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 YZ |
113 | struct btrfs_key key; |
114 | unsigned int level:8; | |
115 | unsigned int key_ready:1; | |
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; | |
125 | }; | |
126 | ||
5d4f98a2 YZ |
127 | struct reloc_control { |
128 | /* block group to relocate */ | |
32da5386 | 129 | struct btrfs_block_group *block_group; |
5d4f98a2 YZ |
130 | /* extent tree */ |
131 | struct btrfs_root *extent_root; | |
132 | /* inode for moving data */ | |
133 | struct inode *data_inode; | |
3fd0a558 YZ |
134 | |
135 | struct btrfs_block_rsv *block_rsv; | |
136 | ||
a26195a5 | 137 | struct btrfs_backref_cache backref_cache; |
3fd0a558 YZ |
138 | |
139 | struct file_extent_cluster cluster; | |
5d4f98a2 YZ |
140 | /* tree blocks have been processed */ |
141 | struct extent_io_tree processed_blocks; | |
142 | /* map start of tree root to corresponding reloc tree */ | |
143 | struct mapping_tree reloc_root_tree; | |
144 | /* list of reloc trees */ | |
145 | struct list_head reloc_roots; | |
d2311e69 QW |
146 | /* list of subvolume trees that get relocated */ |
147 | struct list_head dirty_subvol_roots; | |
3fd0a558 YZ |
148 | /* size of metadata reservation for merging reloc trees */ |
149 | u64 merging_rsv_size; | |
150 | /* size of relocated tree nodes */ | |
151 | u64 nodes_relocated; | |
0647bf56 WS |
152 | /* reserved size for block group relocation*/ |
153 | u64 reserved_bytes; | |
3fd0a558 | 154 | |
5d4f98a2 YZ |
155 | u64 search_start; |
156 | u64 extents_found; | |
3fd0a558 | 157 | |
3fd0a558 YZ |
158 | unsigned int stage:8; |
159 | unsigned int create_reloc_tree:1; | |
160 | unsigned int merge_reloc_tree:1; | |
5d4f98a2 | 161 | unsigned int found_file_extent:1; |
5d4f98a2 YZ |
162 | }; |
163 | ||
164 | /* stages of data relocation */ | |
165 | #define MOVE_DATA_EXTENTS 0 | |
166 | #define UPDATE_DATA_PTRS 1 | |
167 | ||
9569cc20 | 168 | static void mark_block_processed(struct reloc_control *rc, |
a26195a5 | 169 | struct btrfs_backref_node *node) |
9569cc20 QW |
170 | { |
171 | u32 blocksize; | |
172 | ||
173 | if (node->level == 0 || | |
174 | in_range(node->bytenr, rc->block_group->start, | |
175 | rc->block_group->length)) { | |
176 | blocksize = rc->extent_root->fs_info->nodesize; | |
0acd32c2 | 177 | set_extent_bit(&rc->processed_blocks, node->bytenr, |
1d126800 | 178 | node->bytenr + blocksize - 1, EXTENT_DIRTY, NULL); |
9569cc20 QW |
179 | } |
180 | node->processed = 1; | |
181 | } | |
182 | ||
5d4f98a2 YZ |
183 | |
184 | static void mapping_tree_init(struct mapping_tree *tree) | |
185 | { | |
6bef4d31 | 186 | tree->rb_root = RB_ROOT; |
5d4f98a2 YZ |
187 | spin_lock_init(&tree->lock); |
188 | } | |
189 | ||
5d4f98a2 YZ |
190 | /* |
191 | * walk up backref nodes until reach node presents tree root | |
192 | */ | |
a26195a5 QW |
193 | static struct btrfs_backref_node *walk_up_backref( |
194 | struct btrfs_backref_node *node, | |
195 | struct btrfs_backref_edge *edges[], int *index) | |
5d4f98a2 | 196 | { |
a26195a5 | 197 | struct btrfs_backref_edge *edge; |
5d4f98a2 YZ |
198 | int idx = *index; |
199 | ||
200 | while (!list_empty(&node->upper)) { | |
201 | edge = list_entry(node->upper.next, | |
a26195a5 | 202 | struct btrfs_backref_edge, list[LOWER]); |
5d4f98a2 YZ |
203 | edges[idx++] = edge; |
204 | node = edge->node[UPPER]; | |
205 | } | |
3fd0a558 | 206 | BUG_ON(node->detached); |
5d4f98a2 YZ |
207 | *index = idx; |
208 | return node; | |
209 | } | |
210 | ||
211 | /* | |
212 | * walk down backref nodes to find start of next reference path | |
213 | */ | |
a26195a5 QW |
214 | static struct btrfs_backref_node *walk_down_backref( |
215 | struct btrfs_backref_edge *edges[], int *index) | |
5d4f98a2 | 216 | { |
a26195a5 QW |
217 | struct btrfs_backref_edge *edge; |
218 | struct btrfs_backref_node *lower; | |
5d4f98a2 YZ |
219 | int idx = *index; |
220 | ||
221 | while (idx > 0) { | |
222 | edge = edges[idx - 1]; | |
223 | lower = edge->node[LOWER]; | |
224 | if (list_is_last(&edge->list[LOWER], &lower->upper)) { | |
225 | idx--; | |
226 | continue; | |
227 | } | |
228 | edge = list_entry(edge->list[LOWER].next, | |
a26195a5 | 229 | struct btrfs_backref_edge, list[LOWER]); |
5d4f98a2 YZ |
230 | edges[idx - 1] = edge; |
231 | *index = idx; | |
232 | return edge->node[UPPER]; | |
233 | } | |
234 | *index = 0; | |
235 | return NULL; | |
236 | } | |
5d4f98a2 | 237 | |
a26195a5 QW |
238 | static void update_backref_node(struct btrfs_backref_cache *cache, |
239 | struct btrfs_backref_node *node, u64 bytenr) | |
3fd0a558 YZ |
240 | { |
241 | struct rb_node *rb_node; | |
242 | rb_erase(&node->rb_node, &cache->rb_root); | |
243 | node->bytenr = bytenr; | |
e9a28dc5 | 244 | rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node); |
43c04fb1 | 245 | if (rb_node) |
982c92cb | 246 | btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST); |
3fd0a558 YZ |
247 | } |
248 | ||
249 | /* | |
250 | * update backref cache after a transaction commit | |
251 | */ | |
252 | static int update_backref_cache(struct btrfs_trans_handle *trans, | |
a26195a5 | 253 | struct btrfs_backref_cache *cache) |
3fd0a558 | 254 | { |
a26195a5 | 255 | struct btrfs_backref_node *node; |
3fd0a558 YZ |
256 | int level = 0; |
257 | ||
258 | if (cache->last_trans == 0) { | |
259 | cache->last_trans = trans->transid; | |
260 | return 0; | |
261 | } | |
262 | ||
263 | if (cache->last_trans == trans->transid) | |
264 | return 0; | |
265 | ||
266 | /* | |
267 | * detached nodes are used to avoid unnecessary backref | |
268 | * lookup. transaction commit changes the extent tree. | |
269 | * so the detached nodes are no longer useful. | |
270 | */ | |
271 | while (!list_empty(&cache->detached)) { | |
272 | node = list_entry(cache->detached.next, | |
a26195a5 | 273 | struct btrfs_backref_node, list); |
023acb07 | 274 | btrfs_backref_cleanup_node(cache, node); |
3fd0a558 YZ |
275 | } |
276 | ||
277 | while (!list_empty(&cache->changed)) { | |
278 | node = list_entry(cache->changed.next, | |
a26195a5 | 279 | struct btrfs_backref_node, list); |
3fd0a558 YZ |
280 | list_del_init(&node->list); |
281 | BUG_ON(node->pending); | |
282 | update_backref_node(cache, node, node->new_bytenr); | |
283 | } | |
284 | ||
285 | /* | |
286 | * some nodes can be left in the pending list if there were | |
287 | * errors during processing the pending nodes. | |
288 | */ | |
289 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
290 | list_for_each_entry(node, &cache->pending[level], list) { | |
291 | BUG_ON(!node->pending); | |
292 | if (node->bytenr == node->new_bytenr) | |
293 | continue; | |
294 | update_backref_node(cache, node, node->new_bytenr); | |
295 | } | |
296 | } | |
297 | ||
298 | cache->last_trans = 0; | |
299 | return 1; | |
300 | } | |
301 | ||
6282675e QW |
302 | static bool reloc_root_is_dead(struct btrfs_root *root) |
303 | { | |
304 | /* | |
305 | * Pair with set_bit/clear_bit in clean_dirty_subvols and | |
306 | * btrfs_update_reloc_root. We need to see the updated bit before | |
307 | * trying to access reloc_root | |
308 | */ | |
309 | smp_rmb(); | |
310 | if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) | |
311 | return true; | |
312 | return false; | |
313 | } | |
314 | ||
315 | /* | |
316 | * Check if this subvolume tree has valid reloc tree. | |
317 | * | |
318 | * Reloc tree after swap is considered dead, thus not considered as valid. | |
319 | * This is enough for most callers, as they don't distinguish dead reloc root | |
55465730 QW |
320 | * from no reloc root. But btrfs_should_ignore_reloc_root() below is a |
321 | * special case. | |
6282675e QW |
322 | */ |
323 | static bool have_reloc_root(struct btrfs_root *root) | |
324 | { | |
325 | if (reloc_root_is_dead(root)) | |
326 | return false; | |
327 | if (!root->reloc_root) | |
328 | return false; | |
329 | return true; | |
330 | } | |
f2a97a9d | 331 | |
55465730 | 332 | int btrfs_should_ignore_reloc_root(struct btrfs_root *root) |
3fd0a558 YZ |
333 | { |
334 | struct btrfs_root *reloc_root; | |
335 | ||
92a7cc42 | 336 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
3fd0a558 YZ |
337 | return 0; |
338 | ||
6282675e QW |
339 | /* This root has been merged with its reloc tree, we can ignore it */ |
340 | if (reloc_root_is_dead(root)) | |
341 | return 1; | |
342 | ||
3fd0a558 YZ |
343 | reloc_root = root->reloc_root; |
344 | if (!reloc_root) | |
345 | return 0; | |
346 | ||
4d4225fc JB |
347 | if (btrfs_header_generation(reloc_root->commit_root) == |
348 | root->fs_info->running_transaction->transid) | |
3fd0a558 YZ |
349 | return 0; |
350 | /* | |
351 | * if there is reloc tree and it was created in previous | |
352 | * transaction backref lookup can find the reloc tree, | |
353 | * so backref node for the fs tree root is useless for | |
354 | * relocation. | |
355 | */ | |
356 | return 1; | |
357 | } | |
55465730 | 358 | |
5d4f98a2 YZ |
359 | /* |
360 | * find reloc tree by address of tree root | |
361 | */ | |
2433bea5 | 362 | struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) |
5d4f98a2 | 363 | { |
2433bea5 | 364 | struct reloc_control *rc = fs_info->reloc_ctl; |
5d4f98a2 YZ |
365 | struct rb_node *rb_node; |
366 | struct mapping_node *node; | |
367 | struct btrfs_root *root = NULL; | |
368 | ||
2433bea5 | 369 | ASSERT(rc); |
5d4f98a2 | 370 | spin_lock(&rc->reloc_root_tree.lock); |
e9a28dc5 | 371 | rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr); |
5d4f98a2 YZ |
372 | if (rb_node) { |
373 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
0d031dc4 | 374 | root = node->data; |
5d4f98a2 YZ |
375 | } |
376 | spin_unlock(&rc->reloc_root_tree.lock); | |
00246528 | 377 | return btrfs_grab_root(root); |
5d4f98a2 YZ |
378 | } |
379 | ||
29db137b QW |
380 | /* |
381 | * For useless nodes, do two major clean ups: | |
382 | * | |
383 | * - Cleanup the children edges and nodes | |
384 | * If child node is also orphan (no parent) during cleanup, then the child | |
385 | * node will also be cleaned up. | |
386 | * | |
387 | * - Freeing up leaves (level 0), keeps nodes detached | |
388 | * For nodes, the node is still cached as "detached" | |
389 | * | |
390 | * Return false if @node is not in the @useless_nodes list. | |
391 | * Return true if @node is in the @useless_nodes list. | |
392 | */ | |
393 | static bool handle_useless_nodes(struct reloc_control *rc, | |
a26195a5 | 394 | struct btrfs_backref_node *node) |
29db137b | 395 | { |
a26195a5 | 396 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
29db137b QW |
397 | struct list_head *useless_node = &cache->useless_node; |
398 | bool ret = false; | |
399 | ||
400 | while (!list_empty(useless_node)) { | |
a26195a5 | 401 | struct btrfs_backref_node *cur; |
29db137b | 402 | |
a26195a5 | 403 | cur = list_first_entry(useless_node, struct btrfs_backref_node, |
29db137b QW |
404 | list); |
405 | list_del_init(&cur->list); | |
406 | ||
407 | /* Only tree root nodes can be added to @useless_nodes */ | |
408 | ASSERT(list_empty(&cur->upper)); | |
409 | ||
410 | if (cur == node) | |
411 | ret = true; | |
412 | ||
413 | /* The node is the lowest node */ | |
414 | if (cur->lowest) { | |
415 | list_del_init(&cur->lower); | |
416 | cur->lowest = 0; | |
417 | } | |
418 | ||
419 | /* Cleanup the lower edges */ | |
420 | while (!list_empty(&cur->lower)) { | |
a26195a5 QW |
421 | struct btrfs_backref_edge *edge; |
422 | struct btrfs_backref_node *lower; | |
29db137b QW |
423 | |
424 | edge = list_entry(cur->lower.next, | |
a26195a5 | 425 | struct btrfs_backref_edge, list[UPPER]); |
29db137b QW |
426 | list_del(&edge->list[UPPER]); |
427 | list_del(&edge->list[LOWER]); | |
428 | lower = edge->node[LOWER]; | |
741188d3 | 429 | btrfs_backref_free_edge(cache, edge); |
29db137b QW |
430 | |
431 | /* Child node is also orphan, queue for cleanup */ | |
432 | if (list_empty(&lower->upper)) | |
433 | list_add(&lower->list, useless_node); | |
434 | } | |
435 | /* Mark this block processed for relocation */ | |
436 | mark_block_processed(rc, cur); | |
437 | ||
438 | /* | |
439 | * Backref nodes for tree leaves are deleted from the cache. | |
440 | * Backref nodes for upper level tree blocks are left in the | |
441 | * cache to avoid unnecessary backref lookup. | |
442 | */ | |
443 | if (cur->level > 0) { | |
444 | list_add(&cur->list, &cache->detached); | |
445 | cur->detached = 1; | |
446 | } else { | |
447 | rb_erase(&cur->rb_node, &cache->rb_root); | |
741188d3 | 448 | btrfs_backref_free_node(cache, cur); |
29db137b QW |
449 | } |
450 | } | |
451 | return ret; | |
452 | } | |
453 | ||
e7d571c7 QW |
454 | /* |
455 | * Build backref tree for a given tree block. Root of the backref tree | |
456 | * corresponds the tree block, leaves of the backref tree correspond roots of | |
457 | * b-trees that reference the tree block. | |
458 | * | |
459 | * The basic idea of this function is check backrefs of a given block to find | |
460 | * upper level blocks that reference the block, and then check backrefs of | |
461 | * these upper level blocks recursively. The recursion stops when tree root is | |
462 | * reached or backrefs for the block is cached. | |
463 | * | |
464 | * NOTE: if we find that backrefs for a block are cached, we know backrefs for | |
465 | * all upper level blocks that directly/indirectly reference the block are also | |
466 | * cached. | |
467 | */ | |
a26195a5 | 468 | static noinline_for_stack struct btrfs_backref_node *build_backref_tree( |
e7d571c7 QW |
469 | struct reloc_control *rc, struct btrfs_key *node_key, |
470 | int level, u64 bytenr) | |
471 | { | |
472 | struct btrfs_backref_iter *iter; | |
a26195a5 | 473 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
e7d571c7 QW |
474 | /* For searching parent of TREE_BLOCK_REF */ |
475 | struct btrfs_path *path; | |
a26195a5 | 476 | struct btrfs_backref_node *cur; |
a26195a5 QW |
477 | struct btrfs_backref_node *node = NULL; |
478 | struct btrfs_backref_edge *edge; | |
e7d571c7 QW |
479 | int ret; |
480 | int err = 0; | |
5d4f98a2 | 481 | |
d68194b2 | 482 | iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info); |
e7d571c7 QW |
483 | if (!iter) |
484 | return ERR_PTR(-ENOMEM); | |
485 | path = btrfs_alloc_path(); | |
486 | if (!path) { | |
487 | err = -ENOMEM; | |
488 | goto out; | |
489 | } | |
490 | ||
b1818dab | 491 | node = btrfs_backref_alloc_node(cache, bytenr, level); |
e7d571c7 QW |
492 | if (!node) { |
493 | err = -ENOMEM; | |
494 | goto out; | |
5d4f98a2 YZ |
495 | } |
496 | ||
e7d571c7 QW |
497 | node->lowest = 1; |
498 | cur = node; | |
499 | ||
500 | /* Breadth-first search to build backref cache */ | |
501 | do { | |
1b60d2ec QW |
502 | ret = btrfs_backref_add_tree_node(cache, path, iter, node_key, |
503 | cur); | |
e7d571c7 QW |
504 | if (ret < 0) { |
505 | err = ret; | |
506 | goto out; | |
507 | } | |
508 | edge = list_first_entry_or_null(&cache->pending_edge, | |
a26195a5 | 509 | struct btrfs_backref_edge, list[UPPER]); |
e7d571c7 QW |
510 | /* |
511 | * The pending list isn't empty, take the first block to | |
512 | * process | |
513 | */ | |
514 | if (edge) { | |
515 | list_del_init(&edge->list[UPPER]); | |
516 | cur = edge->node[UPPER]; | |
517 | } | |
518 | } while (edge); | |
519 | ||
1f872924 | 520 | /* Finish the upper linkage of newly added edges/nodes */ |
fc997ed0 | 521 | ret = btrfs_backref_finish_upper_links(cache, node); |
1f872924 QW |
522 | if (ret < 0) { |
523 | err = ret; | |
524 | goto out; | |
3fd0a558 | 525 | } |
5d4f98a2 | 526 | |
29db137b QW |
527 | if (handle_useless_nodes(rc, node)) |
528 | node = NULL; | |
5d4f98a2 | 529 | out: |
71f572a9 QW |
530 | btrfs_backref_iter_free(iter); |
531 | btrfs_free_path(path); | |
5d4f98a2 | 532 | if (err) { |
1b23ea18 | 533 | btrfs_backref_error_cleanup(cache, node); |
5d4f98a2 YZ |
534 | return ERR_PTR(err); |
535 | } | |
75bfb9af | 536 | ASSERT(!node || !node->detached); |
84780289 QW |
537 | ASSERT(list_empty(&cache->useless_node) && |
538 | list_empty(&cache->pending_edge)); | |
5d4f98a2 YZ |
539 | return node; |
540 | } | |
541 | ||
3fd0a558 YZ |
542 | /* |
543 | * helper to add backref node for the newly created snapshot. | |
544 | * the backref node is created by cloning backref node that | |
545 | * corresponds to root of source tree | |
546 | */ | |
547 | static int clone_backref_node(struct btrfs_trans_handle *trans, | |
548 | struct reloc_control *rc, | |
549 | struct btrfs_root *src, | |
550 | struct btrfs_root *dest) | |
551 | { | |
552 | struct btrfs_root *reloc_root = src->reloc_root; | |
a26195a5 QW |
553 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
554 | struct btrfs_backref_node *node = NULL; | |
555 | struct btrfs_backref_node *new_node; | |
556 | struct btrfs_backref_edge *edge; | |
557 | struct btrfs_backref_edge *new_edge; | |
3fd0a558 YZ |
558 | struct rb_node *rb_node; |
559 | ||
560 | if (cache->last_trans > 0) | |
561 | update_backref_cache(trans, cache); | |
562 | ||
e9a28dc5 | 563 | rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start); |
3fd0a558 | 564 | if (rb_node) { |
a26195a5 | 565 | node = rb_entry(rb_node, struct btrfs_backref_node, rb_node); |
3fd0a558 YZ |
566 | if (node->detached) |
567 | node = NULL; | |
568 | else | |
569 | BUG_ON(node->new_bytenr != reloc_root->node->start); | |
570 | } | |
571 | ||
572 | if (!node) { | |
e9a28dc5 QW |
573 | rb_node = rb_simple_search(&cache->rb_root, |
574 | reloc_root->commit_root->start); | |
3fd0a558 | 575 | if (rb_node) { |
a26195a5 | 576 | node = rb_entry(rb_node, struct btrfs_backref_node, |
3fd0a558 YZ |
577 | rb_node); |
578 | BUG_ON(node->detached); | |
579 | } | |
580 | } | |
581 | ||
582 | if (!node) | |
583 | return 0; | |
584 | ||
b1818dab QW |
585 | new_node = btrfs_backref_alloc_node(cache, dest->node->start, |
586 | node->level); | |
3fd0a558 YZ |
587 | if (!new_node) |
588 | return -ENOMEM; | |
589 | ||
3fd0a558 | 590 | new_node->lowest = node->lowest; |
6848ad64 | 591 | new_node->checked = 1; |
00246528 | 592 | new_node->root = btrfs_grab_root(dest); |
0b530bc5 | 593 | ASSERT(new_node->root); |
3fd0a558 YZ |
594 | |
595 | if (!node->lowest) { | |
596 | list_for_each_entry(edge, &node->lower, list[UPPER]) { | |
47254d07 | 597 | new_edge = btrfs_backref_alloc_edge(cache); |
3fd0a558 YZ |
598 | if (!new_edge) |
599 | goto fail; | |
600 | ||
f39911e5 QW |
601 | btrfs_backref_link_edge(new_edge, edge->node[LOWER], |
602 | new_node, LINK_UPPER); | |
3fd0a558 | 603 | } |
76b9e23d MX |
604 | } else { |
605 | list_add_tail(&new_node->lower, &cache->leaves); | |
3fd0a558 YZ |
606 | } |
607 | ||
e9a28dc5 QW |
608 | rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr, |
609 | &new_node->rb_node); | |
43c04fb1 | 610 | if (rb_node) |
982c92cb | 611 | btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST); |
3fd0a558 YZ |
612 | |
613 | if (!new_node->lowest) { | |
614 | list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { | |
615 | list_add_tail(&new_edge->list[LOWER], | |
616 | &new_edge->node[LOWER]->upper); | |
617 | } | |
618 | } | |
619 | return 0; | |
620 | fail: | |
621 | while (!list_empty(&new_node->lower)) { | |
622 | new_edge = list_entry(new_node->lower.next, | |
a26195a5 | 623 | struct btrfs_backref_edge, list[UPPER]); |
3fd0a558 | 624 | list_del(&new_edge->list[UPPER]); |
741188d3 | 625 | btrfs_backref_free_edge(cache, new_edge); |
3fd0a558 | 626 | } |
741188d3 | 627 | btrfs_backref_free_node(cache, new_node); |
3fd0a558 YZ |
628 | return -ENOMEM; |
629 | } | |
630 | ||
5d4f98a2 YZ |
631 | /* |
632 | * helper to add 'address of tree root -> reloc tree' mapping | |
