | struct btrfs_trans_handle *trans, |
e7d571c7 QW |
465 | struct reloc_control *rc, struct btrfs_key *node_key, |
466 | int level, u64 bytenr) | |
467 | { | |
468 | struct btrfs_backref_iter *iter; | |
a26195a5 | 469 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
e7d571c7 QW |
470 | /* For searching parent of TREE_BLOCK_REF */ |
471 | struct btrfs_path *path; | |
a26195a5 | 472 | struct btrfs_backref_node *cur; |
a26195a5 QW |
473 | struct btrfs_backref_node *node = NULL; |
474 | struct btrfs_backref_edge *edge; | |
e7d571c7 | 475 | int ret; |
5d4f98a2 | 476 | |
d68194b2 | 477 | iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info); |
e7d571c7 QW |
478 | if (!iter) |
479 | return ERR_PTR(-ENOMEM); | |
480 | path = btrfs_alloc_path(); | |
481 | if (!path) { | |
2daca1e4 | 482 | ret = -ENOMEM; |
e7d571c7 QW |
483 | goto out; |
484 | } | |
485 | ||
b1818dab | 486 | node = btrfs_backref_alloc_node(cache, bytenr, level); |
e7d571c7 | 487 | if (!node) { |
2daca1e4 | 488 | ret = -ENOMEM; |
e7d571c7 | 489 | goto out; |
5d4f98a2 YZ |
490 | } |
491 | ||
e7d571c7 QW |
492 | node->lowest = 1; |
493 | cur = node; | |
494 | ||
495 | /* Breadth-first search to build backref cache */ | |
496 | do { | |
eb96e221 FM |
497 | ret = btrfs_backref_add_tree_node(trans, cache, path, iter, |
498 | node_key, cur); | |
2daca1e4 | 499 | if (ret < 0) |
e7d571c7 | 500 | goto out; |
2daca1e4 | 501 | |
e7d571c7 | 502 | edge = list_first_entry_or_null(&cache->pending_edge, |
a26195a5 | 503 | struct btrfs_backref_edge, list[UPPER]); |
e7d571c7 QW |
504 | /* |
505 | * The pending list isn't empty, take the first block to | |
506 | * process | |
507 | */ | |
508 | if (edge) { | |
509 | list_del_init(&edge->list[UPPER]); | |
510 | cur = edge->node[UPPER]; | |
511 | } | |
512 | } while (edge); | |
513 | ||
1f872924 | 514 | /* Finish the upper linkage of newly added edges/nodes */ |
fc997ed0 | 515 | ret = btrfs_backref_finish_upper_links(cache, node); |
2daca1e4 | 516 | if (ret < 0) |
1f872924 | 517 | goto out; |
5d4f98a2 | 518 | |
29db137b QW |
519 | if (handle_useless_nodes(rc, node)) |
520 | node = NULL; | |
5d4f98a2 | 521 | out: |
56430c14 DS |
522 | btrfs_free_path(iter->path); |
523 | kfree(iter); | |
71f572a9 | 524 | btrfs_free_path(path); |
2daca1e4 | 525 | if (ret) { |
1b23ea18 | 526 | btrfs_backref_error_cleanup(cache, node); |
2daca1e4 | 527 | return ERR_PTR(ret); |
5d4f98a2 | 528 | } |
75bfb9af | 529 | ASSERT(!node || !node->detached); |
84780289 QW |
530 | ASSERT(list_empty(&cache->useless_node) && |
531 | list_empty(&cache->pending_edge)); | |
5d4f98a2 YZ |
532 | return node; |
533 | } | |
534 | ||
3fd0a558 YZ |
535 | /* |
536 | * helper to add backref node for the newly created snapshot. | |
537 | * the backref node is created by cloning backref node that | |
538 | * corresponds to root of source tree | |
539 | */ | |
540 | static int clone_backref_node(struct btrfs_trans_handle *trans, | |
541 | struct reloc_control *rc, | |
ab7c8bbf | 542 | const struct btrfs_root *src, |
3fd0a558 YZ |
543 | struct btrfs_root *dest) |
544 | { | |
545 | struct btrfs_root *reloc_root = src->reloc_root; | |
a26195a5 QW |
546 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
547 | struct btrfs_backref_node *node = NULL; | |
548 | struct btrfs_backref_node *new_node; | |
549 | struct btrfs_backref_edge *edge; | |
550 | struct btrfs_backref_edge *new_edge; | |
3fd0a558 YZ |
551 | struct rb_node *rb_node; |
552 | ||
553 | if (cache->last_trans > 0) | |
554 | update_backref_cache(trans, cache); | |
555 | ||
e9a28dc5 | 556 | rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); |
3fd0a558 | 557 | if (rb_node) { |
a26195a5 | 558 | node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); |
3fd0a558 YZ |
559 | if (node->detached) |
560 | node = NULL; | |
561 | else | |
562 | BUG_ON(node->new_bytenr != reloc_root->node->start); | |
563 | } | |
564 | ||
565 | if (!node) { | |
e9a28dc5 QW |
566 | rb_node = rb_simple_search(&cache->rb_root, |
567 | reloc_root->commit_root->start); | |
3fd0a558 | 568 | if (rb_node) { |
a26195a5 | 569 | node = rb_entry(rb_node, struct btrfs_backref_node, |
3fd0a558 YZ |
570 | rb_node); |
571 | BUG_ON(node->detached); | |
572 | } | |
573 | } | |
574 | ||
575 | if (!node) | |
576 | return 0; | |
577 | ||
b1818dab QW |
578 | new_node = btrfs_backref_alloc_node(cache, dest->node->start, |
579 | node->level); | |
3fd0a558 YZ |
580 | if (!new_node) |
581 | return -ENOMEM; | |
582 | ||
3fd0a558 | 583 | new_node->lowest = node->lowest; |
6848ad64 | 584 | new_node->checked = 1; |
00246528 | 585 | new_node->root = btrfs_grab_root(dest); |
0b530bc5 | 586 | ASSERT(new_node->root); |
3fd0a558 YZ |
587 | |
588 | if (!node->lowest) { | |
589 | list_for_each_entry(edge, &node->lower, list[UPPER]) { | |
47254d07 | 590 | new_edge = btrfs_backref_alloc_edge(cache); |
3fd0a558 YZ |
591 | if (!new_edge) |
592 | goto fail; | |
593 | ||
f39911e5 QW |
594 | btrfs_backref_link_edge(new_edge, edge->node[LOWER], |
595 | new_node, LINK_UPPER); | |
3fd0a558 | 596 | } |
76b9e23d MX |
597 | } else { |
598 | list_add_tail(&new_node->lower, &cache->leaves); | |
3fd0a558 YZ |
599 | } |
600 | ||
e9a28dc5 QW |
601 | rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, |
602 | &new_node->rb_node); | |
43c04fb1 | 603 | if (rb_node) |
982c92cb | 604 | btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); |
3fd0a558 YZ |
605 | |
606 | if (!new_node->lowest) { | |
607 | list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { | |
608 | list_add_tail(&new_edge->list[LOWER], | |
609 | &new_edge->node[LOWER]->upper); | |
610 | } | |
611 | } | |
612 | return 0; | |
613 | fail: | |
614 | while (!list_empty(&new_node->lower)) { | |
615 | new_edge = list_entry(new_node->lower.next, | |
a26195a5 | 616 | struct btrfs_backref_edge, list[UPPER]); |
3fd0a558 | 617 | list_del(&new_edge->list[UPPER]); |
741188d3 | 618 | btrfs_backref_free_edge(cache, new_edge); |
3fd0a558 | 619 | } |
741188d3 | 620 | btrfs_backref_free_node(cache, new_node); |
3fd0a558 YZ |
621 | return -ENOMEM; |
622 | } | |
623 | ||
5d4f98a2 YZ |
624 | /* |
625 | * helper to add 'address of tree root -> reloc tree' mapping | |
626 | */ | |
203f6a87 | 627 | static int __add_reloc_root(struct btrfs_root *root) |
5d4f98a2 | 628 | { |
0b246afa | 629 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
630 | struct rb_node *rb_node; |
631 | struct mapping_node *node; | |
0b246afa | 632 | struct reloc_control *rc = fs_info->reloc_ctl; |
5d4f98a2 YZ |
633 | |
634 | node = kmalloc(sizeof(*node), GFP_NOFS); | |
ffd7b339 JM |
635 | if (!node) |
636 | return -ENOMEM; | |
5d4f98a2 | 637 | |
ea287ab1 | 638 | node->bytenr = root->commit_root->start; |
5d4f98a2 YZ |
639 | node->data = root; |
640 | ||
641 | spin_lock(&rc->reloc_root_tree.lock); | |
e9a28dc5 QW |
642 | rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, |
643 | node->bytenr, &node->rb_node); | |
5d4f98a2 | 644 | spin_unlock(&rc->reloc_root_tree.lock); |
ffd7b339 | 645 | if (rb_node) { |
57a304cf | 646 | btrfs_err(fs_info, |
5d163e0e JM |
647 | "Duplicate root found for start=%llu while inserting into relocation tree", |
648 | node->bytenr); | |
57a304cf | 649 | return -EEXIST; |
ffd7b339 | 650 | } |
5d4f98a2 YZ |
651 | |
652 | list_add_tail(&root->root_list, &rc->reloc_roots); | |
653 | return 0; | |
654 | } | |
655 | ||
656 | /* | |
c974c464 | 657 | * helper to delete the 'address of tree root -> reloc tree' |
5d4f98a2 YZ |
658 | * mapping |
659 | */ | |
c974c464 | 660 | static void __del_reloc_root(struct btrfs_root *root) |
5d4f98a2 | 661 | { |
0b246afa | 662 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
663 | struct rb_node *rb_node; |
664 | struct mapping_node *node = NULL; | |
0b246afa | 665 | struct reloc_control *rc = fs_info->reloc_ctl; |
f44deb74 | 666 | bool put_ref = false; |
5d4f98a2 | 667 | |
65c6e82b | 668 | if (rc && root->node) { |
389305b2 | 669 | spin_lock(&rc->reloc_root_tree.lock); |
e9a28dc5 QW |
670 | rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, |
671 | root->commit_root->start); | |
389305b2 QW |
672 | if (rb_node) { |
673 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
674 | rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); | |
ea287ab1 | 675 | RB_CLEAR_NODE(&node->rb_node); |
389305b2 QW |
676 | } |
677 | spin_unlock(&rc->reloc_root_tree.lock); | |
c78a10ae | 678 | ASSERT(!node || (struct btrfs_root *)node->data == root); |
5d4f98a2 | 679 | } |
5d4f98a2 | 680 | |
f44deb74 JB |
681 | /* |
682 | * We only put the reloc root here if it's on the list. There's a lot | |
683 | * of places where the pattern is to splice the rc->reloc_roots, process | |
684 | * the reloc roots, and then add the reloc root back onto | |
685 | * rc->reloc_roots. If we call __del_reloc_root while it's off of the | |
686 | * list we don't want the reference being dropped, because the guy | |
687 | * messing with the list is in charge of the reference. | |
688 | */ | |
0b246afa | 689 | spin_lock(&fs_info->trans_lock); |
f44deb74 JB |
690 | if (!list_empty(&root->root_list)) { |
691 | put_ref = true; | |
692 | list_del_init(&root->root_list); | |
693 | } | |
0b246afa | 694 | spin_unlock(&fs_info->trans_lock); |
f44deb74 JB |
695 | if (put_ref) |
696 | btrfs_put_root(root); | |
c974c464 WS |
697 | kfree(node); |
698 | } | |
699 | ||
700 | /* | |
701 | * helper to update the 'address of tree root -> reloc tree' | |
702 | * mapping | |
703 | */ | |
ea287ab1 | 704 | static int __update_reloc_root(struct btrfs_root *root) |
c974c464 | 705 | { |
0b246afa | 706 | struct btrfs_fs_info *fs_info = root->fs_info; |
c974c464 WS |
707 | struct rb_node *rb_node; |
708 | struct mapping_node *node = NULL; | |
0b246afa | 709 | struct reloc_control *rc = fs_info->reloc_ctl; |
c974c464 WS |
710 | |
711 | spin_lock(&rc->reloc_root_tree.lock); | |
e9a28dc5 QW |
712 | rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, |
713 | root->commit_root->start); | |
c974c464 WS |
714 | if (rb_node) { |
715 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
716 | rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); | |
5d4f98a2 | 717 | } |
c974c464 WS |
718 | spin_unlock(&rc->reloc_root_tree.lock); |
719 | ||
720 | if (!node) | |
721 | return 0; | |
722 | BUG_ON((struct btrfs_root *)node->data != root); | |
723 | ||
724 | spin_lock(&rc->reloc_root_tree.lock); | |
ea287ab1 | 725 | node->bytenr = root->node->start; |
e9a28dc5 QW |
726 | rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, |
727 | node->bytenr, &node->rb_node); | |
c974c464 WS |
728 | spin_unlock(&rc->reloc_root_tree.lock); |
729 | if (rb_node) | |
982c92cb | 730 | btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); |
5d4f98a2 YZ |
731 | return 0; |
732 | } | |
733 | ||
3fd0a558 YZ |
734 | static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, |
735 | struct btrfs_root *root, u64 objectid) | |
5d4f98a2 | 736 | { |
0b246afa | 737 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
738 | struct btrfs_root *reloc_root; |
739 | struct extent_buffer *eb; | |
740 | struct btrfs_root_item *root_item; | |
741 | struct btrfs_key root_key; | |
84c50ba5 JB |
742 | int ret = 0; |
743 | bool must_abort = false; | |
5d4f98a2 | 744 | |
5d4f98a2 | 745 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); |
84c50ba5 JB |
746 | if (!root_item) |
747 | return ERR_PTR(-ENOMEM); | |
5d4f98a2 YZ |
748 | |
749 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
750 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
3fd0a558 | 751 | root_key.offset = objectid; |
5d4f98a2 | 752 | |
e094f480 | 753 | if (btrfs_root_id(root) == objectid) { |
054570a1 FM |
754 | u64 commit_root_gen; |
755 | ||
3fd0a558 YZ |
756 | /* called by btrfs_init_reloc_root */ |
757 | ret = btrfs_copy_root(trans, root, root->commit_root, &eb, | |
758 | BTRFS_TREE_RELOC_OBJECTID); | |
84c50ba5 JB |
759 | if (ret) |
760 | goto fail; | |
761 | ||
054570a1 FM |
762 | /* |
763 | * Set the last_snapshot field to the generation of the commit | |
764 | * root - like this ctree.c:btrfs_block_can_be_shared() behaves | |
765 | * correctly (returns true) when the relocation root is created | |
766 | * either inside the critical section of a transaction commit | |
767 | * (through transaction.c:qgroup_account_snapshot()) and when | |
768 | * it's created before the transaction commit is started. | |
769 | */ | |
770 | commit_root_gen = btrfs_header_generation(root->commit_root); | |
771 | btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); | |
3fd0a558 YZ |
772 | } else { |
773 | /* | |
774 | * called by btrfs_reloc_post_snapshot_hook. | |
775 | * the source tree is a reloc tree, all tree blocks | |
776 | * modified after it was created have RELOC flag | |
777 | * set in their headers. so it's OK to not update | |
778 | * the 'last_snapshot'. | |
779 | */ | |
780 | ret = btrfs_copy_root(trans, root, root->node, &eb, | |
781 | BTRFS_TREE_RELOC_OBJECTID); | |
84c50ba5 JB |
782 | if (ret) |
783 | goto fail; | |
3fd0a558 | 784 | } |
5d4f98a2 | 785 | |
84c50ba5 JB |
786 | /* |
787 | * We have changed references at this point, we must abort the | |
788 | * transaction if anything fails. | |
789 | */ | |
790 | must_abort = true; | |
791 | ||
5d4f98a2 | 792 | memcpy(root_item, &root->root_item, sizeof(*root_item)); |
5d4f98a2 YZ |
793 | btrfs_set_root_bytenr(root_item, eb->start); |
794 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
795 | btrfs_set_root_generation(root_item, trans->transid); | |
3fd0a558 | 796 | |
e094f480 | 797 | if (btrfs_root_id(root) == objectid) { |
3fd0a558 YZ |
798 | btrfs_set_root_refs(root_item, 0); |
799 | memset(&root_item->drop_progress, 0, | |
800 | sizeof(struct btrfs_disk_key)); | |
c8422684 | 801 | btrfs_set_root_drop_level(root_item, 0); |
3fd0a558 | 802 | } |
5d4f98a2 YZ |
803 | |
804 | btrfs_tree_unlock(eb); | |
805 | free_extent_buffer(eb); | |
806 | ||
0b246afa | 807 | ret = btrfs_insert_root(trans, fs_info->tree_root, |
5d4f98a2 | 808 | &root_key, root_item); |
84c50ba5 JB |
809 | if (ret) |
810 | goto fail; | |
811 | ||
5d4f98a2 YZ |
812 | kfree(root_item); |
813 | ||
3dbf1738 | 814 | reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); |
84c50ba5 JB |
815 | if (IS_ERR(reloc_root)) { |
816 | ret = PTR_ERR(reloc_root); | |
817 | goto abort; | |
818 | } | |
92a7cc42 | 819 | set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); |
5d4f98a2 | 820 | reloc_root->last_trans = trans->transid; |
3fd0a558 | 821 | return reloc_root; |
84c50ba5 JB |
822 | fail: |
823 | kfree(root_item); | |
824 | abort: | |
825 | if (must_abort) | |
826 | btrfs_abort_transaction(trans, ret); | |
827 | return ERR_PTR(ret); | |
3fd0a558 YZ |
828 | } |
829 | ||
830 | /* | |
831 | * create reloc tree for a given fs tree. reloc tree is just a | |
832 | * snapshot of the fs tree with special root objectid. | |
f44deb74 JB |
833 | * |
834 | * The reloc_root comes out of here with two references, one for | |
835 | * root->reloc_root, and another for being on the rc->reloc_roots list. | |
3fd0a558 YZ |
836 | */ |
837 | int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, | |
838 | struct btrfs_root *root) | |
839 | { | |
0b246afa | 840 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 | 841 | struct btrfs_root *reloc_root; |
0b246afa | 842 | struct reloc_control *rc = fs_info->reloc_ctl; |
20dd2cbf | 843 | struct btrfs_block_rsv *rsv; |
3fd0a558 | 844 | int clear_rsv = 0; |
ffd7b339 | 845 | int ret; |
3fd0a558 | 846 | |
aec7db3b | 847 | if (!rc) |
2abc726a JB |
848 | return 0; |
849 | ||
1fac4a54 QW |
850 | /* |
851 | * The subvolume has reloc tree but the swap is finished, no need to | |
852 | * create/update the dead reloc tree | |
853 | */ | |
6282675e | 854 | if (reloc_root_is_dead(root)) |
1fac4a54 QW |
855 | return 0; |
856 | ||
aec7db3b JB |
857 | /* |
858 | * This is subtle but important. We do not do | |
859 | * record_root_in_transaction for reloc roots, instead we record their | |
860 | * corresponding fs root, and then here we update the last trans for the | |
861 | * reloc root. This means that we have to do this for the entire life | |
862 | * of the reloc root, regardless of which stage of the relocation we are | |
863 | * in. | |
864 | */ | |
3fd0a558 YZ |
865 | if (root->reloc_root) { |
866 | reloc_root = root->reloc_root; | |
867 | reloc_root->last_trans = trans->transid; | |
868 | return 0; | |
869 | } | |
870 | ||
aec7db3b JB |
871 | /* |
872 | * We are merging reloc roots, we do not need new reloc trees. Also | |
873 | * reloc trees never need their own reloc tree. | |
874 | */ | |
e094f480 | 875 | if (!rc->create_reloc_tree || btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID) |
aec7db3b JB |
876 | return 0; |
877 | ||
20dd2cbf MX |
878 | if (!trans->reloc_reserved) { |
879 | rsv = trans->block_rsv; | |
3fd0a558 YZ |
880 | trans->block_rsv = rc->block_rsv; |
881 | clear_rsv = 1; | |
882 | } | |
e094f480 | 883 | reloc_root = create_reloc_root(trans, root, btrfs_root_id(root)); |
3fd0a558 | 884 | if (clear_rsv) |
20dd2cbf | 885 | trans->block_rsv = rsv; |
00bb36a0 JB |
886 | if (IS_ERR(reloc_root)) |
887 | return PTR_ERR(reloc_root); | |
5d4f98a2 | 888 | |
ffd7b339 | 889 | ret = __add_reloc_root(reloc_root); |
57a304cf | 890 | ASSERT(ret != -EEXIST); |
00bb36a0 JB |
891 | if (ret) { |
892 | /* Pairs with create_reloc_root */ | |
893 | btrfs_put_root(reloc_root); | |
894 | return ret; | |
895 | } | |
f44deb74 | 896 | root->reloc_root = btrfs_grab_root(reloc_root); |
5d4f98a2 YZ |
897 | return 0; |
898 | } | |
899 | ||
900 | /* | |
901 | * update root item of reloc tree | |
902 | */ | |
903 | int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, | |
904 | struct btrfs_root *root) | |
905 | { | |
0b246afa | 906 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
907 | struct btrfs_root *reloc_root; |
908 | struct btrfs_root_item *root_item; | |
5d4f98a2 YZ |
909 | int ret; |
910 | ||
6282675e | 911 | if (!have_reloc_root(root)) |
592fbcd5 | 912 | return 0; |
5d4f98a2 YZ |
913 | |
914 | reloc_root = root->reloc_root; | |
915 | root_item = &reloc_root->root_item; | |
916 | ||
f44deb74 JB |
917 | /* |
918 | * We are probably ok here, but __del_reloc_root() will drop its ref of | |
919 | * the root. We have the ref for root->reloc_root, but just in case | |
920 | * hold it while we update the reloc root. | |
921 | */ | |
922 | btrfs_grab_root(reloc_root); | |
923 | ||
d2311e69 | 924 | /* root->reloc_root will stay until current relocation finished */ |
0b246afa | 925 | if (fs_info->reloc_ctl->merge_reloc_tree && |
3fd0a558 | 926 | btrfs_root_refs(root_item) == 0) { |
d2311e69 | 927 | set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); |
6282675e QW |
928 | /* |
929 | * Mark the tree as dead before we change reloc_root so | |
930 | * have_reloc_root will not touch it from now on. | |
931 | */ | |
932 | smp_wmb(); | |
c974c464 | 933 | __del_reloc_root(reloc_root); |
5d4f98a2 YZ |
934 | } |
935 | ||
5d4f98a2 | 936 | if (reloc_root->commit_root != reloc_root->node) { |
ea287ab1 | 937 | __update_reloc_root(reloc_root); |
5d4f98a2 YZ |
938 | btrfs_set_root_node(root_item, reloc_root->node); |
939 | free_extent_buffer(reloc_root->commit_root); | |
940 | reloc_root->commit_root = btrfs_root_node(reloc_root); | |
941 | } | |
942 | ||
0b246afa | 943 | ret = btrfs_update_root(trans, fs_info->tree_root, |
5d4f98a2 | 944 | &reloc_root->root_key, root_item); |
f44deb74 | 945 | btrfs_put_root(reloc_root); |
592fbcd5 | 946 | return ret; |
5d4f98a2 YZ |
947 | } |
948 | ||
5d4f98a2 YZ |
949 | /* |
950 | * get new location of data | |
951 | */ | |
952 | static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, | |
953 | u64 bytenr, u64 num_bytes) | |
954 | { | |
955 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
956 | struct btrfs_path *path; | |
957 | struct btrfs_file_extent_item *fi; | |
958 | struct extent_buffer *leaf; | |
959 | int ret; | |
960 | ||
961 | path = btrfs_alloc_path(); | |