633 | */ | |
ffd7b339 | 634 | static int __must_check __add_reloc_root(struct btrfs_root *root) |
5d4f98a2 | 635 | { |
0b246afa | 636 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
637 | struct rb_node *rb_node; |
638 | struct mapping_node *node; | |
0b246afa | 639 | struct reloc_control *rc = fs_info->reloc_ctl; |
5d4f98a2 YZ |
640 | |
641 | node = kmalloc(sizeof(*node), GFP_NOFS); | |
ffd7b339 JM |
642 | if (!node) |
643 | return -ENOMEM; | |
5d4f98a2 | 644 | |
ea287ab1 | 645 | node->bytenr = root->commit_root->start; |
5d4f98a2 YZ |
646 | node->data = root; |
647 | ||
648 | spin_lock(&rc->reloc_root_tree.lock); | |
e9a28dc5 QW |
649 | rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, |
650 | node->bytenr, &node->rb_node); | |
5d4f98a2 | 651 | spin_unlock(&rc->reloc_root_tree.lock); |
ffd7b339 | 652 | if (rb_node) { |
57a304cf | 653 | btrfs_err(fs_info, |
5d163e0e JM |
654 | "Duplicate root found for start=%llu while inserting into relocation tree", |
655 | node->bytenr); | |
57a304cf | 656 | return -EEXIST; |
ffd7b339 | 657 | } |
5d4f98a2 YZ |
658 | |
659 | list_add_tail(&root->root_list, &rc->reloc_roots); | |
660 | return 0; | |
661 | } | |
662 | ||
663 | /* | |
c974c464 | 664 | * helper to delete the 'address of tree root -> reloc tree' |
5d4f98a2 YZ |
665 | * mapping |
666 | */ | |
c974c464 | 667 | static void __del_reloc_root(struct btrfs_root *root) |
5d4f98a2 | 668 | { |
0b246afa | 669 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
670 | struct rb_node *rb_node; |
671 | struct mapping_node *node = NULL; | |
0b246afa | 672 | struct reloc_control *rc = fs_info->reloc_ctl; |
f44deb74 | 673 | bool put_ref = false; |
5d4f98a2 | 674 | |
65c6e82b | 675 | if (rc && root->node) { |
389305b2 | 676 | spin_lock(&rc->reloc_root_tree.lock); |
e9a28dc5 QW |
677 | rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, |
678 | root->commit_root->start); | |
389305b2 QW |
679 | if (rb_node) { |
680 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
681 | rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); | |
ea287ab1 | 682 | RB_CLEAR_NODE(&node->rb_node); |
389305b2 QW |
683 | } |
684 | spin_unlock(&rc->reloc_root_tree.lock); | |
c78a10ae | 685 | ASSERT(!node || (struct btrfs_root *)node->data == root); |
5d4f98a2 | 686 | } |
5d4f98a2 | 687 | |
f44deb74 JB |
688 | /* |
689 | * We only put the reloc root here if it's on the list. There's a lot | |
690 | * of places where the pattern is to splice the rc->reloc_roots, process | |
691 | * the reloc roots, and then add the reloc root back onto | |
692 | * rc->reloc_roots. If we call __del_reloc_root while it's off of the | |
693 | * list we don't want the reference being dropped, because the guy | |
694 | * messing with the list is in charge of the reference. | |
695 | */ | |
0b246afa | 696 | spin_lock(&fs_info->trans_lock); |
f44deb74 JB |
697 | if (!list_empty(&root->root_list)) { |
698 | put_ref = true; | |
699 | list_del_init(&root->root_list); | |
700 | } | |
0b246afa | 701 | spin_unlock(&fs_info->trans_lock); |
f44deb74 JB |
702 | if (put_ref) |
703 | btrfs_put_root(root); | |
c974c464 WS |
704 | kfree(node); |
705 | } | |
706 | ||
707 | /* | |
708 | * helper to update the 'address of tree root -> reloc tree' | |
709 | * mapping | |
710 | */ | |
ea287ab1 | 711 | static int __update_reloc_root(struct btrfs_root *root) |
c974c464 | 712 | { |
0b246afa | 713 | struct btrfs_fs_info *fs_info = root->fs_info; |
c974c464 WS |
714 | struct rb_node *rb_node; |
715 | struct mapping_node *node = NULL; | |
0b246afa | 716 | struct reloc_control *rc = fs_info->reloc_ctl; |
c974c464 WS |
717 | |
718 | spin_lock(&rc->reloc_root_tree.lock); | |
e9a28dc5 QW |
719 | rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, |
720 | root->commit_root->start); | |
c974c464 WS |
721 | if (rb_node) { |
722 | node = rb_entry(rb_node, struct mapping_node, rb_node); | |
723 | rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); | |
5d4f98a2 | 724 | } |
c974c464 WS |
725 | spin_unlock(&rc->reloc_root_tree.lock); |
726 | ||
727 | if (!node) | |
728 | return 0; | |
729 | BUG_ON((struct btrfs_root *)node->data != root); | |
730 | ||
731 | spin_lock(&rc->reloc_root_tree.lock); | |
ea287ab1 | 732 | node->bytenr = root->node->start; |
e9a28dc5 QW |
733 | rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, |
734 | node->bytenr, &node->rb_node); | |
c974c464 WS |
735 | spin_unlock(&rc->reloc_root_tree.lock); |
736 | if (rb_node) | |
982c92cb | 737 | btrfs_backref_panic(fs_info, node->bytenr, -EEXIST); |
5d4f98a2 YZ |
738 | return 0; |
739 | } | |
740 | ||
3fd0a558 YZ |
741 | static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, |
742 | struct btrfs_root *root, u64 objectid) | |
5d4f98a2 | 743 | { |
0b246afa | 744 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
745 | struct btrfs_root *reloc_root; |
746 | struct extent_buffer *eb; | |
747 | struct btrfs_root_item *root_item; | |
748 | struct btrfs_key root_key; | |
84c50ba5 JB |
749 | int ret = 0; |
750 | bool must_abort = false; | |
5d4f98a2 | 751 | |
5d4f98a2 | 752 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); |
84c50ba5 JB |
753 | if (!root_item) |
754 | return ERR_PTR(-ENOMEM); | |
5d4f98a2 YZ |
755 | |
756 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
757 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
3fd0a558 | 758 | root_key.offset = objectid; |
5d4f98a2 | 759 | |
3fd0a558 | 760 | if (root->root_key.objectid == objectid) { |
054570a1 FM |
761 | u64 commit_root_gen; |
762 | ||
3fd0a558 YZ |
763 | /* called by btrfs_init_reloc_root */ |
764 | ret = btrfs_copy_root(trans, root, root->commit_root, &eb, | |
765 | BTRFS_TREE_RELOC_OBJECTID); | |
84c50ba5 JB |
766 | if (ret) |
767 | goto fail; | |
768 | ||
054570a1 FM |
769 | /* |
770 | * Set the last_snapshot field to the generation of the commit | |
771 | * root - like this ctree.c:btrfs_block_can_be_shared() behaves | |
772 | * correctly (returns true) when the relocation root is created | |
773 | * either inside the critical section of a transaction commit | |
774 | * (through transaction.c:qgroup_account_snapshot()) and when | |
775 | * it's created before the transaction commit is started. | |
776 | */ | |
777 | commit_root_gen = btrfs_header_generation(root->commit_root); | |
778 | btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); | |
3fd0a558 YZ |
779 | } else { |
780 | /* | |
781 | * called by btrfs_reloc_post_snapshot_hook. | |
782 | * the source tree is a reloc tree, all tree blocks | |
783 | * modified after it was created have RELOC flag | |
784 | * set in their headers. so it's OK to not update | |
785 | * the 'last_snapshot'. | |
786 | */ | |
787 | ret = btrfs_copy_root(trans, root, root->node, &eb, | |
788 | BTRFS_TREE_RELOC_OBJECTID); | |
84c50ba5 JB |
789 | if (ret) |
790 | goto fail; | |
3fd0a558 | 791 | } |
5d4f98a2 | 792 | |
84c50ba5 JB |
793 | /* |
794 | * We have changed references at this point, we must abort the | |
795 | * transaction if anything fails. | |
796 | */ | |
797 | must_abort = true; | |
798 | ||
5d4f98a2 | 799 | memcpy(root_item, &root->root_item, sizeof(*root_item)); |
5d4f98a2 YZ |
800 | btrfs_set_root_bytenr(root_item, eb->start); |
801 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
802 | btrfs_set_root_generation(root_item, trans->transid); | |
3fd0a558 YZ |
803 | |
804 | if (root->root_key.objectid == objectid) { | |
805 | btrfs_set_root_refs(root_item, 0); | |
806 | memset(&root_item->drop_progress, 0, | |
807 | sizeof(struct btrfs_disk_key)); | |
c8422684 | 808 | btrfs_set_root_drop_level(root_item, 0); |
3fd0a558 | 809 | } |
5d4f98a2 YZ |
810 | |
811 | btrfs_tree_unlock(eb); | |
812 | free_extent_buffer(eb); | |
813 | ||
0b246afa | 814 | ret = btrfs_insert_root(trans, fs_info->tree_root, |
5d4f98a2 | 815 | &root_key, root_item); |
84c50ba5 JB |
816 | if (ret) |
817 | goto fail; | |
818 | ||
5d4f98a2 YZ |
819 | kfree(root_item); |
820 | ||
3dbf1738 | 821 | reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); |
84c50ba5 JB |
822 | if (IS_ERR(reloc_root)) { |
823 | ret = PTR_ERR(reloc_root); | |
824 | goto abort; | |
825 | } | |
92a7cc42 | 826 | set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); |
5d4f98a2 | 827 | reloc_root->last_trans = trans->transid; |
3fd0a558 | 828 | return reloc_root; |
84c50ba5 JB |
829 | fail: |
830 | kfree(root_item); | |
831 | abort: | |
832 | if (must_abort) | |
833 | btrfs_abort_transaction(trans, ret); | |
834 | return ERR_PTR(ret); | |
3fd0a558 YZ |
835 | } |
836 | ||
837 | /* | |
838 | * create reloc tree for a given fs tree. reloc tree is just a | |
839 | * snapshot of the fs tree with special root objectid. | |
f44deb74 JB |
840 | * |
841 | * The reloc_root comes out of here with two references, one for | |
842 | * root->reloc_root, and another for being on the rc->reloc_roots list. | |
3fd0a558 YZ |
843 | */ |
844 | int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, | |
845 | struct btrfs_root *root) | |
846 | { | |
0b246afa | 847 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 | 848 | struct btrfs_root *reloc_root; |
0b246afa | 849 | struct reloc_control *rc = fs_info->reloc_ctl; |
20dd2cbf | 850 | struct btrfs_block_rsv *rsv; |
3fd0a558 | 851 | int clear_rsv = 0; |
ffd7b339 | 852 | int ret; |
3fd0a558 | 853 | |
aec7db3b | 854 | if (!rc) |
2abc726a JB |
855 | return 0; |
856 | ||
1fac4a54 QW |
857 | /* |
858 | * The subvolume has reloc tree but the swap is finished, no need to | |
859 | * create/update the dead reloc tree | |
860 | */ | |
6282675e | 861 | if (reloc_root_is_dead(root)) |
1fac4a54 QW |
862 | return 0; |
863 | ||
aec7db3b JB |
864 | /* |
865 | * This is subtle but important. We do not do | |
866 | * record_root_in_transaction for reloc roots, instead we record their | |
867 | * corresponding fs root, and then here we update the last trans for the | |
868 | * reloc root. This means that we have to do this for the entire life | |
869 | * of the reloc root, regardless of which stage of the relocation we are | |
870 | * in. | |
871 | */ | |
3fd0a558 YZ |
872 | if (root->reloc_root) { |
873 | reloc_root = root->reloc_root; | |
874 | reloc_root->last_trans = trans->transid; | |
875 | return 0; | |
876 | } | |
877 | ||
aec7db3b JB |
878 | /* |
879 | * We are merging reloc roots, we do not need new reloc trees. Also | |
880 | * reloc trees never need their own reloc tree. | |
881 | */ | |
882 | if (!rc->create_reloc_tree || | |
883 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
884 | return 0; | |
885 | ||
20dd2cbf MX |
886 | if (!trans->reloc_reserved) { |
887 | rsv = trans->block_rsv; | |
3fd0a558 YZ |
888 | trans->block_rsv = rc->block_rsv; |
889 | clear_rsv = 1; | |
890 | } | |
891 | reloc_root = create_reloc_root(trans, root, root->root_key.objectid); | |
892 | if (clear_rsv) | |
20dd2cbf | 893 | trans->block_rsv = rsv; |
00bb36a0 JB |
894 | if (IS_ERR(reloc_root)) |
895 | return PTR_ERR(reloc_root); | |
5d4f98a2 | 896 | |
ffd7b339 | 897 | ret = __add_reloc_root(reloc_root); |
57a304cf | 898 | ASSERT(ret != -EEXIST); |
00bb36a0 JB |
899 | if (ret) { |
900 | /* Pairs with create_reloc_root */ | |
901 | btrfs_put_root(reloc_root); | |
902 | return ret; | |
903 | } | |
f44deb74 | 904 | root->reloc_root = btrfs_grab_root(reloc_root); |
5d4f98a2 YZ |
905 | return 0; |
906 | } | |
907 | ||
908 | /* | |
909 | * update root item of reloc tree | |
910 | */ | |
911 | int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, | |
912 | struct btrfs_root *root) | |
913 | { | |
0b246afa | 914 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
915 | struct btrfs_root *reloc_root; |
916 | struct btrfs_root_item *root_item; | |
5d4f98a2 YZ |
917 | int ret; |
918 | ||
6282675e | 919 | if (!have_reloc_root(root)) |
592fbcd5 | 920 | return 0; |
5d4f98a2 YZ |
921 | |
922 | reloc_root = root->reloc_root; | |
923 | root_item = &reloc_root->root_item; | |
924 | ||
f44deb74 JB |
925 | /* |
926 | * We are probably ok here, but __del_reloc_root() will drop its ref of | |
927 | * the root. We have the ref for root->reloc_root, but just in case | |
928 | * hold it while we update the reloc root. | |
929 | */ | |
930 | btrfs_grab_root(reloc_root); | |
931 | ||
d2311e69 | 932 | /* root->reloc_root will stay until current relocation finished */ |
0b246afa | 933 | if (fs_info->reloc_ctl->merge_reloc_tree && |
3fd0a558 | 934 | btrfs_root_refs(root_item) == 0) { |
d2311e69 | 935 | set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); |
6282675e QW |
936 | /* |
937 | * Mark the tree as dead before we change reloc_root so | |
938 | * have_reloc_root will not touch it from now on. | |
939 | */ | |
940 | smp_wmb(); | |
c974c464 | 941 | __del_reloc_root(reloc_root); |
5d4f98a2 YZ |
942 | } |
943 | ||
5d4f98a2 | 944 | if (reloc_root->commit_root != reloc_root->node) { |
ea287ab1 | 945 | __update_reloc_root(reloc_root); |
5d4f98a2 YZ |
946 | btrfs_set_root_node(root_item, reloc_root->node); |
947 | free_extent_buffer(reloc_root->commit_root); | |
948 | reloc_root->commit_root = btrfs_root_node(reloc_root); | |
949 | } | |
950 | ||
0b246afa | 951 | ret = btrfs_update_root(trans, fs_info->tree_root, |
5d4f98a2 | 952 | &reloc_root->root_key, root_item); |
f44deb74 | 953 | btrfs_put_root(reloc_root); |
592fbcd5 | 954 | return ret; |
5d4f98a2 YZ |
955 | } |
956 | ||
957 | /* | |
958 | * helper to find first cached inode with inode number >= objectid | |
959 | * in a subvolume | |
960 | */ | |
961 | static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) | |
962 | { | |
963 | struct rb_node *node; | |
964 | struct rb_node *prev; | |
965 | struct btrfs_inode *entry; | |
966 | struct inode *inode; | |
967 | ||
968 | spin_lock(&root->inode_lock); | |
969 | again: | |
970 | node = root->inode_tree.rb_node; | |
971 | prev = NULL; | |
972 | while (node) { | |
973 | prev = node; | |
974 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
975 | ||
4a0cc7ca | 976 | if (objectid < btrfs_ino(entry)) |
5d4f98a2 | 977 | node = node->rb_left; |
4a0cc7ca | 978 | else if (objectid > btrfs_ino(entry)) |
5d4f98a2 YZ |
979 | node = node->rb_right; |
980 | else | |
981 | break; | |
982 | } | |
983 | if (!node) { | |
984 | while (prev) { | |
985 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4a0cc7ca | 986 | if (objectid <= btrfs_ino(entry)) { |
5d4f98a2 YZ |
987 | node = prev; |
988 | break; | |
989 | } | |
990 | prev = rb_next(prev); | |
991 | } | |
992 | } | |
993 | while (node) { | |
994 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
995 | inode = igrab(&entry->vfs_inode); | |
996 | if (inode) { | |
997 | spin_unlock(&root->inode_lock); | |
998 | return inode; | |
999 | } | |
1000 | ||
4a0cc7ca | 1001 | objectid = btrfs_ino(entry) + 1; |
5d4f98a2 YZ |
1002 | if (cond_resched_lock(&root->inode_lock)) |
1003 | goto again; | |
1004 | ||
1005 | node = rb_next(node); | |
1006 | } | |
1007 | spin_unlock(&root->inode_lock); | |
1008 | return NULL; | |
1009 | } | |
1010 | ||
5d4f98a2 YZ |
1011 | /* |
1012 | * get new location of data | |
1013 | */ | |
1014 | static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, | |
1015 | u64 bytenr, u64 num_bytes) | |
1016 | { | |
1017 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
1018 | struct btrfs_path *path; | |
1019 | struct btrfs_file_extent_item *fi; | |
1020 | struct extent_buffer *leaf; | |
1021 | int ret; | |
1022 | ||
1023 | path = btrfs_alloc_path(); | |
1024 | if (!path) | |
1025 | return -ENOMEM; | |
1026 | ||
1027 | bytenr -= BTRFS_I(reloc_inode)->index_cnt; | |
f85b7379 DS |
1028 | ret = btrfs_lookup_file_extent(NULL, root, path, |
1029 | btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); | |
5d4f98a2 YZ |
1030 | if (ret < 0) |
1031 | goto out; | |
1032 | if (ret > 0) { | |
1033 | ret = -ENOENT; | |
1034 | goto out; | |
1035 | } | |
1036 | ||
1037 | leaf = path->nodes[0]; | |
1038 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1039 | struct btrfs_file_extent_item); | |
1040 | ||
1041 | BUG_ON(btrfs_file_extent_offset(leaf, fi) || | |
1042 | btrfs_file_extent_compression(leaf, fi) || | |
1043 | btrfs_file_extent_encryption(leaf, fi) || | |
1044 | btrfs_file_extent_other_encoding(leaf, fi)); | |
1045 | ||
1046 | if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { | |
83d4cfd4 | 1047 | ret = -EINVAL; |
5d4f98a2 YZ |
1048 | goto out; |
1049 | } | |
1050 | ||
3fd0a558 | 1051 | *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 YZ |
1052 | ret = 0; |
1053 | out: | |
1054 | btrfs_free_path(path); | |
1055 | return ret; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * update file extent items in the tree leaf to point to | |
1060 | * the new locations. | |
1061 | */ | |
3fd0a558 YZ |
1062 | static noinline_for_stack |
1063 | int replace_file_extents(struct btrfs_trans_handle *trans, | |
1064 | struct reloc_control *rc, | |
1065 | struct btrfs_root *root, | |
1066 | struct extent_buffer *leaf) | |
5d4f98a2 | 1067 | { |
0b246afa | 1068 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1069 | struct btrfs_key key; |
1070 | struct btrfs_file_extent_item *fi; | |
1071 | struct inode *inode = NULL; | |
5d4f98a2 YZ |
1072 | u64 parent; |
1073 | u64 bytenr; | |
3fd0a558 | 1074 | u64 new_bytenr = 0; |
5d4f98a2 YZ |
1075 | u64 num_bytes; |
1076 | u64 end; | |
1077 | u32 nritems; | |
1078 | u32 i; | |
83d4cfd4 | 1079 | int ret = 0; |
5d4f98a2 YZ |
1080 | int first = 1; |
1081 | int dirty = 0; | |
1082 | ||
1083 | if (rc->stage != UPDATE_DATA_PTRS) | |
1084 | return 0; | |
1085 | ||
1086 | /* reloc trees always use full backref */ | |
1087 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1088 | parent = leaf->start; | |
1089 | else | |
1090 | parent = 0; | |
1091 | ||
1092 | nritems = btrfs_header_nritems(leaf); | |
1093 | for (i = 0; i < nritems; i++) { | |
82fa113f QW |
1094 | struct btrfs_ref ref = { 0 }; |
1095 | ||
5d4f98a2 YZ |
1096 | cond_resched(); |
1097 | btrfs_item_key_to_cpu(leaf, &key, i); | |
1098 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
1099 | continue; | |
1100 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
1101 | if (btrfs_file_extent_type(leaf, fi) == | |
1102 | BTRFS_FILE_EXTENT_INLINE) | |
1103 | continue; | |
1104 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1105 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
1106 | if (bytenr == 0) | |
1107 | continue; | |
9569cc20 QW |
1108 | if (!in_range(bytenr, rc->block_group->start, |
1109 | rc->block_group->length)) | |
5d4f98a2 YZ |
1110 | continue; |
1111 | ||
1112 | /* | |
fb12489b | 1113 | * if we are modifying block in fs tree, wait for read_folio |
5d4f98a2 YZ |
1114 | * to complete and drop the extent cache |
1115 | */ | |
1116 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { | |
5d4f98a2 YZ |
1117 | if (first) { |
1118 | inode = find_next_inode(root, key.objectid); | |
5d4f98a2 | 1119 | first = 0; |
4a0cc7ca | 1120 | } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { |
e55cf7ca | 1121 | btrfs_add_delayed_iput(BTRFS_I(inode)); |
5d4f98a2 | 1122 | inode = find_next_inode(root, key.objectid); |
5d4f98a2 | 1123 | } |
4a0cc7ca | 1124 | if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { |
9c5c9604 JB |
1125 | struct extent_state *cached_state = NULL; |
1126 | ||
5d4f98a2 YZ |
1127 | end = key.offset + |
1128 | btrfs_file_extent_num_bytes(leaf, fi); | |
1129 | WARN_ON(!IS_ALIGNED(key.offset, | |
0b246afa JM |
1130 | fs_info->sectorsize)); |
1131 | WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); | |
5d4f98a2 YZ |
1132 | end--; |
1133 | ret = try_lock_extent(&BTRFS_I(inode)->io_tree, | |
9c5c9604 JB |
1134 | key.offset, end, |
1135 | &cached_state); | |
5d4f98a2 YZ |
1136 | if (!ret) |
1137 | continue; | |
1138 | ||
4c0c8cfc FM |
1139 | btrfs_drop_extent_map_range(BTRFS_I(inode), |
1140 | key.offset, end, true); | |
5d4f98a2 | 1141 | unlock_extent(&BTRFS_I(inode)->io_tree, |
9c5c9604 | 1142 | key.offset, end, &cached_state); |
5d4f98a2 YZ |
1143 | } |
1144 | } | |
1145 | ||
1146 | ret = get_new_location(rc->data_inode, &new_bytenr, | |
1147 | bytenr, num_bytes); | |
83d4cfd4 JB |
1148 | if (ret) { |
1149 | /* | |
1150 | * Don't have to abort since we've not changed anything | |
1151 | * in the file extent yet. | |
1152 | */ | |
1153 | break; | |
3fd0a558 | 1154 | } |
5d4f98a2 YZ |
1155 | |
1156 | btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); | |
1157 | dirty = 1; | |
1158 | ||
1159 | key.offset -= btrfs_file_extent_offset(leaf, fi); | |
82fa113f QW |
1160 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, |
1161 | num_bytes, parent); | |
82fa113f | 1162 | btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), |
f42c5da6 NB |
1163 | key.objectid, key.offset, |
1164 | root->root_key.objectid, false); | |
82fa113f | 1165 | ret = btrfs_inc_extent_ref(trans, &ref); |
83d4cfd4 | 1166 | if (ret) { |
66642832 | 1167 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 JB |
1168 | break; |
1169 | } | |
5d4f98a2 | 1170 | |
ffd4bb2a QW |
1171 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
1172 | num_bytes, parent); | |
ffd4bb2a | 1173 | btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), |
f42c5da6 NB |
1174 | key.objectid, key.offset, |
1175 | root->root_key.objectid, false); | |
ffd4bb2a | 1176 | ret = btrfs_free_extent(trans, &ref); |
83d4cfd4 | 1177 | if (ret) { |
66642832 | 1178 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 JB |
1179 | break; |
1180 | } | |
5d4f98a2 YZ |
1181 | } |
1182 | if (dirty) | |
1183 | btrfs_mark_buffer_dirty(leaf); | |
3fd0a558 | 1184 | if (inode) |
e55cf7ca | 1185 | btrfs_add_delayed_iput(BTRFS_I(inode)); |
83d4cfd4 | 1186 | return ret; |
5d4f98a2 YZ |
1187 | } |
1188 | ||
1189 | static noinline_for_stack | |
1190 | int memcmp_node_keys(struct extent_buffer *eb, int slot, | |
1191 | struct btrfs_path *path, int level) | |
1192 | { | |
1193 | struct btrfs_disk_key key1; | |
1194 | struct btrfs_disk_key key2; | |
1195 | btrfs_node_key(eb, &key1, slot); | |
1196 | btrfs_node_key(path->nodes[level], &key2, path->slots[level]); | |
1197 | return memcmp(&key1, &key2, sizeof(key1)); | |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * try to replace tree blocks in fs tree with the new blocks | |
1202 | * in reloc tree. tree blocks haven't been modified since the | |
1203 | * reloc tree was create can be replaced. | |
1204 | * | |
1205 | * if a block was replaced, level of the block + 1 is returned. | |
1206 | * if no block got replaced, 0 is returned. if there are other | |
1207 | * errors, a negative error number is returned. | |
1208 | */ | |
3fd0a558 | 1209 | static noinline_for_stack |
3d0174f7 | 1210 | int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, |
3fd0a558 YZ |
1211 | struct btrfs_root *dest, struct btrfs_root *src, |
1212 | struct btrfs_path *path, struct btrfs_key *next_key, | |
1213 | int lowest_level, int max_level) | |
5d4f98a2 | 1214 | { |
0b246afa | 1215 | struct btrfs_fs_info *fs_info = dest->fs_info; |
5d4f98a2 YZ |
1216 | struct extent_buffer *eb; |
1217 | struct extent_buffer *parent; | |
82fa113f | 1218 | struct btrfs_ref ref = { 0 }; |
5d4f98a2 YZ |
1219 | struct btrfs_key key; |
1220 | u64 old_bytenr; | |
1221 | u64 new_bytenr; | |
1222 | u64 old_ptr_gen; | |
1223 | u64 new_ptr_gen; | |
1224 | u64 last_snapshot; | |
1225 | u32 blocksize; | |
3fd0a558 | 1226 | int cow = 0; |
5d4f98a2 YZ |
1227 | int level; |
1228 | int ret; | |
1229 | int slot; | |
1230 | ||
7a9213a9 JB |
1231 | ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); |
1232 | ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
5d4f98a2 YZ |
1233 | |
1234 | last_snapshot = btrfs_root_last_snapshot(&src->root_item); | |
3fd0a558 | 1235 | again: |
5d4f98a2 YZ |
1236 | slot = path->slots[lowest_level]; |
1237 | btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); | |
1238 | ||
1239 | eb = btrfs_lock_root_node(dest); | |
5d4f98a2 YZ |
1240 | level = btrfs_header_level(eb); |
1241 | ||
1242 | if (level < lowest_level) { | |
1243 | btrfs_tree_unlock(eb); | |
1244 | free_extent_buffer(eb); | |
1245 | return 0; | |
1246 | } | |
1247 | ||
3fd0a558 | 1248 | if (cow) { |
9631e4cc JB |
1249 | ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb, |
1250 | BTRFS_NESTING_COW); | |
45b87c5d JB |
1251 | if (ret) { |
1252 | btrfs_tree_unlock(eb); | |
1253 | free_extent_buffer(eb); | |
1254 | return ret; | |
1255 | } | |
3fd0a558 | 1256 | } |
5d4f98a2 YZ |
1257 | |
1258 | if (next_key) { | |
1259 | next_key->objectid = (u64)-1; | |
1260 | next_key->type = (u8)-1; | |
1261 | next_key->offset = (u64)-1; | |
1262 | } | |
1263 | ||
1264 | parent = eb; | |
1265 | while (1) { | |
1266 | level = btrfs_header_level(parent); | |
7a9213a9 | 1267 | ASSERT(level >= lowest_level); |
5d4f98a2 | 1268 | |
fdf8d595 | 1269 | ret = btrfs_bin_search(parent, 0, &key, &slot); |
cbca7d59 FM |
1270 | if (ret < 0) |
1271 | break; | |
5d4f98a2 YZ |
1272 | if (ret && slot > 0) |
1273 | slot--; | |
1274 | ||
1275 | if (next_key && slot + 1 < btrfs_header_nritems(parent)) | |
1276 | btrfs_node_key_to_cpu(parent, next_key, slot + 1); | |
1277 | ||
1278 | old_bytenr = btrfs_node_blockptr(parent, slot); | |
0b246afa | 1279 | blocksize = fs_info->nodesize; |
5d4f98a2 YZ |
1280 | old_ptr_gen = btrfs_node_ptr_generation(parent, slot); |
1281 | ||
1282 | if (level <= max_level) { | |
1283 | eb = path->nodes[level]; | |
1284 | new_bytenr = btrfs_node_blockptr(eb, | |
1285 | path->slots[level]); | |
1286 | new_ptr_gen = btrfs_node_ptr_generation(eb, | |
1287 | path->slots[level]); | |
1288 | } else { | |
1289 | new_bytenr = 0; | |
1290 | new_ptr_gen = 0; | |
1291 | } | |
1292 | ||
fae7f21c | 1293 | if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { |
5d4f98a2 YZ |
1294 | ret = level; |
1295 | break; | |
1296 | } | |
1297 | ||
1298 | if (new_bytenr == 0 || old_ptr_gen > last_snapshot || | |
1299 | memcmp_node_keys(parent, slot, path, level)) { | |
3fd0a558 | 1300 | if (level <= lowest_level) { |
5d4f98a2 YZ |
1301 | ret = 0; |
1302 | break; | |
1303 | } | |
1304 | ||
6b3426be | 1305 | eb = btrfs_read_node_slot(parent, slot); |
64c043de LB |
1306 | if (IS_ERR(eb)) { |
1307 | ret = PTR_ERR(eb); | |
264813ac | 1308 | break; |
416bc658 | 1309 | } |
5d4f98a2 | 1310 | btrfs_tree_lock(eb); |
3fd0a558 YZ |
1311 | if (cow) { |
1312 | ret = btrfs_cow_block(trans, dest, eb, parent, | |
9631e4cc JB |
1313 | slot, &eb, |
1314 | BTRFS_NESTING_COW); | |
45b87c5d JB |
1315 | if (ret) { |
1316 | btrfs_tree_unlock(eb); | |
1317 | free_extent_buffer(eb); | |
1318 | break; | |
1319 | } | |
5d4f98a2 YZ |
1320 | } |
1321 | ||
1322 | btrfs_tree_unlock(parent); | |
1323 | free_extent_buffer(parent); | |
1324 | ||
1325 | parent = eb; | |
1326 | continue; | |
1327 | } | |
1328 | ||
3fd0a558 YZ |
1329 | if (!cow) { |
1330 | btrfs_tree_unlock(parent); | |
1331 | free_extent_buffer(parent); | |
1332 | cow = 1; | |
1333 | goto again; | |
1334 | } | |
1335 | ||
5d4f98a2 YZ |
1336 | btrfs_node_key_to_cpu(path->nodes[level], &key, |
1337 | path->slots[level]); | |
b3b4aa74 | 1338 | btrfs_release_path(path); |
5d4f98a2 YZ |
1339 | |
1340 | path->lowest_level = level; | |
b40130b2 | 1341 | set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); |
5d4f98a2 | 1342 | ret = btrfs_search_slot(trans, src, &key, path, 0, 1); |
b40130b2 | 1343 | clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state); |
5d4f98a2 | 1344 | path->lowest_level = 0; |
0e9873e2 JB |
1345 | if (ret) { |
1346 | if (ret > 0) | |
1347 | ret = -ENOENT; | |
1348 | break; | |
1349 | } | |
5d4f98a2 | 1350 | |
824d8dff QW |
1351 | /* |
1352 | * Info qgroup to trace both subtrees. | |
1353 | * | |
1354 | * We must trace both trees. | |
1355 | * 1) Tree reloc subtree | |
1356 | * If not traced, we will leak data numbers | |
1357 | * 2) Fs subtree | |
1358 | * If not traced, we will double count old data | |
f616f5cd QW |
1359 | * |
1360 | * We don't scan the subtree right now, but only record | |
1361 | * the swapped tree blocks. | |
1362 | * The real subtree rescan is delayed until we have new | |
1363 | * CoW on the subtree root node before transaction commit. | |
824d8dff | 1364 | */ |
370a11b8 QW |
1365 | ret = btrfs_qgroup_add_swapped_blocks(trans, dest, |
1366 | rc->block_group, parent, slot, | |
1367 | path->nodes[level], path->slots[level], | |
1368 | last_snapshot); | |
1369 | if (ret < 0) | |
1370 | break; | |
5d4f98a2 YZ |
1371 | /* |
1372 | * swap blocks in fs tree and reloc tree. | |
1373 | */ | |
1374 | btrfs_set_node_blockptr(parent, slot, new_bytenr); | |
1375 | btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); | |
1376 | btrfs_mark_buffer_dirty(parent); | |
1377 | ||
1378 | btrfs_set_node_blockptr(path->nodes[level], | |
1379 | path->slots[level], old_bytenr); | |
1380 | btrfs_set_node_ptr_generation(path->nodes[level], | |
1381 | path->slots[level], old_ptr_gen); | |
1382 | btrfs_mark_buffer_dirty(path->nodes[level]); | |
1383 | ||
82fa113f QW |
1384 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, |
1385 | blocksize, path->nodes[level]->start); | |
f42c5da6 NB |
1386 | btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, |
1387 | 0, true); | |
82fa113f | 1388 | ret = btrfs_inc_extent_ref(trans, &ref); |
253e258c JB |
1389 | if (ret) { |
1390 | btrfs_abort_transaction(trans, ret); | |
1391 | break; | |
1392 | } | |
82fa113f QW |
1393 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, |
1394 | blocksize, 0); | |
f42c5da6 NB |
1395 | btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0, |
1396 | true); | |
82fa113f | 1397 | ret = btrfs_inc_extent_ref(trans, &ref); |
253e258c JB |
1398 | if (ret) { |
1399 | btrfs_abort_transaction(trans, ret); | |
1400 | break; | |
1401 | } | |
5d4f98a2 | 1402 | |
ffd4bb2a QW |
1403 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, |
1404 | blocksize, path->nodes[level]->start); | |
f42c5da6 NB |
1405 | btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid, |
1406 | 0, true); | |
ffd4bb2a | 1407 | ret = btrfs_free_extent(trans, &ref); |
253e258c JB |
1408 | if (ret) { |
1409 | btrfs_abort_transaction(trans, ret); | |
1410 | break; | |
1411 | } | |
5d4f98a2 | 1412 | |
ffd4bb2a QW |
1413 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, |
1414 | blocksize, 0); | |
f42c5da6 NB |
1415 | btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, |
1416 | 0, true); | |
ffd4bb2a | 1417 | ret = btrfs_free_extent(trans, &ref); |
253e258c JB |
1418 | if (ret) { |
1419 | btrfs_abort_transaction(trans, ret); | |
1420 | break; | |
1421 | } | |
5d4f98a2 YZ |
1422 | |
1423 | btrfs_unlock_up_safe(path, 0); | |
1424 | ||
1425 | ret = level; | |
1426 | break; | |
1427 | } | |
1428 | btrfs_tree_unlock(parent); | |
1429 | free_extent_buffer(parent); | |
1430 | return ret; | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * helper to find next relocated block in reloc tree | |
1435 | */ | |
1436 | static noinline_for_stack | |
1437 | int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, | |
1438 | int *level) | |
1439 | { | |
1440 | struct extent_buffer *eb; | |
1441 | int i; | |
1442 | u64 last_snapshot; | |
1443 | u32 nritems; | |
1444 | ||
1445 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
1446 | ||
1447 | for (i = 0; i < *level; i++) { | |
1448 | free_extent_buffer(path->nodes[i]); | |
1449 | path->nodes[i] = NULL; | |
1450 | } | |
1451 | ||
1452 | for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { | |
1453 | eb = path->nodes[i]; | |
1454 | nritems = btrfs_header_nritems(eb); | |
1455 | while (path->slots[i] + 1 < nritems) { | |
1456 | path->slots[i]++; | |
1457 | if (btrfs_node_ptr_generation(eb, path->slots[i]) <= | |
1458 | last_snapshot) | |
1459 | continue; | |
1460 | ||
1461 | *level = i; | |
1462 | return 0; | |
1463 | } | |
1464 | free_extent_buffer(path->nodes[i]); | |
1465 | path->nodes[i] = NULL; | |
1466 | } | |
1467 | return 1; | |
1468 | } | |
1469 | ||
1470 | /* | |
1471 | * walk down reloc tree to find relocated block of lowest level | |
1472 | */ | |
1473 | static noinline_for_stack | |
1474 | int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, | |
1475 | int *level) | |
1476 | { | |
1477 | struct extent_buffer *eb = NULL; | |
1478 | int i; | |
5d4f98a2 YZ |
1479 | u64 ptr_gen = 0; |
1480 | u64 last_snapshot; | |
5d4f98a2 YZ |
1481 | u32 nritems; |
1482 | ||
1483 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
1484 | ||
1485 | for (i = *level; i > 0; i--) { | |
1486 | eb = path->nodes[i]; | |
1487 | nritems = btrfs_header_nritems(eb); | |
1488 | while (path->slots[i] < nritems) { | |
1489 | ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); | |
1490 | if (ptr_gen > last_snapshot) | |
1491 | break; | |
1492 | path->slots[i]++; | |
1493 | } | |
1494 | if (path->slots[i] >= nritems) { | |
1495 | if (i == *level) | |
1496 | break; | |
1497 | *level = i + 1; | |
1498 | return 0; | |
1499 | } | |
1500 | if (i == 1) { | |
1501 | *level = i; | |
1502 | return 0; | |
1503 | } | |
1504 | ||
8ef385bb JB |
1505 | eb = btrfs_read_node_slot(eb, path->slots[i]); |
1506 | if (IS_ERR(eb)) | |
64c043de | 1507 | return PTR_ERR(eb); |
5d4f98a2 YZ |
1508 | BUG_ON(btrfs_header_level(eb) != i - 1); |
1509 | path->nodes[i - 1] = eb; | |
1510 | path->slots[i - 1] = 0; | |
1511 | } | |
1512 | return 1; | |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * invalidate extent cache for file extents whose key in range of | |
1517 | * [min_key, max_key) | |
1518 | */ | |
1519 | static int invalidate_extent_cache(struct btrfs_root *root, | |
1520 | struct btrfs_key *min_key, | |
1521 | struct btrfs_key *max_key) | |
1522 | { | |
0b246afa | 1523 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1524 | struct inode *inode = NULL; |
1525 | u64 objectid; | |
1526 | u64 start, end; | |
33345d01 | 1527 | u64 ino; |
5d4f98a2 YZ |
1528 | |
1529 | objectid = min_key->objectid; | |
1530 | while (1) { | |
9c5c9604 JB |
1531 | struct extent_state *cached_state = NULL; |
1532 | ||
5d4f98a2 YZ |
1533 | cond_resched(); |
1534 | iput(inode); | |
1535 | ||
1536 | if (objectid > max_key->objectid) | |
1537 | break; | |
1538 | ||
1539 | inode = find_next_inode(root, objectid); | |
1540 | if (!inode) | |
1541 | break; | |
4a0cc7ca | 1542 | ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 1543 | |
33345d01 | 1544 | if (ino > max_key->objectid) { |
5d4f98a2 YZ |
1545 | iput(inode); |
1546 | break; | |
1547 | } | |
1548 | ||
33345d01 | 1549 | objectid = ino + 1; |
5d4f98a2 YZ |
1550 | if (!S_ISREG(inode->i_mode)) |
1551 | continue; | |
1552 | ||
33345d01 | 1553 | if (unlikely(min_key->objectid == ino)) { |
5d4f98a2 YZ |
1554 | if (min_key->type > BTRFS_EXTENT_DATA_KEY) |
1555 | continue; | |
1556 | if (min_key->type < BTRFS_EXTENT_DATA_KEY) | |
1557 | start = 0; | |
1558 | else { | |
1559 | start = min_key->offset; | |
0b246afa | 1560 | WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); |
5d4f98a2 YZ |
1561 | } |
1562 | } else { | |
1563 | start = 0; | |
1564 | } | |
1565 | ||
33345d01 | 1566 | if (unlikely(max_key->objectid == ino)) { |
5d4f98a2 YZ |
1567 | if (max_key->type < BTRFS_EXTENT_DATA_KEY) |
1568 | continue; | |
1569 | if (max_key->type > BTRFS_EXTENT_DATA_KEY) { | |
1570 | end = (u64)-1; | |
1571 | } else { | |
1572 | if (max_key->offset == 0) | |
1573 | continue; | |
1574 | end = max_key->offset; | |
0b246afa | 1575 | WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); |
5d4f98a2 YZ |
1576 | end--; |
1577 | } | |
1578 | } else { | |
1579 | end = (u64)-1; | |
1580 | } | |
1581 | ||
fb12489b | 1582 | /* the lock_extent waits for read_folio to complete */ |
9c5c9604 | 1583 | lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); |
4c0c8cfc | 1584 | btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true); |
9c5c9604 | 1585 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); |
5d4f98a2 YZ |
1586 | } |
1587 | return 0; | |
1588 | } | |
1589 | ||
1590 | static int find_next_key(struct btrfs_path *path, int level, | |
1591 | struct btrfs_key *key) | |
1592 | ||
1593 | { | |
1594 | while (level < BTRFS_MAX_LEVEL) { | |
1595 | if (!path->nodes[level]) | |
1596 | break; | |
1597 | if (path->slots[level] + 1 < | |
1598 | btrfs_header_nritems(path->nodes[level])) { | |
1599 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1600 | path->slots[level] + 1); | |
1601 | return 0; | |
1602 | } | |
1603 | level++; | |
1604 | } | |
1605 | return 1; | |
1606 | } | |
1607 | ||
d2311e69 QW |
1608 | /* |
1609 | * Insert current subvolume into reloc_control::dirty_subvol_roots | |
1610 | */ | |
ac54da6c JB |
1611 | static int insert_dirty_subvol(struct btrfs_trans_handle *trans, |
1612 | struct reloc_control *rc, | |
1613 | struct btrfs_root *root) | |
d2311e69 QW |
1614 | { |
1615 | struct btrfs_root *reloc_root = root->reloc_root; | |
1616 | struct btrfs_root_item *reloc_root_item; | |
7934133f | 1617 | int ret; |
d2311e69 QW |
1618 | |
1619 | /* @root must be a subvolume tree root with a valid reloc tree */ | |
1620 | ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
1621 | ASSERT(reloc_root); | |
1622 | ||
1623 | reloc_root_item = &reloc_root->root_item; | |
1624 | memset(&reloc_root_item->drop_progress, 0, | |
1625 | sizeof(reloc_root_item->drop_progress)); | |
c8422684 | 1626 | btrfs_set_root_drop_level(reloc_root_item, 0); |
d2311e69 | 1627 | btrfs_set_root_refs(reloc_root_item, 0); |
7934133f JB |
1628 | ret = btrfs_update_reloc_root(trans, root); |
1629 | if (ret) | |
1630 | return ret; | |
d2311e69 QW |
1631 | |
1632 | if (list_empty(&root->reloc_dirty_list)) { | |
00246528 | 1633 | btrfs_grab_root(root); |
d2311e69 QW |
1634 | list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); |
1635 | } | |
ac54da6c JB |
1636 | |
1637 | return 0; | |
d2311e69 QW |
1638 | } |
1639 | ||
1640 | static int clean_dirty_subvols(struct reloc_control *rc) | |
1641 | { | |
1642 | struct btrfs_root *root; | |
1643 | struct btrfs_root *next; | |
1644 | int ret = 0; | |
30d40577 | 1645 | int ret2; |
d2311e69 QW |
1646 | |
1647 | list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, | |
1648 | reloc_dirty_list) { | |
30d40577 QW |
1649 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
1650 | /* Merged subvolume, cleanup its reloc root */ | |
1651 | struct btrfs_root *reloc_root = root->reloc_root; | |
d2311e69 | 1652 | |
30d40577 QW |
1653 | list_del_init(&root->reloc_dirty_list); |
1654 | root->reloc_root = NULL; | |
6282675e QW |
1655 | /* |
1656 | * Need barrier to ensure clear_bit() only happens after | |
1657 | * root->reloc_root = NULL. Pairs with have_reloc_root. | |
1658 | */ | |
1659 | smp_wmb(); | |
1fac4a54 | 1660 | clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); |
f28de8d8 | 1661 | if (reloc_root) { |
f44deb74 JB |
1662 | /* |
1663 | * btrfs_drop_snapshot drops our ref we hold for | |
1664 | * ->reloc_root. If it fails however we must | |
1665 | * drop the ref ourselves. | |
1666 | */ | |
f28de8d8 | 1667 | ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); |
f44deb74 JB |
1668 | if (ret2 < 0) { |
1669 | btrfs_put_root(reloc_root); | |
1670 | if (!ret) | |
1671 | ret = ret2; | |
1672 | } | |
f28de8d8 | 1673 | } |
00246528 | 1674 | btrfs_put_root(root); |
30d40577 QW |