962 | if (!path) | |
963 | return -ENOMEM; | |
964 | ||
965 | bytenr -= BTRFS_I(reloc_inode)->index_cnt; | |
f85b7379 DS |
966 | ret = btrfs_lookup_file_extent(NULL, root, path, |
967 | btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); | |
5d4f98a2 YZ |
968 | if (ret < 0) |
969 | goto out; | |
970 | if (ret > 0) { | |
971 | ret = -ENOENT; | |
972 | goto out; | |
973 | } | |
974 | ||
975 | leaf = path->nodes[0]; | |
976 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
977 | struct btrfs_file_extent_item); | |
978 | ||
979 | BUG_ON(btrfs_file_extent_offset(leaf, fi) || | |
980 | btrfs_file_extent_compression(leaf, fi) || | |
981 | btrfs_file_extent_encryption(leaf, fi) || | |
982 | btrfs_file_extent_other_encoding(leaf, fi)); | |
983 | ||
984 | if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { | |
83d4cfd4 | 985 | ret = -EINVAL; |
5d4f98a2 YZ |
986 | goto out; |
987 | } | |
988 | ||
3fd0a558 | 989 | *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 YZ |
990 | ret = 0; |
991 | out: | |
992 | btrfs_free_path(path); | |
993 | return ret; | |
994 | } | |
995 | ||
996 | /* | |
997 | * update file extent items in the tree leaf to point to | |
998 | * the new locations. | |
999 | */ | |
3fd0a558 YZ |
1000 | static noinline_for_stack |
1001 | int replace_file_extents(struct btrfs_trans_handle *trans, | |
1002 | struct reloc_control *rc, | |
1003 | struct btrfs_root *root, | |
1004 | struct extent_buffer *leaf) | |
5d4f98a2 | 1005 | { |
0b246afa | 1006 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1007 | struct btrfs_key key; |
1008 | struct btrfs_file_extent_item *fi; | |
5e485ac6 | 1009 | struct btrfs_inode *inode = NULL; |
5d4f98a2 YZ |
1010 | u64 parent; |
1011 | u64 bytenr; | |
3fd0a558 | 1012 | u64 new_bytenr = 0; |
5d4f98a2 YZ |
1013 | u64 num_bytes; |
1014 | u64 end; | |
1015 | u32 nritems; | |
1016 | u32 i; | |
83d4cfd4 | 1017 | int ret = 0; |
5d4f98a2 YZ |
1018 | int first = 1; |
1019 | int dirty = 0; | |
1020 | ||
1021 | if (rc->stage != UPDATE_DATA_PTRS) | |
1022 | return 0; | |
1023 | ||
1024 | /* reloc trees always use full backref */ | |
e094f480 | 1025 | if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID) |
5d4f98a2 YZ |
1026 | parent = leaf->start; |
1027 | else | |
1028 | parent = 0; | |
1029 | ||
1030 | nritems = btrfs_header_nritems(leaf); | |
1031 | for (i = 0; i < nritems; i++) { | |
82fa113f QW |
1032 | struct btrfs_ref ref = { 0 }; |
1033 | ||
5d4f98a2 YZ |
1034 | cond_resched(); |
1035 | btrfs_item_key_to_cpu(leaf, &key, i); | |
1036 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
1037 | continue; | |
1038 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
1039 | if (btrfs_file_extent_type(leaf, fi) == | |
1040 | BTRFS_FILE_EXTENT_INLINE) | |
1041 | continue; | |
1042 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1043 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
1044 | if (bytenr == 0) | |
1045 | continue; | |
9569cc20 QW |
1046 | if (!in_range(bytenr, rc->block_group->start, |
1047 | rc->block_group->length)) | |
5d4f98a2 YZ |
1048 | continue; |
1049 | ||
1050 | /* | |
fb12489b | 1051 | * if we are modifying block in fs tree, wait for read_folio |
5d4f98a2 YZ |
1052 | * to complete and drop the extent cache |
1053 | */ | |
e094f480 | 1054 | if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) { |
5d4f98a2 | 1055 | if (first) { |
5e485ac6 | 1056 | inode = btrfs_find_first_inode(root, key.objectid); |
5d4f98a2 | 1057 | first = 0; |
5e485ac6 FM |
1058 | } else if (inode && btrfs_ino(inode) < key.objectid) { |
1059 | btrfs_add_delayed_iput(inode); | |
1060 | inode = btrfs_find_first_inode(root, key.objectid); | |
5d4f98a2 | 1061 | } |
5e485ac6 | 1062 | if (inode && btrfs_ino(inode) == key.objectid) { |
9c5c9604 JB |
1063 | struct extent_state *cached_state = NULL; |
1064 | ||
5d4f98a2 YZ |
1065 | end = key.offset + |
1066 | btrfs_file_extent_num_bytes(leaf, fi); | |
1067 | WARN_ON(!IS_ALIGNED(key.offset, | |
0b246afa JM |
1068 | fs_info->sectorsize)); |
1069 | WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); | |
5d4f98a2 | 1070 | end--; |
5d6f0e98 | 1071 | /* Take mmap lock to serialize with reflinks. */ |
5e485ac6 | 1072 | if (!down_read_trylock(&inode->i_mmap_lock)) |
5d6f0e98 | 1073 | continue; |
5e485ac6 FM |
1074 | ret = try_lock_extent(&inode->io_tree, key.offset, |
1075 | end, &cached_state); | |
5d6f0e98 | 1076 | if (!ret) { |
5e485ac6 | 1077 | up_read(&inode->i_mmap_lock); |
5d4f98a2 | 1078 | continue; |
5d6f0e98 | 1079 | } |
5d4f98a2 | 1080 | |
5e485ac6 FM |
1081 | btrfs_drop_extent_map_range(inode, key.offset, end, true); |
1082 | unlock_extent(&inode->io_tree, key.offset, end, | |
1083 | &cached_state); | |
1084 | up_read(&inode->i_mmap_lock); | |
5d4f98a2 YZ |
1085 | } |
1086 | } | |
1087 | ||
1088 | ret = get_new_location(rc->data_inode, &new_bytenr, | |
1089 | bytenr, num_bytes); | |
83d4cfd4 JB |
1090 | if (ret) { |
1091 | /* | |
1092 | * Don't have to abort since we've not changed anything | |
1093 | * in the file extent yet. | |
1094 | */ | |
1095 | break; | |
3fd0a558 | 1096 | } |
5d4f98a2 YZ |
1097 | |
1098 | btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); | |
1099 | dirty = 1; | |
1100 | ||
1101 | key.offset -= btrfs_file_extent_offset(leaf, fi); | |
4d09b4e9 JB |
1102 | ref.action = BTRFS_ADD_DELAYED_REF; |
1103 | ref.bytenr = new_bytenr; | |
12390e42 | 1104 | ref.num_bytes = num_bytes; |
4d09b4e9 | 1105 | ref.parent = parent; |
e094f480 | 1106 | ref.owning_root = btrfs_root_id(root); |
f2e69a77 JB |
1107 | ref.ref_root = btrfs_header_owner(leaf); |
1108 | btrfs_init_data_ref(&ref, key.objectid, key.offset, | |
e094f480 | 1109 | btrfs_root_id(root), false); |
82fa113f | 1110 | ret = btrfs_inc_extent_ref(trans, &ref); |
83d4cfd4 | 1111 | if (ret) { |
66642832 | 1112 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 JB |
1113 | break; |
1114 | } | |
5d4f98a2 | 1115 | |
4d09b4e9 JB |
1116 | ref.action = BTRFS_DROP_DELAYED_REF; |
1117 | ref.bytenr = bytenr; | |
12390e42 | 1118 | ref.num_bytes = num_bytes; |
4d09b4e9 | 1119 | ref.parent = parent; |
e094f480 | 1120 | ref.owning_root = btrfs_root_id(root); |
f2e69a77 JB |
1121 | ref.ref_root = btrfs_header_owner(leaf); |
1122 | btrfs_init_data_ref(&ref, key.objectid, key.offset, | |
e094f480 | 1123 | btrfs_root_id(root), false); |
ffd4bb2a | 1124 | ret = btrfs_free_extent(trans, &ref); |
83d4cfd4 | 1125 | if (ret) { |
66642832 | 1126 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 JB |
1127 | break; |
1128 | } | |
5d4f98a2 YZ |
1129 | } |
1130 | if (dirty) | |
50564b65 | 1131 | btrfs_mark_buffer_dirty(trans, leaf); |
3fd0a558 | 1132 | if (inode) |
5e485ac6 | 1133 | btrfs_add_delayed_iput(inode); |
83d4cfd4 | 1134 | return ret; |
5d4f98a2 YZ |
1135 | } |
1136 | ||
ab7c8bbf DS |
1137 | static noinline_for_stack int memcmp_node_keys(const struct extent_buffer *eb, |
1138 | int slot, const struct btrfs_path *path, | |
1139 | int level) | |
5d4f98a2 YZ |
1140 | { |
1141 | struct btrfs_disk_key key1; | |
1142 | struct btrfs_disk_key key2; | |
1143 | btrfs_node_key(eb, &key1, slot); | |
1144 | btrfs_node_key(path->nodes[level], &key2, path->slots[level]); | |
1145 | return memcmp(&key1, &key2, sizeof(key1)); | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * try to replace tree blocks in fs tree with the new blocks | |
1150 | * in reloc tree. tree blocks haven't been modified since the | |
1151 | * reloc tree was create can be replaced. | |
1152 | * | |
1153 | * if a block was replaced, level of the block + 1 is returned. | |
1154 | * if no block got replaced, 0 is returned. if there are other | |
1155 | * errors, a negative error number is returned. | |
1156 | */ | |
3fd0a558 | 1157 | static noinline_for_stack |
3d0174f7 | 1158 | int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, |
3fd0a558 YZ |
1159 | struct btrfs_root *dest, struct btrfs_root *src, |
1160 | struct btrfs_path *path, struct btrfs_key *next_key, | |
1161 | int lowest_level, int max_level) | |
5d4f98a2 | 1162 | { |
0b246afa | 1163 | struct btrfs_fs_info *fs_info = dest->fs_info; |
5d4f98a2 YZ |
1164 | struct extent_buffer *eb; |
1165 | struct extent_buffer *parent; | |
82fa113f | 1166 | struct btrfs_ref ref = { 0 }; |
5d4f98a2 YZ |
1167 | struct btrfs_key key; |
1168 | u64 old_bytenr; | |
1169 | u64 new_bytenr; | |
1170 | u64 old_ptr_gen; | |
1171 | u64 new_ptr_gen; | |
1172 | u64 last_snapshot; | |
1173 | u32 blocksize; | |
3fd0a558 | 1174 | int cow = 0; |
5d4f98a2 YZ |
1175 | int level; |
1176 | int ret; | |
1177 | int slot; | |
1178 | ||
e094f480 JB |
1179 | ASSERT(btrfs_root_id(src) == BTRFS_TREE_RELOC_OBJECTID); |
1180 | ASSERT(btrfs_root_id(dest) != BTRFS_TREE_RELOC_OBJECTID); | |
5d4f98a2 YZ |
1181 | |
1182 | last_snapshot = btrfs_root_last_snapshot(&src->root_item); | |
3fd0a558 | 1183 | again: |
5d4f98a2 YZ |
1184 | slot = path->slots[lowest_level]; |
1185 | btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); | |
1186 | ||
1187 | eb = btrfs_lock_root_node(dest); | |
5d4f98a2 YZ |
1188 | level = btrfs_header_level(eb); |
1189 | ||
1190 | if (level < lowest_level) { | |
1191 | btrfs_tree_unlock(eb); | |
1192 | free_extent_buffer(eb); | |
1193 | return 0; | |
1194 | } | |
1195 | ||
3fd0a558 | 1196 | if (cow) { |
9631e4cc JB |
1197 | ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, |
1198 | BTRFS_NESTING_COW); | |
45b87c5d JB |
1199 | if (ret) { |
1200 | btrfs_tree_unlock(eb); | |
1201 | free_extent_buffer(eb); | |
1202 | return ret; | |
1203 | } | |
3fd0a558 | 1204 | } |
5d4f98a2 YZ |
1205 | |
1206 | if (next_key) { | |
1207 | next_key->objectid = (u64)-1; | |
1208 | next_key->type = (u8)-1; | |
1209 | next_key->offset = (u64)-1; | |
1210 | } | |
1211 | ||
1212 | parent = eb; | |
1213 | while (1) { | |
1214 | level = btrfs_header_level(parent); | |
7a9213a9 | 1215 | ASSERT(level >= lowest_level); |
5d4f98a2 | 1216 | |
fdf8d595 | 1217 | ret = btrfs_bin_search(parent, 0, &key, &slot); |
cbca7d59 FM |
1218 | if (ret < 0) |
1219 | break; | |
5d4f98a2 YZ |
1220 | if (ret && slot > 0) |
1221 | slot--; | |
1222 | ||
1223 | if (next_key && slot + 1 < btrfs_header_nritems(parent)) | |
1224 | btrfs_node_key_to_cpu(parent, next_key, slot + 1); | |
1225 | ||
1226 | old_bytenr = btrfs_node_blockptr(parent, slot); | |
0b246afa | 1227 | blocksize = fs_info->nodesize; |
5d4f98a2 YZ |
1228 | old_ptr_gen = btrfs_node_ptr_generation(parent, slot); |
1229 | ||
1230 | if (level <= max_level) { | |
1231 | eb = path->nodes[level]; | |
1232 | new_bytenr = btrfs_node_blockptr(eb, | |
1233 | path->slots[level]); | |
1234 | new_ptr_gen = btrfs_node_ptr_generation(eb, | |
1235 | path->slots[level]); | |
1236 | } else { | |
1237 | new_bytenr = 0; | |
1238 | new_ptr_gen = 0; | |
1239 | } | |
1240 | ||
fae7f21c | 1241 | if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { |
5d4f98a2 YZ |
1242 | ret = level; |
1243 | break; | |
1244 | } | |
1245 | ||
1246 | if (new_bytenr == 0 || old_ptr_gen > last_snapshot || | |
1247 | memcmp_node_keys(parent, slot, path, level)) { | |
3fd0a558 | 1248 | if (level <= lowest_level) { |
5d4f98a2 YZ |
1249 | ret = 0; |
1250 | break; | |
1251 | } | |
1252 | ||
6b3426be | 1253 | eb = btrfs_read_node_slot(parent, slot); |
64c043de LB |
1254 | if (IS_ERR(eb)) { |
1255 | ret = PTR_ERR(eb); | |
264813ac | 1256 | break; |
416bc658 | 1257 | } |
5d4f98a2 | 1258 | btrfs_tree_lock(eb); |
3fd0a558 YZ |
1259 | if (cow) { |
1260 | ret = btrfs_cow_block(trans, dest, eb, parent, | |
9631e4cc JB |
1261 | slot, &eb, |
1262 | BTRFS_NESTING_COW); | |
45b87c5d JB |
1263 | if (ret) { |
1264 | btrfs_tree_unlock(eb); | |
1265 | free_extent_buffer(eb); | |
1266 | break; | |
1267 | } | |
5d4f98a2 YZ |
1268 | } |
1269 | ||
1270 | btrfs_tree_unlock(parent); | |
1271 | free_extent_buffer(parent); | |
1272 | ||
1273 | parent = eb; | |
1274 | continue; | |
1275 | } | |
1276 | ||
3fd0a558 YZ |
1277 | if (!cow) { |
1278 | btrfs_tree_unlock(parent); | |
1279 | free_extent_buffer(parent); | |
1280 | cow = 1; | |
1281 | goto again; | |
1282 | } | |
1283 | ||
5d4f98a2 YZ |
1284 | btrfs_node_key_to_cpu(path->nodes[level], &key, |
1285 | path->slots[level]); | |
b3b4aa74 | 1286 | btrfs_release_path(path); |
5d4f98a2 YZ |
1287 | |
1288 | path->lowest_level = level; | |
b40130b2 | 1289 | set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); |
5d4f98a2 | 1290 | ret = btrfs_search_slot(trans, src, &key, path, 0, 1); |
b40130b2 | 1291 | clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); |
5d4f98a2 | 1292 | path->lowest_level = 0; |
0e9873e2 JB |
1293 | if (ret) { |
1294 | if (ret > 0) | |
1295 | ret = -ENOENT; | |
1296 | break; | |
1297 | } | |
5d4f98a2 | 1298 | |
824d8dff QW |
1299 | /* |
1300 | * Info qgroup to trace both subtrees. | |
1301 | * | |
1302 | * We must trace both trees. | |
1303 | * 1) Tree reloc subtree | |
1304 | * If not traced, we will leak data numbers | |
1305 | * 2) Fs subtree | |
1306 | * If not traced, we will double count old data | |
f616f5cd QW |
1307 | * |
1308 | * We don't scan the subtree right now, but only record | |
1309 | * the swapped tree blocks. | |
1310 | * The real subtree rescan is delayed until we have new | |
1311 | * CoW on the subtree root node before transaction commit. | |
824d8dff | 1312 | */ |
370a11b8 QW |
1313 | ret = btrfs_qgroup_add_swapped_blocks(trans, dest, |
1314 | rc->block_group, parent, slot, | |
1315 | path->nodes[level], path->slots[level], | |
1316 | last_snapshot); | |
1317 | if (ret < 0) | |
1318 | break; | |
5d4f98a2 YZ |
1319 | /* |
1320 | * swap blocks in fs tree and reloc tree. | |
1321 | */ | |
1322 | btrfs_set_node_blockptr(parent, slot, new_bytenr); | |
1323 | btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); | |
50564b65 | 1324 | btrfs_mark_buffer_dirty(trans, parent); |
5d4f98a2 YZ |
1325 | |
1326 | btrfs_set_node_blockptr(path->nodes[level], | |
1327 | path->slots[level], old_bytenr); | |
1328 | btrfs_set_node_ptr_generation(path->nodes[level], | |
1329 | path->slots[level], old_ptr_gen); | |
50564b65 | 1330 | btrfs_mark_buffer_dirty(trans, path->nodes[level]); |
5d4f98a2 | 1331 | |
4d09b4e9 JB |
1332 | ref.action = BTRFS_ADD_DELAYED_REF; |
1333 | ref.bytenr = old_bytenr; | |
12390e42 | 1334 | ref.num_bytes = blocksize; |
4d09b4e9 | 1335 | ref.parent = path->nodes[level]->start; |
e094f480 JB |
1336 | ref.owning_root = btrfs_root_id(src); |
1337 | ref.ref_root = btrfs_root_id(src); | |
f2e69a77 | 1338 | btrfs_init_tree_ref(&ref, level - 1, 0, true); |
82fa113f | 1339 | ret = btrfs_inc_extent_ref(trans, &ref); |
253e258c JB |
1340 | if (ret) { |
1341 | btrfs_abort_transaction(trans, ret); | |
1342 | break; | |
1343 | } | |
4d09b4e9 JB |
1344 | |
1345 | ref.action = BTRFS_ADD_DELAYED_REF; | |
1346 | ref.bytenr = new_bytenr; | |
12390e42 | 1347 | ref.num_bytes = blocksize; |
4d09b4e9 | 1348 | ref.parent = 0; |
e094f480 JB |
1349 | ref.owning_root = btrfs_root_id(dest); |
1350 | ref.ref_root = btrfs_root_id(dest); | |
f2e69a77 | 1351 | btrfs_init_tree_ref(&ref, level - 1, 0, true); |
82fa113f | 1352 | ret = btrfs_inc_extent_ref(trans, &ref); |
253e258c JB |
1353 | if (ret) { |
1354 | btrfs_abort_transaction(trans, ret); | |
1355 | break; | |
1356 | } | |
5d4f98a2 | 1357 | |
457cb1dd | 1358 | /* We don't know the real owning_root, use 0. */ |
4d09b4e9 JB |
1359 | ref.action = BTRFS_DROP_DELAYED_REF; |
1360 | ref.bytenr = new_bytenr; | |
12390e42 | 1361 | ref.num_bytes = blocksize; |
4d09b4e9 JB |
1362 | ref.parent = path->nodes[level]->start; |
1363 | ref.owning_root = 0; | |
e094f480 | 1364 | ref.ref_root = btrfs_root_id(src); |
f2e69a77 | 1365 | btrfs_init_tree_ref(&ref, level - 1, 0, true); |
ffd4bb2a | 1366 | ret = btrfs_free_extent(trans, &ref); |
253e258c JB |
1367 | if (ret) { |
1368 | btrfs_abort_transaction(trans, ret); | |
1369 | break; | |
1370 | } | |
5d4f98a2 | 1371 | |
457cb1dd | 1372 | /* We don't know the real owning_root, use 0. */ |
4d09b4e9 JB |
1373 | ref.action = BTRFS_DROP_DELAYED_REF; |
1374 | ref.bytenr = old_bytenr; | |
12390e42 | 1375 | ref.num_bytes = blocksize; |
4d09b4e9 JB |
1376 | ref.parent = 0; |
1377 | ref.owning_root = 0; | |
e094f480 | 1378 | ref.ref_root = btrfs_root_id(dest); |
f2e69a77 | 1379 | btrfs_init_tree_ref(&ref, level - 1, 0, true); |
ffd4bb2a | 1380 | ret = btrfs_free_extent(trans, &ref); |
253e258c JB |
1381 | if (ret) { |
1382 | btrfs_abort_transaction(trans, ret); | |
1383 | break; | |
1384 | } | |
5d4f98a2 YZ |
1385 | |
1386 | btrfs_unlock_up_safe(path, 0); | |
1387 | ||
1388 | ret = level; | |
1389 | break; | |
1390 | } | |
1391 | btrfs_tree_unlock(parent); | |
1392 | free_extent_buffer(parent); | |
1393 | return ret; | |
1394 | } | |
1395 | ||
1396 | /* | |
1397 | * helper to find next relocated block in reloc tree | |
1398 | */ | |
1399 | static noinline_for_stack | |
1400 | int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, | |
1401 | int *level) | |
1402 | { | |
1403 | struct extent_buffer *eb; | |
1404 | int i; | |
1405 | u64 last_snapshot; | |
1406 | u32 nritems; | |
1407 | ||
1408 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
1409 | ||
1410 | for (i = 0; i < *level; i++) { | |
1411 | free_extent_buffer(path->nodes[i]); | |
1412 | path->nodes[i] = NULL; | |
1413 | } | |
1414 | ||
1415 | for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { | |
1416 | eb = path->nodes[i]; | |
1417 | nritems = btrfs_header_nritems(eb); | |
1418 | while (path->slots[i] + 1 < nritems) { | |
1419 | path->slots[i]++; | |
1420 | if (btrfs_node_ptr_generation(eb, path->slots[i]) <= | |
1421 | last_snapshot) | |
1422 | continue; | |
1423 | ||
1424 | *level = i; | |
1425 | return 0; | |
1426 | } | |
1427 | free_extent_buffer(path->nodes[i]); | |
1428 | path->nodes[i] = NULL; | |
1429 | } | |
1430 | return 1; | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * walk down reloc tree to find relocated block of lowest level | |
1435 | */ | |
1436 | static noinline_for_stack | |
1437 | int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, | |
1438 | int *level) | |
1439 | { | |
1440 | struct extent_buffer *eb = NULL; | |
1441 | int i; | |
5d4f98a2 YZ |
1442 | u64 ptr_gen = 0; |
1443 | u64 last_snapshot; | |
5d4f98a2 YZ |
1444 | u32 nritems; |
1445 | ||
1446 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
1447 | ||
1448 | for (i = *level; i > 0; i--) { | |
1449 | eb = path->nodes[i]; | |
1450 | nritems = btrfs_header_nritems(eb); | |
1451 | while (path->slots[i] < nritems) { | |
1452 | ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); | |
1453 | if (ptr_gen > last_snapshot) | |
1454 | break; | |
1455 | path->slots[i]++; | |
1456 | } | |
1457 | if (path->slots[i] >= nritems) { | |
1458 | if (i == *level) | |
1459 | break; | |
1460 | *level = i + 1; | |
1461 | return 0; | |
1462 | } | |
1463 | if (i == 1) { | |
1464 | *level = i; | |
1465 | return 0; | |
1466 | } | |
1467 | ||
8ef385bb JB |
1468 | eb = btrfs_read_node_slot(eb, path->slots[i]); |
1469 | if (IS_ERR(eb)) | |
64c043de | 1470 | return PTR_ERR(eb); |
5d4f98a2 YZ |
1471 | BUG_ON(btrfs_header_level(eb) != i - 1); |
1472 | path->nodes[i - 1] = eb; | |
1473 | path->slots[i - 1] = 0; | |
1474 | } | |
1475 | return 1; | |
1476 | } | |
1477 | ||
1478 | /* | |
1479 | * invalidate extent cache for file extents whose key in range of | |
1480 | * [min_key, max_key) | |
1481 | */ | |
1482 | static int invalidate_extent_cache(struct btrfs_root *root, | |
ab7c8bbf DS |
1483 | const struct btrfs_key *min_key, |
1484 | const struct btrfs_key *max_key) | |
5d4f98a2 | 1485 | { |
0b246afa | 1486 | struct btrfs_fs_info *fs_info = root->fs_info; |
5e485ac6 | 1487 | struct btrfs_inode *inode = NULL; |
5d4f98a2 YZ |
1488 | u64 objectid; |
1489 | u64 start, end; | |
33345d01 | 1490 | u64 ino; |
5d4f98a2 YZ |
1491 | |
1492 | objectid = min_key->objectid; | |
1493 | while (1) { | |
9c5c9604 JB |
1494 | struct extent_state *cached_state = NULL; |
1495 | ||
5d4f98a2 | 1496 | cond_resched(); |
5e485ac6 FM |
1497 | if (inode) |
1498 | iput(&inode->vfs_inode); | |
5d4f98a2 YZ |
1499 | |
1500 | if (objectid > max_key->objectid) | |
1501 | break; | |
1502 | ||
5e485ac6 | 1503 | inode = btrfs_find_first_inode(root, objectid); |
5d4f98a2 YZ |
1504 | if (!inode) |
1505 | break; | |
5e485ac6 | 1506 | ino = btrfs_ino(inode); |
5d4f98a2 | 1507 | |
33345d01 | 1508 | if (ino > max_key->objectid) { |
5e485ac6 | 1509 | iput(&inode->vfs_inode); |
5d4f98a2 YZ |
1510 | break; |
1511 | } | |
1512 | ||
33345d01 | 1513 | objectid = ino + 1; |
5e485ac6 | 1514 | if (!S_ISREG(inode->vfs_inode.i_mode)) |
5d4f98a2 YZ |
1515 | continue; |
1516 | ||
33345d01 | 1517 | if (unlikely(min_key->objectid == ino)) { |
5d4f98a2 YZ |
1518 | if (min_key->type > BTRFS_EXTENT_DATA_KEY) |
1519 | continue; | |
1520 | if (min_key->type < BTRFS_EXTENT_DATA_KEY) | |
1521 | start = 0; | |
1522 | else { | |
1523 | start = min_key->offset; | |
0b246afa | 1524 | WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); |
5d4f98a2 YZ |
1525 | } |
1526 | } else { | |
1527 | start = 0; | |
1528 | } | |
1529 | ||
33345d01 | 1530 | if (unlikely(max_key->objectid == ino)) { |
5d4f98a2 YZ |
1531 | if (max_key->type < BTRFS_EXTENT_DATA_KEY) |
1532 | continue; | |
1533 | if (max_key->type > BTRFS_EXTENT_DATA_KEY) { | |
1534 | end = (u64)-1; | |
1535 | } else { | |
1536 | if (max_key->offset == 0) | |
1537 | continue; | |
1538 | end = max_key->offset; | |
0b246afa | 1539 | WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); |
5d4f98a2 YZ |
1540 | end--; |
1541 | } | |
1542 | } else { | |
1543 | end = (u64)-1; | |
1544 | } | |
1545 | ||
fb12489b | 1546 | /* the lock_extent waits for read_folio to complete */ |
5e485ac6 FM |
1547 | lock_extent(&inode->io_tree, start, end, &cached_state); |
1548 | btrfs_drop_extent_map_range(inode, start, end, true); | |
1549 | unlock_extent(&inode->io_tree, start, end, &cached_state); | |
5d4f98a2 YZ |
1550 | } |
1551 | return 0; | |
1552 | } | |
1553 | ||
1554 | static int find_next_key(struct btrfs_path *path, int level, | |
1555 | struct btrfs_key *key) | |
1556 | ||
1557 | { | |
1558 | while (level < BTRFS_MAX_LEVEL) { | |
1559 | if (!path->nodes[level]) | |
1560 | break; | |
1561 | if (path->slots[level] + 1 < | |
1562 | btrfs_header_nritems(path->nodes[level])) { | |
1563 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1564 | path->slots[level] + 1); | |
1565 | return 0; | |
1566 | } | |
1567 | level++; | |
1568 | } | |
1569 | return 1; | |
1570 | } | |
1571 | ||
d2311e69 QW |
1572 | /* |
1573 | * Insert current subvolume into reloc_control::dirty_subvol_roots | |
1574 | */ | |
ac54da6c JB |
1575 | static int insert_dirty_subvol(struct btrfs_trans_handle *trans, |
1576 | struct reloc_control *rc, | |
1577 | struct btrfs_root *root) | |
d2311e69 QW |
1578 | { |
1579 | struct btrfs_root *reloc_root = root->reloc_root; | |
1580 | struct btrfs_root_item *reloc_root_item; | |
7934133f | 1581 | int ret; |
d2311e69 QW |
1582 | |
1583 | /* @root must be a subvolume tree root with a valid reloc tree */ | |
e094f480 | 1584 | ASSERT(btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID); |
d2311e69 QW |
1585 | ASSERT(reloc_root); |
1586 | ||
1587 | reloc_root_item = &reloc_root->root_item; | |
1588 | memset(&reloc_root_item->drop_progress, 0, | |
1589 | sizeof(reloc_root_item->drop_progress)); | |
c8422684 | 1590 | btrfs_set_root_drop_level(reloc_root_item, 0); |
d2311e69 | 1591 | btrfs_set_root_refs(reloc_root_item, 0); |
7934133f JB |
1592 | ret = btrfs_update_reloc_root(trans, root); |
1593 | if (ret) | |
1594 | return ret; | |
d2311e69 QW |
1595 | |
1596 | if (list_empty(&root->reloc_dirty_list)) { | |
00246528 | 1597 | btrfs_grab_root(root); |
d2311e69 QW |
1598 | list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); |
1599 | } | |
ac54da6c JB |
1600 | |
1601 | return 0; | |
d2311e69 QW |
1602 | } |
1603 | ||
1604 | static int clean_dirty_subvols(struct reloc_control *rc) | |
1605 | { | |
1606 | struct btrfs_root *root; | |
1607 | struct btrfs_root *next; | |
1608 | int ret = 0; | |
30d40577 | 1609 | int ret2; |
d2311e69 QW |
1610 | |
1611 | list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, | |
1612 | reloc_dirty_list) { | |
e094f480 | 1613 | if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) { |
30d40577 QW |
1614 | /* Merged subvolume, cleanup its reloc root */ |
1615 | struct btrfs_root *reloc_root = root->reloc_root; | |
d2311e69 | 1616 | |
30d40577 QW |
1617 | list_del_init(&root->reloc_dirty_list); |
1618 | root->reloc_root = NULL; | |
6282675e QW |
1619 | /* |
1620 | * Need barrier to ensure clear_bit() only happens after | |
1621 | * root->reloc_root = NULL. Pairs with have_reloc_root. | |
1622 | */ | |
1623 | smp_wmb(); | |
1fac4a54 | 1624 | clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); |
f28de8d8 | 1625 | if (reloc_root) { |
f44deb74 JB |
1626 | /* |
1627 | * btrfs_drop_snapshot drops our ref we hold for | |
1628 | * ->reloc_root. If it fails however we must | |
1629 | * drop the ref ourselves. | |
1630 | */ | |
f28de8d8 | 1631 | ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); |
f44deb74 JB |
1632 | if (ret2 < 0) { |
1633 | btrfs_put_root(reloc_root); | |
1634 | if (!ret) | |
1635 | ret = ret2; | |
1636 | } | |
f28de8d8 | 1637 | } |
00246528 | 1638 | btrfs_put_root(root); |
30d40577 QW |
1639 | } else { |
1640 | /* Orphan reloc tree, just clean it up */ | |
0078a9f9 | 1641 | ret2 = btrfs_drop_snapshot(root, 0, 1); |
f44deb74 JB |
1642 | if (ret2 < 0) { |
1643 | btrfs_put_root(root); | |
1644 | if (!ret) | |
1645 | ret = ret2; | |
1646 | } | |
d2311e69 | 1647 | } |
d2311e69 QW |
1648 | } |
1649 | return ret; | |
1650 | } | |
1651 | ||
5d4f98a2 YZ |
1652 | /* |
1653 | * merge the relocated tree blocks in reloc tree with corresponding | |
1654 | * fs tree. | |
1655 | */ | |
1656 | static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, | |
1657 | struct btrfs_root *root) | |
1658 | { | |
0b246afa | 1659 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
1660 | struct btrfs_key key; |
1661 | struct btrfs_key next_key; | |
9e6a0c52 | 1662 | struct btrfs_trans_handle *trans = NULL; |
5d4f98a2 YZ |
1663 | struct btrfs_root *reloc_root; |
1664 | struct btrfs_root_item *root_item; | |
1665 | struct btrfs_path *path; | |
3fd0a558 | 1666 | struct extent_buffer *leaf; |
fca3a45d | 1667 | int reserve_level; |
5d4f98a2 YZ |
1668 | int level; |
1669 | int max_level; | |
1670 | int replaced = 0; | |
c6a592f2 | 1671 | int ret = 0; |
3fd0a558 | 1672 | u32 min_reserved; |
5d4f98a2 YZ |
1673 | |
1674 | path = btrfs_alloc_path(); | |
1675 | if (!path) | |
1676 | return -ENOMEM; | |
e4058b54 | 1677 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
1678 | |
1679 | reloc_root = root->reloc_root; | |
1680 | root_item = &reloc_root->root_item; | |
1681 | ||
1682 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { | |
1683 | level = btrfs_root_level(root_item); | |
67439dad | 1684 | atomic_inc(&reloc_root->node->refs); |
5d4f98a2 YZ |
1685 | path->nodes[level] = reloc_root->node; |
1686 | path->slots[level] = 0; | |
1687 | } else { | |
1688 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); | |
1689 | ||
c8422684 | 1690 | level = btrfs_root_drop_level(root_item); |
5d4f98a2 YZ |
1691 | BUG_ON(level == 0); |
1692 | path->lowest_level = level; | |
1693 | ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); | |
33c66f43 | 1694 | path->lowest_level = 0; |
5d4f98a2 YZ |
1695 | if (ret < 0) { |
1696 | btrfs_free_path(path); | |
1697 | return ret; | |
1698 | } | |
1699 | ||
1700 | btrfs_node_key_to_cpu(path->nodes[level], &next_key, | |
1701 | path->slots[level]); | |
1702 | WARN_ON(memcmp(&key, &next_key, sizeof(key))); | |
1703 | ||
1704 | btrfs_unlock_up_safe(path, 0); | |
1705 | } | |
1706 | ||
44d354ab QW |
1707 | /* |
1708 | * In merge_reloc_root(), we modify the upper level pointer to swap the | |
1709 | * tree blocks between reloc tree and subvolume tree. Thus for tree | |
1710 | * block COW, we COW at most from level 1 to root level for each tree. | |
1711 | * | |
1712 | * Thus the needed metadata size is at most root_level * nodesize, | |
1713 | * and * 2 since we have two trees to COW. | |
1714 | */ | |
fca3a45d JB |
1715 | reserve_level = max_t(int, 1, btrfs_root_level(root_item)); |
1716 | min_reserved = fs_info->nodesize * reserve_level * 2; | |
3fd0a558 | 1717 | memset(&next_key, 0, sizeof(next_key)); |
5d4f98a2 | 1718 | |
3fd0a558 | 1719 | while (1) { |
9270501c JB |
1720 | ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, |
1721 | min_reserved, | |
44d354ab | 1722 | BTRFS_RESERVE_FLUSH_LIMIT); |
c6a592f2 | 1723 | if (ret) |
9e6a0c52 | 1724 | goto out; |
9e6a0c52 JB |
1725 | trans = btrfs_start_transaction(root, 0); |
1726 | if (IS_ERR(trans)) { | |
c6a592f2 | 1727 | ret = PTR_ERR(trans); |
9e6a0c52 JB |
1728 | trans = NULL; |
1729 | goto out; | |
1730 | } | |
2abc726a JB |
1731 | |
1732 | /* | |
1733 | * At this point we no longer have a reloc_control, so we can't | |
1734 | * depend on btrfs_init_reloc_root to update our last_trans. | |
1735 | * | |
1736 | * But that's ok, we started the trans handle on our | |
1737 | * corresponding fs_root, which means it's been added to the | |
1738 | * dirty list. At commit time we'll still call | |
1739 | * btrfs_update_reloc_root() and update our root item | |
1740 | * appropriately. | |
1741 | */ | |
1742 | reloc_root->last_trans = trans->transid; | |
9e6a0c52 | 1743 | trans->block_rsv = rc->block_rsv; |
5d4f98a2 | 1744 | |
5d4f98a2 | 1745 | replaced = 0; |
5d4f98a2 YZ |
1746 | max_level = level; |
1747 | ||
1748 | ret = walk_down_reloc_tree(reloc_root, path, &level); | |
c6a592f2 | 1749 | if (ret < 0) |
5d4f98a2 | 1750 | goto out; |
5d4f98a2 YZ |
1751 | if (ret > 0) |
1752 | break; | |
1753 | ||
1754 | if (!find_next_key(path, level, &key) && | |
1755 | btrfs_comp_cpu_keys(&next_key, &key) >= 0) { | |
1756 | ret = 0; | |
5d4f98a2 | 1757 | } else { |
3d0174f7 | 1758 | ret = replace_path(trans, rc, root, reloc_root, path, |
3fd0a558 | 1759 | &next_key, level, max_level); |
5d4f98a2 | 1760 | } |
c6a592f2 | 1761 | if (ret < 0) |
5d4f98a2 | 1762 | goto out; |
5d4f98a2 YZ |
1763 | if (ret > 0) { |
1764 | level = ret; | |
1765 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
1766 | path->slots[level]); | |
1767 | replaced = 1; | |
5d4f98a2 YZ |
1768 | } |
1769 | ||
1770 | ret = walk_up_reloc_tree(reloc_root, path, &level); | |
1771 | if (ret > 0) | |
1772 | break; | |
1773 | ||
1774 | BUG_ON(level == 0); | |
1775 | /* | |
1776 | * save the merging progress in the drop_progress. | |
1777 | * this is OK since root refs == 1 in this case. | |
1778 | */ | |
1779 | btrfs_node_key(path->nodes[level], &root_item->drop_progress, | |
1780 | path->slots[level]); | |
c8422684 | 1781 | btrfs_set_root_drop_level(root_item, level); |
5d4f98a2 | 1782 | |
3a45bb20 | 1783 | btrfs_end_transaction_throttle(trans); |
9e6a0c52 | 1784 | trans = NULL; |
5d4f98a2 | 1785 | |
2ff7e61e | 1786 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 YZ |
1787 | |
1788 | if (replaced && rc->stage == UPDATE_DATA_PTRS) | |
1789 | invalidate_extent_cache(root, &key, &next_key); | |
1790 | } | |
1791 | ||
1792 | /* | |
1793 | * handle the case only one block in the fs tree need to be | |
1794 | * relocated and the block is tree root. | |
1795 | */ | |
1796 | leaf = btrfs_lock_root_node(root); | |
9631e4cc JB |
1797 | ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, |
1798 | BTRFS_NESTING_COW); | |
5d4f98a2 YZ |
1799 | btrfs_tree_unlock(leaf); |
1800 | free_extent_buffer(leaf); | |
5d4f98a2 YZ |
1801 | out: |
1802 | btrfs_free_path(path); | |
1803 | ||
ac54da6c JB |
1804 | if (ret == 0) { |
1805 | ret = insert_dirty_subvol(trans, rc, root); | |
1806 | if (ret) | |
1807 | btrfs_abort_transaction(trans, ret); | |
1808 | } | |
5d4f98a2 | 1809 | |
9e6a0c52 | 1810 | if (trans) |
3a45bb20 | 1811 | btrfs_end_transaction_throttle(trans); |
5d4f98a2 | 1812 | |
2ff7e61e | 1813 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 | 1814 | |
5d4f98a2 YZ |
1815 | if (replaced && rc->stage == UPDATE_DATA_PTRS) |
1816 | invalidate_extent_cache(root, &key, &next_key); | |
1817 | ||
c6a592f2 | 1818 | return ret; |
5d4f98a2 YZ |
1819 | } |
1820 | ||
3fd0a558 YZ |
1821 | static noinline_for_stack |
1822 | int prepare_to_merge(struct reloc_control *rc, int err) | |
5d4f98a2 | 1823 | { |
3fd0a558 | 1824 | struct btrfs_root *root = rc->extent_root; |
0b246afa | 1825 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1826 | struct btrfs_root *reloc_root; |
3fd0a558 YZ |
1827 | struct btrfs_trans_handle *trans; |
1828 | LIST_HEAD(reloc_roots); | |
1829 | u64 num_bytes = 0; | |
1830 | int ret; | |
3fd0a558 | 1831 | |
0b246afa JM |
1832 | mutex_lock(&fs_info->reloc_mutex); |
1833 | rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; | |
3fd0a558 | 1834 | rc->merging_rsv_size += rc->nodes_relocated * 2; |
0b246afa | 1835 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f | 1836 | |
3fd0a558 YZ |
1837 | again: |
1838 | if (!err) { | |
1839 | num_bytes = rc->merging_rsv_size; | |
9270501c | 1840 | ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes, |
08e007d2 | 1841 | BTRFS_RESERVE_FLUSH_ALL); |
3fd0a558 YZ |
1842 | if (ret) |
1843 | err = ret; | |
1844 | } | |
1845 | ||
7a7eaa40 | 1846 | trans = btrfs_join_transaction(rc->extent_root); |
3612b495 TI |
1847 | if (IS_ERR(trans)) { |
1848 | if (!err) | |
2ff7e61e | 1849 | btrfs_block_rsv_release(fs_info, rc->block_rsv, |
63f018be | 1850 | num_bytes, NULL); |
3612b495 TI |
1851 | return PTR_ERR(trans); |
1852 | } | |
3fd0a558 YZ |
1853 | |
1854 | if (!err) { | |
1855 | if (num_bytes != rc->merging_rsv_size) { | |
3a45bb20 | 1856 | btrfs_end_transaction(trans); |
2ff7e61e | 1857 | btrfs_block_rsv_release(fs_info, rc->block_rsv, |
63f018be | 1858 | num_bytes, NULL); |
3fd0a558 YZ |
1859 | goto again; |
1860 | } | |
1861 | } | |
5d4f98a2 | 1862 | |
d23d42e3 | 1863 | rc->merge_reloc_tree = true; |
3fd0a558 YZ |
1864 | |
1865 | while (!list_empty(&rc->reloc_roots)) { | |
1866 | reloc_root = list_entry(rc->reloc_roots.next, | |
1867 | struct btrfs_root, root_list); | |
1868 | list_del_init(&reloc_root->root_list); | |
5d4f98a2 | 1869 | |
a820feb5 DS |
1870 | root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, |
1871 | false); | |
e0b085b0 JB |
1872 | if (IS_ERR(root)) { |
1873 | /* | |
1874 | * Even if we have an error we need this reloc root | |
1875 | * back on our list so we can clean up properly. | |
1876 | */ | |
1877 | list_add(&reloc_root->root_list, &reloc_roots); | |
1878 | btrfs_abort_transaction(trans, (int)PTR_ERR(root)); | |
1879 | if (!err) | |
1880 | err = PTR_ERR(root); | |
1881 | break; | |
1882 | } | |
05d7ce50 QW |
1883 | |
1884 | if (unlikely(root->reloc_root != reloc_root)) { | |
1885 | if (root->reloc_root) { | |
1886 | btrfs_err(fs_info, | |
1887 | "reloc tree mismatch, root %lld has reloc root key (%lld %u %llu) gen %llu, expect reloc root key (%lld %u %llu) gen %llu", | |
e094f480 JB |
1888 | btrfs_root_id(root), |
1889 | btrfs_root_id(root->reloc_root), | |
05d7ce50 QW |
1890 | root->reloc_root->root_key.type, |
1891 | root->reloc_root->root_key.offset, | |
1892 | btrfs_root_generation( | |
1893 | &root->reloc_root->root_item), | |
e094f480 | 1894 | btrfs_root_id(reloc_root), |
05d7ce50 QW |
1895 | reloc_root->root_key.type, |
1896 | reloc_root->root_key.offset, | |
1897 | btrfs_root_generation( | |
1898 | &reloc_root->root_item)); | |
1899 | } else { | |
1900 | btrfs_err(fs_info, | |
1901 | "reloc tree mismatch, root %lld has no reloc root, expect reloc root key (%lld %u %llu) gen %llu", | |
e094f480 JB |
1902 | btrfs_root_id(root), |
1903 | btrfs_root_id(reloc_root), | |
05d7ce50 QW |
1904 | reloc_root->root_key.type, |
1905 | reloc_root->root_key.offset, | |
1906 | btrfs_root_generation( | |
1907 | &reloc_root->root_item)); | |
1908 | } | |
1909 | list_add(&reloc_root->root_list, &reloc_roots); | |
1910 | btrfs_put_root(root); | |
1911 | btrfs_abort_transaction(trans, -EUCLEAN); | |
1912 | if (!err) | |
1913 | err = -EUCLEAN; | |
1914 | break; | |
1915 | } | |
5d4f98a2 | 1916 | |
3fd0a558 YZ |
1917 | /* |
1918 | * set reference count to 1, so btrfs_recover_relocation | |
1919 | * knows it should resumes merging | |
1920 | */ | |
1921 | if (!err) | |
1922 | btrfs_set_root_refs(&reloc_root->root_item, 1); | |
bbae13f8 | 1923 | ret = btrfs_update_reloc_root(trans, root); |
5d4f98a2 | 1924 | |
bbae13f8 JB |
1925 | /* |
1926 | * Even if we have an error we need this reloc root back on our | |
1927 | * list so we can clean up properly. | |
1928 | */ | |
3fd0a558 | 1929 | list_add(&reloc_root->root_list, &reloc_roots); |
00246528 | 1930 | btrfs_put_root(root); |
bbae13f8 JB |
1931 | |
1932 | if (ret) { | |
1933 | btrfs_abort_transaction(trans, ret); | |
1934 | if (!err) | |
1935 | err = ret; | |
1936 | break; | |
1937 | } | |
3fd0a558 | 1938 | } |
5d4f98a2 | 1939 | |
3fd0a558 | 1940 | list_splice(&reloc_roots, &rc->reloc_roots); |
5d4f98a2 | 1941 | |
3fd0a558 | 1942 | if (!err) |
fb686c68 | 1943 | err = btrfs_commit_transaction(trans); |
3fd0a558 | 1944 | else |
3a45bb20 | 1945 | btrfs_end_transaction(trans); |
3fd0a558 | 1946 | return err; |
5d4f98a2 YZ |
1947 | } |
1948 | ||
aca1bba6 LB |
1949 | static noinline_for_stack |
1950 | void free_reloc_roots(struct list_head *list) | |
1951 | { | |
a7571232 | 1952 | struct btrfs_root *reloc_root, *tmp; |
aca1bba6 | 1953 | |
a7571232 | 1954 | list_for_each_entry_safe(reloc_root, tmp, list, root_list) |
bb166d72 | 1955 | __del_reloc_root(reloc_root); |
aca1bba6 LB |
1956 | } |
1957 | ||
3fd0a558 | 1958 | static noinline_for_stack |
94404e82 | 1959 | void merge_reloc_roots(struct reloc_control *rc) |
5d4f98a2 | 1960 | { |
0b246afa | 1961 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 | 1962 | struct btrfs_root *root; |
3fd0a558 YZ |
1963 | struct btrfs_root *reloc_root; |
1964 | LIST_HEAD(reloc_roots); | |
1965 | int found = 0; | |
aca1bba6 | 1966 | int ret = 0; |
3fd0a558 YZ |
1967 | again: |
1968 | root = rc->extent_root; | |
7585717f CM |
1969 | |
1970 | /* | |
1971 | * this serializes us with btrfs_record_root_in_transaction, | |
1972 | * we have to make sure nobody is in the middle of | |
1973 | * adding their roots to the list while we are | |
1974 | * doing this splice | |
1975 | */ | |
0b246afa | 1976 | mutex_lock(&fs_info->reloc_mutex); |
3fd0a558 | 1977 | list_splice_init(&rc->reloc_roots, &reloc_roots); |
0b246afa | 1978 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 | 1979 | |
3fd0a558 YZ |
1980 | while (!list_empty(&reloc_roots)) { |
1981 | found = 1; | |
1982 | reloc_root = list_entry(reloc_roots.next, | |
1983 | struct btrfs_root, root_list); | |
5d4f98a2 | 1984 | |
a820feb5 DS |
1985 | root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, |
1986 | false); | |
3fd0a558 | 1987 | if (btrfs_root_refs(&reloc_root->root_item) > 0) { |
05d7ce50 | 1988 | if (WARN_ON(IS_ERR(root))) { |
24213fa4 JB |
1989 | /* |
1990 | * For recovery we read the fs roots on mount, | |
1991 | * and if we didn't find the root then we marked | |
1992 | * the reloc root as a garbage root. For normal | |
1993 | * relocation obviously the root should exist in | |
1994 | * memory. However there's no reason we can't | |
1995 | * handle the error properly here just in case. | |
1996 | */ | |
24213fa4 JB |
1997 | ret = PTR_ERR(root); |
1998 | goto out; | |
1999 | } | |
05d7ce50 | 2000 | if (WARN_ON(root->reloc_root != reloc_root)) { |
24213fa4 | 2001 | /* |
05d7ce50 QW |
2002 | * This can happen if on-disk metadata has some |
2003 | * corruption, e.g. bad reloc tree key offset. | |
24213fa4 | 2004 | */ |
24213fa4 JB |
2005 | ret = -EINVAL; |
2006 | goto out; | |
2007 | } | |
3fd0a558 | 2008 | ret = merge_reloc_root(rc, root); |
00246528 | 2009 | btrfs_put_root(root); |
b37b39cd | 2010 | if (ret) { |
25e293c2 WS |
2011 | if (list_empty(&reloc_root->root_list)) |
2012 | list_add_tail(&reloc_root->root_list, | |
2013 | &reloc_roots); | |
aca1bba6 | 2014 | goto out; |