1675 | } else { |
1676 | /* Orphan reloc tree, just clean it up */ | |
0078a9f9 | 1677 | ret2 = btrfs_drop_snapshot(root, 0, 1); |
f44deb74 JB |
1678 | if (ret2 < 0) { |
1679 | btrfs_put_root(root); | |
1680 | if (!ret) | |
1681 | ret = ret2; | |
1682 | } | |
d2311e69 | 1683 | } |
d2311e69 QW |
1684 | } |
1685 | return ret; | |
1686 | } | |
1687 | ||
5d4f98a2 YZ |
1688 | /* |
1689 | * merge the relocated tree blocks in reloc tree with corresponding | |
1690 | * fs tree. | |
1691 | */ | |
1692 | static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, | |
1693 | struct btrfs_root *root) | |
1694 | { | |
0b246afa | 1695 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
1696 | struct btrfs_key key; |
1697 | struct btrfs_key next_key; | |
9e6a0c52 | 1698 | struct btrfs_trans_handle *trans = NULL; |
5d4f98a2 YZ |
1699 | struct btrfs_root *reloc_root; |
1700 | struct btrfs_root_item *root_item; | |
1701 | struct btrfs_path *path; | |
3fd0a558 | 1702 | struct extent_buffer *leaf; |
fca3a45d | 1703 | int reserve_level; |
5d4f98a2 YZ |
1704 | int level; |
1705 | int max_level; | |
1706 | int replaced = 0; | |
c6a592f2 | 1707 | int ret = 0; |
3fd0a558 | 1708 | u32 min_reserved; |
5d4f98a2 YZ |
1709 | |
1710 | path = btrfs_alloc_path(); | |
1711 | if (!path) | |
1712 | return -ENOMEM; | |
e4058b54 | 1713 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
1714 | |
1715 | reloc_root = root->reloc_root; | |
1716 | root_item = &reloc_root->root_item; | |
1717 | ||
1718 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { | |
1719 | level = btrfs_root_level(root_item); | |
67439dad | 1720 | atomic_inc(&reloc_root->node->refs); |
5d4f98a2 YZ |
1721 | path->nodes[level] = reloc_root->node; |
1722 | path->slots[level] = 0; | |
1723 | } else { | |
1724 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); | |
1725 | ||
c8422684 | 1726 | level = btrfs_root_drop_level(root_item); |
5d4f98a2 YZ |
1727 | BUG_ON(level == 0); |
1728 | path->lowest_level = level; | |
1729 | ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); | |
33c66f43 | 1730 | path->lowest_level = 0; |
5d4f98a2 YZ |
1731 | if (ret < 0) { |
1732 | btrfs_free_path(path); | |
1733 | return ret; | |
1734 | } | |
1735 | ||
1736 | btrfs_node_key_to_cpu(path->nodes[level], &next_key, | |
1737 | path->slots[level]); | |
1738 | WARN_ON(memcmp(&key, &next_key, sizeof(key))); | |
1739 | ||
1740 | btrfs_unlock_up_safe(path, 0); | |
1741 | } | |
1742 | ||
44d354ab QW |
1743 | /* |
1744 | * In merge_reloc_root(), we modify the upper level pointer to swap the | |
1745 | * tree blocks between reloc tree and subvolume tree. Thus for tree | |
1746 | * block COW, we COW at most from level 1 to root level for each tree. | |
1747 | * | |
1748 | * Thus the needed metadata size is at most root_level * nodesize, | |
1749 | * and * 2 since we have two trees to COW. | |
1750 | */ | |
fca3a45d JB |
1751 | reserve_level = max_t(int, 1, btrfs_root_level(root_item)); |
1752 | min_reserved = fs_info->nodesize * reserve_level * 2; | |
3fd0a558 | 1753 | memset(&next_key, 0, sizeof(next_key)); |
5d4f98a2 | 1754 | |
3fd0a558 | 1755 | while (1) { |
9270501c JB |
1756 | ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, |
1757 | min_reserved, | |
44d354ab | 1758 | BTRFS_RESERVE_FLUSH_LIMIT); |
c6a592f2 | 1759 | if (ret) |
9e6a0c52 | 1760 | goto out; |
9e6a0c52 JB |
1761 | trans = btrfs_start_transaction(root, 0); |
1762 | if (IS_ERR(trans)) { | |
c6a592f2 | 1763 | ret = PTR_ERR(trans); |
9e6a0c52 JB |
1764 | trans = NULL; |
1765 | goto out; | |
1766 | } | |
2abc726a JB |
1767 | |
1768 | /* | |
1769 | * At this point we no longer have a reloc_control, so we can't | |
1770 | * depend on btrfs_init_reloc_root to update our last_trans. | |
1771 | * | |
1772 | * But that's ok, we started the trans handle on our | |
1773 | * corresponding fs_root, which means it's been added to the | |
1774 | * dirty list. At commit time we'll still call | |
1775 | * btrfs_update_reloc_root() and update our root item | |
1776 | * appropriately. | |
1777 | */ | |
1778 | reloc_root->last_trans = trans->transid; | |
9e6a0c52 | 1779 | trans->block_rsv = rc->block_rsv; |
5d4f98a2 | 1780 | |
5d4f98a2 | 1781 | replaced = 0; |
5d4f98a2 YZ |
1782 | max_level = level; |
1783 | ||
1784 | ret = walk_down_reloc_tree(reloc_root, path, &level); | |
c6a592f2 | 1785 | if (ret < 0) |
5d4f98a2 | 1786 | goto out; |
5d4f98a2 YZ |
1787 | if (ret > 0) |
1788 | break; | |
1789 | ||
1790 | if (!find_next_key(path, level, &key) && | |
1791 | btrfs_comp_cpu_keys(&next_key, &key) >= 0) { | |
1792 | ret = 0; | |
5d4f98a2 | 1793 | } else { |
3d0174f7 | 1794 | ret = replace_path(trans, rc, root, reloc_root, path, |
3fd0a558 | 1795 | &next_key, level, max_level); |
5d4f98a2 | 1796 | } |
c6a592f2 | 1797 | if (ret < 0) |
5d4f98a2 | 1798 | goto out; |
5d4f98a2 YZ |
1799 | if (ret > 0) { |
1800 | level = ret; | |
1801 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
1802 | path->slots[level]); | |
1803 | replaced = 1; | |
5d4f98a2 YZ |
1804 | } |
1805 | ||
1806 | ret = walk_up_reloc_tree(reloc_root, path, &level); | |
1807 | if (ret > 0) | |
1808 | break; | |
1809 | ||
1810 | BUG_ON(level == 0); | |
1811 | /* | |
1812 | * save the merging progress in the drop_progress. | |
1813 | * this is OK since root refs == 1 in this case. | |
1814 | */ | |
1815 | btrfs_node_key(path->nodes[level], &root_item->drop_progress, | |
1816 | path->slots[level]); | |
c8422684 | 1817 | btrfs_set_root_drop_level(root_item, level); |
5d4f98a2 | 1818 | |
3a45bb20 | 1819 | btrfs_end_transaction_throttle(trans); |
9e6a0c52 | 1820 | trans = NULL; |
5d4f98a2 | 1821 | |
2ff7e61e | 1822 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 YZ |
1823 | |
1824 | if (replaced && rc->stage == UPDATE_DATA_PTRS) | |
1825 | invalidate_extent_cache(root, &key, &next_key); | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * handle the case only one block in the fs tree need to be | |
1830 | * relocated and the block is tree root. | |
1831 | */ | |
1832 | leaf = btrfs_lock_root_node(root); | |
9631e4cc JB |
1833 | ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf, |
1834 | BTRFS_NESTING_COW); | |
5d4f98a2 YZ |
1835 | btrfs_tree_unlock(leaf); |
1836 | free_extent_buffer(leaf); | |
5d4f98a2 YZ |
1837 | out: |
1838 | btrfs_free_path(path); | |
1839 | ||
ac54da6c JB |
1840 | if (ret == 0) { |
1841 | ret = insert_dirty_subvol(trans, rc, root); | |
1842 | if (ret) | |
1843 | btrfs_abort_transaction(trans, ret); | |
1844 | } | |
5d4f98a2 | 1845 | |
9e6a0c52 | 1846 | if (trans) |
3a45bb20 | 1847 | btrfs_end_transaction_throttle(trans); |
5d4f98a2 | 1848 | |
2ff7e61e | 1849 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 | 1850 | |
5d4f98a2 YZ |
1851 | if (replaced && rc->stage == UPDATE_DATA_PTRS) |
1852 | invalidate_extent_cache(root, &key, &next_key); | |
1853 | ||
c6a592f2 | 1854 | return ret; |
5d4f98a2 YZ |
1855 | } |
1856 | ||
3fd0a558 YZ |
1857 | static noinline_for_stack |
1858 | int prepare_to_merge(struct reloc_control *rc, int err) | |
5d4f98a2 | 1859 | { |
3fd0a558 | 1860 | struct btrfs_root *root = rc->extent_root; |
0b246afa | 1861 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1862 | struct btrfs_root *reloc_root; |
3fd0a558 YZ |
1863 | struct btrfs_trans_handle *trans; |
1864 | LIST_HEAD(reloc_roots); | |
1865 | u64 num_bytes = 0; | |
1866 | int ret; | |
3fd0a558 | 1867 | |
0b246afa JM |
1868 | mutex_lock(&fs_info->reloc_mutex); |
1869 | rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; | |
3fd0a558 | 1870 | rc->merging_rsv_size += rc->nodes_relocated * 2; |
0b246afa | 1871 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f | 1872 | |
3fd0a558 YZ |
1873 | again: |
1874 | if (!err) { | |
1875 | num_bytes = rc->merging_rsv_size; | |
9270501c | 1876 | ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes, |
08e007d2 | 1877 | BTRFS_RESERVE_FLUSH_ALL); |
3fd0a558 YZ |
1878 | if (ret) |
1879 | err = ret; | |
1880 | } | |
1881 | ||
7a7eaa40 | 1882 | trans = btrfs_join_transaction(rc->extent_root); |
3612b495 TI |
1883 | if (IS_ERR(trans)) { |
1884 | if (!err) | |
2ff7e61e | 1885 | btrfs_block_rsv_release(fs_info, rc->block_rsv, |
63f018be | 1886 | num_bytes, NULL); |
3612b495 TI |
1887 | return PTR_ERR(trans); |
1888 | } | |
3fd0a558 YZ |
1889 | |
1890 | if (!err) { | |
1891 | if (num_bytes != rc->merging_rsv_size) { | |
3a45bb20 | 1892 | btrfs_end_transaction(trans); |
2ff7e61e | 1893 | btrfs_block_rsv_release(fs_info, rc->block_rsv, |
63f018be | 1894 | num_bytes, NULL); |
3fd0a558 YZ |
1895 | goto again; |
1896 | } | |
1897 | } | |
5d4f98a2 | 1898 | |
3fd0a558 YZ |
1899 | rc->merge_reloc_tree = 1; |
1900 | ||
1901 | while (!list_empty(&rc->reloc_roots)) { | |
1902 | reloc_root = list_entry(rc->reloc_roots.next, | |
1903 | struct btrfs_root, root_list); | |
1904 | list_del_init(&reloc_root->root_list); | |
5d4f98a2 | 1905 | |
a820feb5 DS |
1906 | root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, |
1907 | false); | |
e0b085b0 JB |
1908 | if (IS_ERR(root)) { |
1909 | /* | |
1910 | * Even if we have an error we need this reloc root | |
1911 | * back on our list so we can clean up properly. | |
1912 | */ | |
1913 | list_add(&reloc_root->root_list, &reloc_roots); | |
1914 | btrfs_abort_transaction(trans, (int)PTR_ERR(root)); | |
1915 | if (!err) | |
1916 | err = PTR_ERR(root); | |
1917 | break; | |
1918 | } | |
1919 | ASSERT(root->reloc_root == reloc_root); | |
5d4f98a2 | 1920 | |
3fd0a558 YZ |
1921 | /* |
1922 | * set reference count to 1, so btrfs_recover_relocation | |
1923 | * knows it should resumes merging | |
1924 | */ | |
1925 | if (!err) | |
1926 | btrfs_set_root_refs(&reloc_root->root_item, 1); | |
bbae13f8 | 1927 | ret = btrfs_update_reloc_root(trans, root); |
5d4f98a2 | 1928 | |
bbae13f8 JB |
1929 | /* |
1930 | * Even if we have an error we need this reloc root back on our | |
1931 | * list so we can clean up properly. | |
1932 | */ | |
3fd0a558 | 1933 | list_add(&reloc_root->root_list, &reloc_roots); |
00246528 | 1934 | btrfs_put_root(root); |
bbae13f8 JB |
1935 | |
1936 | if (ret) { | |
1937 | btrfs_abort_transaction(trans, ret); | |
1938 | if (!err) | |
1939 | err = ret; | |
1940 | break; | |
1941 | } | |
3fd0a558 | 1942 | } |
5d4f98a2 | 1943 | |
3fd0a558 | 1944 | list_splice(&reloc_roots, &rc->reloc_roots); |
5d4f98a2 | 1945 | |
3fd0a558 | 1946 | if (!err) |
fb686c68 | 1947 | err = btrfs_commit_transaction(trans); |
3fd0a558 | 1948 | else |
3a45bb20 | 1949 | btrfs_end_transaction(trans); |
3fd0a558 | 1950 | return err; |
5d4f98a2 YZ |
1951 | } |
1952 | ||
aca1bba6 LB |
1953 | static noinline_for_stack |
1954 | void free_reloc_roots(struct list_head *list) | |
1955 | { | |
a7571232 | 1956 | struct btrfs_root *reloc_root, *tmp; |
aca1bba6 | 1957 | |
a7571232 | 1958 | list_for_each_entry_safe(reloc_root, tmp, list, root_list) |
bb166d72 | 1959 | __del_reloc_root(reloc_root); |
aca1bba6 LB |
1960 | } |
1961 | ||
3fd0a558 | 1962 | static noinline_for_stack |
94404e82 | 1963 | void merge_reloc_roots(struct reloc_control *rc) |
5d4f98a2 | 1964 | { |
0b246afa | 1965 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 | 1966 | struct btrfs_root *root; |
3fd0a558 YZ |
1967 | struct btrfs_root *reloc_root; |
1968 | LIST_HEAD(reloc_roots); | |
1969 | int found = 0; | |
aca1bba6 | 1970 | int ret = 0; |
3fd0a558 YZ |
1971 | again: |
1972 | root = rc->extent_root; | |
7585717f CM |
1973 | |
1974 | /* | |
1975 | * this serializes us with btrfs_record_root_in_transaction, | |
1976 | * we have to make sure nobody is in the middle of | |
1977 | * adding their roots to the list while we are | |
1978 | * doing this splice | |
1979 | */ | |
0b246afa | 1980 | mutex_lock(&fs_info->reloc_mutex); |
3fd0a558 | 1981 | list_splice_init(&rc->reloc_roots, &reloc_roots); |
0b246afa | 1982 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 | 1983 | |
3fd0a558 YZ |
1984 | while (!list_empty(&reloc_roots)) { |
1985 | found = 1; | |
1986 | reloc_root = list_entry(reloc_roots.next, | |
1987 | struct btrfs_root, root_list); | |
5d4f98a2 | 1988 | |
a820feb5 DS |
1989 | root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, |
1990 | false); | |
3fd0a558 | 1991 | if (btrfs_root_refs(&reloc_root->root_item) > 0) { |
24213fa4 JB |
1992 | if (IS_ERR(root)) { |
1993 | /* | |
1994 | * For recovery we read the fs roots on mount, | |
1995 | * and if we didn't find the root then we marked | |
1996 | * the reloc root as a garbage root. For normal | |
1997 | * relocation obviously the root should exist in | |
1998 | * memory. However there's no reason we can't | |
1999 | * handle the error properly here just in case. | |
2000 | */ | |
2001 | ASSERT(0); | |
2002 | ret = PTR_ERR(root); | |
2003 | goto out; | |
2004 | } | |
2005 | if (root->reloc_root != reloc_root) { | |
2006 | /* | |
2007 | * This is actually impossible without something | |
2008 | * going really wrong (like weird race condition | |
2009 | * or cosmic rays). | |
2010 | */ | |
2011 | ASSERT(0); | |
2012 | ret = -EINVAL; | |
2013 | goto out; | |
2014 | } | |
3fd0a558 | 2015 | ret = merge_reloc_root(rc, root); |
00246528 | 2016 | btrfs_put_root(root); |
b37b39cd | 2017 | if (ret) { |
25e293c2 WS |
2018 | if (list_empty(&reloc_root->root_list)) |
2019 | list_add_tail(&reloc_root->root_list, | |
2020 | &reloc_roots); | |
aca1bba6 | 2021 | goto out; |
b37b39cd | 2022 | } |
3fd0a558 | 2023 | } else { |
51415b6c QW |
2024 | if (!IS_ERR(root)) { |
2025 | if (root->reloc_root == reloc_root) { | |
2026 | root->reloc_root = NULL; | |
2027 | btrfs_put_root(reloc_root); | |
2028 | } | |
1dae7e0e QW |
2029 | clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, |
2030 | &root->state); | |
51415b6c QW |
2031 | btrfs_put_root(root); |
2032 | } | |
2033 | ||
3fd0a558 | 2034 | list_del_init(&reloc_root->root_list); |
30d40577 QW |
2035 | /* Don't forget to queue this reloc root for cleanup */ |
2036 | list_add_tail(&reloc_root->reloc_dirty_list, | |
2037 | &rc->dirty_subvol_roots); | |
3fd0a558 | 2038 | } |
5d4f98a2 YZ |
2039 | } |
2040 | ||
3fd0a558 YZ |
2041 | if (found) { |
2042 | found = 0; | |
2043 | goto again; | |
2044 | } | |
aca1bba6 LB |
2045 | out: |
2046 | if (ret) { | |
0b246afa | 2047 | btrfs_handle_fs_error(fs_info, ret, NULL); |
a7571232 | 2048 | free_reloc_roots(&reloc_roots); |
467bb1d2 WS |
2049 | |
2050 | /* new reloc root may be added */ | |
0b246afa | 2051 | mutex_lock(&fs_info->reloc_mutex); |
467bb1d2 | 2052 | list_splice_init(&rc->reloc_roots, &reloc_roots); |
0b246afa | 2053 | mutex_unlock(&fs_info->reloc_mutex); |
a7571232 | 2054 | free_reloc_roots(&reloc_roots); |
aca1bba6 LB |
2055 | } |
2056 | ||
7b7b7431 JB |
2057 | /* |
2058 | * We used to have | |
2059 | * | |
2060 | * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); | |
2061 | * | |
2062 | * here, but it's wrong. If we fail to start the transaction in | |
2063 | * prepare_to_merge() we will have only 0 ref reloc roots, none of which | |
2064 | * have actually been removed from the reloc_root_tree rb tree. This is | |
2065 | * fine because we're bailing here, and we hold a reference on the root | |
2066 | * for the list that holds it, so these roots will be cleaned up when we | |
2067 | * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root | |
2068 | * will be cleaned up on unmount. | |
2069 | * | |
2070 | * The remaining nodes will be cleaned up by free_reloc_control. | |
2071 | */ | |
5d4f98a2 YZ |
2072 | } |
2073 | ||
2074 | static void free_block_list(struct rb_root *blocks) | |
2075 | { | |
2076 | struct tree_block *block; | |
2077 | struct rb_node *rb_node; | |
2078 | while ((rb_node = rb_first(blocks))) { | |
2079 | block = rb_entry(rb_node, struct tree_block, rb_node); | |
2080 | rb_erase(rb_node, blocks); | |
2081 | kfree(block); | |
2082 | } | |
2083 | } | |
2084 | ||
2085 | static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, | |
2086 | struct btrfs_root *reloc_root) | |
2087 | { | |
0b246afa | 2088 | struct btrfs_fs_info *fs_info = reloc_root->fs_info; |
5d4f98a2 | 2089 | struct btrfs_root *root; |
442b1ac5 | 2090 | int ret; |
5d4f98a2 YZ |
2091 | |
2092 | if (reloc_root->last_trans == trans->transid) | |
2093 | return 0; | |
2094 | ||
a820feb5 | 2095 | root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); |
404bccbc JB |
2096 | |
2097 | /* | |
2098 | * This should succeed, since we can't have a reloc root without having | |
2099 | * already looked up the actual root and created the reloc root for this | |
2100 | * root. | |
2101 | * | |
2102 | * However if there's some sort of corruption where we have a ref to a | |
2103 | * reloc root without a corresponding root this could return ENOENT. | |
2104 | */ | |
2105 | if (IS_ERR(root)) { | |
2106 | ASSERT(0); | |
2107 | return PTR_ERR(root); | |
2108 | } | |
2109 | if (root->reloc_root != reloc_root) { | |
2110 | ASSERT(0); | |
2111 | btrfs_err(fs_info, | |
2112 | "root %llu has two reloc roots associated with it", | |
2113 | reloc_root->root_key.offset); | |
2114 | btrfs_put_root(root); | |
2115 | return -EUCLEAN; | |
2116 | } | |
442b1ac5 | 2117 | ret = btrfs_record_root_in_trans(trans, root); |
00246528 | 2118 | btrfs_put_root(root); |
5d4f98a2 | 2119 | |
442b1ac5 | 2120 | return ret; |
5d4f98a2 YZ |
2121 | } |
2122 | ||
3fd0a558 YZ |
2123 | static noinline_for_stack |
2124 | struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, | |
2125 | struct reloc_control *rc, | |
a26195a5 QW |
2126 | struct btrfs_backref_node *node, |
2127 | struct btrfs_backref_edge *edges[]) | |
5d4f98a2 | 2128 | { |
a26195a5 | 2129 | struct btrfs_backref_node *next; |
5d4f98a2 | 2130 | struct btrfs_root *root; |
3fd0a558 | 2131 | int index = 0; |
92de551b | 2132 | int ret; |
3fd0a558 | 2133 | |
5d4f98a2 YZ |
2134 | next = node; |
2135 | while (1) { | |
2136 | cond_resched(); | |
2137 | next = walk_up_backref(next, edges, &index); | |
2138 | root = next->root; | |
8ee66afe JB |
2139 | |
2140 | /* | |
2141 | * If there is no root, then our references for this block are | |
2142 | * incomplete, as we should be able to walk all the way up to a | |
2143 | * block that is owned by a root. | |
2144 | * | |
2145 | * This path is only for SHAREABLE roots, so if we come upon a | |
2146 | * non-SHAREABLE root then we have backrefs that resolve | |
2147 | * improperly. | |
2148 | * | |
2149 | * Both of these cases indicate file system corruption, or a bug | |
2150 | * in the backref walking code. | |
2151 | */ | |
2152 | if (!root) { | |
2153 | ASSERT(0); | |
2154 | btrfs_err(trans->fs_info, | |
2155 | "bytenr %llu doesn't have a backref path ending in a root", | |
2156 | node->bytenr); | |
2157 | return ERR_PTR(-EUCLEAN); | |
2158 | } | |
2159 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { | |
2160 | ASSERT(0); | |
2161 | btrfs_err(trans->fs_info, | |
2162 | "bytenr %llu has multiple refs with one ending in a non-shareable root", | |
2163 | node->bytenr); | |
2164 | return ERR_PTR(-EUCLEAN); | |
2165 | } | |
5d4f98a2 YZ |
2166 | |
2167 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
92de551b JB |
2168 | ret = record_reloc_root_in_trans(trans, root); |
2169 | if (ret) | |
2170 | return ERR_PTR(ret); | |
5d4f98a2 YZ |
2171 | break; |
2172 | } | |
2173 | ||
92de551b JB |
2174 | ret = btrfs_record_root_in_trans(trans, root); |
2175 | if (ret) | |
2176 | return ERR_PTR(ret); | |
3fd0a558 YZ |
2177 | root = root->reloc_root; |
2178 | ||
39200e59 JB |
2179 | /* |
2180 | * We could have raced with another thread which failed, so | |
2181 | * root->reloc_root may not be set, return ENOENT in this case. | |