b37b39cd | 2015 | } |
3fd0a558 | 2016 | } else { |
51415b6c QW |
2017 | if (!IS_ERR(root)) { |
2018 | if (root->reloc_root == reloc_root) { | |
2019 | root->reloc_root = NULL; | |
2020 | btrfs_put_root(reloc_root); | |
2021 | } | |
1dae7e0e QW |
2022 | clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, |
2023 | &root->state); | |
51415b6c QW |
2024 | btrfs_put_root(root); |
2025 | } | |
2026 | ||
3fd0a558 | 2027 | list_del_init(&reloc_root->root_list); |
30d40577 QW |
2028 | /* Don't forget to queue this reloc root for cleanup */ |
2029 | list_add_tail(&reloc_root->reloc_dirty_list, | |
2030 | &rc->dirty_subvol_roots); | |
3fd0a558 | 2031 | } |
5d4f98a2 YZ |
2032 | } |
2033 | ||
3fd0a558 YZ |
2034 | if (found) { |
2035 | found = 0; | |
2036 | goto again; | |
2037 | } | |
aca1bba6 LB |
2038 | out: |
2039 | if (ret) { | |
0b246afa | 2040 | btrfs_handle_fs_error(fs_info, ret, NULL); |
a7571232 | 2041 | free_reloc_roots(&reloc_roots); |
467bb1d2 WS |
2042 | |
2043 | /* new reloc root may be added */ | |
0b246afa | 2044 | mutex_lock(&fs_info->reloc_mutex); |
467bb1d2 | 2045 | list_splice_init(&rc->reloc_roots, &reloc_roots); |
0b246afa | 2046 | mutex_unlock(&fs_info->reloc_mutex); |
a7571232 | 2047 | free_reloc_roots(&reloc_roots); |
aca1bba6 LB |
2048 | } |
2049 | ||
7b7b7431 JB |
2050 | /* |
2051 | * We used to have | |
2052 | * | |
2053 | * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); | |
2054 | * | |
2055 | * here, but it's wrong. If we fail to start the transaction in | |
2056 | * prepare_to_merge() we will have only 0 ref reloc roots, none of which | |
2057 | * have actually been removed from the reloc_root_tree rb tree. This is | |
2058 | * fine because we're bailing here, and we hold a reference on the root | |
2059 | * for the list that holds it, so these roots will be cleaned up when we | |
2060 | * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root | |
2061 | * will be cleaned up on unmount. | |
2062 | * | |
2063 | * The remaining nodes will be cleaned up by free_reloc_control. | |
2064 | */ | |
5d4f98a2 YZ |
2065 | } |
2066 | ||
2067 | static void free_block_list(struct rb_root *blocks) | |
2068 | { | |
2069 | struct tree_block *block; | |
2070 | struct rb_node *rb_node; | |
2071 | while ((rb_node = rb_first(blocks))) { | |
2072 | block = rb_entry(rb_node, struct tree_block, rb_node); | |
2073 | rb_erase(rb_node, blocks); | |
2074 | kfree(block); | |
2075 | } | |
2076 | } | |
2077 | ||
2078 | static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, | |
2079 | struct btrfs_root *reloc_root) | |
2080 | { | |
0b246afa | 2081 | struct btrfs_fs_info *fs_info = reloc_root->fs_info; |
5d4f98a2 | 2082 | struct btrfs_root *root; |
442b1ac5 | 2083 | int ret; |
5d4f98a2 YZ |
2084 | |
2085 | if (reloc_root->last_trans == trans->transid) | |
2086 | return 0; | |
2087 | ||
a820feb5 | 2088 | root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); |
404bccbc JB |
2089 | |
2090 | /* | |
2091 | * This should succeed, since we can't have a reloc root without having | |
2092 | * already looked up the actual root and created the reloc root for this | |
2093 | * root. | |
2094 | * | |
2095 | * However if there's some sort of corruption where we have a ref to a | |
2096 | * reloc root without a corresponding root this could return ENOENT. | |
2097 | */ | |
2098 | if (IS_ERR(root)) { | |
2099 | ASSERT(0); | |
2100 | return PTR_ERR(root); | |
2101 | } | |
2102 | if (root->reloc_root != reloc_root) { | |
2103 | ASSERT(0); | |
2104 | btrfs_err(fs_info, | |
2105 | "root %llu has two reloc roots associated with it", | |
2106 | reloc_root->root_key.offset); | |
2107 | btrfs_put_root(root); | |
2108 | return -EUCLEAN; | |
2109 | } | |
442b1ac5 | 2110 | ret = btrfs_record_root_in_trans(trans, root); |
00246528 | 2111 | btrfs_put_root(root); |
5d4f98a2 | 2112 | |
442b1ac5 | 2113 | return ret; |
5d4f98a2 YZ |
2114 | } |
2115 | ||
3fd0a558 YZ |
2116 | static noinline_for_stack |
2117 | struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, | |
2118 | struct reloc_control *rc, | |
a26195a5 QW |
2119 | struct btrfs_backref_node *node, |
2120 | struct btrfs_backref_edge *edges[]) | |
5d4f98a2 | 2121 | { |
a26195a5 | 2122 | struct btrfs_backref_node *next; |
5d4f98a2 | 2123 | struct btrfs_root *root; |
3fd0a558 | 2124 | int index = 0; |
92de551b | 2125 | int ret; |
3fd0a558 | 2126 | |
5d4f98a2 YZ |
2127 | next = node; |
2128 | while (1) { | |
2129 | cond_resched(); | |
2130 | next = walk_up_backref(next, edges, &index); | |
2131 | root = next->root; | |
8ee66afe JB |
2132 | |
2133 | /* | |
2134 | * If there is no root, then our references for this block are | |
2135 | * incomplete, as we should be able to walk all the way up to a | |
2136 | * block that is owned by a root. | |
2137 | * | |
2138 | * This path is only for SHAREABLE roots, so if we come upon a | |
2139 | * non-SHAREABLE root then we have backrefs that resolve | |
2140 | * improperly. | |
2141 | * | |
2142 | * Both of these cases indicate file system corruption, or a bug | |
2143 | * in the backref walking code. | |
2144 | */ | |
2145 | if (!root) { | |
2146 | ASSERT(0); | |
2147 | btrfs_err(trans->fs_info, | |
2148 | "bytenr %llu doesn't have a backref path ending in a root", | |
2149 | node->bytenr); | |
2150 | return ERR_PTR(-EUCLEAN); | |
2151 | } | |
2152 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { | |
2153 | ASSERT(0); | |
2154 | btrfs_err(trans->fs_info, | |
2155 | "bytenr %llu has multiple refs with one ending in a non-shareable root", | |
2156 | node->bytenr); | |
2157 | return ERR_PTR(-EUCLEAN); | |
2158 | } | |
5d4f98a2 | 2159 | |
e094f480 | 2160 | if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID) { |
92de551b JB |
2161 | ret = record_reloc_root_in_trans(trans, root); |
2162 | if (ret) | |
2163 | return ERR_PTR(ret); | |
5d4f98a2 YZ |
2164 | break; |
2165 | } | |
2166 | ||
92de551b JB |
2167 | ret = btrfs_record_root_in_trans(trans, root); |
2168 | if (ret) | |
2169 | return ERR_PTR(ret); | |
3fd0a558 YZ |
2170 | root = root->reloc_root; |
2171 | ||
39200e59 JB |
2172 | /* |
2173 | * We could have raced with another thread which failed, so | |
2174 | * root->reloc_root may not be set, return ENOENT in this case. | |
2175 | */ | |
2176 | if (!root) | |
2177 | return ERR_PTR(-ENOENT); | |
2178 | ||
3fd0a558 | 2179 | if (next->new_bytenr != root->node->start) { |
8ee66afe JB |
2180 | /* |
2181 | * We just created the reloc root, so we shouldn't have | |
2182 | * ->new_bytenr set and this shouldn't be in the changed | |
2183 | * list. If it is then we have multiple roots pointing | |
2184 | * at the same bytenr which indicates corruption, or | |
2185 | * we've made a mistake in the backref walking code. | |
2186 | */ | |
2187 | ASSERT(next->new_bytenr == 0); | |
2188 | ASSERT(list_empty(&next->list)); | |
2189 | if (next->new_bytenr || !list_empty(&next->list)) { | |
2190 | btrfs_err(trans->fs_info, | |
2191 | "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", | |
2192 | node->bytenr, next->bytenr); | |
2193 | return ERR_PTR(-EUCLEAN); | |
2194 | } | |
2195 | ||
3fd0a558 | 2196 | next->new_bytenr = root->node->start; |
00246528 JB |
2197 | btrfs_put_root(next->root); |
2198 | next->root = btrfs_grab_root(root); | |
0b530bc5 | 2199 | ASSERT(next->root); |
3fd0a558 YZ |
2200 | list_add_tail(&next->list, |
2201 | &rc->backref_cache.changed); | |
9569cc20 | 2202 | mark_block_processed(rc, next); |
5d4f98a2 YZ |
2203 | break; |
2204 | } | |
2205 | ||
3fd0a558 | 2206 | WARN_ON(1); |
5d4f98a2 YZ |
2207 | root = NULL; |
2208 | next = walk_down_backref(edges, &index); | |
2209 | if (!next || next->level <= node->level) | |
2210 | break; | |
2211 | } | |
cbdc2ebc JB |
2212 | if (!root) { |
2213 | /* | |
2214 | * This can happen if there's fs corruption or if there's a bug | |
2215 | * in the backref lookup code. | |
2216 | */ | |
2217 | ASSERT(0); | |
2218 | return ERR_PTR(-ENOENT); | |
2219 | } | |
5d4f98a2 | 2220 | |
3fd0a558 YZ |
2221 | next = node; |
2222 | /* setup backref node path for btrfs_reloc_cow_block */ | |
2223 | while (1) { | |
2224 | rc->backref_cache.path[next->level] = next; | |
2225 | if (--index < 0) | |
2226 | break; | |
2227 | next = edges[index]->node[UPPER]; | |
5d4f98a2 | 2228 | } |
5d4f98a2 YZ |
2229 | return root; |
2230 | } | |
2231 | ||
3fd0a558 | 2232 | /* |
92a7cc42 QW |
2233 | * Select a tree root for relocation. |
2234 | * | |
2235 | * Return NULL if the block is not shareable. We should use do_relocation() in | |
2236 | * this case. | |
2237 | * | |
2238 | * Return a tree root pointer if the block is shareable. | |
2239 | * Return -ENOENT if the block is root of reloc tree. | |
3fd0a558 | 2240 | */ |
5d4f98a2 | 2241 | static noinline_for_stack |
a26195a5 | 2242 | struct btrfs_root *select_one_root(struct btrfs_backref_node *node) |
5d4f98a2 | 2243 | { |
a26195a5 | 2244 | struct btrfs_backref_node *next; |
3fd0a558 YZ |
2245 | struct btrfs_root *root; |
2246 | struct btrfs_root *fs_root = NULL; | |
a26195a5 | 2247 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; |
3fd0a558 YZ |
2248 | int index = 0; |
2249 | ||
2250 | next = node; | |
2251 | while (1) { | |
2252 | cond_resched(); | |
2253 | next = walk_up_backref(next, edges, &index); | |
2254 | root = next->root; | |
8717cf44 JB |
2255 | |
2256 | /* | |
2257 | * This can occur if we have incomplete extent refs leading all | |
2258 | * the way up a particular path, in this case return -EUCLEAN. | |
2259 | */ | |
2260 | if (!root) | |
2261 | return ERR_PTR(-EUCLEAN); | |
3fd0a558 | 2262 | |
92a7cc42 QW |
2263 | /* No other choice for non-shareable tree */ |
2264 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) | |
3fd0a558 YZ |
2265 | return root; |
2266 | ||
e094f480 | 2267 | if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) |
3fd0a558 YZ |
2268 | fs_root = root; |
2269 | ||
2270 | if (next != node) | |
2271 | return NULL; | |
2272 | ||
2273 | next = walk_down_backref(edges, &index); | |
2274 | if (!next || next->level <= node->level) | |
2275 | break; | |
2276 | } | |
2277 | ||
2278 | if (!fs_root) | |
2279 | return ERR_PTR(-ENOENT); | |
2280 | return fs_root; | |
5d4f98a2 YZ |
2281 | } |
2282 | ||
30704a0d NA |
2283 | static noinline_for_stack u64 calcu_metadata_size(struct reloc_control *rc, |
2284 | struct btrfs_backref_node *node) | |
5d4f98a2 | 2285 | { |
0b246afa | 2286 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
a26195a5 QW |
2287 | struct btrfs_backref_node *next = node; |
2288 | struct btrfs_backref_edge *edge; | |
2289 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
3fd0a558 YZ |
2290 | u64 num_bytes = 0; |
2291 | int index = 0; | |
2292 | ||
30704a0d | 2293 | BUG_ON(node->processed); |
3fd0a558 YZ |
2294 | |
2295 | while (next) { | |
2296 | cond_resched(); | |
2297 | while (1) { | |
30704a0d | 2298 | if (next->processed) |
3fd0a558 YZ |
2299 | break; |
2300 | ||
0b246afa | 2301 | num_bytes += fs_info->nodesize; |
3fd0a558 YZ |
2302 | |
2303 | if (list_empty(&next->upper)) | |
2304 | break; | |
2305 | ||
2306 | edge = list_entry(next->upper.next, | |
a26195a5 | 2307 | struct btrfs_backref_edge, list[LOWER]); |
3fd0a558 YZ |
2308 | edges[index++] = edge; |
2309 | next = edge->node[UPPER]; | |
2310 | } | |
2311 | next = walk_down_backref(edges, &index); | |
2312 | } | |
2313 | return num_bytes; | |
5d4f98a2 YZ |
2314 | } |
2315 | ||
3fd0a558 YZ |
2316 | static int reserve_metadata_space(struct btrfs_trans_handle *trans, |
2317 | struct reloc_control *rc, | |
a26195a5 | 2318 | struct btrfs_backref_node *node) |
5d4f98a2 | 2319 | { |
3fd0a558 | 2320 | struct btrfs_root *root = rc->extent_root; |
da17066c | 2321 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 YZ |
2322 | u64 num_bytes; |
2323 | int ret; | |
0647bf56 | 2324 | u64 tmp; |
3fd0a558 | 2325 | |
30704a0d | 2326 | num_bytes = calcu_metadata_size(rc, node) * 2; |
5d4f98a2 | 2327 | |
3fd0a558 | 2328 | trans->block_rsv = rc->block_rsv; |
0647bf56 | 2329 | rc->reserved_bytes += num_bytes; |
8ca17f0f JB |
2330 | |
2331 | /* | |
2332 | * We are under a transaction here so we can only do limited flushing. | |
2333 | * If we get an enospc just kick back -EAGAIN so we know to drop the | |
2334 | * transaction and try to refill when we can flush all the things. | |
2335 | */ | |
9270501c JB |
2336 | ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes, |
2337 | BTRFS_RESERVE_FLUSH_LIMIT); | |
3fd0a558 | 2338 | if (ret) { |
da17066c | 2339 | tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; |
8ca17f0f JB |
2340 | while (tmp <= rc->reserved_bytes) |
2341 | tmp <<= 1; | |
2342 | /* | |
2343 | * only one thread can access block_rsv at this point, | |
2344 | * so we don't need hold lock to protect block_rsv. | |
2345 | * we expand more reservation size here to allow enough | |
52042d8e | 2346 | * space for relocation and we will return earlier in |
8ca17f0f JB |
2347 | * enospc case. |
2348 | */ | |
da17066c JM |
2349 | rc->block_rsv->size = tmp + fs_info->nodesize * |
2350 | RELOCATION_RESERVED_NODES; | |
8ca17f0f | 2351 | return -EAGAIN; |
5d4f98a2 | 2352 | } |
3fd0a558 | 2353 | |
3fd0a558 YZ |
2354 | return 0; |
2355 | } | |
2356 | ||
5d4f98a2 YZ |
2357 | /* |
2358 | * relocate a block tree, and then update pointers in upper level | |
2359 | * blocks that reference the block to point to the new location. | |
2360 | * | |
2361 | * if called by link_to_upper, the block has already been relocated. | |
2362 | * in that case this function just updates pointers. | |
2363 | */ | |
2364 | static int do_relocation(struct btrfs_trans_handle *trans, | |
3fd0a558 | 2365 | struct reloc_control *rc, |
a26195a5 | 2366 | struct btrfs_backref_node *node, |
5d4f98a2 YZ |
2367 | struct btrfs_key *key, |
2368 | struct btrfs_path *path, int lowest) | |
2369 | { | |
a26195a5 QW |
2370 | struct btrfs_backref_node *upper; |
2371 | struct btrfs_backref_edge *edge; | |
2372 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
5d4f98a2 YZ |
2373 | struct btrfs_root *root; |
2374 | struct extent_buffer *eb; | |
2375 | u32 blocksize; | |
2376 | u64 bytenr; | |
5d4f98a2 | 2377 | int slot; |
8df01fdd | 2378 | int ret = 0; |
5d4f98a2 | 2379 | |
ffe30dd8 JB |
2380 | /* |
2381 | * If we are lowest then this is the first time we're processing this | |
2382 | * block, and thus shouldn't have an eb associated with it yet. | |
2383 | */ | |
2384 | ASSERT(!lowest || !node->eb); | |
5d4f98a2 YZ |
2385 | |
2386 | path->lowest_level = node->level + 1; | |
3fd0a558 | 2387 | rc->backref_cache.path[node->level] = node; |
5d4f98a2 YZ |
2388 | list_for_each_entry(edge, &node->upper, list[LOWER]) { |
2389 | cond_resched(); | |
5d4f98a2 YZ |
2390 | |
2391 | upper = edge->node[UPPER]; | |
dc4103f9 | 2392 | root = select_reloc_root(trans, rc, upper, edges); |
cbdc2ebc JB |
2393 | if (IS_ERR(root)) { |
2394 | ret = PTR_ERR(root); | |
2395 | goto next; | |
2396 | } | |
3fd0a558 YZ |
2397 | |
2398 | if (upper->eb && !upper->locked) { | |
2399 | if (!lowest) { | |
fdf8d595 | 2400 | ret = btrfs_bin_search(upper->eb, 0, key, &slot); |
8df01fdd | 2401 | if (ret < 0) |
cbca7d59 | 2402 | goto next; |
3fd0a558 YZ |
2403 | BUG_ON(ret); |
2404 | bytenr = btrfs_node_blockptr(upper->eb, slot); | |
2405 | if (node->eb->start == bytenr) | |
2406 | goto next; | |
2407 | } | |
b0fe7078 | 2408 | btrfs_backref_drop_node_buffer(upper); |
3fd0a558 | 2409 | } |
5d4f98a2 YZ |
2410 | |
2411 | if (!upper->eb) { | |
2412 | ret = btrfs_search_slot(trans, root, key, path, 0, 1); | |
3561b9db | 2413 | if (ret) { |
8df01fdd NB |
2414 | if (ret > 0) |
2415 | ret = -ENOENT; | |
3561b9db LB |
2416 | |
2417 | btrfs_release_path(path); | |
5d4f98a2 YZ |
2418 | break; |
2419 | } | |
5d4f98a2 | 2420 | |
3fd0a558 YZ |
2421 | if (!upper->eb) { |
2422 | upper->eb = path->nodes[upper->level]; | |
2423 | path->nodes[upper->level] = NULL; | |
2424 | } else { | |
2425 | BUG_ON(upper->eb != path->nodes[upper->level]); | |
2426 | } | |
5d4f98a2 | 2427 | |
3fd0a558 YZ |
2428 | upper->locked = 1; |
2429 | path->locks[upper->level] = 0; | |
5d4f98a2 | 2430 | |
3fd0a558 | 2431 | slot = path->slots[upper->level]; |
b3b4aa74 | 2432 | btrfs_release_path(path); |
5d4f98a2 | 2433 | } else { |
fdf8d595 | 2434 | ret = btrfs_bin_search(upper->eb, 0, key, &slot); |
8df01fdd | 2435 | if (ret < 0) |
cbca7d59 | 2436 | goto next; |
5d4f98a2 YZ |
2437 | BUG_ON(ret); |
2438 | } | |
2439 | ||
2440 | bytenr = btrfs_node_blockptr(upper->eb, slot); | |
3fd0a558 | 2441 | if (lowest) { |
4547f4d8 LB |
2442 | if (bytenr != node->bytenr) { |
2443 | btrfs_err(root->fs_info, | |
2444 | "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", | |
2445 | bytenr, node->bytenr, slot, | |
2446 | upper->eb->start); | |
8df01fdd | 2447 | ret = -EIO; |
4547f4d8 LB |
2448 | goto next; |
2449 | } | |
5d4f98a2 | 2450 | } else { |
3fd0a558 YZ |
2451 | if (node->eb->start == bytenr) |
2452 | goto next; | |
5d4f98a2 YZ |
2453 | } |
2454 | ||
da17066c | 2455 | blocksize = root->fs_info->nodesize; |
c9752536 | 2456 | eb = btrfs_read_node_slot(upper->eb, slot); |
64c043de | 2457 | if (IS_ERR(eb)) { |
8df01fdd | 2458 | ret = PTR_ERR(eb); |
64c043de | 2459 | goto next; |
97d9a8a4 | 2460 | } |
5d4f98a2 | 2461 | btrfs_tree_lock(eb); |
5d4f98a2 YZ |
2462 | |
2463 | if (!node->eb) { | |
2464 | ret = btrfs_cow_block(trans, root, eb, upper->eb, | |
9631e4cc | 2465 | slot, &eb, BTRFS_NESTING_COW); |
3fd0a558 YZ |
2466 | btrfs_tree_unlock(eb); |
2467 | free_extent_buffer(eb); | |
8df01fdd | 2468 | if (ret < 0) |
3fd0a558 | 2469 | goto next; |
ffe30dd8 JB |
2470 | /* |
2471 | * We've just COWed this block, it should have updated | |
2472 | * the correct backref node entry. | |
2473 | */ | |
2474 | ASSERT(node->eb == eb); | |
5d4f98a2 | 2475 | } else { |
4d09b4e9 JB |
2476 | struct btrfs_ref ref = { |
2477 | .action = BTRFS_ADD_DELAYED_REF, | |
2478 | .bytenr = node->eb->start, | |
12390e42 | 2479 | .num_bytes = blocksize, |
4d09b4e9 JB |
2480 | .parent = upper->eb->start, |
2481 | .owning_root = btrfs_header_owner(upper->eb), | |
f2e69a77 | 2482 | .ref_root = btrfs_header_owner(upper->eb), |
4d09b4e9 JB |
2483 | }; |
2484 | ||
5d4f98a2 YZ |
2485 | btrfs_set_node_blockptr(upper->eb, slot, |
2486 | node->eb->start); | |
2487 | btrfs_set_node_ptr_generation(upper->eb, slot, | |
2488 | trans->transid); | |
50564b65 | 2489 | btrfs_mark_buffer_dirty(trans, upper->eb); |
5d4f98a2 | 2490 | |
82fa113f | 2491 | btrfs_init_tree_ref(&ref, node->level, |
e094f480 | 2492 | btrfs_root_id(root), false); |
82fa113f | 2493 | ret = btrfs_inc_extent_ref(trans, &ref); |
eb6b7fb4 JB |
2494 | if (!ret) |
2495 | ret = btrfs_drop_subtree(trans, root, eb, | |
2496 | upper->eb); | |
2497 | if (ret) | |
2498 | btrfs_abort_transaction(trans, ret); | |
5d4f98a2 | 2499 | } |
3fd0a558 YZ |
2500 | next: |
2501 | if (!upper->pending) | |
b0fe7078 | 2502 | btrfs_backref_drop_node_buffer(upper); |
3fd0a558 | 2503 | else |
b0fe7078 | 2504 | btrfs_backref_unlock_node_buffer(upper); |
8df01fdd | 2505 | if (ret) |
3fd0a558 | 2506 | break; |
5d4f98a2 | 2507 | } |
3fd0a558 | 2508 | |
8df01fdd | 2509 | if (!ret && node->pending) { |
b0fe7078 | 2510 | btrfs_backref_drop_node_buffer(node); |
3fd0a558 YZ |
2511 | list_move_tail(&node->list, &rc->backref_cache.changed); |
2512 | node->pending = 0; | |
2513 | } | |
2514 | ||
5d4f98a2 | 2515 | path->lowest_level = 0; |
ffe30dd8 JB |
2516 | |
2517 | /* | |
2518 | * We should have allocated all of our space in the block rsv and thus | |
2519 | * shouldn't ENOSPC. | |
2520 | */ | |
2521 | ASSERT(ret != -ENOSPC); | |
8df01fdd | 2522 | return ret; |
5d4f98a2 YZ |
2523 | } |
2524 | ||
2525 | static int link_to_upper(struct btrfs_trans_handle *trans, | |
3fd0a558 | 2526 | struct reloc_control *rc, |
a26195a5 | 2527 | struct btrfs_backref_node *node, |
5d4f98a2 YZ |
2528 | struct btrfs_path *path) |
2529 | { | |
2530 | struct btrfs_key key; | |
5d4f98a2 YZ |
2531 | |
2532 | btrfs_node_key_to_cpu(node->eb, &key, 0); | |
3fd0a558 | 2533 | return do_relocation(trans, rc, node, &key, path, 0); |
5d4f98a2 YZ |
2534 | } |
2535 | ||
2536 | static int finish_pending_nodes(struct btrfs_trans_handle *trans, | |
3fd0a558 YZ |
2537 | struct reloc_control *rc, |
2538 | struct btrfs_path *path, int err) | |
5d4f98a2 | 2539 | { |
3fd0a558 | 2540 | LIST_HEAD(list); |
a26195a5 QW |
2541 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
2542 | struct btrfs_backref_node *node; | |
5d4f98a2 YZ |
2543 | int level; |
2544 | int ret; | |
5d4f98a2 YZ |
2545 | |
2546 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
2547 | while (!list_empty(&cache->pending[level])) { | |
2548 | node = list_entry(cache->pending[level].next, | |
a26195a5 | 2549 | struct btrfs_backref_node, list); |
3fd0a558 YZ |
2550 | list_move_tail(&node->list, &list); |
2551 | BUG_ON(!node->pending); | |
5d4f98a2 | 2552 | |
3fd0a558 YZ |
2553 | if (!err) { |
2554 | ret = link_to_upper(trans, rc, node, path); | |
2555 | if (ret < 0) | |
2556 | err = ret; | |
2557 | } | |
5d4f98a2 | 2558 | } |
3fd0a558 | 2559 | list_splice_init(&list, &cache->pending[level]); |
5d4f98a2 | 2560 | } |
5d4f98a2 YZ |
2561 | return err; |
2562 | } | |
2563 | ||
5d4f98a2 YZ |
2564 | /* |
2565 | * mark a block and all blocks directly/indirectly reference the block | |
2566 | * as processed. | |
2567 | */ | |
2568 | static void update_processed_blocks(struct reloc_control *rc, | |
a26195a5 | 2569 | struct btrfs_backref_node *node) |
5d4f98a2 | 2570 | { |
a26195a5 QW |
2571 | struct btrfs_backref_node *next = node; |
2572 | struct btrfs_backref_edge *edge; | |
2573 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
5d4f98a2 YZ |
2574 | int index = 0; |
2575 | ||
2576 | while (next) { | |
2577 | cond_resched(); | |
2578 | while (1) { | |
2579 | if (next->processed) | |
2580 | break; | |
2581 | ||
9569cc20 | 2582 | mark_block_processed(rc, next); |
5d4f98a2 YZ |
2583 | |
2584 | if (list_empty(&next->upper)) | |
2585 | break; | |
2586 | ||
2587 | edge = list_entry(next->upper.next, | |
a26195a5 | 2588 | struct btrfs_backref_edge, list[LOWER]); |
5d4f98a2 YZ |
2589 | edges[index++] = edge; |
2590 | next = edge->node[UPPER]; | |
2591 | } | |
2592 | next = walk_down_backref(edges, &index); | |
2593 | } | |
2594 | } | |
2595 | ||
7476dfda | 2596 | static int tree_block_processed(u64 bytenr, struct reloc_control *rc) |
3fd0a558 | 2597 | { |
da17066c | 2598 | u32 blocksize = rc->extent_root->fs_info->nodesize; |
7476dfda | 2599 | |
3fd0a558 | 2600 | if (test_range_bit(&rc->processed_blocks, bytenr, |
893fe243 | 2601 | bytenr + blocksize - 1, EXTENT_DIRTY, NULL)) |
3fd0a558 YZ |
2602 | return 1; |
2603 | return 0; | |
5d4f98a2 YZ |
2604 | } |
2605 | ||
2ff7e61e | 2606 | static int get_tree_block_key(struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2607 | struct tree_block *block) |
2608 | { | |
789d6a3a QW |
2609 | struct btrfs_tree_parent_check check = { |
2610 | .level = block->level, | |
2611 | .owner_root = block->owner, | |
2612 | .transid = block->key.offset | |
2613 | }; | |
5d4f98a2 YZ |
2614 | struct extent_buffer *eb; |
2615 | ||
789d6a3a | 2616 | eb = read_tree_block(fs_info, block->bytenr, &check); |
4eb150d6 | 2617 | if (IS_ERR(eb)) |
64c043de | 2618 | return PTR_ERR(eb); |
4eb150d6 | 2619 | if (!extent_buffer_uptodate(eb)) { |
416bc658 JB |
2620 | free_extent_buffer(eb); |
2621 | return -EIO; | |
2622 | } | |
5d4f98a2 YZ |
2623 | if (block->level == 0) |
2624 | btrfs_item_key_to_cpu(eb, &block->key, 0); | |
2625 | else | |
2626 | btrfs_node_key_to_cpu(eb, &block->key, 0); | |
2627 | free_extent_buffer(eb); | |
a3bb700f | 2628 | block->key_ready = true; |
5d4f98a2 YZ |
2629 | return 0; |
2630 | } | |
2631 | ||
5d4f98a2 YZ |
2632 | /* |
2633 | * helper function to relocate a tree block | |
2634 | */ | |
2635 | static int relocate_tree_block(struct btrfs_trans_handle *trans, | |
2636 | struct reloc_control *rc, | |
a26195a5 | 2637 | struct btrfs_backref_node *node, |
5d4f98a2 YZ |
2638 | struct btrfs_key *key, |
2639 | struct btrfs_path *path) | |
2640 | { | |
2641 | struct btrfs_root *root; | |
3fd0a558 YZ |
2642 | int ret = 0; |
2643 | ||
2644 | if (!node) | |
2645 | return 0; | |
5d4f98a2 | 2646 | |
5f6b2e5c JB |
2647 | /* |
2648 | * If we fail here we want to drop our backref_node because we are going | |
2649 | * to start over and regenerate the tree for it. | |
2650 | */ | |
2651 | ret = reserve_metadata_space(trans, rc, node); | |
2652 | if (ret) | |
2653 | goto out; | |
2654 | ||
3fd0a558 | 2655 | BUG_ON(node->processed); |
147d256e | 2656 | root = select_one_root(node); |
8717cf44 JB |
2657 | if (IS_ERR(root)) { |
2658 | ret = PTR_ERR(root); | |
2659 | ||
2660 | /* See explanation in select_one_root for the -EUCLEAN case. */ | |
2661 | ASSERT(ret == -ENOENT); | |
2662 | if (ret == -ENOENT) { | |
2663 | ret = 0; | |
2664 | update_processed_blocks(rc, node); | |
2665 | } | |
3fd0a558 | 2666 | goto out; |
5d4f98a2 YZ |
2667 | } |
2668 | ||
3fd0a558 | 2669 | if (root) { |
92a7cc42 | 2670 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
1c7bfa15 JB |
2671 | /* |
2672 | * This block was the root block of a root, and this is | |
2673 | * the first time we're processing the block and thus it | |
2674 | * should not have had the ->new_bytenr modified and | |
2675 | * should have not been included on the changed list. | |
2676 | * | |
2677 | * However in the case of corruption we could have | |
2678 | * multiple refs pointing to the same block improperly, | |
2679 | * and thus we would trip over these checks. ASSERT() | |
2680 | * for the developer case, because it could indicate a | |
2681 | * bug in the backref code, however error out for a | |
2682 | * normal user in the case of corruption. | |
2683 | */ | |
2684 | ASSERT(node->new_bytenr == 0); | |
2685 | ASSERT(list_empty(&node->list)); | |
2686 | if (node->new_bytenr || !list_empty(&node->list)) { | |
2687 | btrfs_err(root->fs_info, | |
2688 | "bytenr %llu has improper references to it", | |
2689 | node->bytenr); | |
2690 | ret = -EUCLEAN; | |
2691 | goto out; | |
2692 | } | |
d18c7bd9 JB |
2693 | ret = btrfs_record_root_in_trans(trans, root); |
2694 | if (ret) | |
2695 | goto out; | |
39200e59 JB |
2696 | /* |
2697 | * Another thread could have failed, need to check if we | |
2698 | * have reloc_root actually set. | |
2699 | */ | |
2700 | if (!root->reloc_root) { | |
2701 | ret = -ENOENT; | |
2702 | goto out; | |
2703 | } | |
3fd0a558 YZ |
2704 | root = root->reloc_root; |
2705 | node->new_bytenr = root->node->start; | |
00246528 JB |
2706 | btrfs_put_root(node->root); |
2707 | node->root = btrfs_grab_root(root); | |
0b530bc5 | 2708 | ASSERT(node->root); |
3fd0a558 YZ |
2709 | list_add_tail(&node->list, &rc->backref_cache.changed); |
2710 | } else { | |
2711 | path->lowest_level = node->level; | |
2bb2e00e FM |
2712 | if (root == root->fs_info->chunk_root) |
2713 | btrfs_reserve_chunk_metadata(trans, false); | |
3fd0a558 | 2714 | ret = btrfs_search_slot(trans, root, key, path, 0, 1); |
b3b4aa74 | 2715 | btrfs_release_path(path); |
2bb2e00e FM |
2716 | if (root == root->fs_info->chunk_root) |
2717 | btrfs_trans_release_chunk_metadata(trans); | |
3fd0a558 YZ |
2718 | if (ret > 0) |
2719 | ret = 0; | |
2720 | } | |
2721 | if (!ret) | |
2722 | update_processed_blocks(rc, node); | |
2723 | } else { | |
2724 | ret = do_relocation(trans, rc, node, key, path, 1); | |
2725 | } | |
5d4f98a2 | 2726 | out: |
0647bf56 | 2727 | if (ret || node->level == 0 || node->cowonly) |
023acb07 | 2728 | btrfs_backref_cleanup_node(&rc->backref_cache, node); |
5d4f98a2 YZ |
2729 | return ret; |
2730 | } | |
2731 | ||
2732 | /* | |
2733 | * relocate a list of blocks | |
2734 | */ | |
2735 | static noinline_for_stack | |
2736 | int relocate_tree_blocks(struct btrfs_trans_handle *trans, | |
2737 | struct reloc_control *rc, struct rb_root *blocks) | |
2738 | { | |
2ff7e61e | 2739 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
a26195a5 | 2740 | struct btrfs_backref_node *node; |
5d4f98a2 YZ |
2741 | struct btrfs_path *path; |
2742 | struct tree_block *block; | |
98ff7b94 | 2743 | struct tree_block *next; |
acde0e86 | 2744 | int ret = 0; |
5d4f98a2 YZ |
2745 | |
2746 | path = btrfs_alloc_path(); | |
e1a12670 | 2747 | if (!path) { |
acde0e86 | 2748 | ret = -ENOMEM; |
34c2b290 | 2749 | goto out_free_blocks; |
e1a12670 | 2750 | } |
5d4f98a2 | 2751 | |
98ff7b94 QW |
2752 | /* Kick in readahead for tree blocks with missing keys */ |
2753 | rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { | |
5d4f98a2 | 2754 | if (!block->key_ready) |
f7ba2d37 JB |
2755 | btrfs_readahead_tree_block(fs_info, block->bytenr, |
2756 | block->owner, 0, | |
3fbaf258 | 2757 | block->level); |
5d4f98a2 YZ |
2758 | } |
2759 | ||
98ff7b94 QW |
2760 | /* Get first keys */ |
2761 | rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { | |
34c2b290 | 2762 | if (!block->key_ready) { |
acde0e86 AJ |
2763 | ret = get_tree_block_key(fs_info, block); |
2764 | if (ret) | |
34c2b290 DS |
2765 | goto out_free_path; |
2766 | } | |
5d4f98a2 YZ |
2767 | } |
2768 | ||
98ff7b94 QW |
2769 | /* Do tree relocation */ |
2770 | rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { | |
eb96e221 | 2771 | node = build_backref_tree(trans, rc, &block->key, |
5d4f98a2 YZ |
2772 | block->level, block->bytenr); |
2773 | if (IS_ERR(node)) { | |
acde0e86 | 2774 | ret = PTR_ERR(node); |
5d4f98a2 YZ |
2775 | goto out; |
2776 | } | |
2777 | ||
2778 | ret = relocate_tree_block(trans, rc, node, &block->key, | |
2779 | path); | |
acde0e86 | 2780 | if (ret < 0) |
50dbbb71 | 2781 | break; |
5d4f98a2 | 2782 | } |
5d4f98a2 | 2783 | out: |
acde0e86 | 2784 | ret = finish_pending_nodes(trans, rc, path, ret); |
5d4f98a2 | 2785 | |
34c2b290 | 2786 | out_free_path: |
5d4f98a2 | 2787 | btrfs_free_path(path); |
34c2b290 | 2788 | out_free_blocks: |
e1a12670 | 2789 | free_block_list(blocks); |
acde0e86 | 2790 | return ret; |
5d4f98a2 YZ |
2791 | } |
2792 | ||
056d9bec NB |
2793 | static noinline_for_stack int prealloc_file_extent_cluster( |
2794 | struct btrfs_inode *inode, | |
ab7c8bbf | 2795 | const struct file_extent_cluster *cluster) |
efa56464 YZ |
2796 | { |
2797 | u64 alloc_hint = 0; | |
2798 | u64 start; | |
2799 | u64 end; | |
056d9bec | 2800 | u64 offset = inode->index_cnt; |
efa56464 | 2801 | u64 num_bytes; |
4e9d0d01 | 2802 | int nr; |
efa56464 | 2803 | int ret = 0; |
9d9ea1e6 | 2804 | u64 i_size = i_size_read(&inode->vfs_inode); |
dcb40c19 WX |
2805 | u64 prealloc_start = cluster->start - offset; |
2806 | u64 prealloc_end = cluster->end - offset; | |
214e61d0 | 2807 | u64 cur_offset = prealloc_start; |
efa56464 | 2808 | |
9d9ea1e6 QW |
2809 | /* |
2810 | * For subpage case, previous i_size may not be aligned to PAGE_SIZE. | |
2811 | * This means the range [i_size, PAGE_END + 1) is filled with zeros by | |
2812 | * btrfs_do_readpage() call of previously relocated file cluster. | |
2813 | * | |
2814 | * If the current cluster starts in the above range, btrfs_do_readpage() | |
a16c2c48 | 2815 | * will skip the read, and relocate_one_folio() will later writeback |
9d9ea1e6 QW |
2816 | * the padding zeros as new data, causing data corruption. |
2817 | * | |
2818 | * Here we have to manually invalidate the range (i_size, PAGE_END + 1). | |
2819 | */ | |
ce394a7f | 2820 | if (!PAGE_ALIGNED(i_size)) { |
9d9ea1e6 QW |
2821 | struct address_space *mapping = inode->vfs_inode.i_mapping; |
2822 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
2823 | const u32 sectorsize = fs_info->sectorsize; | |
8d6e5f9a | 2824 | struct folio *folio; |
9d9ea1e6 QW |
2825 | |
2826 | ASSERT(sectorsize < PAGE_SIZE); | |
2827 | ASSERT(IS_ALIGNED(i_size, sectorsize)); | |
2828 | ||
2829 | /* | |
2830 | * Subpage can't handle page with DIRTY but without UPTODATE | |
2831 | * bit as it can lead to the following deadlock: | |
2832 | * | |
fb12489b | 2833 | * btrfs_read_folio() |
9d9ea1e6 QW |
2834 | * | Page already *locked* |
2835 | * |- btrfs_lock_and_flush_ordered_range() | |
2836 | * |- btrfs_start_ordered_extent() | |
2837 | * |- extent_write_cache_pages() | |
2838 | * |- lock_page() | |
2839 | * We try to lock the page we already hold. | |
2840 | * | |
2841 | * Here we just writeback the whole data reloc inode, so that | |
2842 | * we will be ensured to have no dirty range in the page, and | |
2843 | * are safe to clear the uptodate bits. | |
2844 | * | |
2845 | * This shouldn't cause too much overhead, as we need to write | |
2846 | * the data back anyway. | |
2847 | */ | |
2848 | ret = filemap_write_and_wait(mapping); | |
2849 | if (ret < 0) | |
2850 | return ret; | |
2851 | ||
2852 | clear_extent_bits(&inode->io_tree, i_size, | |
2853 | round_up(i_size, PAGE_SIZE) - 1, | |
2854 | EXTENT_UPTODATE); | |
8d6e5f9a | 2855 | folio = filemap_lock_folio(mapping, i_size >> PAGE_SHIFT); |
9d9ea1e6 QW |
2856 | /* |
2857 | * If page is freed we don't need to do anything then, as we | |
2858 | * will re-read the whole page anyway. | |
2859 | */ | |
8d6e5f9a GR |
2860 | if (!IS_ERR(folio)) { |
2861 | btrfs_subpage_clear_uptodate(fs_info, folio, i_size, | |
9d9ea1e6 | 2862 | round_up(i_size, PAGE_SIZE) - i_size); |
8d6e5f9a GR |
2863 | folio_unlock(folio); |
2864 | folio_put(folio); | |
9d9ea1e6 QW |
2865 | } |
2866 | } | |
2867 | ||
efa56464 | 2868 | BUG_ON(cluster->start != cluster->boundary[0]); |
056d9bec | 2869 | ret = btrfs_alloc_data_chunk_ondemand(inode, |
a89ef455 | 2870 | prealloc_end + 1 - prealloc_start); |
efa56464 | 2871 | if (ret) |
214e61d0 | 2872 | return ret; |
efa56464 | 2873 | |
29b6352b | 2874 | btrfs_inode_lock(inode, 0); |
4e9d0d01 | 2875 | for (nr = 0; nr < cluster->nr; nr++) { |
9c5c9604 JB |
2876 | struct extent_state *cached_state = NULL; |
2877 | ||
efa56464 YZ |
2878 | start = cluster->boundary[nr] - offset; |
2879 | if (nr + 1 < cluster->nr) | |
2880 | end = cluster->boundary[nr + 1] - 1 - offset; | |
2881 | else | |
2882 | end = cluster->end - offset; | |
2883 | ||
9c5c9604 | 2884 | lock_extent(&inode->io_tree, start, end, &cached_state); |
efa56464 | 2885 | num_bytes = end + 1 - start; |
056d9bec | 2886 | ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, |
efa56464 YZ |
2887 | num_bytes, num_bytes, |
2888 | end + 1, &alloc_hint); | |
18513091 | 2889 | cur_offset = end + 1; |
9c5c9604 | 2890 | unlock_extent(&inode->io_tree, start, end, &cached_state); |
efa56464 YZ |
2891 | if (ret) |
2892 | break; | |
efa56464 | 2893 | } |
e5d4d75b | 2894 | btrfs_inode_unlock(inode, 0); |
214e61d0 | 2895 | |
18513091 | 2896 | if (cur_offset < prealloc_end) |
056d9bec | 2897 | btrfs_free_reserved_data_space_noquota(inode->root->fs_info, |
a89ef455 | 2898 | prealloc_end + 1 - cur_offset); |
efa56464 YZ |
2899 | return ret; |
2900 | } | |
2901 | ||
4b01c44f JT |
2902 | static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode, |
2903 | u64 start, u64 end, u64 block_start) | |
0257bb82 | 2904 | { |
0257bb82 | 2905 | struct extent_map *em; |
9c5c9604 | 2906 | struct extent_state *cached_state = NULL; |
0257bb82 YZ |
2907 | int ret = 0; |
2908 | ||
172ddd60 | 2909 | em = alloc_extent_map(); |
0257bb82 YZ |
2910 | if (!em) |
2911 | return -ENOMEM; | |
2912 | ||
2913 | em->start = start; | |
2914 | em->len = end + 1 - start; | |
2915 | em->block_len = em->len; | |
2916 | em->block_start = block_start; | |
f86f7a75 | 2917 | em->flags |= EXTENT_FLAG_PINNED; |
0257bb82 | 2918 | |
9c5c9604 | 2919 | lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); |
a1ba4c08 | 2920 | ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false); |
9c5c9604 | 2921 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); |
a1ba4c08 FM |
2922 | free_extent_map(em); |
2923 | ||
0257bb82 YZ |
2924 | return ret; |
2925 | } | |
2926 | ||
726a3421 | 2927 | /* |
907d2710 | 2928 | * Allow error injection to test balance/relocation cancellation |
726a3421 | 2929 | */ |
ab7c8bbf | 2930 | noinline int btrfs_should_cancel_balance(const struct btrfs_fs_info *fs_info) |
726a3421 | 2931 | { |
5cb502f4 | 2932 | return atomic_read(&fs_info->balance_cancel_req) || |
907d2710 | 2933 | atomic_read(&fs_info->reloc_cancel_req) || |
5cb502f4 | 2934 | fatal_signal_pending(current); |
726a3421 QW |
2935 | } |
2936 | ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); | |
2937 | ||
ab7c8bbf | 2938 | static u64 get_cluster_boundary_end(const struct file_extent_cluster *cluster, |
c2832898 QW |
2939 | int cluster_nr) |
2940 | { | |
2941 | /* Last extent, use cluster end directly */ | |
2942 | if (cluster_nr >= cluster->nr - 1) | |
2943 | return cluster->end; | |
2944 | ||
2945 | /* Use next boundary start*/ | |
2946 | return cluster->boundary[cluster_nr + 1] - 1; | |
2947 | } | |
2948 | ||
a16c2c48 GR |
2949 | static int relocate_one_folio(struct inode *inode, struct file_ra_state *ra, |
2950 | const struct file_extent_cluster *cluster, | |
2951 | int *cluster_nr, unsigned long index) | |
5d4f98a2 | 2952 | { |
41044b41 | 2953 | struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); |
f47960f4 QW |
2954 | u64 offset = BTRFS_I(inode)->index_cnt; |
2955 | const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; | |
2956 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
a16c2c48 GR |
2957 | struct folio *folio; |
2958 | u64 folio_start; | |
2959 | u64 folio_end; | |
c2832898 | 2960 | u64 cur; |
f47960f4 QW |
2961 | int ret; |
2962 | ||
a16c2c48 GR |
2963 | ASSERT(index <= last_index); |
2964 | folio = filemap_lock_folio(inode->i_mapping, index); | |
2965 | if (IS_ERR(folio)) { | |
f47960f4 | 2966 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, |
a16c2c48 GR |
2967 | index, last_index + 1 - index); |
2968 | folio = __filemap_get_folio(inode->i_mapping, index, | |
2969 | FGP_LOCK | FGP_ACCESSED | FGP_CREAT, mask); | |
2970 | if (IS_ERR(folio)) | |
2971 | return PTR_ERR(folio); | |
f47960f4 | 2972 | } |
f47960f4 | 2973 | |
a16c2c48 GR |
2974 | WARN_ON(folio_order(folio)); |
2975 | ||
2976 | if (folio_test_readahead(folio)) | |
2ebdd1df | 2977 | page_cache_async_readahead(inode->i_mapping, ra, NULL, |
a16c2c48 GR |
2978 | folio, index, |
2979 | last_index + 1 - index); | |
f47960f4 | 2980 | |
a16c2c48 GR |
2981 | if (!folio_test_uptodate(folio)) { |
2982 | btrfs_read_folio(NULL, folio); | |
2983 | folio_lock(folio); | |
2984 | if (!folio_test_uptodate(folio)) { | |
f47960f4 | 2985 | ret = -EIO; |
a16c2c48 | 2986 | goto release_folio; |
f47960f4 QW |
2987 | } |
2988 | } | |
2989 | ||
e7f1326c | 2990 | /* |
a16c2c48 GR |
2991 | * We could have lost folio private when we dropped the lock to read the |
2992 | * folio above, make sure we set_page_extent_mapped here so we have any | |
e7f1326c JB |
2993 | * of the subpage blocksize stuff we need in place. |
2994 | */ | |
a16c2c48 | 2995 | ret = set_folio_extent_mapped(folio); |
e7f1326c | 2996 | if (ret < 0) |
a16c2c48 | 2997 | goto release_folio; |
e7f1326c | 2998 | |
a16c2c48 GR |
2999 | folio_start = folio_pos(folio); |
3000 | folio_end = folio_start + PAGE_SIZE - 1; | |
f47960f4 | 3001 | |
c2832898 QW |
3002 | /* |
3003 | * Start from the cluster, as for subpage case, the cluster can start | |
a16c2c48 | 3004 | * inside the folio. |
c2832898 | 3005 | */ |
a16c2c48 GR |
3006 | cur = max(folio_start, cluster->boundary[*cluster_nr] - offset); |
3007 | while (cur <= folio_end) { | |
9c5c9604 | 3008 | struct extent_state *cached_state = NULL; |
c2832898 QW |
3009 | u64 extent_start = cluster->boundary[*cluster_nr] - offset; |
3010 | u64 extent_end = get_cluster_boundary_end(cluster, | |
3011 | *cluster_nr) - offset; | |
a16c2c48 GR |
3012 | u64 clamped_start = max(folio_start, extent_start); |
3013 | u64 clamped_end = min(folio_end, extent_end); | |
c2832898 QW |
3014 | u32 clamped_len = clamped_end + 1 - clamped_start; |
3015 | ||
3016 | /* Reserve metadata for this range */ | |
3017 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), | |
d4135134 FM |
3018 | clamped_len, clamped_len, |
3019 | false); | |
c2832898 | 3020 | if (ret) |
a16c2c48 | 3021 | goto release_folio; |