2182 | */ | |
2183 | if (!root) | |
2184 | return ERR_PTR(-ENOENT); | |
2185 | ||
3fd0a558 | 2186 | if (next->new_bytenr != root->node->start) { |
8ee66afe JB |
2187 | /* |
2188 | * We just created the reloc root, so we shouldn't have | |
2189 | * ->new_bytenr set and this shouldn't be in the changed | |
2190 | * list. If it is then we have multiple roots pointing | |
2191 | * at the same bytenr which indicates corruption, or | |
2192 | * we've made a mistake in the backref walking code. | |
2193 | */ | |
2194 | ASSERT(next->new_bytenr == 0); | |
2195 | ASSERT(list_empty(&next->list)); | |
2196 | if (next->new_bytenr || !list_empty(&next->list)) { | |
2197 | btrfs_err(trans->fs_info, | |
2198 | "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu", | |
2199 | node->bytenr, next->bytenr); | |
2200 | return ERR_PTR(-EUCLEAN); | |
2201 | } | |
2202 | ||
3fd0a558 | 2203 | next->new_bytenr = root->node->start; |
00246528 JB |
2204 | btrfs_put_root(next->root); |
2205 | next->root = btrfs_grab_root(root); | |
0b530bc5 | 2206 | ASSERT(next->root); |
3fd0a558 YZ |
2207 | list_add_tail(&next->list, |
2208 | &rc->backref_cache.changed); | |
9569cc20 | 2209 | mark_block_processed(rc, next); |
5d4f98a2 YZ |
2210 | break; |
2211 | } | |
2212 | ||
3fd0a558 | 2213 | WARN_ON(1); |
5d4f98a2 YZ |
2214 | root = NULL; |
2215 | next = walk_down_backref(edges, &index); | |
2216 | if (!next || next->level <= node->level) | |
2217 | break; | |
2218 | } | |
cbdc2ebc JB |
2219 | if (!root) { |
2220 | /* | |
2221 | * This can happen if there's fs corruption or if there's a bug | |
2222 | * in the backref lookup code. | |
2223 | */ | |
2224 | ASSERT(0); | |
2225 | return ERR_PTR(-ENOENT); | |
2226 | } | |
5d4f98a2 | 2227 | |
3fd0a558 YZ |
2228 | next = node; |
2229 | /* setup backref node path for btrfs_reloc_cow_block */ | |
2230 | while (1) { | |
2231 | rc->backref_cache.path[next->level] = next; | |
2232 | if (--index < 0) | |
2233 | break; | |
2234 | next = edges[index]->node[UPPER]; | |
5d4f98a2 | 2235 | } |
5d4f98a2 YZ |
2236 | return root; |
2237 | } | |
2238 | ||
3fd0a558 | 2239 | /* |
92a7cc42 QW |
2240 | * Select a tree root for relocation. |
2241 | * | |
2242 | * Return NULL if the block is not shareable. We should use do_relocation() in | |
2243 | * this case. | |
2244 | * | |
2245 | * Return a tree root pointer if the block is shareable. | |
2246 | * Return -ENOENT if the block is root of reloc tree. | |
3fd0a558 | 2247 | */ |
5d4f98a2 | 2248 | static noinline_for_stack |
a26195a5 | 2249 | struct btrfs_root *select_one_root(struct btrfs_backref_node *node) |
5d4f98a2 | 2250 | { |
a26195a5 | 2251 | struct btrfs_backref_node *next; |
3fd0a558 YZ |
2252 | struct btrfs_root *root; |
2253 | struct btrfs_root *fs_root = NULL; | |
a26195a5 | 2254 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; |
3fd0a558 YZ |
2255 | int index = 0; |
2256 | ||
2257 | next = node; | |
2258 | while (1) { | |
2259 | cond_resched(); | |
2260 | next = walk_up_backref(next, edges, &index); | |
2261 | root = next->root; | |
8717cf44 JB |
2262 | |
2263 | /* | |
2264 | * This can occur if we have incomplete extent refs leading all | |
2265 | * the way up a particular path, in this case return -EUCLEAN. | |
2266 | */ | |
2267 | if (!root) | |
2268 | return ERR_PTR(-EUCLEAN); | |
3fd0a558 | 2269 | |
92a7cc42 QW |
2270 | /* No other choice for non-shareable tree */ |
2271 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) | |
3fd0a558 YZ |
2272 | return root; |
2273 | ||
2274 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) | |
2275 | fs_root = root; | |
2276 | ||
2277 | if (next != node) | |
2278 | return NULL; | |
2279 | ||
2280 | next = walk_down_backref(edges, &index); | |
2281 | if (!next || next->level <= node->level) | |
2282 | break; | |
2283 | } | |
2284 | ||
2285 | if (!fs_root) | |
2286 | return ERR_PTR(-ENOENT); | |
2287 | return fs_root; | |
5d4f98a2 YZ |
2288 | } |
2289 | ||
2290 | static noinline_for_stack | |
3fd0a558 | 2291 | u64 calcu_metadata_size(struct reloc_control *rc, |
a26195a5 | 2292 | struct btrfs_backref_node *node, int reserve) |
5d4f98a2 | 2293 | { |
0b246afa | 2294 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
a26195a5 QW |
2295 | struct btrfs_backref_node *next = node; |
2296 | struct btrfs_backref_edge *edge; | |
2297 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
3fd0a558 YZ |
2298 | u64 num_bytes = 0; |
2299 | int index = 0; | |
2300 | ||
2301 | BUG_ON(reserve && node->processed); | |
2302 | ||
2303 | while (next) { | |
2304 | cond_resched(); | |
2305 | while (1) { | |
2306 | if (next->processed && (reserve || next != node)) | |
2307 | break; | |
2308 | ||
0b246afa | 2309 | num_bytes += fs_info->nodesize; |
3fd0a558 YZ |
2310 | |
2311 | if (list_empty(&next->upper)) | |
2312 | break; | |
2313 | ||
2314 | edge = list_entry(next->upper.next, | |
a26195a5 | 2315 | struct btrfs_backref_edge, list[LOWER]); |
3fd0a558 YZ |
2316 | edges[index++] = edge; |
2317 | next = edge->node[UPPER]; | |
2318 | } | |
2319 | next = walk_down_backref(edges, &index); | |
2320 | } | |
2321 | return num_bytes; | |
5d4f98a2 YZ |
2322 | } |
2323 | ||
3fd0a558 YZ |
2324 | static int reserve_metadata_space(struct btrfs_trans_handle *trans, |
2325 | struct reloc_control *rc, | |
a26195a5 | 2326 | struct btrfs_backref_node *node) |
5d4f98a2 | 2327 | { |
3fd0a558 | 2328 | struct btrfs_root *root = rc->extent_root; |
da17066c | 2329 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 YZ |
2330 | u64 num_bytes; |
2331 | int ret; | |
0647bf56 | 2332 | u64 tmp; |
3fd0a558 YZ |
2333 | |
2334 | num_bytes = calcu_metadata_size(rc, node, 1) * 2; | |
5d4f98a2 | 2335 | |
3fd0a558 | 2336 | trans->block_rsv = rc->block_rsv; |
0647bf56 | 2337 | rc->reserved_bytes += num_bytes; |
8ca17f0f JB |
2338 | |
2339 | /* | |
2340 | * We are under a transaction here so we can only do limited flushing. | |
2341 | * If we get an enospc just kick back -EAGAIN so we know to drop the | |
2342 | * transaction and try to refill when we can flush all the things. | |
2343 | */ | |
9270501c JB |
2344 | ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes, |
2345 | BTRFS_RESERVE_FLUSH_LIMIT); | |
3fd0a558 | 2346 | if (ret) { |
da17066c | 2347 | tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; |
8ca17f0f JB |
2348 | while (tmp <= rc->reserved_bytes) |
2349 | tmp <<= 1; | |
2350 | /* | |
2351 | * only one thread can access block_rsv at this point, | |
2352 | * so we don't need hold lock to protect block_rsv. | |
2353 | * we expand more reservation size here to allow enough | |
52042d8e | 2354 | * space for relocation and we will return earlier in |
8ca17f0f JB |
2355 | * enospc case. |
2356 | */ | |
da17066c JM |
2357 | rc->block_rsv->size = tmp + fs_info->nodesize * |
2358 | RELOCATION_RESERVED_NODES; | |
8ca17f0f | 2359 | return -EAGAIN; |
5d4f98a2 | 2360 | } |
3fd0a558 | 2361 | |
3fd0a558 YZ |
2362 | return 0; |
2363 | } | |
2364 | ||
5d4f98a2 YZ |
2365 | /* |
2366 | * relocate a block tree, and then update pointers in upper level | |
2367 | * blocks that reference the block to point to the new location. | |
2368 | * | |
2369 | * if called by link_to_upper, the block has already been relocated. | |
2370 | * in that case this function just updates pointers. | |
2371 | */ | |
2372 | static int do_relocation(struct btrfs_trans_handle *trans, | |
3fd0a558 | 2373 | struct reloc_control *rc, |
a26195a5 | 2374 | struct btrfs_backref_node *node, |
5d4f98a2 YZ |
2375 | struct btrfs_key *key, |
2376 | struct btrfs_path *path, int lowest) | |
2377 | { | |
a26195a5 QW |
2378 | struct btrfs_backref_node *upper; |
2379 | struct btrfs_backref_edge *edge; | |
2380 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
5d4f98a2 YZ |
2381 | struct btrfs_root *root; |
2382 | struct extent_buffer *eb; | |
2383 | u32 blocksize; | |
2384 | u64 bytenr; | |
5d4f98a2 | 2385 | int slot; |
8df01fdd | 2386 | int ret = 0; |
5d4f98a2 | 2387 | |
ffe30dd8 JB |
2388 | /* |
2389 | * If we are lowest then this is the first time we're processing this | |
2390 | * block, and thus shouldn't have an eb associated with it yet. | |
2391 | */ | |
2392 | ASSERT(!lowest || !node->eb); | |
5d4f98a2 YZ |
2393 | |
2394 | path->lowest_level = node->level + 1; | |
3fd0a558 | 2395 | rc->backref_cache.path[node->level] = node; |
5d4f98a2 | 2396 | list_for_each_entry(edge, &node->upper, list[LOWER]) { |
82fa113f | 2397 | struct btrfs_ref ref = { 0 }; |
581c1760 | 2398 | |
5d4f98a2 | 2399 | cond_resched(); |
5d4f98a2 YZ |
2400 | |
2401 | upper = edge->node[UPPER]; | |
dc4103f9 | 2402 | root = select_reloc_root(trans, rc, upper, edges); |
cbdc2ebc JB |
2403 | if (IS_ERR(root)) { |
2404 | ret = PTR_ERR(root); | |
2405 | goto next; | |
2406 | } | |
3fd0a558 YZ |
2407 | |
2408 | if (upper->eb && !upper->locked) { | |
2409 | if (!lowest) { | |
fdf8d595 | 2410 | ret = btrfs_bin_search(upper->eb, 0, key, &slot); |
8df01fdd | 2411 | if (ret < 0) |
cbca7d59 | 2412 | goto next; |
3fd0a558 YZ |
2413 | BUG_ON(ret); |
2414 | bytenr = btrfs_node_blockptr(upper->eb, slot); | |
2415 | if (node->eb->start == bytenr) | |
2416 | goto next; | |
2417 | } | |
b0fe7078 | 2418 | btrfs_backref_drop_node_buffer(upper); |
3fd0a558 | 2419 | } |
5d4f98a2 YZ |
2420 | |
2421 | if (!upper->eb) { | |
2422 | ret = btrfs_search_slot(trans, root, key, path, 0, 1); | |
3561b9db | 2423 | if (ret) { |
8df01fdd NB |
2424 | if (ret > 0) |
2425 | ret = -ENOENT; | |
3561b9db LB |
2426 | |
2427 | btrfs_release_path(path); | |
5d4f98a2 YZ |
2428 | break; |
2429 | } | |
5d4f98a2 | 2430 | |
3fd0a558 YZ |
2431 | if (!upper->eb) { |
2432 | upper->eb = path->nodes[upper->level]; | |
2433 | path->nodes[upper->level] = NULL; | |
2434 | } else { | |
2435 | BUG_ON(upper->eb != path->nodes[upper->level]); | |
2436 | } | |
5d4f98a2 | 2437 | |
3fd0a558 YZ |
2438 | upper->locked = 1; |
2439 | path->locks[upper->level] = 0; | |
5d4f98a2 | 2440 | |
3fd0a558 | 2441 | slot = path->slots[upper->level]; |
b3b4aa74 | 2442 | btrfs_release_path(path); |
5d4f98a2 | 2443 | } else { |
fdf8d595 | 2444 | ret = btrfs_bin_search(upper->eb, 0, key, &slot); |
8df01fdd | 2445 | if (ret < 0) |
cbca7d59 | 2446 | goto next; |
5d4f98a2 YZ |
2447 | BUG_ON(ret); |
2448 | } | |
2449 | ||
2450 | bytenr = btrfs_node_blockptr(upper->eb, slot); | |
3fd0a558 | 2451 | if (lowest) { |
4547f4d8 LB |
2452 | if (bytenr != node->bytenr) { |
2453 | btrfs_err(root->fs_info, | |
2454 | "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", | |
2455 | bytenr, node->bytenr, slot, | |
2456 | upper->eb->start); | |
8df01fdd | 2457 | ret = -EIO; |
4547f4d8 LB |
2458 | goto next; |
2459 | } | |
5d4f98a2 | 2460 | } else { |
3fd0a558 YZ |
2461 | if (node->eb->start == bytenr) |
2462 | goto next; | |
5d4f98a2 YZ |
2463 | } |
2464 | ||
da17066c | 2465 | blocksize = root->fs_info->nodesize; |
c9752536 | 2466 | eb = btrfs_read_node_slot(upper->eb, slot); |
64c043de | 2467 | if (IS_ERR(eb)) { |
8df01fdd | 2468 | ret = PTR_ERR(eb); |
64c043de | 2469 | goto next; |
97d9a8a4 | 2470 | } |
5d4f98a2 | 2471 | btrfs_tree_lock(eb); |
5d4f98a2 YZ |
2472 | |
2473 | if (!node->eb) { | |
2474 | ret = btrfs_cow_block(trans, root, eb, upper->eb, | |
9631e4cc | 2475 | slot, &eb, BTRFS_NESTING_COW); |
3fd0a558 YZ |
2476 | btrfs_tree_unlock(eb); |
2477 | free_extent_buffer(eb); | |
8df01fdd | 2478 | if (ret < 0) |
3fd0a558 | 2479 | goto next; |
ffe30dd8 JB |
2480 | /* |
2481 | * We've just COWed this block, it should have updated | |
2482 | * the correct backref node entry. | |
2483 | */ | |
2484 | ASSERT(node->eb == eb); | |
5d4f98a2 YZ |
2485 | } else { |
2486 | btrfs_set_node_blockptr(upper->eb, slot, | |
2487 | node->eb->start); | |
2488 | btrfs_set_node_ptr_generation(upper->eb, slot, | |
2489 | trans->transid); | |
2490 | btrfs_mark_buffer_dirty(upper->eb); | |
2491 | ||
82fa113f QW |
2492 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, |
2493 | node->eb->start, blocksize, | |
2494 | upper->eb->start); | |
82fa113f | 2495 | btrfs_init_tree_ref(&ref, node->level, |
f42c5da6 NB |
2496 | btrfs_header_owner(upper->eb), |
2497 | root->root_key.objectid, false); | |
82fa113f | 2498 | ret = btrfs_inc_extent_ref(trans, &ref); |
eb6b7fb4 JB |
2499 | if (!ret) |
2500 | ret = btrfs_drop_subtree(trans, root, eb, | |
2501 | upper->eb); | |
2502 | if (ret) | |
2503 | btrfs_abort_transaction(trans, ret); | |
5d4f98a2 | 2504 | } |
3fd0a558 YZ |
2505 | next: |
2506 | if (!upper->pending) | |
b0fe7078 | 2507 | btrfs_backref_drop_node_buffer(upper); |
3fd0a558 | 2508 | else |
b0fe7078 | 2509 | btrfs_backref_unlock_node_buffer(upper); |
8df01fdd | 2510 | if (ret) |
3fd0a558 | 2511 | break; |
5d4f98a2 | 2512 | } |
3fd0a558 | 2513 | |
8df01fdd | 2514 | if (!ret && node->pending) { |
b0fe7078 | 2515 | btrfs_backref_drop_node_buffer(node); |
3fd0a558 YZ |
2516 | list_move_tail(&node->list, &rc->backref_cache.changed); |
2517 | node->pending = 0; | |
2518 | } | |
2519 | ||
5d4f98a2 | 2520 | path->lowest_level = 0; |
ffe30dd8 JB |
2521 | |
2522 | /* | |
2523 | * We should have allocated all of our space in the block rsv and thus | |
2524 | * shouldn't ENOSPC. | |
2525 | */ | |
2526 | ASSERT(ret != -ENOSPC); | |
8df01fdd | 2527 | return ret; |
5d4f98a2 YZ |
2528 | } |
2529 | ||
2530 | static int link_to_upper(struct btrfs_trans_handle *trans, | |
3fd0a558 | 2531 | struct reloc_control *rc, |
a26195a5 | 2532 | struct btrfs_backref_node *node, |
5d4f98a2 YZ |
2533 | struct btrfs_path *path) |
2534 | { | |
2535 | struct btrfs_key key; | |
5d4f98a2 YZ |
2536 | |
2537 | btrfs_node_key_to_cpu(node->eb, &key, 0); | |
3fd0a558 | 2538 | return do_relocation(trans, rc, node, &key, path, 0); |
5d4f98a2 YZ |
2539 | } |
2540 | ||
2541 | static int finish_pending_nodes(struct btrfs_trans_handle *trans, | |
3fd0a558 YZ |
2542 | struct reloc_control *rc, |
2543 | struct btrfs_path *path, int err) | |
5d4f98a2 | 2544 | { |
3fd0a558 | 2545 | LIST_HEAD(list); |
a26195a5 QW |
2546 | struct btrfs_backref_cache *cache = &rc->backref_cache; |
2547 | struct btrfs_backref_node *node; | |
5d4f98a2 YZ |
2548 | int level; |
2549 | int ret; | |
5d4f98a2 YZ |
2550 | |
2551 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { | |
2552 | while (!list_empty(&cache->pending[level])) { | |
2553 | node = list_entry(cache->pending[level].next, | |
a26195a5 | 2554 | struct btrfs_backref_node, list); |
3fd0a558 YZ |
2555 | list_move_tail(&node->list, &list); |
2556 | BUG_ON(!node->pending); | |
5d4f98a2 | 2557 | |
3fd0a558 YZ |
2558 | if (!err) { |
2559 | ret = link_to_upper(trans, rc, node, path); | |
2560 | if (ret < 0) | |
2561 | err = ret; | |
2562 | } | |
5d4f98a2 | 2563 | } |
3fd0a558 | 2564 | list_splice_init(&list, &cache->pending[level]); |
5d4f98a2 | 2565 | } |
5d4f98a2 YZ |
2566 | return err; |
2567 | } | |
2568 | ||
5d4f98a2 YZ |
2569 | /* |
2570 | * mark a block and all blocks directly/indirectly reference the block | |
2571 | * as processed. | |
2572 | */ | |
2573 | static void update_processed_blocks(struct reloc_control *rc, | |
a26195a5 | 2574 | struct btrfs_backref_node *node) |
5d4f98a2 | 2575 | { |
a26195a5 QW |
2576 | struct btrfs_backref_node *next = node; |
2577 | struct btrfs_backref_edge *edge; | |
2578 | struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; | |
5d4f98a2 YZ |
2579 | int index = 0; |
2580 | ||
2581 | while (next) { | |
2582 | cond_resched(); | |
2583 | while (1) { | |
2584 | if (next->processed) | |
2585 | break; | |
2586 | ||
9569cc20 | 2587 | mark_block_processed(rc, next); |
5d4f98a2 YZ |
2588 | |
2589 | if (list_empty(&next->upper)) | |
2590 | break; | |
2591 | ||
2592 | edge = list_entry(next->upper.next, | |
a26195a5 | 2593 | struct btrfs_backref_edge, list[LOWER]); |
5d4f98a2 YZ |
2594 | edges[index++] = edge; |
2595 | next = edge->node[UPPER]; | |
2596 | } | |
2597 | next = walk_down_backref(edges, &index); | |
2598 | } | |
2599 | } | |
2600 | ||
7476dfda | 2601 | static int tree_block_processed(u64 bytenr, struct reloc_control *rc) |
3fd0a558 | 2602 | { |
da17066c | 2603 | u32 blocksize = rc->extent_root->fs_info->nodesize; |
7476dfda | 2604 | |
3fd0a558 YZ |
2605 | if (test_range_bit(&rc->processed_blocks, bytenr, |
2606 | bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) | |
2607 | return 1; | |
2608 | return 0; | |
5d4f98a2 YZ |
2609 | } |
2610 | ||
2ff7e61e | 2611 | static int get_tree_block_key(struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2612 | struct tree_block *block) |
2613 | { | |
789d6a3a QW |
2614 | struct btrfs_tree_parent_check check = { |
2615 | .level = block->level, | |
2616 | .owner_root = block->owner, | |
2617 | .transid = block->key.offset | |
2618 | }; | |
5d4f98a2 YZ |
2619 | struct extent_buffer *eb; |
2620 | ||
789d6a3a | 2621 | eb = read_tree_block(fs_info, block->bytenr, &check); |
4eb150d6 | 2622 | if (IS_ERR(eb)) |
64c043de | 2623 | return PTR_ERR(eb); |
4eb150d6 | 2624 | if (!extent_buffer_uptodate(eb)) { |
416bc658 JB |
2625 | free_extent_buffer(eb); |
2626 | return -EIO; | |
2627 | } | |
5d4f98a2 YZ |
2628 | if (block->level == 0) |
2629 | btrfs_item_key_to_cpu(eb, &block->key, 0); | |
2630 | else | |
2631 | btrfs_node_key_to_cpu(eb, &block->key, 0); | |
2632 | free_extent_buffer(eb); | |
2633 | block->key_ready = 1; | |
2634 | return 0; | |
2635 | } | |
2636 | ||
5d4f98a2 YZ |
2637 | /* |
2638 | * helper function to relocate a tree block | |
2639 | */ | |
2640 | static int relocate_tree_block(struct btrfs_trans_handle *trans, | |
2641 | struct reloc_control *rc, | |
a26195a5 | 2642 | struct btrfs_backref_node *node, |
5d4f98a2 YZ |
2643 | struct btrfs_key *key, |
2644 | struct btrfs_path *path) | |
2645 | { | |
2646 | struct btrfs_root *root; | |
3fd0a558 YZ |
2647 | int ret = 0; |
2648 | ||
2649 | if (!node) | |
2650 | return 0; | |
5d4f98a2 | 2651 | |
5f6b2e5c JB |
2652 | /* |
2653 | * If we fail here we want to drop our backref_node because we are going | |
2654 | * to start over and regenerate the tree for it. | |
2655 | */ | |
2656 | ret = reserve_metadata_space(trans, rc, node); | |
2657 | if (ret) | |
2658 | goto out; | |
2659 | ||
3fd0a558 | 2660 | BUG_ON(node->processed); |
147d256e | 2661 | root = select_one_root(node); |
8717cf44 JB |
2662 | if (IS_ERR(root)) { |
2663 | ret = PTR_ERR(root); | |
2664 | ||
2665 | /* See explanation in select_one_root for the -EUCLEAN case. */ | |
2666 | ASSERT(ret == -ENOENT); | |
2667 | if (ret == -ENOENT) { | |
2668 | ret = 0; | |
2669 | update_processed_blocks(rc, node); | |
2670 | } | |
3fd0a558 | 2671 | goto out; |
5d4f98a2 YZ |
2672 | } |
2673 | ||
3fd0a558 | 2674 | if (root) { |
92a7cc42 | 2675 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
1c7bfa15 JB |
2676 | /* |
2677 | * This block was the root block of a root, and this is | |
2678 | * the first time we're processing the block and thus it | |
2679 | * should not have had the ->new_bytenr modified and | |
2680 | * should have not been included on the changed list. | |
2681 | * | |
2682 | * However in the case of corruption we could have | |
2683 | * multiple refs pointing to the same block improperly, | |
2684 | * and thus we would trip over these checks. ASSERT() | |
2685 | * for the developer case, because it could indicate a | |
2686 | * bug in the backref code, however error out for a | |
2687 | * normal user in the case of corruption. | |
2688 | */ | |
2689 | ASSERT(node->new_bytenr == 0); | |
2690 | ASSERT(list_empty(&node->list)); | |
2691 | if (node->new_bytenr || !list_empty(&node->list)) { | |
2692 | btrfs_err(root->fs_info, | |
2693 | "bytenr %llu has improper references to it", | |