f47960f4 | 3022 | |
c2832898 | 3023 | /* Mark the range delalloc and dirty for later writeback */ |
9c5c9604 JB |
3024 | lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, |
3025 | &cached_state); | |
c2832898 | 3026 | ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, |
9c5c9604 | 3027 | clamped_end, 0, &cached_state); |
c2832898 | 3028 | if (ret) { |
9c5c9604 JB |
3029 | clear_extent_bit(&BTRFS_I(inode)->io_tree, |
3030 | clamped_start, clamped_end, | |
3031 | EXTENT_LOCKED | EXTENT_BOUNDARY, | |
3032 | &cached_state); | |
c2832898 QW |
3033 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
3034 | clamped_len, true); | |
3035 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
3036 | clamped_len); | |
a16c2c48 | 3037 | goto release_folio; |
c2832898 | 3038 | } |
a16c2c48 | 3039 | btrfs_folio_set_dirty(fs_info, folio, clamped_start, clamped_len); |
f47960f4 | 3040 | |
c2832898 | 3041 | /* |
a16c2c48 | 3042 | * Set the boundary if it's inside the folio. |
c2832898 QW |
3043 | * Data relocation requires the destination extents to have the |
3044 | * same size as the source. | |
3045 | * EXTENT_BOUNDARY bit prevents current extent from being merged | |
3046 | * with previous extent. | |
3047 | */ | |
a16c2c48 | 3048 | if (in_range(cluster->boundary[*cluster_nr] - offset, folio_start, PAGE_SIZE)) { |
c2832898 QW |
3049 | u64 boundary_start = cluster->boundary[*cluster_nr] - |
3050 | offset; | |
3051 | u64 boundary_end = boundary_start + | |
3052 | fs_info->sectorsize - 1; | |
3053 | ||
0acd32c2 DS |
3054 | set_extent_bit(&BTRFS_I(inode)->io_tree, |
3055 | boundary_start, boundary_end, | |
1d126800 | 3056 | EXTENT_BOUNDARY, NULL); |
c2832898 | 3057 | } |
9c5c9604 JB |
3058 | unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, |
3059 | &cached_state); | |
c2832898 QW |
3060 | btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len); |
3061 | cur += clamped_len; | |
3062 | ||
3063 | /* Crossed extent end, go to next extent */ | |
3064 | if (cur >= extent_end) { | |
3065 | (*cluster_nr)++; | |
3066 | /* Just finished the last extent of the cluster, exit. */ | |
3067 | if (*cluster_nr >= cluster->nr) | |
3068 | break; | |
3069 | } | |
f47960f4 | 3070 | } |
a16c2c48 GR |
3071 | folio_unlock(folio); |
3072 | folio_put(folio); | |
f47960f4 | 3073 | |
f47960f4 QW |
3074 | balance_dirty_pages_ratelimited(inode->i_mapping); |
3075 | btrfs_throttle(fs_info); | |
3076 | if (btrfs_should_cancel_balance(fs_info)) | |
3077 | ret = -ECANCELED; | |
3078 | return ret; | |
3079 | ||
a16c2c48 GR |
3080 | release_folio: |
3081 | folio_unlock(folio); | |
3082 | folio_put(folio); | |
f47960f4 QW |
3083 | return ret; |
3084 | } | |
3085 | ||
3086 | static int relocate_file_extent_cluster(struct inode *inode, | |
ab7c8bbf | 3087 | const struct file_extent_cluster *cluster) |
f47960f4 | 3088 | { |
0257bb82 YZ |
3089 | u64 offset = BTRFS_I(inode)->index_cnt; |
3090 | unsigned long index; | |
5d4f98a2 | 3091 | unsigned long last_index; |
5d4f98a2 | 3092 | struct file_ra_state *ra; |
f47960f4 | 3093 | int cluster_nr = 0; |
5d4f98a2 YZ |
3094 | int ret = 0; |
3095 | ||
0257bb82 YZ |
3096 | if (!cluster->nr) |
3097 | return 0; | |
3098 | ||
5d4f98a2 YZ |
3099 | ra = kzalloc(sizeof(*ra), GFP_NOFS); |
3100 | if (!ra) | |
3101 | return -ENOMEM; | |
3102 | ||
056d9bec | 3103 | ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster); |
efa56464 YZ |
3104 | if (ret) |
3105 | goto out; | |
0257bb82 | 3106 | |
efa56464 | 3107 | file_ra_state_init(ra, inode->i_mapping); |
5d4f98a2 | 3108 | |
4b01c44f | 3109 | ret = setup_relocation_extent_mapping(inode, cluster->start - offset, |
0257bb82 | 3110 | cluster->end - offset, cluster->start); |
5d4f98a2 | 3111 | if (ret) |
efa56464 | 3112 | goto out; |
5d4f98a2 | 3113 | |
09cbfeaf | 3114 | last_index = (cluster->end - offset) >> PAGE_SHIFT; |
f47960f4 QW |
3115 | for (index = (cluster->start - offset) >> PAGE_SHIFT; |
3116 | index <= last_index && !ret; index++) | |
a16c2c48 | 3117 | ret = relocate_one_folio(inode, ra, cluster, &cluster_nr, index); |
f47960f4 QW |
3118 | if (ret == 0) |
3119 | WARN_ON(cluster_nr != cluster->nr); | |
efa56464 | 3120 | out: |
5d4f98a2 | 3121 | kfree(ra); |
5d4f98a2 YZ |
3122 | return ret; |
3123 | } | |
3124 | ||
ab7c8bbf DS |
3125 | static noinline_for_stack int relocate_data_extent(struct inode *inode, |
3126 | const struct btrfs_key *extent_key, | |
3127 | struct file_extent_cluster *cluster) | |
5d4f98a2 | 3128 | { |
0257bb82 | 3129 | int ret; |
2672a051 | 3130 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5d4f98a2 | 3131 | |
0257bb82 YZ |
3132 | if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { |
3133 | ret = relocate_file_extent_cluster(inode, cluster); | |
3134 | if (ret) | |
3135 | return ret; | |
3136 | cluster->nr = 0; | |
5d4f98a2 | 3137 | } |
5d4f98a2 | 3138 | |
2672a051 BB |
3139 | /* |
3140 | * Under simple quotas, we set root->relocation_src_root when we find | |
3141 | * the extent. If adjacent extents have different owners, we can't merge | |
3142 | * them while relocating. Handle this by storing the owning root that | |
3143 | * started a cluster and if we see an extent from a different root break | |
3144 | * cluster formation (just like the above case of non-adjacent extents). | |
3145 | * | |
3146 | * Without simple quotas, relocation_src_root is always 0, so we should | |
3147 | * never see a mismatch, and it should have no effect on relocation | |
3148 | * clusters. | |
3149 | */ | |
3150 | if (cluster->nr > 0 && cluster->owning_root != root->relocation_src_root) { | |
3151 | u64 tmp = root->relocation_src_root; | |
3152 | ||
3153 | /* | |
3154 | * root->relocation_src_root is the state that actually affects | |
3155 | * the preallocation we do here, so set it to the root owning | |
3156 | * the cluster we need to relocate. | |
3157 | */ | |
3158 | root->relocation_src_root = cluster->owning_root; | |
3159 | ret = relocate_file_extent_cluster(inode, cluster); | |
3160 | if (ret) | |
3161 | return ret; | |
3162 | cluster->nr = 0; | |
3163 | /* And reset it back for the current extent's owning root. */ | |
3164 | root->relocation_src_root = tmp; | |
3165 | } | |
3166 | ||
3167 | if (!cluster->nr) { | |
0257bb82 | 3168 | cluster->start = extent_key->objectid; |
2672a051 BB |
3169 | cluster->owning_root = root->relocation_src_root; |
3170 | } | |
0257bb82 YZ |
3171 | else |
3172 | BUG_ON(cluster->nr >= MAX_EXTENTS); | |
3173 | cluster->end = extent_key->objectid + extent_key->offset - 1; | |
3174 | cluster->boundary[cluster->nr] = extent_key->objectid; | |
3175 | cluster->nr++; | |
3176 | ||
3177 | if (cluster->nr >= MAX_EXTENTS) { | |
3178 | ret = relocate_file_extent_cluster(inode, cluster); | |
3179 | if (ret) | |
3180 | return ret; | |
3181 | cluster->nr = 0; | |
3182 | } | |
3183 | return 0; | |
5d4f98a2 YZ |
3184 | } |
3185 | ||
5d4f98a2 YZ |
3186 | /* |
3187 | * helper to add a tree block to the list. | |
3188 | * the major work is getting the generation and level of the block | |
3189 | */ | |
3190 | static int add_tree_block(struct reloc_control *rc, | |
ab7c8bbf | 3191 | const struct btrfs_key *extent_key, |
5d4f98a2 YZ |
3192 | struct btrfs_path *path, |
3193 | struct rb_root *blocks) | |
3194 | { | |
3195 | struct extent_buffer *eb; | |
3196 | struct btrfs_extent_item *ei; | |
3197 | struct btrfs_tree_block_info *bi; | |
3198 | struct tree_block *block; | |
3199 | struct rb_node *rb_node; | |
3200 | u32 item_size; | |
3201 | int level = -1; | |
7fdf4b60 | 3202 | u64 generation; |
f7ba2d37 | 3203 | u64 owner = 0; |
5d4f98a2 YZ |
3204 | |
3205 | eb = path->nodes[0]; | |
3212fa14 | 3206 | item_size = btrfs_item_size(eb, path->slots[0]); |
5d4f98a2 | 3207 | |
3173a18f JB |
3208 | if (extent_key->type == BTRFS_METADATA_ITEM_KEY || |
3209 | item_size >= sizeof(*ei) + sizeof(*bi)) { | |
f7ba2d37 JB |
3210 | unsigned long ptr = 0, end; |
3211 | ||
5d4f98a2 YZ |
3212 | ei = btrfs_item_ptr(eb, path->slots[0], |
3213 | struct btrfs_extent_item); | |
f7ba2d37 | 3214 | end = (unsigned long)ei + item_size; |
3173a18f JB |
3215 | if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { |
3216 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3217 | level = btrfs_tree_block_level(eb, bi); | |
f7ba2d37 | 3218 | ptr = (unsigned long)(bi + 1); |
3173a18f JB |
3219 | } else { |
3220 | level = (int)extent_key->offset; | |
f7ba2d37 | 3221 | ptr = (unsigned long)(ei + 1); |
3173a18f | 3222 | } |
5d4f98a2 | 3223 | generation = btrfs_extent_generation(eb, ei); |
f7ba2d37 JB |
3224 | |
3225 | /* | |
3226 | * We're reading random blocks without knowing their owner ahead | |
3227 | * of time. This is ok most of the time, as all reloc roots and | |
3228 | * fs roots have the same lock type. However normal trees do | |
3229 | * not, and the only way to know ahead of time is to read the | |
3230 | * inline ref offset. We know it's an fs root if | |
3231 | * | |
3232 | * 1. There's more than one ref. | |
3233 | * 2. There's a SHARED_DATA_REF_KEY set. | |
3234 | * 3. FULL_BACKREF is set on the flags. | |
3235 | * | |
3236 | * Otherwise it's safe to assume that the ref offset == the | |
3237 | * owner of this block, so we can use that when calling | |
3238 | * read_tree_block. | |
3239 | */ | |
3240 | if (btrfs_extent_refs(eb, ei) == 1 && | |
3241 | !(btrfs_extent_flags(eb, ei) & | |
3242 | BTRFS_BLOCK_FLAG_FULL_BACKREF) && | |
3243 | ptr < end) { | |
3244 | struct btrfs_extent_inline_ref *iref; | |
3245 | int type; | |
3246 | ||
3247 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3248 | type = btrfs_get_extent_inline_ref_type(eb, iref, | |
3249 | BTRFS_REF_TYPE_BLOCK); | |
3250 | if (type == BTRFS_REF_TYPE_INVALID) | |
3251 | return -EINVAL; | |
3252 | if (type == BTRFS_TREE_BLOCK_REF_KEY) | |
3253 | owner = btrfs_extent_inline_ref_offset(eb, iref); | |
3254 | } | |
5d4f98a2 | 3255 | } else { |
182741d2 QW |
3256 | btrfs_print_leaf(eb); |
3257 | btrfs_err(rc->block_group->fs_info, | |
3258 | "unrecognized tree backref at tree block %llu slot %u", | |
3259 | eb->start, path->slots[0]); | |
3260 | btrfs_release_path(path); | |
3261 | return -EUCLEAN; | |
5d4f98a2 YZ |
3262 | } |
3263 | ||
b3b4aa74 | 3264 | btrfs_release_path(path); |
5d4f98a2 YZ |
3265 | |
3266 | BUG_ON(level == -1); | |
3267 | ||
3268 | block = kmalloc(sizeof(*block), GFP_NOFS); | |
3269 | if (!block) | |
3270 | return -ENOMEM; | |
3271 | ||
3272 | block->bytenr = extent_key->objectid; | |
da17066c | 3273 | block->key.objectid = rc->extent_root->fs_info->nodesize; |
5d4f98a2 YZ |
3274 | block->key.offset = generation; |
3275 | block->level = level; | |
a3bb700f | 3276 | block->key_ready = false; |
f7ba2d37 | 3277 | block->owner = owner; |
5d4f98a2 | 3278 | |
e9a28dc5 | 3279 | rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); |
43c04fb1 | 3280 | if (rb_node) |
982c92cb QW |
3281 | btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, |
3282 | -EEXIST); | |
5d4f98a2 YZ |
3283 | |
3284 | return 0; | |
3285 | } | |
3286 | ||
3287 | /* | |
3288 | * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY | |
3289 | */ | |
3290 | static int __add_tree_block(struct reloc_control *rc, | |
3291 | u64 bytenr, u32 blocksize, | |
3292 | struct rb_root *blocks) | |
3293 | { | |
0b246afa | 3294 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3295 | struct btrfs_path *path; |
3296 | struct btrfs_key key; | |
3297 | int ret; | |
0b246afa | 3298 | bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
5d4f98a2 | 3299 | |
7476dfda | 3300 | if (tree_block_processed(bytenr, rc)) |
5d4f98a2 YZ |
3301 | return 0; |
3302 | ||
e9a28dc5 | 3303 | if (rb_simple_search(blocks, bytenr)) |
5d4f98a2 YZ |
3304 | return 0; |
3305 | ||
3306 | path = btrfs_alloc_path(); | |
3307 | if (!path) | |
3308 | return -ENOMEM; | |
aee68ee5 | 3309 | again: |
5d4f98a2 | 3310 | key.objectid = bytenr; |
aee68ee5 JB |
3311 | if (skinny) { |
3312 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3313 | key.offset = (u64)-1; | |
3314 | } else { | |
3315 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3316 | key.offset = blocksize; | |
3317 | } | |
5d4f98a2 YZ |
3318 | |
3319 | path->search_commit_root = 1; | |
3320 | path->skip_locking = 1; | |
3321 | ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); | |
3322 | if (ret < 0) | |
3323 | goto out; | |
5d4f98a2 | 3324 | |
aee68ee5 JB |
3325 | if (ret > 0 && skinny) { |
3326 | if (path->slots[0]) { | |
3327 | path->slots[0]--; | |
3328 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3329 | path->slots[0]); | |
3330 | if (key.objectid == bytenr && | |
3331 | (key.type == BTRFS_METADATA_ITEM_KEY || | |
3332 | (key.type == BTRFS_EXTENT_ITEM_KEY && | |
3333 | key.offset == blocksize))) | |
3334 | ret = 0; | |
3335 | } | |
3336 | ||
3337 | if (ret) { | |
3338 | skinny = false; | |
3339 | btrfs_release_path(path); | |
3340 | goto again; | |
3341 | } | |
3173a18f | 3342 | } |
cdccee99 LB |
3343 | if (ret) { |
3344 | ASSERT(ret == 1); | |
3345 | btrfs_print_leaf(path->nodes[0]); | |
3346 | btrfs_err(fs_info, | |
3347 | "tree block extent item (%llu) is not found in extent tree", | |
3348 | bytenr); | |
3349 | WARN_ON(1); | |
3350 | ret = -EINVAL; | |
3351 | goto out; | |
3352 | } | |
3173a18f | 3353 | |
5d4f98a2 YZ |
3354 | ret = add_tree_block(rc, &key, path, blocks); |
3355 | out: | |
3356 | btrfs_free_path(path); | |
3357 | return ret; | |
3358 | } | |
3359 | ||
0af3d00b | 3360 | static int delete_block_group_cache(struct btrfs_fs_info *fs_info, |
32da5386 | 3361 | struct btrfs_block_group *block_group, |
1bbc621e CM |
3362 | struct inode *inode, |
3363 | u64 ino) | |
0af3d00b | 3364 | { |
0af3d00b JB |
3365 | struct btrfs_root *root = fs_info->tree_root; |
3366 | struct btrfs_trans_handle *trans; | |
0af3d00b JB |
3367 | int ret = 0; |
3368 | ||
3369 | if (inode) | |
3370 | goto truncate; | |
3371 | ||
0202e83f | 3372 | inode = btrfs_iget(fs_info->sb, ino, root); |
2e19f1f9 | 3373 | if (IS_ERR(inode)) |
0af3d00b | 3374 | return -ENOENT; |
0af3d00b JB |
3375 | |
3376 | truncate: | |
2ff7e61e | 3377 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
7b61cd92 MX |
3378 | &fs_info->global_block_rsv); |
3379 | if (ret) | |
3380 | goto out; | |
3381 | ||
7a7eaa40 | 3382 | trans = btrfs_join_transaction(root); |
0af3d00b | 3383 | if (IS_ERR(trans)) { |
3612b495 | 3384 | ret = PTR_ERR(trans); |
0af3d00b JB |
3385 | goto out; |
3386 | } | |
3387 | ||
77ab86bf | 3388 | ret = btrfs_truncate_free_space_cache(trans, block_group, inode); |
0af3d00b | 3389 | |
3a45bb20 | 3390 | btrfs_end_transaction(trans); |
2ff7e61e | 3391 | btrfs_btree_balance_dirty(fs_info); |
0af3d00b JB |
3392 | out: |
3393 | iput(inode); | |
3394 | return ret; | |
3395 | } | |
3396 | ||
5d4f98a2 | 3397 | /* |
19b546d7 QW |
3398 | * Locate the free space cache EXTENT_DATA in root tree leaf and delete the |
3399 | * cache inode, to avoid free space cache data extent blocking data relocation. | |
5d4f98a2 | 3400 | */ |
19b546d7 QW |
3401 | static int delete_v1_space_cache(struct extent_buffer *leaf, |
3402 | struct btrfs_block_group *block_group, | |
3403 | u64 data_bytenr) | |
5d4f98a2 | 3404 | { |
19b546d7 QW |
3405 | u64 space_cache_ino; |
3406 | struct btrfs_file_extent_item *ei; | |
5d4f98a2 | 3407 | struct btrfs_key key; |
19b546d7 QW |
3408 | bool found = false; |
3409 | int i; | |
5d4f98a2 YZ |
3410 | int ret; |
3411 | ||
19b546d7 QW |
3412 | if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) |
3413 | return 0; | |
5d4f98a2 | 3414 | |
19b546d7 | 3415 | for (i = 0; i < btrfs_header_nritems(leaf); i++) { |
50e31ef4 QW |
3416 | u8 type; |
3417 | ||
19b546d7 QW |
3418 | btrfs_item_key_to_cpu(leaf, &key, i); |
3419 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
3420 | continue; | |
3421 | ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
50e31ef4 QW |
3422 | type = btrfs_file_extent_type(leaf, ei); |
3423 | ||
3424 | if ((type == BTRFS_FILE_EXTENT_REG || | |
3425 | type == BTRFS_FILE_EXTENT_PREALLOC) && | |
19b546d7 QW |
3426 | btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { |
3427 | found = true; | |
3428 | space_cache_ino = key.objectid; | |
5d4f98a2 | 3429 | break; |
5d4f98a2 | 3430 | } |
5d4f98a2 | 3431 | } |
19b546d7 QW |
3432 | if (!found) |
3433 | return -ENOENT; | |
3434 | ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, | |
3435 | space_cache_ino); | |
3436 | return ret; | |
5d4f98a2 YZ |
3437 | } |
3438 | ||
3439 | /* | |
2c016dc2 | 3440 | * helper to find all tree blocks that reference a given data extent |
5d4f98a2 | 3441 | */ |
ab7c8bbf DS |
3442 | static noinline_for_stack int add_data_references(struct reloc_control *rc, |
3443 | const struct btrfs_key *extent_key, | |
3444 | struct btrfs_path *path, | |
3445 | struct rb_root *blocks) | |
5d4f98a2 | 3446 | { |
a2c8d27e | 3447 | struct btrfs_backref_walk_ctx ctx = { 0 }; |
19b546d7 QW |
3448 | struct ulist_iterator leaf_uiter; |
3449 | struct ulist_node *ref_node = NULL; | |
a2c8d27e | 3450 | const u32 blocksize = rc->extent_root->fs_info->nodesize; |
647f63bd | 3451 | int ret = 0; |
5d4f98a2 | 3452 | |
19b546d7 | 3453 | btrfs_release_path(path); |
a2c8d27e FM |
3454 | |
3455 | ctx.bytenr = extent_key->objectid; | |
0cad8f14 | 3456 | ctx.skip_inode_ref_list = true; |
a2c8d27e FM |
3457 | ctx.fs_info = rc->extent_root->fs_info; |
3458 | ||
3459 | ret = btrfs_find_all_leafs(&ctx); | |
19b546d7 QW |
3460 | if (ret < 0) |
3461 | return ret; | |
5d4f98a2 | 3462 | |
19b546d7 | 3463 | ULIST_ITER_INIT(&leaf_uiter); |
a2c8d27e | 3464 | while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) { |
789d6a3a | 3465 | struct btrfs_tree_parent_check check = { 0 }; |
19b546d7 | 3466 | struct extent_buffer *eb; |
5d4f98a2 | 3467 | |
789d6a3a | 3468 | eb = read_tree_block(ctx.fs_info, ref_node->val, &check); |
19b546d7 QW |
3469 | if (IS_ERR(eb)) { |
3470 | ret = PTR_ERR(eb); | |
5d4f98a2 | 3471 | break; |
5d4f98a2 | 3472 | } |
19b546d7 QW |
3473 | ret = delete_v1_space_cache(eb, rc->block_group, |
3474 | extent_key->objectid); | |
3475 | free_extent_buffer(eb); | |
3476 | if (ret < 0) | |
3477 | break; | |
3478 | ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); | |
3479 | if (ret < 0) | |
5d4f98a2 | 3480 | break; |
5d4f98a2 | 3481 | } |
19b546d7 | 3482 | if (ret < 0) |
5d4f98a2 | 3483 | free_block_list(blocks); |
a2c8d27e | 3484 | ulist_free(ctx.refs); |
19b546d7 | 3485 | return ret; |
5d4f98a2 YZ |
3486 | } |
3487 | ||
3488 | /* | |
2c016dc2 | 3489 | * helper to find next unprocessed extent |
5d4f98a2 YZ |
3490 | */ |
3491 | static noinline_for_stack | |
147d256e | 3492 | int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, |
3fd0a558 | 3493 | struct btrfs_key *extent_key) |
5d4f98a2 | 3494 | { |
0b246afa | 3495 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3496 | struct btrfs_key key; |
3497 | struct extent_buffer *leaf; | |
3498 | u64 start, end, last; | |
3499 | int ret; | |
3500 | ||
b3470b5d | 3501 | last = rc->block_group->start + rc->block_group->length; |
5d4f98a2 | 3502 | while (1) { |
e5860f82 FM |
3503 | bool block_found; |
3504 | ||
5d4f98a2 YZ |
3505 | cond_resched(); |
3506 | if (rc->search_start >= last) { | |
3507 | ret = 1; | |
3508 | break; | |
3509 | } | |
3510 | ||
3511 | key.objectid = rc->search_start; | |
3512 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3513 | key.offset = 0; | |
3514 | ||
3515 | path->search_commit_root = 1; | |
3516 | path->skip_locking = 1; | |
3517 | ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, | |
3518 | 0, 0); | |
3519 | if (ret < 0) | |
3520 | break; | |
3521 | next: | |
3522 | leaf = path->nodes[0]; | |
3523 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3524 | ret = btrfs_next_leaf(rc->extent_root, path); | |
3525 | if (ret != 0) | |
3526 | break; | |
3527 | leaf = path->nodes[0]; | |
3528 | } | |
3529 | ||
3530 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3531 | if (key.objectid >= last) { | |
3532 | ret = 1; | |
3533 | break; | |
3534 | } | |
3535 | ||
3173a18f JB |
3536 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
3537 | key.type != BTRFS_METADATA_ITEM_KEY) { | |
3538 | path->slots[0]++; | |
3539 | goto next; | |
3540 | } | |
3541 | ||
3542 | if (key.type == BTRFS_EXTENT_ITEM_KEY && | |
5d4f98a2 YZ |
3543 | key.objectid + key.offset <= rc->search_start) { |
3544 | path->slots[0]++; | |
3545 | goto next; | |
3546 | } | |
3547 | ||
3173a18f | 3548 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
0b246afa | 3549 | key.objectid + fs_info->nodesize <= |
3173a18f JB |
3550 | rc->search_start) { |
3551 | path->slots[0]++; | |
3552 | goto next; | |
3553 | } | |
3554 | ||
e5860f82 FM |
3555 | block_found = find_first_extent_bit(&rc->processed_blocks, |
3556 | key.objectid, &start, &end, | |
3557 | EXTENT_DIRTY, NULL); | |
5d4f98a2 | 3558 | |
e5860f82 | 3559 | if (block_found && start <= key.objectid) { |
b3b4aa74 | 3560 | btrfs_release_path(path); |
5d4f98a2 YZ |
3561 | rc->search_start = end + 1; |
3562 | } else { | |
3173a18f JB |
3563 | if (key.type == BTRFS_EXTENT_ITEM_KEY) |
3564 | rc->search_start = key.objectid + key.offset; | |
3565 | else | |
3566 | rc->search_start = key.objectid + | |
0b246afa | 3567 | fs_info->nodesize; |
3fd0a558 | 3568 | memcpy(extent_key, &key, sizeof(key)); |
5d4f98a2 YZ |
3569 | return 0; |
3570 | } | |
3571 | } | |
b3b4aa74 | 3572 | btrfs_release_path(path); |
5d4f98a2 YZ |
3573 | return ret; |
3574 | } | |
3575 | ||
3576 | static void set_reloc_control(struct reloc_control *rc) | |
3577 | { | |
3578 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; | |
7585717f CM |
3579 | |
3580 | mutex_lock(&fs_info->reloc_mutex); | |
5d4f98a2 | 3581 | fs_info->reloc_ctl = rc; |
7585717f | 3582 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 YZ |
3583 | } |
3584 | ||
3585 | static void unset_reloc_control(struct reloc_control *rc) | |
3586 | { | |
3587 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; | |
7585717f CM |
3588 | |
3589 | mutex_lock(&fs_info->reloc_mutex); | |
5d4f98a2 | 3590 | fs_info->reloc_ctl = NULL; |
7585717f | 3591 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 YZ |
3592 | } |
3593 | ||
3fd0a558 YZ |
3594 | static noinline_for_stack |
3595 | int prepare_to_relocate(struct reloc_control *rc) | |
3596 | { | |
3597 | struct btrfs_trans_handle *trans; | |
ac2fabac | 3598 | int ret; |
3fd0a558 | 3599 | |
2ff7e61e | 3600 | rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, |
66d8f3dd | 3601 | BTRFS_BLOCK_RSV_TEMP); |
3fd0a558 YZ |
3602 | if (!rc->block_rsv) |
3603 | return -ENOMEM; | |
3604 | ||
3fd0a558 | 3605 | memset(&rc->cluster, 0, sizeof(rc->cluster)); |
b3470b5d | 3606 | rc->search_start = rc->block_group->start; |
3fd0a558 YZ |
3607 | rc->extents_found = 0; |
3608 | rc->nodes_relocated = 0; | |
3609 | rc->merging_rsv_size = 0; | |
0647bf56 | 3610 | rc->reserved_bytes = 0; |
da17066c | 3611 | rc->block_rsv->size = rc->extent_root->fs_info->nodesize * |
0647bf56 | 3612 | RELOCATION_RESERVED_NODES; |
9270501c | 3613 | ret = btrfs_block_rsv_refill(rc->extent_root->fs_info, |
ac2fabac JB |
3614 | rc->block_rsv, rc->block_rsv->size, |
3615 | BTRFS_RESERVE_FLUSH_ALL); | |
3616 | if (ret) | |
3617 | return ret; | |
3fd0a558 | 3618 | |
d23d42e3 | 3619 | rc->create_reloc_tree = true; |
3fd0a558 YZ |
3620 | set_reloc_control(rc); |
3621 | ||
7a7eaa40 | 3622 | trans = btrfs_join_transaction(rc->extent_root); |
28818947 LB |
3623 | if (IS_ERR(trans)) { |
3624 | unset_reloc_control(rc); | |
3625 | /* | |
3626 | * extent tree is not a ref_cow tree and has no reloc_root to | |
3627 | * cleanup. And callers are responsible to free the above | |
3628 | * block rsv. | |
3629 | */ | |
3630 | return PTR_ERR(trans); | |
3631 | } | |
85f02d6c ZF |
3632 | |
3633 | ret = btrfs_commit_transaction(trans); | |
3634 | if (ret) | |
3635 | unset_reloc_control(rc); | |
3636 | ||
3637 | return ret; | |
3fd0a558 | 3638 | } |
76dda93c | 3639 | |
5d4f98a2 YZ |
3640 | static noinline_for_stack int relocate_block_group(struct reloc_control *rc) |
3641 | { | |
2ff7e61e | 3642 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3643 | struct rb_root blocks = RB_ROOT; |
3644 | struct btrfs_key key; | |
3645 | struct btrfs_trans_handle *trans = NULL; | |
3646 | struct btrfs_path *path; | |
3647 | struct btrfs_extent_item *ei; | |
5d4f98a2 | 3648 | u64 flags; |
5d4f98a2 YZ |
3649 | int ret; |
3650 | int err = 0; | |
c87f08ca | 3651 | int progress = 0; |
5d4f98a2 YZ |
3652 | |
3653 | path = btrfs_alloc_path(); | |
3fd0a558 | 3654 | if (!path) |
5d4f98a2 | 3655 | return -ENOMEM; |
e4058b54 | 3656 | path->reada = READA_FORWARD; |
5d4f98a2 | 3657 | |
3fd0a558 YZ |
3658 | ret = prepare_to_relocate(rc); |
3659 | if (ret) { | |
3660 | err = ret; | |
3661 | goto out_free; | |
3662 | } | |
5d4f98a2 YZ |
3663 | |
3664 | while (1) { | |
0647bf56 | 3665 | rc->reserved_bytes = 0; |
9270501c JB |
3666 | ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, |
3667 | rc->block_rsv->size, | |
3668 | BTRFS_RESERVE_FLUSH_ALL); | |
0647bf56 WS |
3669 | if (ret) { |
3670 | err = ret; | |
3671 | break; | |
3672 | } | |
c87f08ca | 3673 | progress++; |
a22285a6 | 3674 | trans = btrfs_start_transaction(rc->extent_root, 0); |
0f788c58 LB |
3675 | if (IS_ERR(trans)) { |
3676 | err = PTR_ERR(trans); | |
3677 | trans = NULL; | |
3678 | break; | |
3679 | } | |
c87f08ca | 3680 | restart: |
3fd0a558 | 3681 | if (update_backref_cache(trans, &rc->backref_cache)) { |
3a45bb20 | 3682 | btrfs_end_transaction(trans); |
42a657f5 | 3683 | trans = NULL; |
3fd0a558 YZ |
3684 | continue; |
3685 | } | |
3686 | ||
147d256e | 3687 | ret = find_next_extent(rc, path, &key); |
5d4f98a2 YZ |
3688 | if (ret < 0) |
3689 | err = ret; | |
3690 | if (ret != 0) | |
3691 | break; | |
3692 | ||
3693 | rc->extents_found++; | |
3694 | ||
3695 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3696 | struct btrfs_extent_item); | |
24cd6389 | 3697 | flags = btrfs_extent_flags(path->nodes[0], ei); |
5d4f98a2 | 3698 | |
2672a051 BB |
3699 | /* |
3700 | * If we are relocating a simple quota owned extent item, we | |
3701 | * need to note the owner on the reloc data root so that when | |
3702 | * we allocate the replacement item, we can attribute it to the | |
3703 | * correct eventual owner (rather than the reloc data root). | |
3704 | */ | |
3705 | if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) { | |
3706 | struct btrfs_root *root = BTRFS_I(rc->data_inode)->root; | |
3707 | u64 owning_root_id = btrfs_get_extent_owner_root(fs_info, | |
3708 | path->nodes[0], | |
3709 | path->slots[0]); | |
3710 | ||
3711 | root->relocation_src_root = owning_root_id; | |
3712 | } | |
3713 | ||
5d4f98a2 YZ |
3714 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
3715 | ret = add_tree_block(rc, &key, path, &blocks); | |
3716 | } else if (rc->stage == UPDATE_DATA_PTRS && | |
3fd0a558 | 3717 | (flags & BTRFS_EXTENT_FLAG_DATA)) { |
5d4f98a2 YZ |
3718 | ret = add_data_references(rc, &key, path, &blocks); |
3719 | } else { | |
b3b4aa74 | 3720 | btrfs_release_path(path); |
5d4f98a2 YZ |
3721 | ret = 0; |
3722 | } | |
3723 | if (ret < 0) { | |
3fd0a558 | 3724 | err = ret; |
5d4f98a2 YZ |
3725 | break; |
3726 | } | |
3727 | ||
3728 | if (!RB_EMPTY_ROOT(&blocks)) { | |
3729 | ret = relocate_tree_blocks(trans, rc, &blocks); | |
3730 | if (ret < 0) { | |
3fd0a558 YZ |
3731 | if (ret != -EAGAIN) { |
3732 | err = ret; | |
3733 | break; | |
3734 | } | |
3735 | rc->extents_found--; | |
3736 | rc->search_start = key.objectid; | |
3737 | } | |
3738 | } | |
3739 | ||
3a45bb20 | 3740 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 3741 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 | 3742 | trans = NULL; |
5d4f98a2 YZ |
3743 | |
3744 | if (rc->stage == MOVE_DATA_EXTENTS && | |
3745 | (flags & BTRFS_EXTENT_FLAG_DATA)) { | |
d23d42e3 | 3746 | rc->found_file_extent = true; |
0257bb82 | 3747 | ret = relocate_data_extent(rc->data_inode, |
3fd0a558 | 3748 | &key, &rc->cluster); |
5d4f98a2 YZ |
3749 | if (ret < 0) { |
3750 | err = ret; | |
3751 | break; | |
3752 | } | |
3753 | } | |
f31ea088 QW |
3754 | if (btrfs_should_cancel_balance(fs_info)) { |
3755 | err = -ECANCELED; | |
3756 | break; | |
3757 | } | |
5d4f98a2 | 3758 | } |
c87f08ca | 3759 | if (trans && progress && err == -ENOSPC) { |
43a7e99d | 3760 | ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); |
9689457b | 3761 | if (ret == 1) { |
c87f08ca CM |
3762 | err = 0; |
3763 | progress = 0; | |
3764 | goto restart; | |
3765 | } | |
3766 | } | |
3fd0a558 | 3767 | |
b3b4aa74 | 3768 | btrfs_release_path(path); |
91166212 | 3769 | clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); |
5d4f98a2 YZ |
3770 | |
3771 | if (trans) { | |
3a45bb20 | 3772 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 3773 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 YZ |
3774 | } |
3775 | ||
0257bb82 | 3776 | if (!err) { |
3fd0a558 YZ |
3777 | ret = relocate_file_extent_cluster(rc->data_inode, |
3778 | &rc->cluster); | |
0257bb82 YZ |
3779 | if (ret < 0) |
3780 | err = ret; | |
3781 | } | |
3782 | ||
d23d42e3 | 3783 | rc->create_reloc_tree = false; |
3fd0a558 | 3784 | set_reloc_control(rc); |
0257bb82 | 3785 | |
13fe1bdb | 3786 | btrfs_backref_release_cache(&rc->backref_cache); |
63f018be | 3787 | btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); |
5d4f98a2 | 3788 | |
7f913c7c QW |
3789 | /* |
3790 | * Even in the case when the relocation is cancelled, we should all go | |
3791 | * through prepare_to_merge() and merge_reloc_roots(). | |
3792 | * | |
3793 | * For error (including cancelled balance), prepare_to_merge() will | |
3794 | * mark all reloc trees orphan, then queue them for cleanup in | |
3795 | * merge_reloc_roots() | |
3796 | */ | |
3fd0a558 | 3797 | err = prepare_to_merge(rc, err); |
5d4f98a2 YZ |
3798 | |
3799 | merge_reloc_roots(rc); | |
3800 | ||
d23d42e3 | 3801 | rc->merge_reloc_tree = false; |
5d4f98a2 | 3802 | unset_reloc_control(rc); |
63f018be | 3803 | btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); |
5d4f98a2 YZ |
3804 | |
3805 | /* get rid of pinned extents */ | |
7a7eaa40 | 3806 | trans = btrfs_join_transaction(rc->extent_root); |
62b99540 | 3807 | if (IS_ERR(trans)) { |
3612b495 | 3808 | err = PTR_ERR(trans); |
62b99540 QW |
3809 | goto out_free; |
3810 | } | |
fb686c68 JB |
3811 | ret = btrfs_commit_transaction(trans); |
3812 | if (ret && !err) | |
3813 | err = ret; | |
6217b0fa | 3814 | out_free: |
d2311e69 QW |
3815 | ret = clean_dirty_subvols(rc); |
3816 | if (ret < 0 && !err) | |
3817 | err = ret; | |
2ff7e61e | 3818 | btrfs_free_block_rsv(fs_info, rc->block_rsv); |
3fd0a558 | 3819 | btrfs_free_path(path); |
5d4f98a2 YZ |
3820 | return err; |
3821 | } | |
3822 | ||
3823 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, | |
0257bb82 | 3824 | struct btrfs_root *root, u64 objectid) |
5d4f98a2 YZ |
3825 | { |
3826 | struct btrfs_path *path; | |
3827 | struct btrfs_inode_item *item; | |
3828 | struct extent_buffer *leaf; | |
3829 | int ret; | |
3830 | ||
3831 | path = btrfs_alloc_path(); | |
3832 | if (!path) | |
3833 | return -ENOMEM; | |
3834 | ||
3835 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); | |
3836 | if (ret) | |
3837 | goto out; | |
3838 | ||
3839 | leaf = path->nodes[0]; | |
3840 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
b159fa28 | 3841 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
5d4f98a2 | 3842 | btrfs_set_inode_generation(leaf, item, 1); |
0257bb82 | 3843 | btrfs_set_inode_size(leaf, item, 0); |
5d4f98a2 | 3844 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); |
960a3166 JT |
3845 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | |
3846 | BTRFS_INODE_PREALLOC); | |
50564b65 | 3847 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
3848 | out: |
3849 | btrfs_free_path(path); | |
3850 | return ret; | |
3851 | } | |
3852 | ||
790c1b8c JB |
3853 | static void delete_orphan_inode(struct btrfs_trans_handle *trans, |
3854 | struct btrfs_root *root, u64 objectid) | |
3855 | { | |
3856 | struct btrfs_path *path; | |
3857 | struct btrfs_key key; | |
3858 | int ret = 0; | |
3859 | ||
3860 | path = btrfs_alloc_path(); | |
3861 | if (!path) { | |
3862 | ret = -ENOMEM; | |
3863 | goto out; | |
3864 | } | |
3865 | ||
3866 | key.objectid = objectid; | |
3867 | key.type = BTRFS_INODE_ITEM_KEY; | |
3868 | key.offset = 0; | |
3869 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3870 | if (ret) { | |
3871 | if (ret > 0) | |
3872 | ret = -ENOENT; | |
3873 | goto out; | |
3874 | } | |
3875 | ret = btrfs_del_item(trans, root, path); | |
3876 | out: | |
3877 | if (ret) | |
3878 | btrfs_abort_transaction(trans, ret); | |
3879 | btrfs_free_path(path); | |
3880 | } | |
3881 | ||
5d4f98a2 YZ |
3882 | /* |
3883 | * helper to create inode for data relocation. | |
3884 | * the inode is in data relocation tree and its link count is 0 | |
3885 | */ | |
ab7c8bbf DS |
3886 | static noinline_for_stack struct inode *create_reloc_inode( |
3887 | struct btrfs_fs_info *fs_info, | |
3888 | const struct btrfs_block_group *group) | |
5d4f98a2 YZ |
3889 | { |
3890 | struct inode *inode = NULL; | |
3891 | struct btrfs_trans_handle *trans; | |
3892 | struct btrfs_root *root; | |
4624900d | 3893 | u64 objectid; |
04e4e189 | 3894 | int ret = 0; |
5d4f98a2 | 3895 | |
aeb935a4 | 3896 | root = btrfs_grab_root(fs_info->data_reloc_root); |
a22285a6 | 3897 | trans = btrfs_start_transaction(root, 6); |
76deacf0 | 3898 | if (IS_ERR(trans)) { |
00246528 | 3899 | btrfs_put_root(root); |
3fd0a558 | 3900 | return ERR_CAST(trans); |
76deacf0 | 3901 | } |
5d4f98a2 | 3902 | |
04e4e189 AJ |
3903 | ret = btrfs_get_free_objectid(root, &objectid); |
3904 | if (ret) | |
5d4f98a2 YZ |
3905 | goto out; |
3906 | ||
04e4e189 AJ |
3907 | ret = __insert_orphan_inode(trans, root, objectid); |
3908 | if (ret) | |
790c1b8c | 3909 | goto out; |
5d4f98a2 | 3910 | |
0202e83f | 3911 | inode = btrfs_iget(fs_info->sb, objectid, root); |
790c1b8c JB |
3912 | if (IS_ERR(inode)) { |
3913 | delete_orphan_inode(trans, root, objectid); | |
04e4e189 | 3914 | ret = PTR_ERR(inode); |
790c1b8c JB |
3915 | inode = NULL; |
3916 | goto out; | |
3917 | } | |
b3470b5d | 3918 | BTRFS_I(inode)->index_cnt = group->start; |
5d4f98a2 | 3919 | |
04e4e189 | 3920 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
5d4f98a2 | 3921 | out: |
00246528 | 3922 | btrfs_put_root(root); |
3a45bb20 | 3923 | btrfs_end_transaction(trans); |
2ff7e61e | 3924 | btrfs_btree_balance_dirty(fs_info); |
04e4e189 | 3925 | if (ret) { |
8aa1e49e | 3926 | iput(inode); |
04e4e189 | 3927 | inode = ERR_PTR(ret); |
5d4f98a2 YZ |
3928 | } |
3929 | return inode; | |
3930 | } | |
3931 | ||
907d2710 DS |
3932 | /* |
3933 | * Mark start of chunk relocation that is cancellable. Check if the cancellation | |
3934 | * has been requested meanwhile and don't start in that case. | |
3935 | * | |
3936 | * Return: | |
3937 | * 0 success | |
3938 | * -EINPROGRESS operation is already in progress, that's probably a bug | |
3939 | * -ECANCELED cancellation request was set before the operation started | |
3940 | */ | |
3941 | static int reloc_chunk_start(struct btrfs_fs_info *fs_info) | |
3942 | { | |
3943 | if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { | |
3944 | /* This should not happen */ | |
3945 | btrfs_err(fs_info, "reloc already running, cannot start"); | |
3946 | return -EINPROGRESS; | |
3947 | } | |
3948 | ||
3949 | if (atomic_read(&fs_info->reloc_cancel_req) > 0) { | |
3950 | btrfs_info(fs_info, "chunk relocation canceled on start"); | |
3951 | /* | |
3952 | * On cancel, clear all requests but let the caller mark | |
3953 | * the end after cleanup operations. | |
3954 | */ | |
3955 | atomic_set(&fs_info->reloc_cancel_req, 0); | |
3956 | return -ECANCELED; | |
3957 | } | |
3958 | return 0; | |
3959 | } | |
3960 | ||
3961 | /* | |
3962 | * Mark end of chunk relocation that is cancellable and wake any waiters. | |
3963 | */ | |
3964 | static void reloc_chunk_end(struct btrfs_fs_info *fs_info) | |
3965 | { | |
3966 | /* Requested after start, clear bit first so any waiters can continue */ | |
3967 | if (atomic_read(&fs_info->reloc_cancel_req) > 0) | |
3968 | btrfs_info(fs_info, "chunk relocation canceled during operation"); | |
3969 | clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); | |
3970 | atomic_set(&fs_info->reloc_cancel_req, 0); | |
3971 | } | |
3972 | ||
c258d6e3 | 3973 | static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) |
3fd0a558 YZ |
3974 | { |
3975 | struct reloc_control *rc; | |
3976 | ||
3977 | rc = kzalloc(sizeof(*rc), GFP_NOFS); | |
3978 | if (!rc) | |
3979 | return NULL; | |
3980 | ||
3981 | INIT_LIST_HEAD(&rc->reloc_roots); | |
d2311e69 | 3982 | INIT_LIST_HEAD(&rc->dirty_subvol_roots); |
c71d3c69 | 3983 | btrfs_backref_init_cache(fs_info, &rc->backref_cache, true); |
733fa44d DS |
3984 | rc->reloc_root_tree.rb_root = RB_ROOT; |
3985 | spin_lock_init(&rc->reloc_root_tree.lock); | |
35da5a7e | 3986 | extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS); |
3fd0a558 YZ |
3987 | return rc; |
3988 | } | |
3989 | ||
1a0afa0e JB |
3990 | static void free_reloc_control(struct reloc_control *rc) |
3991 | { | |
3992 | struct mapping_node *node, *tmp; | |
3993 | ||
3994 | free_reloc_roots(&rc->reloc_roots); | |
3995 | rbtree_postorder_for_each_entry_safe(node, tmp, | |
3996 | &rc->reloc_root_tree.rb_root, rb_node) | |
3997 | kfree(node); | |
3998 | ||
3999 | kfree(rc); | |
4000 | } | |
4001 | ||
ebce0e01 AB |
4002 | /* |
4003 | * Print the block group being relocated | |
4004 | */ | |
4005 | static void describe_relocation(struct btrfs_fs_info *fs_info, | |
32da5386 | 4006 | struct btrfs_block_group *block_group) |
ebce0e01 | 4007 | { |
f89e09cf | 4008 | char buf[128] = {'\0'}; |
ebce0e01 | 4009 | |
f89e09cf | 4010 | btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); |
ebce0e01 AB |
4011 | |
4012 | btrfs_info(fs_info, | |
4013 | "relocating block group %llu flags %s", | |
b3470b5d | 4014 | block_group->start, buf); |
ebce0e01 AB |
4015 | } |
4016 | ||
8daf07cf | 4017 | static const char *stage_to_string(enum reloc_stage stage) |
430640e3 QW |
4018 | { |
4019 | if (stage == MOVE_DATA_EXTENTS) | |
4020 | return "move data extents"; | |
4021 | if (stage == UPDATE_DATA_PTRS) | |
4022 | return "update data pointers"; | |
4023 | return "unknown"; | |
4024 | } | |
4025 | ||
5d4f98a2 YZ |
4026 | /* |
4027 | * function to relocate all extents in a block group. | |
4028 | */ | |
6bccf3ab | 4029 | int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) |
5d4f98a2 | 4030 | { |
32da5386 | 4031 | struct btrfs_block_group *bg; |
29cbcf40 | 4032 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start); |
5d4f98a2 | 4033 | struct reloc_control *rc; |
0af3d00b JB |
4034 | struct inode *inode; |
4035 | struct btrfs_path *path; | |
5d4f98a2 | 4036 | int ret; |
f0486c68 | 4037 | int rw = 0; |
5d4f98a2 YZ |
4038 | int err = 0; |
4039 | ||
b4be6aef JB |
4040 | /* |
4041 | * This only gets set if we had a half-deleted snapshot on mount. We | |
4042 | * cannot allow relocation to start while we're still trying to clean up | |
4043 | * these pending deletions. | |
4044 | */ | |
4045 | ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE); | |
4046 | if (ret) | |
4047 | return ret; | |
4048 | ||
4049 | /* We may have been woken up by close_ctree, so bail if we're closing. */ | |
4050 | if (btrfs_fs_closing(fs_info)) | |
4051 | return -EINTR; | |
4052 | ||
eede2bf3 OS |
4053 | bg = btrfs_lookup_block_group(fs_info, group_start); |
4054 | if (!bg) | |
4055 | return -ENOENT; | |
4056 | ||
0320b353 NA |
4057 | /* |
4058 | * Relocation of a data block group creates ordered extents. Without | |
4059 | * sb_start_write(), we can freeze the filesystem while unfinished | |
4060 | * ordered extents are left. Such ordered extents can cause a deadlock | |
4061 | * e.g. when syncfs() is waiting for their completion but they can't | |
4062 | * finish because they block when joining a transaction, due to the | |