2694 | node->bytenr); | |
2695 | ret = -EUCLEAN; | |
2696 | goto out; | |
2697 | } | |
d18c7bd9 JB |
2698 | ret = btrfs_record_root_in_trans(trans, root); |
2699 | if (ret) | |
2700 | goto out; | |
39200e59 JB |
2701 | /* |
2702 | * Another thread could have failed, need to check if we | |
2703 | * have reloc_root actually set. | |
2704 | */ | |
2705 | if (!root->reloc_root) { | |
2706 | ret = -ENOENT; | |
2707 | goto out; | |
2708 | } | |
3fd0a558 YZ |
2709 | root = root->reloc_root; |
2710 | node->new_bytenr = root->node->start; | |
00246528 JB |
2711 | btrfs_put_root(node->root); |
2712 | node->root = btrfs_grab_root(root); | |
0b530bc5 | 2713 | ASSERT(node->root); |
3fd0a558 YZ |
2714 | list_add_tail(&node->list, &rc->backref_cache.changed); |
2715 | } else { | |
2716 | path->lowest_level = node->level; | |
2bb2e00e FM |
2717 | if (root == root->fs_info->chunk_root) |
2718 | btrfs_reserve_chunk_metadata(trans, false); | |
3fd0a558 | 2719 | ret = btrfs_search_slot(trans, root, key, path, 0, 1); |
b3b4aa74 | 2720 | btrfs_release_path(path); |
2bb2e00e FM |
2721 | if (root == root->fs_info->chunk_root) |
2722 | btrfs_trans_release_chunk_metadata(trans); | |
3fd0a558 YZ |
2723 | if (ret > 0) |
2724 | ret = 0; | |
2725 | } | |
2726 | if (!ret) | |
2727 | update_processed_blocks(rc, node); | |
2728 | } else { | |
2729 | ret = do_relocation(trans, rc, node, key, path, 1); | |
2730 | } | |
5d4f98a2 | 2731 | out: |
0647bf56 | 2732 | if (ret || node->level == 0 || node->cowonly) |
023acb07 | 2733 | btrfs_backref_cleanup_node(&rc->backref_cache, node); |
5d4f98a2 YZ |
2734 | return ret; |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * relocate a list of blocks | |
2739 | */ | |
2740 | static noinline_for_stack | |
2741 | int relocate_tree_blocks(struct btrfs_trans_handle *trans, | |
2742 | struct reloc_control *rc, struct rb_root *blocks) | |
2743 | { | |
2ff7e61e | 2744 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
a26195a5 | 2745 | struct btrfs_backref_node *node; |
5d4f98a2 YZ |
2746 | struct btrfs_path *path; |
2747 | struct tree_block *block; | |
98ff7b94 | 2748 | struct tree_block *next; |
5d4f98a2 YZ |
2749 | int ret; |
2750 | int err = 0; | |
2751 | ||
2752 | path = btrfs_alloc_path(); | |
e1a12670 LB |
2753 | if (!path) { |
2754 | err = -ENOMEM; | |
34c2b290 | 2755 | goto out_free_blocks; |
e1a12670 | 2756 | } |
5d4f98a2 | 2757 | |
98ff7b94 QW |
2758 | /* Kick in readahead for tree blocks with missing keys */ |
2759 | rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { | |
5d4f98a2 | 2760 | if (!block->key_ready) |
f7ba2d37 JB |
2761 | btrfs_readahead_tree_block(fs_info, block->bytenr, |
2762 | block->owner, 0, | |
3fbaf258 | 2763 | block->level); |
5d4f98a2 YZ |
2764 | } |
2765 | ||
98ff7b94 QW |
2766 | /* Get first keys */ |
2767 | rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { | |
34c2b290 | 2768 | if (!block->key_ready) { |
2ff7e61e | 2769 | err = get_tree_block_key(fs_info, block); |
34c2b290 DS |
2770 | if (err) |
2771 | goto out_free_path; | |
2772 | } | |
5d4f98a2 YZ |
2773 | } |
2774 | ||
98ff7b94 QW |
2775 | /* Do tree relocation */ |
2776 | rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { | |
3fd0a558 | 2777 | node = build_backref_tree(rc, &block->key, |
5d4f98a2 YZ |
2778 | block->level, block->bytenr); |
2779 | if (IS_ERR(node)) { | |
2780 | err = PTR_ERR(node); | |
2781 | goto out; | |
2782 | } | |
2783 | ||
2784 | ret = relocate_tree_block(trans, rc, node, &block->key, | |
2785 | path); | |
2786 | if (ret < 0) { | |
50dbbb71 JB |
2787 | err = ret; |
2788 | break; | |
5d4f98a2 | 2789 | } |
5d4f98a2 | 2790 | } |
5d4f98a2 | 2791 | out: |
3fd0a558 | 2792 | err = finish_pending_nodes(trans, rc, path, err); |
5d4f98a2 | 2793 | |
34c2b290 | 2794 | out_free_path: |
5d4f98a2 | 2795 | btrfs_free_path(path); |
34c2b290 | 2796 | out_free_blocks: |
e1a12670 | 2797 | free_block_list(blocks); |
5d4f98a2 YZ |
2798 | return err; |
2799 | } | |
2800 | ||
056d9bec NB |
2801 | static noinline_for_stack int prealloc_file_extent_cluster( |
2802 | struct btrfs_inode *inode, | |
2803 | struct file_extent_cluster *cluster) | |
efa56464 YZ |
2804 | { |
2805 | u64 alloc_hint = 0; | |
2806 | u64 start; | |
2807 | u64 end; | |
056d9bec | 2808 | u64 offset = inode->index_cnt; |
efa56464 | 2809 | u64 num_bytes; |
4e9d0d01 | 2810 | int nr; |
efa56464 | 2811 | int ret = 0; |
9d9ea1e6 | 2812 | u64 i_size = i_size_read(&inode->vfs_inode); |
dcb40c19 WX |
2813 | u64 prealloc_start = cluster->start - offset; |
2814 | u64 prealloc_end = cluster->end - offset; | |
214e61d0 | 2815 | u64 cur_offset = prealloc_start; |
efa56464 | 2816 | |
9d9ea1e6 QW |
2817 | /* |
2818 | * For subpage case, previous i_size may not be aligned to PAGE_SIZE. | |
2819 | * This means the range [i_size, PAGE_END + 1) is filled with zeros by | |
2820 | * btrfs_do_readpage() call of previously relocated file cluster. | |
2821 | * | |
2822 | * If the current cluster starts in the above range, btrfs_do_readpage() | |
2823 | * will skip the read, and relocate_one_page() will later writeback | |
2824 | * the padding zeros as new data, causing data corruption. | |
2825 | * | |
2826 | * Here we have to manually invalidate the range (i_size, PAGE_END + 1). | |
2827 | */ | |
ce394a7f | 2828 | if (!PAGE_ALIGNED(i_size)) { |
9d9ea1e6 QW |
2829 | struct address_space *mapping = inode->vfs_inode.i_mapping; |
2830 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
2831 | const u32 sectorsize = fs_info->sectorsize; | |
2832 | struct page *page; | |
2833 | ||
2834 | ASSERT(sectorsize < PAGE_SIZE); | |
2835 | ASSERT(IS_ALIGNED(i_size, sectorsize)); | |
2836 | ||
2837 | /* | |
2838 | * Subpage can't handle page with DIRTY but without UPTODATE | |
2839 | * bit as it can lead to the following deadlock: | |
2840 | * | |
fb12489b | 2841 | * btrfs_read_folio() |
9d9ea1e6 QW |
2842 | * | Page already *locked* |
2843 | * |- btrfs_lock_and_flush_ordered_range() | |
2844 | * |- btrfs_start_ordered_extent() | |
2845 | * |- extent_write_cache_pages() | |
2846 | * |- lock_page() | |
2847 | * We try to lock the page we already hold. | |
2848 | * | |
2849 | * Here we just writeback the whole data reloc inode, so that | |
2850 | * we will be ensured to have no dirty range in the page, and | |
2851 | * are safe to clear the uptodate bits. | |
2852 | * | |
2853 | * This shouldn't cause too much overhead, as we need to write | |
2854 | * the data back anyway. | |
2855 | */ | |
2856 | ret = filemap_write_and_wait(mapping); | |
2857 | if (ret < 0) | |
2858 | return ret; | |
2859 | ||
2860 | clear_extent_bits(&inode->io_tree, i_size, | |
2861 | round_up(i_size, PAGE_SIZE) - 1, | |
2862 | EXTENT_UPTODATE); | |
2863 | page = find_lock_page(mapping, i_size >> PAGE_SHIFT); | |
2864 | /* | |
2865 | * If page is freed we don't need to do anything then, as we | |
2866 | * will re-read the whole page anyway. | |
2867 | */ | |
2868 | if (page) { | |
2869 | btrfs_subpage_clear_uptodate(fs_info, page, i_size, | |
2870 | round_up(i_size, PAGE_SIZE) - i_size); | |
2871 | unlock_page(page); | |
2872 | put_page(page); | |
2873 | } | |
2874 | } | |
2875 | ||
efa56464 | 2876 | BUG_ON(cluster->start != cluster->boundary[0]); |
056d9bec | 2877 | ret = btrfs_alloc_data_chunk_ondemand(inode, |
a89ef455 | 2878 | prealloc_end + 1 - prealloc_start); |
efa56464 | 2879 | if (ret) |
214e61d0 | 2880 | return ret; |
efa56464 | 2881 | |
29b6352b | 2882 | btrfs_inode_lock(inode, 0); |
4e9d0d01 | 2883 | for (nr = 0; nr < cluster->nr; nr++) { |
9c5c9604 JB |
2884 | struct extent_state *cached_state = NULL; |
2885 | ||
efa56464 YZ |
2886 | start = cluster->boundary[nr] - offset; |
2887 | if (nr + 1 < cluster->nr) | |
2888 | end = cluster->boundary[nr + 1] - 1 - offset; | |
2889 | else | |
2890 | end = cluster->end - offset; | |
2891 | ||
9c5c9604 | 2892 | lock_extent(&inode->io_tree, start, end, &cached_state); |
efa56464 | 2893 | num_bytes = end + 1 - start; |
056d9bec | 2894 | ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start, |
efa56464 YZ |
2895 | num_bytes, num_bytes, |
2896 | end + 1, &alloc_hint); | |
18513091 | 2897 | cur_offset = end + 1; |
9c5c9604 | 2898 | unlock_extent(&inode->io_tree, start, end, &cached_state); |
efa56464 YZ |
2899 | if (ret) |
2900 | break; | |
efa56464 | 2901 | } |
e5d4d75b | 2902 | btrfs_inode_unlock(inode, 0); |
214e61d0 | 2903 | |
18513091 | 2904 | if (cur_offset < prealloc_end) |
056d9bec | 2905 | btrfs_free_reserved_data_space_noquota(inode->root->fs_info, |
a89ef455 | 2906 | prealloc_end + 1 - cur_offset); |
efa56464 YZ |
2907 | return ret; |
2908 | } | |
2909 | ||
4b01c44f JT |
2910 | static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode, |
2911 | u64 start, u64 end, u64 block_start) | |
0257bb82 | 2912 | { |
0257bb82 | 2913 | struct extent_map *em; |
9c5c9604 | 2914 | struct extent_state *cached_state = NULL; |
0257bb82 YZ |
2915 | int ret = 0; |
2916 | ||
172ddd60 | 2917 | em = alloc_extent_map(); |
0257bb82 YZ |
2918 | if (!em) |
2919 | return -ENOMEM; | |
2920 | ||
2921 | em->start = start; | |
2922 | em->len = end + 1 - start; | |
2923 | em->block_len = em->len; | |
2924 | em->block_start = block_start; | |
0257bb82 YZ |
2925 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
2926 | ||
9c5c9604 | 2927 | lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); |
a1ba4c08 | 2928 | ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false); |
9c5c9604 | 2929 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state); |
a1ba4c08 FM |
2930 | free_extent_map(em); |
2931 | ||
0257bb82 YZ |
2932 | return ret; |
2933 | } | |
2934 | ||
726a3421 | 2935 | /* |
907d2710 | 2936 | * Allow error injection to test balance/relocation cancellation |
726a3421 | 2937 | */ |
1fec12a5 | 2938 | noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) |
726a3421 | 2939 | { |
5cb502f4 | 2940 | return atomic_read(&fs_info->balance_cancel_req) || |
907d2710 | 2941 | atomic_read(&fs_info->reloc_cancel_req) || |
5cb502f4 | 2942 | fatal_signal_pending(current); |
726a3421 QW |
2943 | } |
2944 | ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); | |
2945 | ||
c2832898 QW |
2946 | static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster, |
2947 | int cluster_nr) | |
2948 | { | |
2949 | /* Last extent, use cluster end directly */ | |
2950 | if (cluster_nr >= cluster->nr - 1) | |
2951 | return cluster->end; | |
2952 | ||
2953 | /* Use next boundary start*/ | |
2954 | return cluster->boundary[cluster_nr + 1] - 1; | |
2955 | } | |
2956 | ||
f47960f4 QW |
2957 | static int relocate_one_page(struct inode *inode, struct file_ra_state *ra, |
2958 | struct file_extent_cluster *cluster, | |
2959 | int *cluster_nr, unsigned long page_index) | |
5d4f98a2 | 2960 | { |
2ff7e61e | 2961 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f47960f4 QW |
2962 | u64 offset = BTRFS_I(inode)->index_cnt; |
2963 | const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT; | |
2964 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); | |
2965 | struct page *page; | |
5d4f98a2 YZ |
2966 | u64 page_start; |
2967 | u64 page_end; | |
c2832898 | 2968 | u64 cur; |
f47960f4 QW |
2969 | int ret; |
2970 | ||
2971 | ASSERT(page_index <= last_index); | |
f47960f4 QW |
2972 | page = find_lock_page(inode->i_mapping, page_index); |
2973 | if (!page) { | |
2974 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, | |
2975 | page_index, last_index + 1 - page_index); | |
2976 | page = find_or_create_page(inode->i_mapping, page_index, mask); | |
c2832898 QW |
2977 | if (!page) |
2978 | return -ENOMEM; | |
f47960f4 QW |
2979 | } |
2980 | ret = set_page_extent_mapped(page); | |
2981 | if (ret < 0) | |
2982 | goto release_page; | |
2983 | ||
2984 | if (PageReadahead(page)) | |
2ebdd1df MWO |
2985 | page_cache_async_readahead(inode->i_mapping, ra, NULL, |
2986 | page_folio(page), page_index, | |
2987 | last_index + 1 - page_index); | |
f47960f4 QW |
2988 | |
2989 | if (!PageUptodate(page)) { | |
fb12489b | 2990 | btrfs_read_folio(NULL, page_folio(page)); |
f47960f4 QW |
2991 | lock_page(page); |
2992 | if (!PageUptodate(page)) { | |
2993 | ret = -EIO; | |
2994 | goto release_page; | |
2995 | } | |
2996 | } | |
2997 | ||
2998 | page_start = page_offset(page); | |
2999 | page_end = page_start + PAGE_SIZE - 1; | |
3000 | ||
c2832898 QW |
3001 | /* |
3002 | * Start from the cluster, as for subpage case, the cluster can start | |
3003 | * inside the page. | |
3004 | */ | |
3005 | cur = max(page_start, cluster->boundary[*cluster_nr] - offset); | |
3006 | while (cur <= page_end) { | |
9c5c9604 | 3007 | struct extent_state *cached_state = NULL; |
c2832898 QW |
3008 | u64 extent_start = cluster->boundary[*cluster_nr] - offset; |
3009 | u64 extent_end = get_cluster_boundary_end(cluster, | |
3010 | *cluster_nr) - offset; | |
3011 | u64 clamped_start = max(page_start, extent_start); | |
3012 | u64 clamped_end = min(page_end, extent_end); | |
3013 | u32 clamped_len = clamped_end + 1 - clamped_start; | |
3014 | ||
3015 | /* Reserve metadata for this range */ | |
3016 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), | |
d4135134 FM |
3017 | clamped_len, clamped_len, |
3018 | false); | |
c2832898 QW |
3019 | if (ret) |
3020 | goto release_page; | |
f47960f4 | 3021 | |
c2832898 | 3022 | /* Mark the range delalloc and dirty for later writeback */ |
9c5c9604 JB |
3023 | lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end, |
3024 | &cached_state); | |
c2832898 | 3025 | ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start, |
9c5c9604 | 3026 | clamped_end, 0, &cached_state); |
c2832898 | 3027 | if (ret) { |
9c5c9604 JB |
3028 | clear_extent_bit(&BTRFS_I(inode)->io_tree, |
3029 | clamped_start, clamped_end, | |
3030 | EXTENT_LOCKED | EXTENT_BOUNDARY, | |
3031 | &cached_state); | |
c2832898 QW |
3032 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
3033 | clamped_len, true); | |
3034 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
3035 | clamped_len); | |
3036 | goto release_page; | |
3037 | } | |
3038 | btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len); | |
f47960f4 | 3039 | |
c2832898 QW |
3040 | /* |
3041 | * Set the boundary if it's inside the page. | |
3042 | * Data relocation requires the destination extents to have the | |
3043 | * same size as the source. | |
3044 | * EXTENT_BOUNDARY bit prevents current extent from being merged | |
3045 | * with previous extent. | |
3046 | */ | |
3047 | if (in_range(cluster->boundary[*cluster_nr] - offset, | |
3048 | page_start, PAGE_SIZE)) { | |
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 | } |
f47960f4 QW |
3071 | unlock_page(page); |
3072 | put_page(page); | |
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 | ||
3080 | release_page: | |
3081 | unlock_page(page); | |
3082 | put_page(page); | |
f47960f4 QW |
3083 | return ret; |
3084 | } | |
3085 | ||
3086 | static int relocate_file_extent_cluster(struct inode *inode, | |
3087 | struct file_extent_cluster *cluster) | |
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++) | |
3117 | ret = relocate_one_page(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 | ||
3125 | static noinline_for_stack | |
0257bb82 YZ |
3126 | int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, |
3127 | struct file_extent_cluster *cluster) | |
5d4f98a2 | 3128 | { |
0257bb82 | 3129 | int ret; |
5d4f98a2 | 3130 | |
0257bb82 YZ |
3131 | if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { |
3132 | ret = relocate_file_extent_cluster(inode, cluster); | |
3133 | if (ret) | |
3134 | return ret; | |
3135 | cluster->nr = 0; | |
5d4f98a2 | 3136 | } |
5d4f98a2 | 3137 | |
0257bb82 YZ |
3138 | if (!cluster->nr) |
3139 | cluster->start = extent_key->objectid; | |
3140 | else | |
3141 | BUG_ON(cluster->nr >= MAX_EXTENTS); | |
3142 | cluster->end = extent_key->objectid + extent_key->offset - 1; | |
3143 | cluster->boundary[cluster->nr] = extent_key->objectid; | |
3144 | cluster->nr++; | |
3145 | ||
3146 | if (cluster->nr >= MAX_EXTENTS) { | |
3147 | ret = relocate_file_extent_cluster(inode, cluster); | |
3148 | if (ret) | |
3149 | return ret; | |
3150 | cluster->nr = 0; | |
3151 | } | |
3152 | return 0; | |
5d4f98a2 YZ |
3153 | } |
3154 | ||
5d4f98a2 YZ |
3155 | /* |
3156 | * helper to add a tree block to the list. | |
3157 | * the major work is getting the generation and level of the block | |
3158 | */ | |
3159 | static int add_tree_block(struct reloc_control *rc, | |
3160 | struct btrfs_key *extent_key, | |
3161 | struct btrfs_path *path, | |
3162 | struct rb_root *blocks) | |
3163 | { | |
3164 | struct extent_buffer *eb; | |
3165 | struct btrfs_extent_item *ei; | |
3166 | struct btrfs_tree_block_info *bi; | |
3167 | struct tree_block *block; | |
3168 | struct rb_node *rb_node; | |
3169 | u32 item_size; | |
3170 | int level = -1; | |
7fdf4b60 | 3171 | u64 generation; |
f7ba2d37 | 3172 | u64 owner = 0; |
5d4f98a2 YZ |
3173 | |
3174 | eb = path->nodes[0]; | |
3212fa14 | 3175 | item_size = btrfs_item_size(eb, path->slots[0]); |
5d4f98a2 | 3176 | |
3173a18f JB |
3177 | if (extent_key->type == BTRFS_METADATA_ITEM_KEY || |
3178 | item_size >= sizeof(*ei) + sizeof(*bi)) { | |
f7ba2d37 JB |
3179 | unsigned long ptr = 0, end; |
3180 | ||
5d4f98a2 YZ |
3181 | ei = btrfs_item_ptr(eb, path->slots[0], |
3182 | struct btrfs_extent_item); | |
f7ba2d37 | 3183 | end = (unsigned long)ei + item_size; |
3173a18f JB |
3184 | if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { |
3185 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3186 | level = btrfs_tree_block_level(eb, bi); | |
f7ba2d37 | 3187 | ptr = (unsigned long)(bi + 1); |
3173a18f JB |
3188 | } else { |
3189 | level = (int)extent_key->offset; | |
f7ba2d37 | 3190 | ptr = (unsigned long)(ei + 1); |
3173a18f | 3191 | } |
5d4f98a2 | 3192 | generation = btrfs_extent_generation(eb, ei); |
f7ba2d37 JB |
3193 | |
3194 | /* | |
3195 | * We're reading random blocks without knowing their owner ahead | |
3196 | * of time. This is ok most of the time, as all reloc roots and | |
3197 | * fs roots have the same lock type. However normal trees do | |
3198 | * not, and the only way to know ahead of time is to read the | |
3199 | * inline ref offset. We know it's an fs root if | |
3200 | * | |
3201 | * 1. There's more than one ref. | |
3202 | * 2. There's a SHARED_DATA_REF_KEY set. | |
3203 | * 3. FULL_BACKREF is set on the flags. | |
3204 | * | |
3205 | * Otherwise it's safe to assume that the ref offset == the | |
3206 | * owner of this block, so we can use that when calling | |
3207 | * read_tree_block. | |
3208 | */ | |
3209 | if (btrfs_extent_refs(eb, ei) == 1 && | |
3210 | !(btrfs_extent_flags(eb, ei) & | |
3211 | BTRFS_BLOCK_FLAG_FULL_BACKREF) && | |
3212 | ptr < end) { | |
3213 | struct btrfs_extent_inline_ref *iref; | |
3214 | int type; | |
3215 | ||
3216 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3217 | type = btrfs_get_extent_inline_ref_type(eb, iref, | |
3218 | BTRFS_REF_TYPE_BLOCK); | |
3219 | if (type == BTRFS_REF_TYPE_INVALID) | |
3220 | return -EINVAL; | |
3221 | if (type == BTRFS_TREE_BLOCK_REF_KEY) | |
3222 | owner = btrfs_extent_inline_ref_offset(eb, iref); | |
3223 | } | |
6d8ff4e4 | 3224 | } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { |
ba3c2b19 NB |
3225 | btrfs_print_v0_err(eb->fs_info); |
3226 | btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); | |
3227 | return -EINVAL; | |
5d4f98a2 | 3228 | } else { |
5d4f98a2 | 3229 | BUG(); |
5d4f98a2 YZ |
3230 | } |
3231 | ||
b3b4aa74 | 3232 | btrfs_release_path(path); |
5d4f98a2 YZ |
3233 | |
3234 | BUG_ON(level == -1); | |
3235 | ||
3236 | block = kmalloc(sizeof(*block), GFP_NOFS); | |
3237 | if (!block) | |
3238 | return -ENOMEM; | |
3239 | ||
3240 | block->bytenr = extent_key->objectid; | |
da17066c | 3241 | block->key.objectid = rc->extent_root->fs_info->nodesize; |
5d4f98a2 YZ |
3242 | block->key.offset = generation; |
3243 | block->level = level; | |
3244 | block->key_ready = 0; | |
f7ba2d37 | 3245 | block->owner = owner; |
5d4f98a2 | 3246 | |
e9a28dc5 | 3247 | rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node); |
43c04fb1 | 3248 | if (rb_node) |
982c92cb QW |
3249 | btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr, |
3250 | -EEXIST); | |
5d4f98a2 YZ |
3251 | |
3252 | return 0; | |
3253 | } | |
3254 | ||
3255 | /* | |
3256 | * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY | |
3257 | */ | |
3258 | static int __add_tree_block(struct reloc_control *rc, | |
3259 | u64 bytenr, u32 blocksize, | |
3260 | struct rb_root *blocks) | |
3261 | { | |
0b246afa | 3262 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3263 | struct btrfs_path *path; |
3264 | struct btrfs_key key; | |
3265 | int ret; | |
0b246afa | 3266 | bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
5d4f98a2 | 3267 | |
7476dfda | 3268 | if (tree_block_processed(bytenr, rc)) |
5d4f98a2 YZ |
3269 | return 0; |
3270 | ||
e9a28dc5 | 3271 | if (rb_simple_search(blocks, bytenr)) |
5d4f98a2 YZ |
3272 | return 0; |
3273 | ||
3274 | path = btrfs_alloc_path(); | |
3275 | if (!path) | |
3276 | return -ENOMEM; | |
aee68ee5 | 3277 | again: |
5d4f98a2 | 3278 | key.objectid = bytenr; |
aee68ee5 JB |
3279 | if (skinny) { |
3280 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3281 | key.offset = (u64)-1; | |
3282 | } else { | |
3283 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3284 | key.offset = blocksize; | |
3285 | } | |
5d4f98a2 YZ |
3286 | |
3287 | path->search_commit_root = 1; | |
3288 | path->skip_locking = 1; | |
3289 | ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); | |
3290 | if (ret < 0) | |
3291 | goto out; | |
5d4f98a2 | 3292 | |
aee68ee5 JB |
3293 | if (ret > 0 && skinny) { |
3294 | if (path->slots[0]) { | |
3295 | path->slots[0]--; | |
3296 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3297 | path->slots[0]); | |
3298 | if (key.objectid == bytenr && | |
3299 | (key.type == BTRFS_METADATA_ITEM_KEY || | |
3300 | (key.type == BTRFS_EXTENT_ITEM_KEY && | |
3301 | key.offset == blocksize))) | |
3302 | ret = 0; | |
3303 | } | |
3304 | ||
3305 | if (ret) { | |
3306 | skinny = false; | |
3307 | btrfs_release_path(path); | |
3308 | goto again; | |
3309 | } | |
3173a18f | 3310 | } |
cdccee99 LB |
3311 | if (ret) { |
3312 | ASSERT(ret == 1); | |
3313 | btrfs_print_leaf(path->nodes[0]); | |
3314 | btrfs_err(fs_info, | |
3315 | "tree block extent item (%llu) is not found in extent tree", | |
3316 | bytenr); | |
3317 | WARN_ON(1); | |
3318 | ret = -EINVAL; | |
3319 | goto out; | |
3320 | } | |
3173a18f | 3321 | |
5d4f98a2 YZ |
3322 | ret = add_tree_block(rc, &key, path, blocks); |
3323 | out: | |
3324 | btrfs_free_path(path); | |
3325 | return ret; | |
3326 | } | |
3327 | ||
0af3d00b | 3328 | static int delete_block_group_cache(struct btrfs_fs_info *fs_info, |
32da5386 | 3329 | struct btrfs_block_group *block_group, |
1bbc621e CM |
3330 | struct inode *inode, |
3331 | u64 ino) | |
0af3d00b | 3332 | { |
0af3d00b JB |
3333 | struct btrfs_root *root = fs_info->tree_root; |
3334 | struct btrfs_trans_handle *trans; | |
0af3d00b JB |
3335 | int ret = 0; |
3336 | ||
3337 | if (inode) | |
3338 | goto truncate; | |
3339 | ||
0202e83f | 3340 | inode = btrfs_iget(fs_info->sb, ino, root); |
2e19f1f9 | 3341 | if (IS_ERR(inode)) |
0af3d00b | 3342 | return -ENOENT; |
0af3d00b JB |
3343 | |
3344 | truncate: | |
2ff7e61e | 3345 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
7b61cd92 MX |
3346 | &fs_info->global_block_rsv); |
3347 | if (ret) | |
3348 | goto out; | |
3349 | ||
7a7eaa40 | 3350 | trans = btrfs_join_transaction(root); |
0af3d00b | 3351 | if (IS_ERR(trans)) { |
3612b495 | 3352 | ret = PTR_ERR(trans); |
0af3d00b JB |
3353 | goto out; |
3354 | } | |
3355 | ||
77ab86bf | 3356 | ret = btrfs_truncate_free_space_cache(trans, block_group, inode); |
0af3d00b | 3357 | |
3a45bb20 | 3358 | btrfs_end_transaction(trans); |
2ff7e61e | 3359 | btrfs_btree_balance_dirty(fs_info); |
0af3d00b JB |
3360 | out: |
3361 | iput(inode); | |
3362 | return ret; | |
3363 | } | |
3364 | ||
5d4f98a2 | 3365 | /* |
19b546d7 QW |
3366 | * Locate the free space cache EXTENT_DATA in root tree leaf and delete the |
3367 | * cache inode, to avoid free space cache data extent blocking data relocation. | |
5d4f98a2 | 3368 | */ |
19b546d7 QW |
3369 | static int delete_v1_space_cache(struct extent_buffer *leaf, |
3370 | struct btrfs_block_group *block_group, | |
3371 | u64 data_bytenr) | |
5d4f98a2 | 3372 | { |
19b546d7 QW |
3373 | u64 space_cache_ino; |
3374 | struct btrfs_file_extent_item *ei; | |
5d4f98a2 | 3375 | struct btrfs_key key; |
19b546d7 QW |
3376 | bool found = false; |
3377 | int i; | |
5d4f98a2 YZ |
3378 | int ret; |
3379 | ||
19b546d7 QW |
3380 | if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) |
3381 | return 0; | |
5d4f98a2 | 3382 | |
19b546d7 | 3383 | for (i = 0; i < btrfs_header_nritems(leaf); i++) { |
50e31ef4 QW |
3384 | u8 type; |
3385 | ||
19b546d7 QW |
3386 | btrfs_item_key_to_cpu(leaf, &key, i); |
3387 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
3388 | continue; | |
3389 | ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
50e31ef4 QW |
3390 | type = btrfs_file_extent_type(leaf, ei); |
3391 | ||
3392 | if ((type == BTRFS_FILE_EXTENT_REG || | |
3393 | type == BTRFS_FILE_EXTENT_PREALLOC) && | |
19b546d7 QW |
3394 | btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { |
3395 | found = true; | |
3396 | space_cache_ino = key.objectid; | |
5d4f98a2 | 3397 | break; |
5d4f98a2 | 3398 | } |
5d4f98a2 | 3399 | } |
19b546d7 QW |
3400 | if (!found) |
3401 | return -ENOENT; | |
3402 | ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, | |
3403 | space_cache_ino); | |
3404 | return ret; | |
5d4f98a2 YZ |
3405 | } |
3406 | ||
3407 | /* | |
2c016dc2 | 3408 | * helper to find all tree blocks that reference a given data extent |
5d4f98a2 YZ |
3409 | */ |
3410 | static noinline_for_stack | |
3411 | int add_data_references(struct reloc_control *rc, | |
3412 | struct btrfs_key *extent_key, | |
3413 | struct btrfs_path *path, | |
3414 | struct rb_root *blocks) | |
3415 | { | |
a2c8d27e | 3416 | struct btrfs_backref_walk_ctx ctx = { 0 }; |
19b546d7 QW |
3417 | struct ulist_iterator leaf_uiter; |
3418 | struct ulist_node *ref_node = NULL; | |
a2c8d27e | 3419 | const u32 blocksize = rc->extent_root->fs_info->nodesize; |
647f63bd | 3420 | int ret = 0; |
5d4f98a2 | 3421 | |
19b546d7 | 3422 | btrfs_release_path(path); |
a2c8d27e FM |
3423 | |
3424 | ctx.bytenr = extent_key->objectid; | |
0cad8f14 | 3425 | ctx.skip_inode_ref_list = true; |
a2c8d27e FM |
3426 | ctx.fs_info = rc->extent_root->fs_info; |
3427 | ||
3428 | ret = btrfs_find_all_leafs(&ctx); | |
19b546d7 QW |
3429 | if (ret < 0) |
3430 | return ret; | |
5d4f98a2 | 3431 | |
19b546d7 | 3432 | ULIST_ITER_INIT(&leaf_uiter); |
a2c8d27e | 3433 | while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) { |
789d6a3a | 3434 | struct btrfs_tree_parent_check check = { 0 }; |
19b546d7 | 3435 | struct extent_buffer *eb; |
5d4f98a2 | 3436 | |
789d6a3a | 3437 | eb = read_tree_block(ctx.fs_info, ref_node->val, &check); |
19b546d7 QW |
3438 | if (IS_ERR(eb)) { |
3439 | ret = PTR_ERR(eb); | |
5d4f98a2 | 3440 | break; |
5d4f98a2 | 3441 | } |
19b546d7 QW |
3442 | ret = delete_v1_space_cache(eb, rc->block_group, |
3443 | extent_key->objectid); | |
3444 | free_extent_buffer(eb); | |
3445 | if (ret < 0) | |
3446 | break; | |
3447 | ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); | |
3448 | if (ret < 0) | |
5d4f98a2 | 3449 | break; |
5d4f98a2 | 3450 | } |
19b546d7 | 3451 | if (ret < 0) |
5d4f98a2 | 3452 | free_block_list(blocks); |
a2c8d27e | 3453 | ulist_free(ctx.refs); |
19b546d7 | 3454 | return ret; |
5d4f98a2 YZ |
3455 | } |
3456 | ||
3457 | /* | |
2c016dc2 | 3458 | * helper to find next unprocessed extent |
5d4f98a2 YZ |
3459 | */ |
3460 | static noinline_for_stack | |
147d256e | 3461 | int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, |
3fd0a558 | 3462 | struct btrfs_key *extent_key) |
5d4f98a2 | 3463 | { |
0b246afa | 3464 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3465 | struct btrfs_key key; |
3466 | struct extent_buffer *leaf; | |
3467 | u64 start, end, last; | |
3468 | int ret; | |
3469 | ||
b3470b5d | 3470 | last = rc->block_group->start + rc->block_group->length; |
5d4f98a2 YZ |
3471 | while (1) { |
3472 | cond_resched(); | |
3473 | if (rc->search_start >= last) { | |
3474 | ret = 1; | |
3475 | break; | |
3476 | } | |
3477 | ||
3478 | key.objectid = rc->search_start; | |
3479 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3480 | key.offset = 0; | |
3481 | ||
3482 | path->search_commit_root = 1; | |
3483 | path->skip_locking = 1; | |
3484 | ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, | |
3485 | 0, 0); | |
3486 | if (ret < 0) | |
3487 | break; | |
3488 | next: | |
3489 | leaf = path->nodes[0]; | |
3490 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3491 | ret = btrfs_next_leaf(rc->extent_root, path); | |
3492 | if (ret != 0) | |
3493 | break; | |
3494 | leaf = path->nodes[0]; | |
3495 | } | |
3496 | ||
3497 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3498 | if (key.objectid >= last) { | |
3499 | ret = 1; | |
3500 | break; | |
3501 | } | |
3502 | ||
3173a18f JB |
3503 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
3504 | key.type != BTRFS_METADATA_ITEM_KEY) { | |
3505 | path->slots[0]++; | |
3506 | goto next; | |
3507 | } | |
3508 | ||
3509 | if (key.type == BTRFS_EXTENT_ITEM_KEY && | |
5d4f98a2 YZ |
3510 | key.objectid + key.offset <= rc->search_start) { |
3511 | path->slots[0]++; | |
3512 | goto next; | |
3513 | } | |
3514 | ||
3173a18f | 3515 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
0b246afa | 3516 | key.objectid + fs_info->nodesize <= |
3173a18f JB |
3517 | rc->search_start) { |
3518 | path->slots[0]++; | |
3519 | goto next; | |
3520 | } | |
3521 | ||
5d4f98a2 YZ |
3522 | ret = find_first_extent_bit(&rc->processed_blocks, |
3523 | key.objectid, &start, &end, | |
e6138876 | 3524 | EXTENT_DIRTY, NULL); |
5d4f98a2 YZ |
3525 | |
3526 | if (ret == 0 && start <= key.objectid) { | |
b3b4aa74 | 3527 | btrfs_release_path(path); |
5d4f98a2 YZ |
3528 | rc->search_start = end + 1; |
3529 | } else { | |
3173a18f JB |
3530 | if (key.type == BTRFS_EXTENT_ITEM_KEY) |
3531 | rc->search_start = key.objectid + key.offset; | |
3532 | else | |
3533 | rc->search_start = key.objectid + | |
0b246afa | 3534 | fs_info->nodesize; |
3fd0a558 | 3535 | memcpy(extent_key, &key, sizeof(key)); |
5d4f98a2 YZ |
3536 | return 0; |
3537 | } | |
3538 | } | |
b3b4aa74 | 3539 | btrfs_release_path(path); |
5d4f98a2 YZ |
3540 | return ret; |
3541 | } | |
3542 | ||
3543 | static void set_reloc_control(struct reloc_control *rc) | |
3544 | { | |
3545 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; | |
7585717f CM |
3546 | |
3547 | mutex_lock(&fs_info->reloc_mutex); | |
5d4f98a2 | 3548 | fs_info->reloc_ctl = rc; |
7585717f | 3549 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 YZ |
3550 | } |
3551 | ||
3552 | static void unset_reloc_control(struct reloc_control *rc) | |
3553 | { | |
3554 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; | |
7585717f CM |
3555 | |
3556 | mutex_lock(&fs_info->reloc_mutex); | |
5d4f98a2 | 3557 | fs_info->reloc_ctl = NULL; |
7585717f | 3558 | mutex_unlock(&fs_info->reloc_mutex); |
5d4f98a2 YZ |
3559 | } |
3560 | ||
3fd0a558 YZ |
3561 | static noinline_for_stack |
3562 | int prepare_to_relocate(struct reloc_control *rc) | |
3563 | { | |
3564 | struct btrfs_trans_handle *trans; | |
ac2fabac | 3565 | int ret; |
3fd0a558 | 3566 | |
2ff7e61e | 3567 | rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, |
66d8f3dd | 3568 | BTRFS_BLOCK_RSV_TEMP); |
3fd0a558 YZ |
3569 | if (!rc->block_rsv) |
3570 | return -ENOMEM; | |
3571 | ||
3fd0a558 | 3572 | memset(&rc->cluster, 0, sizeof(rc->cluster)); |
b3470b5d | 3573 | rc->search_start = rc->block_group->start; |
3fd0a558 YZ |
3574 | rc->extents_found = 0; |
3575 | rc->nodes_relocated = 0; | |
3576 | rc->merging_rsv_size = 0; | |
0647bf56 | 3577 | rc->reserved_bytes = 0; |
da17066c | 3578 | rc->block_rsv->size = rc->extent_root->fs_info->nodesize * |
0647bf56 | 3579 | RELOCATION_RESERVED_NODES; |
9270501c | 3580 | ret = btrfs_block_rsv_refill(rc->extent_root->fs_info, |
ac2fabac JB |
3581 | rc->block_rsv, rc->block_rsv->size, |
3582 | BTRFS_RESERVE_FLUSH_ALL); | |
3583 | if (ret) | |
3584 | return ret; | |
3fd0a558 YZ |
3585 | |
3586 | rc->create_reloc_tree = 1; | |
3587 | set_reloc_control(rc); | |
3588 | ||
7a7eaa40 | 3589 | trans = btrfs_join_transaction(rc->extent_root); |
28818947 LB |
3590 | if (IS_ERR(trans)) { |
3591 | unset_reloc_control(rc); | |
3592 | /* | |
3593 | * extent tree is not a ref_cow tree and has no reloc_root to | |
3594 | * cleanup. And callers are responsible to free the above | |
3595 | * block rsv. | |
3596 | */ | |
3597 | return PTR_ERR(trans); | |
3598 | } | |
85f02d6c ZF |
3599 | |
3600 | ret = btrfs_commit_transaction(trans); | |
3601 | if (ret) | |
3602 | unset_reloc_control(rc); | |
3603 | ||
3604 | return ret; | |
3fd0a558 | 3605 | } |
76dda93c | 3606 | |
5d4f98a2 YZ |
3607 | static noinline_for_stack int relocate_block_group(struct reloc_control *rc) |
3608 | { | |
2ff7e61e | 3609 | struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; |
5d4f98a2 YZ |
3610 | struct rb_root blocks = RB_ROOT; |
3611 | struct btrfs_key key; | |
3612 | struct btrfs_trans_handle *trans = NULL; | |
3613 | struct btrfs_path *path; | |
3614 | struct btrfs_extent_item *ei; | |
5d4f98a2 | 3615 | u64 flags; |
5d4f98a2 YZ |
3616 | int ret; |
3617 | int err = 0; | |
c87f08ca | 3618 | int progress = 0; |
5d4f98a2 YZ |
3619 | |
3620 | path = btrfs_alloc_path(); | |
3fd0a558 | 3621 | if (!path) |
5d4f98a2 | 3622 | return -ENOMEM; |
e4058b54 | 3623 | path->reada = READA_FORWARD; |
5d4f98a2 | 3624 | |
3fd0a558 YZ |
3625 | ret = prepare_to_relocate(rc); |
3626 | if (ret) { | |
3627 | err = ret; | |
3628 | goto out_free; | |
3629 | } | |
5d4f98a2 YZ |
3630 | |
3631 | while (1) { | |
0647bf56 | 3632 | rc->reserved_bytes = 0; |
9270501c JB |
3633 | ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, |
3634 | rc->block_rsv->size, | |
3635 | BTRFS_RESERVE_FLUSH_ALL); | |
0647bf56 WS |
3636 | if (ret) { |
3637 | err = ret; | |
3638 | break; | |
3639 | } | |
c87f08ca | 3640 | progress++; |
a22285a6 | 3641 | trans = btrfs_start_transaction(rc->extent_root, 0); |
0f788c58 LB |
3642 | if (IS_ERR(trans)) { |
3643 | err = PTR_ERR(trans); | |
3644 | trans = NULL; | |
3645 | break; | |
3646 | } | |
c87f08ca | 3647 | restart: |
3fd0a558 | 3648 | if (update_backref_cache(trans, &rc->backref_cache)) { |
3a45bb20 | 3649 | btrfs_end_transaction(trans); |
42a657f5 | 3650 | trans = NULL; |
3fd0a558 YZ |
3651 | continue; |
3652 | } | |
3653 | ||
147d256e | 3654 | ret = find_next_extent(rc, path, &key); |
5d4f98a2 YZ |
3655 | if (ret < 0) |
3656 | err = ret; | |
3657 | if (ret != 0) | |
3658 | break; | |
3659 | ||
3660 | rc->extents_found++; | |
3661 | ||
3662 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3663 | struct btrfs_extent_item); | |
24cd6389 | 3664 | flags = btrfs_extent_flags(path->nodes[0], ei); |
5d4f98a2 YZ |
3665 | |
3666 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
3667 | ret = add_tree_block(rc, &key, path, &blocks); | |
3668 | } else if (rc->stage == UPDATE_DATA_PTRS && | |
3fd0a558 | 3669 | (flags & BTRFS_EXTENT_FLAG_DATA)) { |
5d4f98a2 YZ |
3670 | ret = add_data_references(rc, &key, path, &blocks); |
3671 | } else { | |
b3b4aa74 | 3672 | btrfs_release_path(path); |
5d4f98a2 YZ |
3673 | ret = 0; |
3674 | } | |
3675 | if (ret < 0) { | |
3fd0a558 | 3676 | err = ret; |
5d4f98a2 YZ |
3677 | break; |
3678 | } | |
3679 | ||
3680 | if (!RB_EMPTY_ROOT(&blocks)) { | |
3681 | ret = relocate_tree_blocks(trans, rc, &blocks); | |
3682 | if (ret < 0) { | |
3fd0a558 YZ |
3683 | if (ret != -EAGAIN) { |
3684 | err = ret; | |
3685 | break; | |
3686 | } | |
3687 | rc->extents_found--; | |
3688 | rc->search_start = key.objectid; | |
3689 | } | |
3690 | } | |
3691 | ||
3a45bb20 | 3692 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 3693 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 | 3694 | trans = NULL; |
5d4f98a2 YZ |
3695 | |
3696 | if (rc->stage == MOVE_DATA_EXTENTS && | |
3697 | (flags & BTRFS_EXTENT_FLAG_DATA)) { | |
3698 | rc->found_file_extent = 1; | |
0257bb82 | 3699 | ret = relocate_data_extent(rc->data_inode, |
3fd0a558 | 3700 | &key, &rc->cluster); |
5d4f98a2 YZ |
3701 | if (ret < 0) { |
3702 | err = ret; | |
3703 | break; | |
3704 | } | |
3705 | } | |
f31ea088 QW |
3706 | if (btrfs_should_cancel_balance(fs_info)) { |
3707 | err = -ECANCELED; | |
3708 | break; | |
3709 | } | |
5d4f98a2 | 3710 | } |
c87f08ca | 3711 | if (trans && progress && err == -ENOSPC) { |
43a7e99d | 3712 | ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); |
9689457b | 3713 | if (ret == 1) { |
c87f08ca CM |
3714 | err = 0; |
3715 | progress = 0; | |
3716 | goto restart; | |
3717 | } | |
3718 | } | |
3fd0a558 | 3719 | |
b3b4aa74 | 3720 | btrfs_release_path(path); |
91166212 | 3721 | clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); |
5d4f98a2 YZ |
3722 | |
3723 | if (trans) { | |
3a45bb20 | 3724 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 3725 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 YZ |
3726 | } |
3727 | ||
0257bb82 | 3728 | if (!err) { |
3fd0a558 YZ |
3729 | ret = relocate_file_extent_cluster(rc->data_inode, |
3730 | &rc->cluster); | |
0257bb82 YZ |
3731 | if (ret < 0) |
3732 | err = ret; | |
3733 | } | |
3734 | ||
3fd0a558 YZ |
3735 | rc->create_reloc_tree = 0; |
3736 | set_reloc_control(rc); | |
0257bb82 | 3737 | |
13fe1bdb | 3738 | btrfs_backref_release_cache(&rc->backref_cache); |
63f018be | 3739 | btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); |
5d4f98a2 | 3740 | |
7f913c7c QW |
3741 | /* |
3742 | * Even in the case when the relocation is cancelled, we should all go | |
3743 | * through prepare_to_merge() and merge_reloc_roots(). | |
3744 | * | |
3745 | * For error (including cancelled balance), prepare_to_merge() will | |
3746 | * mark all reloc trees orphan, then queue them for cleanup in | |
3747 | * merge_reloc_roots() | |
3748 | */ | |
3fd0a558 | 3749 | err = prepare_to_merge(rc, err); |
5d4f98a2 YZ |
3750 | |
3751 | merge_reloc_roots(rc); | |
3752 | ||
3fd0a558 | 3753 | rc->merge_reloc_tree = 0; |
5d4f98a2 | 3754 | unset_reloc_control(rc); |
63f018be | 3755 | btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); |
5d4f98a2 YZ |
3756 | |
3757 | /* get rid of pinned extents */ | |
7a7eaa40 | 3758 | trans = btrfs_join_transaction(rc->extent_root); |
62b99540 | 3759 | if (IS_ERR(trans)) { |
3612b495 | 3760 | err = PTR_ERR(trans); |
62b99540 QW |
3761 | goto out_free; |
3762 | } | |
fb686c68 JB |
3763 | ret = btrfs_commit_transaction(trans); |
3764 | if (ret && !err) | |
3765 | err = ret; | |
6217b0fa | 3766 | out_free: |
d2311e69 QW |
3767 | ret = clean_dirty_subvols(rc); |
3768 | if (ret < 0 && !err) | |
3769 | err = ret; | |
2ff7e61e | 3770 | btrfs_free_block_rsv(fs_info, rc->block_rsv); |
3fd0a558 | 3771 | btrfs_free_path(path); |
5d4f98a2 YZ |