4063 | * fact that the freeze locks are being held in write mode. | |
4064 | */ | |
4065 | if (bg->flags & BTRFS_BLOCK_GROUP_DATA) | |
4066 | ASSERT(sb_write_started(fs_info->sb)); | |
4067 | ||
eede2bf3 OS |
4068 | if (btrfs_pinned_by_swapfile(fs_info, bg)) { |
4069 | btrfs_put_block_group(bg); | |
4070 | return -ETXTBSY; | |
4071 | } | |
4072 | ||
c258d6e3 | 4073 | rc = alloc_reloc_control(fs_info); |
eede2bf3 OS |
4074 | if (!rc) { |
4075 | btrfs_put_block_group(bg); | |
5d4f98a2 | 4076 | return -ENOMEM; |
eede2bf3 | 4077 | } |
5d4f98a2 | 4078 | |
907d2710 DS |
4079 | ret = reloc_chunk_start(fs_info); |
4080 | if (ret < 0) { | |
4081 | err = ret; | |
4082 | goto out_put_bg; | |
4083 | } | |
4084 | ||
f0486c68 | 4085 | rc->extent_root = extent_root; |
eede2bf3 | 4086 | rc->block_group = bg; |
5d4f98a2 | 4087 | |
b12de528 | 4088 | ret = btrfs_inc_block_group_ro(rc->block_group, true); |
868f401a Z |
4089 | if (ret) { |
4090 | err = ret; | |
4091 | goto out; | |
f0486c68 | 4092 | } |
868f401a | 4093 | rw = 1; |
f0486c68 | 4094 | |
0af3d00b JB |
4095 | path = btrfs_alloc_path(); |
4096 | if (!path) { | |
4097 | err = -ENOMEM; | |
4098 | goto out; | |
4099 | } | |
4100 | ||
7949f339 | 4101 | inode = lookup_free_space_inode(rc->block_group, path); |
0af3d00b JB |
4102 | btrfs_free_path(path); |
4103 | ||
4104 | if (!IS_ERR(inode)) | |
1bbc621e | 4105 | ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); |
0af3d00b JB |
4106 | else |
4107 | ret = PTR_ERR(inode); | |
4108 | ||
4109 | if (ret && ret != -ENOENT) { | |
4110 | err = ret; | |
4111 | goto out; | |
4112 | } | |
4113 | ||
5d4f98a2 YZ |
4114 | rc->data_inode = create_reloc_inode(fs_info, rc->block_group); |
4115 | if (IS_ERR(rc->data_inode)) { | |
4116 | err = PTR_ERR(rc->data_inode); | |
4117 | rc->data_inode = NULL; | |
4118 | goto out; | |
4119 | } | |
4120 | ||
0b246afa | 4121 | describe_relocation(fs_info, rc->block_group); |
5d4f98a2 | 4122 | |
9cfa3e34 | 4123 | btrfs_wait_block_group_reservations(rc->block_group); |
f78c436c | 4124 | btrfs_wait_nocow_writers(rc->block_group); |
6374e57a | 4125 | btrfs_wait_ordered_roots(fs_info, U64_MAX, |
b3470b5d DS |
4126 | rc->block_group->start, |
4127 | rc->block_group->length); | |
5d4f98a2 | 4128 | |
7ae9bd18 NA |
4129 | ret = btrfs_zone_finish(rc->block_group); |
4130 | WARN_ON(ret && ret != -EAGAIN); | |
4131 | ||
5d4f98a2 | 4132 | while (1) { |
8daf07cf | 4133 | enum reloc_stage finishes_stage; |
430640e3 | 4134 | |
76dda93c | 4135 | mutex_lock(&fs_info->cleaner_mutex); |
5d4f98a2 | 4136 | ret = relocate_block_group(rc); |
76dda93c | 4137 | mutex_unlock(&fs_info->cleaner_mutex); |
ff612ba7 | 4138 | if (ret < 0) |
5d4f98a2 | 4139 | err = ret; |
5d4f98a2 | 4140 | |
430640e3 | 4141 | finishes_stage = rc->stage; |
ff612ba7 JB |
4142 | /* |
4143 | * We may have gotten ENOSPC after we already dirtied some | |
4144 | * extents. If writeout happens while we're relocating a | |
4145 | * different block group we could end up hitting the | |
4146 | * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in | |
4147 | * btrfs_reloc_cow_block. Make sure we write everything out | |
4148 | * properly so we don't trip over this problem, and then break | |
4149 | * out of the loop if we hit an error. | |
4150 | */ | |
5d4f98a2 | 4151 | if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { |
0ef8b726 JB |
4152 | ret = btrfs_wait_ordered_range(rc->data_inode, 0, |
4153 | (u64)-1); | |
ff612ba7 | 4154 | if (ret) |
0ef8b726 | 4155 | err = ret; |
5d4f98a2 YZ |
4156 | invalidate_mapping_pages(rc->data_inode->i_mapping, |
4157 | 0, -1); | |
4158 | rc->stage = UPDATE_DATA_PTRS; | |
5d4f98a2 | 4159 | } |
ff612ba7 JB |
4160 | |
4161 | if (err < 0) | |
4162 | goto out; | |
4163 | ||
4164 | if (rc->extents_found == 0) | |
4165 | break; | |
4166 | ||
430640e3 QW |
4167 | btrfs_info(fs_info, "found %llu extents, stage: %s", |
4168 | rc->extents_found, stage_to_string(finishes_stage)); | |
5d4f98a2 YZ |
4169 | } |
4170 | ||
5d4f98a2 YZ |
4171 | WARN_ON(rc->block_group->pinned > 0); |
4172 | WARN_ON(rc->block_group->reserved > 0); | |
bf38be65 | 4173 | WARN_ON(rc->block_group->used > 0); |
5d4f98a2 | 4174 | out: |
f0486c68 | 4175 | if (err && rw) |
2ff7e61e | 4176 | btrfs_dec_block_group_ro(rc->block_group); |
5d4f98a2 | 4177 | iput(rc->data_inode); |
907d2710 DS |
4178 | out_put_bg: |
4179 | btrfs_put_block_group(bg); | |
4180 | reloc_chunk_end(fs_info); | |
1a0afa0e | 4181 | free_reloc_control(rc); |
5d4f98a2 YZ |
4182 | return err; |
4183 | } | |
4184 | ||
76dda93c YZ |
4185 | static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) |
4186 | { | |
0b246afa | 4187 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c | 4188 | struct btrfs_trans_handle *trans; |
79787eaa | 4189 | int ret, err; |
76dda93c | 4190 | |
0b246afa | 4191 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
4192 | if (IS_ERR(trans)) |
4193 | return PTR_ERR(trans); | |
76dda93c YZ |
4194 | |
4195 | memset(&root->root_item.drop_progress, 0, | |
4196 | sizeof(root->root_item.drop_progress)); | |
c8422684 | 4197 | btrfs_set_root_drop_level(&root->root_item, 0); |
76dda93c | 4198 | btrfs_set_root_refs(&root->root_item, 0); |
0b246afa | 4199 | ret = btrfs_update_root(trans, fs_info->tree_root, |
76dda93c | 4200 | &root->root_key, &root->root_item); |
76dda93c | 4201 | |
3a45bb20 | 4202 | err = btrfs_end_transaction(trans); |
79787eaa JM |
4203 | if (err) |
4204 | return err; | |
4205 | return ret; | |
76dda93c YZ |
4206 | } |
4207 | ||
5d4f98a2 YZ |
4208 | /* |
4209 | * recover relocation interrupted by system crash. | |
4210 | * | |
4211 | * this function resumes merging reloc trees with corresponding fs trees. | |
4212 | * this is important for keeping the sharing of tree blocks | |
4213 | */ | |
7eefae6b | 4214 | int btrfs_recover_relocation(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
4215 | { |
4216 | LIST_HEAD(reloc_roots); | |
4217 | struct btrfs_key key; | |
4218 | struct btrfs_root *fs_root; | |
4219 | struct btrfs_root *reloc_root; | |
4220 | struct btrfs_path *path; | |
4221 | struct extent_buffer *leaf; | |
4222 | struct reloc_control *rc = NULL; | |
4223 | struct btrfs_trans_handle *trans; | |
4224 | int ret; | |
4225 | int err = 0; | |
4226 | ||
4227 | path = btrfs_alloc_path(); | |
4228 | if (!path) | |
4229 | return -ENOMEM; | |
e4058b54 | 4230 | path->reada = READA_BACK; |
5d4f98a2 YZ |
4231 | |
4232 | key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
4233 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4234 | key.offset = (u64)-1; | |
4235 | ||
4236 | while (1) { | |
0b246afa | 4237 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, |
5d4f98a2 YZ |
4238 | path, 0, 0); |
4239 | if (ret < 0) { | |
4240 | err = ret; | |
4241 | goto out; | |
4242 | } | |
4243 | if (ret > 0) { | |
4244 | if (path->slots[0] == 0) | |
4245 | break; | |
4246 | path->slots[0]--; | |
4247 | } | |
4248 | leaf = path->nodes[0]; | |
4249 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4250 | btrfs_release_path(path); |
5d4f98a2 YZ |
4251 | |
4252 | if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || | |
4253 | key.type != BTRFS_ROOT_ITEM_KEY) | |
4254 | break; | |
4255 | ||
7eefae6b | 4256 | reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key); |
5d4f98a2 YZ |
4257 | if (IS_ERR(reloc_root)) { |
4258 | err = PTR_ERR(reloc_root); | |
4259 | goto out; | |
4260 | } | |
4261 | ||
92a7cc42 | 4262 | set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); |
5d4f98a2 YZ |
4263 | list_add(&reloc_root->root_list, &reloc_roots); |
4264 | ||
4265 | if (btrfs_root_refs(&reloc_root->root_item) > 0) { | |
a820feb5 DS |
4266 | fs_root = btrfs_get_fs_root(fs_info, |
4267 | reloc_root->root_key.offset, false); | |
5d4f98a2 | 4268 | if (IS_ERR(fs_root)) { |
76dda93c YZ |
4269 | ret = PTR_ERR(fs_root); |
4270 | if (ret != -ENOENT) { | |
4271 | err = ret; | |
4272 | goto out; | |
4273 | } | |
79787eaa JM |
4274 | ret = mark_garbage_root(reloc_root); |
4275 | if (ret < 0) { | |
4276 | err = ret; | |
4277 | goto out; | |
4278 | } | |
932fd26d | 4279 | } else { |
00246528 | 4280 | btrfs_put_root(fs_root); |
5d4f98a2 YZ |
4281 | } |
4282 | } | |
4283 | ||
4284 | if (key.offset == 0) | |
4285 | break; | |
4286 | ||
4287 | key.offset--; | |
4288 | } | |
b3b4aa74 | 4289 | btrfs_release_path(path); |
5d4f98a2 YZ |
4290 | |
4291 | if (list_empty(&reloc_roots)) | |
4292 | goto out; | |
4293 | ||
c258d6e3 | 4294 | rc = alloc_reloc_control(fs_info); |
5d4f98a2 YZ |
4295 | if (!rc) { |
4296 | err = -ENOMEM; | |
4297 | goto out; | |
4298 | } | |
4299 | ||
907d2710 DS |
4300 | ret = reloc_chunk_start(fs_info); |
4301 | if (ret < 0) { | |
4302 | err = ret; | |
4303 | goto out_end; | |
4304 | } | |
4305 | ||
29cbcf40 | 4306 | rc->extent_root = btrfs_extent_root(fs_info, 0); |
5d4f98a2 YZ |
4307 | |
4308 | set_reloc_control(rc); | |
4309 | ||
7a7eaa40 | 4310 | trans = btrfs_join_transaction(rc->extent_root); |
3612b495 | 4311 | if (IS_ERR(trans)) { |
3612b495 | 4312 | err = PTR_ERR(trans); |
fb2d83ee | 4313 | goto out_unset; |
3612b495 | 4314 | } |
3fd0a558 | 4315 | |
d23d42e3 | 4316 | rc->merge_reloc_tree = true; |
3fd0a558 | 4317 | |
5d4f98a2 YZ |
4318 | while (!list_empty(&reloc_roots)) { |
4319 | reloc_root = list_entry(reloc_roots.next, | |
4320 | struct btrfs_root, root_list); | |
4321 | list_del(&reloc_root->root_list); | |
4322 | ||
4323 | if (btrfs_root_refs(&reloc_root->root_item) == 0) { | |
4324 | list_add_tail(&reloc_root->root_list, | |
4325 | &rc->reloc_roots); | |
4326 | continue; | |
4327 | } | |
4328 | ||
a820feb5 DS |
4329 | fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, |
4330 | false); | |
79787eaa JM |
4331 | if (IS_ERR(fs_root)) { |
4332 | err = PTR_ERR(fs_root); | |
ca1aa281 | 4333 | list_add_tail(&reloc_root->root_list, &reloc_roots); |
1402d17d | 4334 | btrfs_end_transaction(trans); |
fb2d83ee | 4335 | goto out_unset; |
79787eaa | 4336 | } |
5d4f98a2 | 4337 | |
ffd7b339 | 4338 | err = __add_reloc_root(reloc_root); |
57a304cf | 4339 | ASSERT(err != -EEXIST); |
3c925863 JB |
4340 | if (err) { |
4341 | list_add_tail(&reloc_root->root_list, &reloc_roots); | |
4342 | btrfs_put_root(fs_root); | |
4343 | btrfs_end_transaction(trans); | |
4344 | goto out_unset; | |
4345 | } | |
f44deb74 | 4346 | fs_root->reloc_root = btrfs_grab_root(reloc_root); |
00246528 | 4347 | btrfs_put_root(fs_root); |
5d4f98a2 YZ |
4348 | } |
4349 | ||
3a45bb20 | 4350 | err = btrfs_commit_transaction(trans); |
79787eaa | 4351 | if (err) |
fb2d83ee | 4352 | goto out_unset; |
5d4f98a2 YZ |
4353 | |
4354 | merge_reloc_roots(rc); | |
4355 | ||
4356 | unset_reloc_control(rc); | |
4357 | ||
7a7eaa40 | 4358 | trans = btrfs_join_transaction(rc->extent_root); |
62b99540 | 4359 | if (IS_ERR(trans)) { |
3612b495 | 4360 | err = PTR_ERR(trans); |
6217b0fa | 4361 | goto out_clean; |
62b99540 | 4362 | } |
3a45bb20 | 4363 | err = btrfs_commit_transaction(trans); |
6217b0fa | 4364 | out_clean: |
d2311e69 QW |
4365 | ret = clean_dirty_subvols(rc); |
4366 | if (ret < 0 && !err) | |
4367 | err = ret; | |
fb2d83ee JB |
4368 | out_unset: |
4369 | unset_reloc_control(rc); | |
907d2710 DS |
4370 | out_end: |
4371 | reloc_chunk_end(fs_info); | |
1a0afa0e | 4372 | free_reloc_control(rc); |
3612b495 | 4373 | out: |
a7571232 | 4374 | free_reloc_roots(&reloc_roots); |
aca1bba6 | 4375 | |
5d4f98a2 YZ |
4376 | btrfs_free_path(path); |
4377 | ||
4378 | if (err == 0) { | |
4379 | /* cleanup orphan inode in data relocation tree */ | |
aeb935a4 QW |
4380 | fs_root = btrfs_grab_root(fs_info->data_reloc_root); |
4381 | ASSERT(fs_root); | |
4382 | err = btrfs_orphan_cleanup(fs_root); | |
4383 | btrfs_put_root(fs_root); | |
5d4f98a2 YZ |
4384 | } |
4385 | return err; | |
4386 | } | |
4387 | ||
4388 | /* | |
4389 | * helper to add ordered checksum for data relocation. | |
4390 | * | |
4391 | * cloning checksum properly handles the nodatasum extents. | |
4392 | * it also saves CPU time to re-calculate the checksum. | |
4393 | */ | |
34bfaf15 | 4394 | int btrfs_reloc_clone_csums(struct btrfs_ordered_extent *ordered) |
5d4f98a2 | 4395 | { |
34bfaf15 | 4396 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); |
7bfa9535 | 4397 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
34bfaf15 CH |
4398 | u64 disk_bytenr = ordered->file_offset + inode->index_cnt; |
4399 | struct btrfs_root *csum_root = btrfs_csum_root(fs_info, disk_bytenr); | |
5d4f98a2 | 4400 | LIST_HEAD(list); |
34bfaf15 | 4401 | int ret; |
5d4f98a2 | 4402 | |
97e38239 | 4403 | ret = btrfs_lookup_csums_list(csum_root, disk_bytenr, |
34bfaf15 | 4404 | disk_bytenr + ordered->num_bytes - 1, |
afcb8062 | 4405 | &list, false); |
aa5ccf29 JB |
4406 | if (ret < 0) { |
4407 | btrfs_mark_ordered_extent_error(ordered); | |
34bfaf15 | 4408 | return ret; |
aa5ccf29 | 4409 | } |
5d4f98a2 YZ |
4410 | |
4411 | while (!list_empty(&list)) { | |
34bfaf15 CH |
4412 | struct btrfs_ordered_sum *sums = |
4413 | list_entry(list.next, struct btrfs_ordered_sum, list); | |
4414 | ||
5d4f98a2 YZ |
4415 | list_del_init(&sums->list); |
4416 | ||
4577b014 JB |
4417 | /* |
4418 | * We need to offset the new_bytenr based on where the csum is. | |
4419 | * We need to do this because we will read in entire prealloc | |
4420 | * extents but we may have written to say the middle of the | |
4421 | * prealloc extent, so we need to make sure the csum goes with | |
4422 | * the right disk offset. | |
4423 | * | |
4424 | * We can do this because the data reloc inode refers strictly | |
4425 | * to the on disk bytes, so we don't have to worry about | |
4426 | * disk_len vs real len like with real inodes since it's all | |
4427 | * disk length. | |
4428 | */ | |
34bfaf15 | 4429 | sums->logical = ordered->disk_bytenr + sums->logical - disk_bytenr; |
f9756261 | 4430 | btrfs_add_ordered_sum(ordered, sums); |
5d4f98a2 | 4431 | } |
34bfaf15 CH |
4432 | |
4433 | return 0; | |
5d4f98a2 | 4434 | } |
3fd0a558 | 4435 | |
83d4cfd4 | 4436 | int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, |
ab7c8bbf DS |
4437 | struct btrfs_root *root, |
4438 | const struct extent_buffer *buf, | |
83d4cfd4 | 4439 | struct extent_buffer *cow) |
3fd0a558 | 4440 | { |
0b246afa | 4441 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 | 4442 | struct reloc_control *rc; |
a26195a5 | 4443 | struct btrfs_backref_node *node; |
3fd0a558 YZ |
4444 | int first_cow = 0; |
4445 | int level; | |
83d4cfd4 | 4446 | int ret = 0; |
3fd0a558 | 4447 | |
0b246afa | 4448 | rc = fs_info->reloc_ctl; |
3fd0a558 | 4449 | if (!rc) |
83d4cfd4 | 4450 | return 0; |
3fd0a558 | 4451 | |
37f00a6d | 4452 | BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); |
3fd0a558 YZ |
4453 | |
4454 | level = btrfs_header_level(buf); | |
4455 | if (btrfs_header_generation(buf) <= | |
4456 | btrfs_root_last_snapshot(&root->root_item)) | |
4457 | first_cow = 1; | |
4458 | ||
e094f480 | 4459 | if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID && rc->create_reloc_tree) { |
3fd0a558 YZ |
4460 | WARN_ON(!first_cow && level == 0); |
4461 | ||
4462 | node = rc->backref_cache.path[level]; | |
4463 | BUG_ON(node->bytenr != buf->start && | |
4464 | node->new_bytenr != buf->start); | |
4465 | ||
b0fe7078 | 4466 | btrfs_backref_drop_node_buffer(node); |
67439dad | 4467 | atomic_inc(&cow->refs); |
3fd0a558 YZ |
4468 | node->eb = cow; |
4469 | node->new_bytenr = cow->start; | |
4470 | ||
4471 | if (!node->pending) { | |
4472 | list_move_tail(&node->list, | |
4473 | &rc->backref_cache.pending[level]); | |
4474 | node->pending = 1; | |
4475 | } | |
4476 | ||
4477 | if (first_cow) | |
9569cc20 | 4478 | mark_block_processed(rc, node); |
3fd0a558 YZ |
4479 | |
4480 | if (first_cow && level > 0) | |
4481 | rc->nodes_relocated += buf->len; | |
4482 | } | |
4483 | ||
83d4cfd4 | 4484 | if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) |
3fd0a558 | 4485 | ret = replace_file_extents(trans, rc, root, cow); |
83d4cfd4 | 4486 | return ret; |
3fd0a558 YZ |
4487 | } |
4488 | ||
4489 | /* | |
4490 | * called before creating snapshot. it calculates metadata reservation | |
01327610 | 4491 | * required for relocating tree blocks in the snapshot |
3fd0a558 | 4492 | */ |
147d256e | 4493 | void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, |
3fd0a558 YZ |
4494 | u64 *bytes_to_reserve) |
4495 | { | |
10995c04 QW |
4496 | struct btrfs_root *root = pending->root; |
4497 | struct reloc_control *rc = root->fs_info->reloc_ctl; | |
3fd0a558 | 4498 | |
6282675e | 4499 | if (!rc || !have_reloc_root(root)) |
3fd0a558 YZ |
4500 | return; |
4501 | ||
3fd0a558 YZ |
4502 | if (!rc->merge_reloc_tree) |
4503 | return; | |
4504 | ||
4505 | root = root->reloc_root; | |
4506 | BUG_ON(btrfs_root_refs(&root->root_item) == 0); | |
4507 | /* | |
4508 | * relocation is in the stage of merging trees. the space | |
4509 | * used by merging a reloc tree is twice the size of | |
4510 | * relocated tree nodes in the worst case. half for cowing | |
4511 | * the reloc tree, half for cowing the fs tree. the space | |
4512 | * used by cowing the reloc tree will be freed after the | |
4513 | * tree is dropped. if we create snapshot, cowing the fs | |
4514 | * tree may use more space than it frees. so we need | |
4515 | * reserve extra space. | |
4516 | */ | |
4517 | *bytes_to_reserve += rc->nodes_relocated; | |
4518 | } | |
4519 | ||
4520 | /* | |
4521 | * called after snapshot is created. migrate block reservation | |
4522 | * and create reloc root for the newly created snapshot | |
f44deb74 JB |
4523 | * |
4524 | * This is similar to btrfs_init_reloc_root(), we come out of here with two | |
4525 | * references held on the reloc_root, one for root->reloc_root and one for | |
4526 | * rc->reloc_roots. | |
3fd0a558 | 4527 | */ |
49b25e05 | 4528 | int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, |
3fd0a558 YZ |
4529 | struct btrfs_pending_snapshot *pending) |
4530 | { | |
4531 | struct btrfs_root *root = pending->root; | |
4532 | struct btrfs_root *reloc_root; | |
4533 | struct btrfs_root *new_root; | |
10995c04 | 4534 | struct reloc_control *rc = root->fs_info->reloc_ctl; |
3fd0a558 YZ |
4535 | int ret; |
4536 | ||
6282675e | 4537 | if (!rc || !have_reloc_root(root)) |
49b25e05 | 4538 | return 0; |
3fd0a558 YZ |
4539 | |
4540 | rc = root->fs_info->reloc_ctl; | |
4541 | rc->merging_rsv_size += rc->nodes_relocated; | |
4542 | ||
4543 | if (rc->merge_reloc_tree) { | |
4544 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
4545 | rc->block_rsv, | |
3a584174 | 4546 | rc->nodes_relocated, true); |
49b25e05 JM |
4547 | if (ret) |
4548 | return ret; | |
3fd0a558 YZ |
4549 | } |
4550 | ||
4551 | new_root = pending->snap; | |
e094f480 | 4552 | reloc_root = create_reloc_root(trans, root->reloc_root, btrfs_root_id(new_root)); |
49b25e05 JM |
4553 | if (IS_ERR(reloc_root)) |
4554 | return PTR_ERR(reloc_root); | |
3fd0a558 | 4555 | |
ffd7b339 | 4556 | ret = __add_reloc_root(reloc_root); |
57a304cf | 4557 | ASSERT(ret != -EEXIST); |
3c925863 JB |
4558 | if (ret) { |
4559 | /* Pairs with create_reloc_root */ | |
4560 | btrfs_put_root(reloc_root); | |
4561 | return ret; | |
4562 | } | |
f44deb74 | 4563 | new_root->reloc_root = btrfs_grab_root(reloc_root); |
3fd0a558 | 4564 | |
49b25e05 | 4565 | if (rc->create_reloc_tree) |
3fd0a558 | 4566 | ret = clone_backref_node(trans, rc, root, reloc_root); |
49b25e05 | 4567 | return ret; |
3fd0a558 | 4568 | } |
b9a9a850 QW |
4569 | |
4570 | /* | |
4571 | * Get the current bytenr for the block group which is being relocated. | |
4572 | * | |
4573 | * Return U64_MAX if no running relocation. | |
4574 | */ | |
ab7c8bbf | 4575 | u64 btrfs_get_reloc_bg_bytenr(const struct btrfs_fs_info *fs_info) |
b9a9a850 QW |
4576 | { |
4577 | u64 logical = U64_MAX; | |
4578 | ||
4579 | lockdep_assert_held(&fs_info->reloc_mutex); | |
4580 | ||
4581 | if (fs_info->reloc_ctl && fs_info->reloc_ctl->block_group) | |
4582 | logical = fs_info->reloc_ctl->block_group->start; | |
4583 | return logical; | |
4584 | } |