3772 | return err; |
3773 | } | |
3774 | ||
3775 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, | |
0257bb82 | 3776 | struct btrfs_root *root, u64 objectid) |
5d4f98a2 YZ |
3777 | { |
3778 | struct btrfs_path *path; | |
3779 | struct btrfs_inode_item *item; | |
3780 | struct extent_buffer *leaf; | |
3781 | int ret; | |
3782 | ||
3783 | path = btrfs_alloc_path(); | |
3784 | if (!path) | |
3785 | return -ENOMEM; | |
3786 | ||
3787 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); | |
3788 | if (ret) | |
3789 | goto out; | |
3790 | ||
3791 | leaf = path->nodes[0]; | |
3792 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
b159fa28 | 3793 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
5d4f98a2 | 3794 | btrfs_set_inode_generation(leaf, item, 1); |
0257bb82 | 3795 | btrfs_set_inode_size(leaf, item, 0); |
5d4f98a2 | 3796 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); |
960a3166 JT |
3797 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | |
3798 | BTRFS_INODE_PREALLOC); | |
5d4f98a2 | 3799 | btrfs_mark_buffer_dirty(leaf); |
5d4f98a2 YZ |
3800 | out: |
3801 | btrfs_free_path(path); | |
3802 | return ret; | |
3803 | } | |
3804 | ||
790c1b8c JB |
3805 | static void delete_orphan_inode(struct btrfs_trans_handle *trans, |
3806 | struct btrfs_root *root, u64 objectid) | |
3807 | { | |
3808 | struct btrfs_path *path; | |
3809 | struct btrfs_key key; | |
3810 | int ret = 0; | |
3811 | ||
3812 | path = btrfs_alloc_path(); | |
3813 | if (!path) { | |
3814 | ret = -ENOMEM; | |
3815 | goto out; | |
3816 | } | |
3817 | ||
3818 | key.objectid = objectid; | |
3819 | key.type = BTRFS_INODE_ITEM_KEY; | |
3820 | key.offset = 0; | |
3821 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3822 | if (ret) { | |
3823 | if (ret > 0) | |
3824 | ret = -ENOENT; | |
3825 | goto out; | |
3826 | } | |
3827 | ret = btrfs_del_item(trans, root, path); | |
3828 | out: | |
3829 | if (ret) | |
3830 | btrfs_abort_transaction(trans, ret); | |
3831 | btrfs_free_path(path); | |
3832 | } | |
3833 | ||
5d4f98a2 YZ |
3834 | /* |
3835 | * helper to create inode for data relocation. | |
3836 | * the inode is in data relocation tree and its link count is 0 | |
3837 | */ | |
3fd0a558 YZ |
3838 | static noinline_for_stack |
3839 | struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, | |
32da5386 | 3840 | struct btrfs_block_group *group) |
5d4f98a2 YZ |
3841 | { |
3842 | struct inode *inode = NULL; | |
3843 | struct btrfs_trans_handle *trans; | |
3844 | struct btrfs_root *root; | |
4624900d | 3845 | u64 objectid; |
5d4f98a2 YZ |
3846 | int err = 0; |
3847 | ||
aeb935a4 | 3848 | root = btrfs_grab_root(fs_info->data_reloc_root); |
a22285a6 | 3849 | trans = btrfs_start_transaction(root, 6); |
76deacf0 | 3850 | if (IS_ERR(trans)) { |
00246528 | 3851 | btrfs_put_root(root); |
3fd0a558 | 3852 | return ERR_CAST(trans); |
76deacf0 | 3853 | } |
5d4f98a2 | 3854 | |
543068a2 | 3855 | err = btrfs_get_free_objectid(root, &objectid); |
5d4f98a2 YZ |
3856 | if (err) |
3857 | goto out; | |
3858 | ||
0257bb82 | 3859 | err = __insert_orphan_inode(trans, root, objectid); |
790c1b8c JB |
3860 | if (err) |
3861 | goto out; | |
5d4f98a2 | 3862 | |
0202e83f | 3863 | inode = btrfs_iget(fs_info->sb, objectid, root); |
790c1b8c JB |
3864 | if (IS_ERR(inode)) { |
3865 | delete_orphan_inode(trans, root, objectid); | |
3866 | err = PTR_ERR(inode); | |
3867 | inode = NULL; | |
3868 | goto out; | |
3869 | } | |
b3470b5d | 3870 | BTRFS_I(inode)->index_cnt = group->start; |
5d4f98a2 | 3871 | |
73f2e545 | 3872 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
5d4f98a2 | 3873 | out: |
00246528 | 3874 | btrfs_put_root(root); |
3a45bb20 | 3875 | btrfs_end_transaction(trans); |
2ff7e61e | 3876 | btrfs_btree_balance_dirty(fs_info); |
5d4f98a2 | 3877 | if (err) { |
8aa1e49e | 3878 | iput(inode); |
5d4f98a2 YZ |
3879 | inode = ERR_PTR(err); |
3880 | } | |
3881 | return inode; | |
3882 | } | |
3883 | ||
907d2710 DS |
3884 | /* |
3885 | * Mark start of chunk relocation that is cancellable. Check if the cancellation | |
3886 | * has been requested meanwhile and don't start in that case. | |
3887 | * | |
3888 | * Return: | |
3889 | * 0 success | |
3890 | * -EINPROGRESS operation is already in progress, that's probably a bug | |
3891 | * -ECANCELED cancellation request was set before the operation started | |
3892 | */ | |
3893 | static int reloc_chunk_start(struct btrfs_fs_info *fs_info) | |
3894 | { | |
3895 | if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) { | |
3896 | /* This should not happen */ | |
3897 | btrfs_err(fs_info, "reloc already running, cannot start"); | |
3898 | return -EINPROGRESS; | |
3899 | } | |
3900 | ||
3901 | if (atomic_read(&fs_info->reloc_cancel_req) > 0) { | |
3902 | btrfs_info(fs_info, "chunk relocation canceled on start"); | |
3903 | /* | |
3904 | * On cancel, clear all requests but let the caller mark | |
3905 | * the end after cleanup operations. | |
3906 | */ | |
3907 | atomic_set(&fs_info->reloc_cancel_req, 0); | |
3908 | return -ECANCELED; | |
3909 | } | |
3910 | return 0; | |
3911 | } | |
3912 | ||
3913 | /* | |
3914 | * Mark end of chunk relocation that is cancellable and wake any waiters. | |
3915 | */ | |
3916 | static void reloc_chunk_end(struct btrfs_fs_info *fs_info) | |
3917 | { | |
3918 | /* Requested after start, clear bit first so any waiters can continue */ | |
3919 | if (atomic_read(&fs_info->reloc_cancel_req) > 0) | |
3920 | btrfs_info(fs_info, "chunk relocation canceled during operation"); | |
3921 | clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags); | |
3922 | atomic_set(&fs_info->reloc_cancel_req, 0); | |
3923 | } | |
3924 | ||
c258d6e3 | 3925 | static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) |
3fd0a558 YZ |
3926 | { |
3927 | struct reloc_control *rc; | |
3928 | ||
3929 | rc = kzalloc(sizeof(*rc), GFP_NOFS); | |
3930 | if (!rc) | |
3931 | return NULL; | |
3932 | ||
3933 | INIT_LIST_HEAD(&rc->reloc_roots); | |
d2311e69 | 3934 | INIT_LIST_HEAD(&rc->dirty_subvol_roots); |
584fb121 | 3935 | btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1); |
3fd0a558 | 3936 | mapping_tree_init(&rc->reloc_root_tree); |
35da5a7e | 3937 | extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS); |
3fd0a558 YZ |
3938 | return rc; |
3939 | } | |
3940 | ||
1a0afa0e JB |
3941 | static void free_reloc_control(struct reloc_control *rc) |
3942 | { | |
3943 | struct mapping_node *node, *tmp; | |
3944 | ||
3945 | free_reloc_roots(&rc->reloc_roots); | |
3946 | rbtree_postorder_for_each_entry_safe(node, tmp, | |
3947 | &rc->reloc_root_tree.rb_root, rb_node) | |
3948 | kfree(node); | |
3949 | ||
3950 | kfree(rc); | |
3951 | } | |
3952 | ||
ebce0e01 AB |
3953 | /* |
3954 | * Print the block group being relocated | |
3955 | */ | |
3956 | static void describe_relocation(struct btrfs_fs_info *fs_info, | |
32da5386 | 3957 | struct btrfs_block_group *block_group) |
ebce0e01 | 3958 | { |
f89e09cf | 3959 | char buf[128] = {'\0'}; |
ebce0e01 | 3960 | |
f89e09cf | 3961 | btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); |
ebce0e01 AB |
3962 | |
3963 | btrfs_info(fs_info, | |
3964 | "relocating block group %llu flags %s", | |
b3470b5d | 3965 | block_group->start, buf); |
ebce0e01 AB |
3966 | } |
3967 | ||
430640e3 QW |
3968 | static const char *stage_to_string(int stage) |
3969 | { | |
3970 | if (stage == MOVE_DATA_EXTENTS) | |
3971 | return "move data extents"; | |
3972 | if (stage == UPDATE_DATA_PTRS) | |
3973 | return "update data pointers"; | |
3974 | return "unknown"; | |
3975 | } | |
3976 | ||
5d4f98a2 YZ |
3977 | /* |
3978 | * function to relocate all extents in a block group. | |
3979 | */ | |
6bccf3ab | 3980 | int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) |
5d4f98a2 | 3981 | { |
32da5386 | 3982 | struct btrfs_block_group *bg; |
29cbcf40 | 3983 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start); |
5d4f98a2 | 3984 | struct reloc_control *rc; |
0af3d00b JB |
3985 | struct inode *inode; |
3986 | struct btrfs_path *path; | |
5d4f98a2 | 3987 | int ret; |
f0486c68 | 3988 | int rw = 0; |
5d4f98a2 YZ |
3989 | int err = 0; |
3990 | ||
b4be6aef JB |
3991 | /* |
3992 | * This only gets set if we had a half-deleted snapshot on mount. We | |
3993 | * cannot allow relocation to start while we're still trying to clean up | |
3994 | * these pending deletions. | |
3995 | */ | |
3996 | ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE); | |
3997 | if (ret) | |
3998 | return ret; | |
3999 | ||
4000 | /* We may have been woken up by close_ctree, so bail if we're closing. */ | |
4001 | if (btrfs_fs_closing(fs_info)) | |
4002 | return -EINTR; | |
4003 | ||
eede2bf3 OS |
4004 | bg = btrfs_lookup_block_group(fs_info, group_start); |
4005 | if (!bg) | |
4006 | return -ENOENT; | |
4007 | ||
0320b353 NA |
4008 | /* |
4009 | * Relocation of a data block group creates ordered extents. Without | |
4010 | * sb_start_write(), we can freeze the filesystem while unfinished | |
4011 | * ordered extents are left. Such ordered extents can cause a deadlock | |
4012 | * e.g. when syncfs() is waiting for their completion but they can't | |
4013 | * finish because they block when joining a transaction, due to the | |
4014 | * fact that the freeze locks are being held in write mode. | |
4015 | */ | |
4016 | if (bg->flags & BTRFS_BLOCK_GROUP_DATA) | |
4017 | ASSERT(sb_write_started(fs_info->sb)); | |
4018 | ||
eede2bf3 OS |
4019 | if (btrfs_pinned_by_swapfile(fs_info, bg)) { |
4020 | btrfs_put_block_group(bg); | |
4021 | return -ETXTBSY; | |
4022 | } | |
4023 | ||
c258d6e3 | 4024 | rc = alloc_reloc_control(fs_info); |
eede2bf3 OS |
4025 | if (!rc) { |
4026 | btrfs_put_block_group(bg); | |
5d4f98a2 | 4027 | return -ENOMEM; |
eede2bf3 | 4028 | } |
5d4f98a2 | 4029 | |
907d2710 DS |
4030 | ret = reloc_chunk_start(fs_info); |
4031 | if (ret < 0) { | |
4032 | err = ret; | |
4033 | goto out_put_bg; | |
4034 | } | |
4035 | ||
f0486c68 | 4036 | rc->extent_root = extent_root; |
eede2bf3 | 4037 | rc->block_group = bg; |
5d4f98a2 | 4038 | |
b12de528 | 4039 | ret = btrfs_inc_block_group_ro(rc->block_group, true); |
868f401a Z |
4040 | if (ret) { |
4041 | err = ret; | |
4042 | goto out; | |
f0486c68 | 4043 | } |
868f401a | 4044 | rw = 1; |
f0486c68 | 4045 | |
0af3d00b JB |
4046 | path = btrfs_alloc_path(); |
4047 | if (!path) { | |
4048 | err = -ENOMEM; | |
4049 | goto out; | |
4050 | } | |
4051 | ||
7949f339 | 4052 | inode = lookup_free_space_inode(rc->block_group, path); |
0af3d00b JB |
4053 | btrfs_free_path(path); |
4054 | ||
4055 | if (!IS_ERR(inode)) | |
1bbc621e | 4056 | ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); |
0af3d00b JB |
4057 | else |
4058 | ret = PTR_ERR(inode); | |
4059 | ||
4060 | if (ret && ret != -ENOENT) { | |
4061 | err = ret; | |
4062 | goto out; | |
4063 | } | |
4064 | ||
5d4f98a2 YZ |
4065 | rc->data_inode = create_reloc_inode(fs_info, rc->block_group); |
4066 | if (IS_ERR(rc->data_inode)) { | |
4067 | err = PTR_ERR(rc->data_inode); | |
4068 | rc->data_inode = NULL; | |
4069 | goto out; | |
4070 | } | |
4071 | ||
0b246afa | 4072 | describe_relocation(fs_info, rc->block_group); |
5d4f98a2 | 4073 | |
9cfa3e34 | 4074 | btrfs_wait_block_group_reservations(rc->block_group); |
f78c436c | 4075 | btrfs_wait_nocow_writers(rc->block_group); |
6374e57a | 4076 | btrfs_wait_ordered_roots(fs_info, U64_MAX, |
b3470b5d DS |
4077 | rc->block_group->start, |
4078 | rc->block_group->length); | |
5d4f98a2 | 4079 | |
7ae9bd18 NA |
4080 | ret = btrfs_zone_finish(rc->block_group); |
4081 | WARN_ON(ret && ret != -EAGAIN); | |
4082 | ||
5d4f98a2 | 4083 | while (1) { |
430640e3 QW |
4084 | int finishes_stage; |
4085 | ||
76dda93c | 4086 | mutex_lock(&fs_info->cleaner_mutex); |
5d4f98a2 | 4087 | ret = relocate_block_group(rc); |
76dda93c | 4088 | mutex_unlock(&fs_info->cleaner_mutex); |
ff612ba7 | 4089 | if (ret < 0) |
5d4f98a2 | 4090 | err = ret; |
5d4f98a2 | 4091 | |
430640e3 | 4092 | finishes_stage = rc->stage; |
ff612ba7 JB |
4093 | /* |
4094 | * We may have gotten ENOSPC after we already dirtied some | |
4095 | * extents. If writeout happens while we're relocating a | |
4096 | * different block group we could end up hitting the | |
4097 | * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in | |
4098 | * btrfs_reloc_cow_block. Make sure we write everything out | |
4099 | * properly so we don't trip over this problem, and then break | |
4100 | * out of the loop if we hit an error. | |
4101 | */ | |
5d4f98a2 | 4102 | if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { |
0ef8b726 JB |
4103 | ret = btrfs_wait_ordered_range(rc->data_inode, 0, |
4104 | (u64)-1); | |
ff612ba7 | 4105 | if (ret) |
0ef8b726 | 4106 | err = ret; |
5d4f98a2 YZ |
4107 | invalidate_mapping_pages(rc->data_inode->i_mapping, |
4108 | 0, -1); | |
4109 | rc->stage = UPDATE_DATA_PTRS; | |
5d4f98a2 | 4110 | } |
ff612ba7 JB |
4111 | |
4112 | if (err < 0) | |
4113 | goto out; | |
4114 | ||
4115 | if (rc->extents_found == 0) | |
4116 | break; | |
4117 | ||
430640e3 QW |
4118 | btrfs_info(fs_info, "found %llu extents, stage: %s", |
4119 | rc->extents_found, stage_to_string(finishes_stage)); | |
5d4f98a2 YZ |
4120 | } |
4121 | ||
5d4f98a2 YZ |
4122 | WARN_ON(rc->block_group->pinned > 0); |
4123 | WARN_ON(rc->block_group->reserved > 0); | |
bf38be65 | 4124 | WARN_ON(rc->block_group->used > 0); |
5d4f98a2 | 4125 | out: |
f0486c68 | 4126 | if (err && rw) |
2ff7e61e | 4127 | btrfs_dec_block_group_ro(rc->block_group); |
5d4f98a2 | 4128 | iput(rc->data_inode); |
907d2710 DS |
4129 | out_put_bg: |
4130 | btrfs_put_block_group(bg); | |
4131 | reloc_chunk_end(fs_info); | |
1a0afa0e | 4132 | free_reloc_control(rc); |
5d4f98a2 YZ |
4133 | return err; |
4134 | } | |
4135 | ||
76dda93c YZ |
4136 | static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) |
4137 | { | |
0b246afa | 4138 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c | 4139 | struct btrfs_trans_handle *trans; |
79787eaa | 4140 | int ret, err; |
76dda93c | 4141 | |
0b246afa | 4142 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
4143 | if (IS_ERR(trans)) |
4144 | return PTR_ERR(trans); | |
76dda93c YZ |
4145 | |
4146 | memset(&root->root_item.drop_progress, 0, | |
4147 | sizeof(root->root_item.drop_progress)); | |
c8422684 | 4148 | btrfs_set_root_drop_level(&root->root_item, 0); |
76dda93c | 4149 | btrfs_set_root_refs(&root->root_item, 0); |
0b246afa | 4150 | ret = btrfs_update_root(trans, fs_info->tree_root, |
76dda93c | 4151 | &root->root_key, &root->root_item); |
76dda93c | 4152 | |
3a45bb20 | 4153 | err = btrfs_end_transaction(trans); |
79787eaa JM |
4154 | if (err) |
4155 | return err; | |
4156 | return ret; | |
76dda93c YZ |
4157 | } |
4158 | ||
5d4f98a2 YZ |
4159 | /* |
4160 | * recover relocation interrupted by system crash. | |
4161 | * | |
4162 | * this function resumes merging reloc trees with corresponding fs trees. | |
4163 | * this is important for keeping the sharing of tree blocks | |
4164 | */ | |
7eefae6b | 4165 | int btrfs_recover_relocation(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
4166 | { |
4167 | LIST_HEAD(reloc_roots); | |
4168 | struct btrfs_key key; | |
4169 | struct btrfs_root *fs_root; | |
4170 | struct btrfs_root *reloc_root; | |
4171 | struct btrfs_path *path; | |
4172 | struct extent_buffer *leaf; | |
4173 | struct reloc_control *rc = NULL; | |
4174 | struct btrfs_trans_handle *trans; | |
4175 | int ret; | |
4176 | int err = 0; | |
4177 | ||
4178 | path = btrfs_alloc_path(); | |
4179 | if (!path) | |
4180 | return -ENOMEM; | |
e4058b54 | 4181 | path->reada = READA_BACK; |
5d4f98a2 YZ |
4182 | |
4183 | key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
4184 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4185 | key.offset = (u64)-1; | |
4186 | ||
4187 | while (1) { | |
0b246afa | 4188 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, |
5d4f98a2 YZ |
4189 | path, 0, 0); |
4190 | if (ret < 0) { | |
4191 | err = ret; | |
4192 | goto out; | |
4193 | } | |
4194 | if (ret > 0) { | |
4195 | if (path->slots[0] == 0) | |
4196 | break; | |
4197 | path->slots[0]--; | |
4198 | } | |
4199 | leaf = path->nodes[0]; | |
4200 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4201 | btrfs_release_path(path); |
5d4f98a2 YZ |
4202 | |
4203 | if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || | |
4204 | key.type != BTRFS_ROOT_ITEM_KEY) | |
4205 | break; | |
4206 | ||
7eefae6b | 4207 | reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key); |
5d4f98a2 YZ |
4208 | if (IS_ERR(reloc_root)) { |
4209 | err = PTR_ERR(reloc_root); | |
4210 | goto out; | |
4211 | } | |
4212 | ||
92a7cc42 | 4213 | set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state); |
5d4f98a2 YZ |
4214 | list_add(&reloc_root->root_list, &reloc_roots); |
4215 | ||
4216 | if (btrfs_root_refs(&reloc_root->root_item) > 0) { | |
a820feb5 DS |
4217 | fs_root = btrfs_get_fs_root(fs_info, |
4218 | reloc_root->root_key.offset, false); | |
5d4f98a2 | 4219 | if (IS_ERR(fs_root)) { |
76dda93c YZ |
4220 | ret = PTR_ERR(fs_root); |
4221 | if (ret != -ENOENT) { | |
4222 | err = ret; | |
4223 | goto out; | |
4224 | } | |
79787eaa JM |
4225 | ret = mark_garbage_root(reloc_root); |
4226 | if (ret < 0) { | |
4227 | err = ret; | |
4228 | goto out; | |
4229 | } | |
932fd26d | 4230 | } else { |
00246528 | 4231 | btrfs_put_root(fs_root); |
5d4f98a2 YZ |
4232 | } |
4233 | } | |
4234 | ||
4235 | if (key.offset == 0) | |
4236 | break; | |
4237 | ||
4238 | key.offset--; | |
4239 | } | |
b3b4aa74 | 4240 | btrfs_release_path(path); |
5d4f98a2 YZ |
4241 | |
4242 | if (list_empty(&reloc_roots)) | |
4243 | goto out; | |
4244 | ||
c258d6e3 | 4245 | rc = alloc_reloc_control(fs_info); |
5d4f98a2 YZ |
4246 | if (!rc) { |
4247 | err = -ENOMEM; | |
4248 | goto out; | |
4249 | } | |
4250 | ||
907d2710 DS |
4251 | ret = reloc_chunk_start(fs_info); |
4252 | if (ret < 0) { | |
4253 | err = ret; | |
4254 | goto out_end; | |
4255 | } | |
4256 | ||
29cbcf40 | 4257 | rc->extent_root = btrfs_extent_root(fs_info, 0); |
5d4f98a2 YZ |
4258 | |
4259 | set_reloc_control(rc); | |
4260 | ||
7a7eaa40 | 4261 | trans = btrfs_join_transaction(rc->extent_root); |
3612b495 | 4262 | if (IS_ERR(trans)) { |
3612b495 | 4263 | err = PTR_ERR(trans); |
fb2d83ee | 4264 | goto out_unset; |
3612b495 | 4265 | } |
3fd0a558 YZ |
4266 | |
4267 | rc->merge_reloc_tree = 1; | |
4268 | ||
5d4f98a2 YZ |
4269 | while (!list_empty(&reloc_roots)) { |
4270 | reloc_root = list_entry(reloc_roots.next, | |
4271 | struct btrfs_root, root_list); | |
4272 | list_del(&reloc_root->root_list); | |
4273 | ||
4274 | if (btrfs_root_refs(&reloc_root->root_item) == 0) { | |
4275 | list_add_tail(&reloc_root->root_list, | |
4276 | &rc->reloc_roots); | |
4277 | continue; | |
4278 | } | |
4279 | ||
a820feb5 DS |
4280 | fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, |
4281 | false); | |
79787eaa JM |
4282 | if (IS_ERR(fs_root)) { |
4283 | err = PTR_ERR(fs_root); | |
ca1aa281 | 4284 | list_add_tail(&reloc_root->root_list, &reloc_roots); |
1402d17d | 4285 | btrfs_end_transaction(trans); |
fb2d83ee | 4286 | goto out_unset; |
79787eaa | 4287 | } |
5d4f98a2 | 4288 | |
ffd7b339 | 4289 | err = __add_reloc_root(reloc_root); |
57a304cf | 4290 | ASSERT(err != -EEXIST); |
3c925863 JB |
4291 | if (err) { |
4292 | list_add_tail(&reloc_root->root_list, &reloc_roots); | |
4293 | btrfs_put_root(fs_root); | |
4294 | btrfs_end_transaction(trans); | |
4295 | goto out_unset; | |
4296 | } | |
f44deb74 | 4297 | fs_root->reloc_root = btrfs_grab_root(reloc_root); |
00246528 | 4298 | btrfs_put_root(fs_root); |
5d4f98a2 YZ |
4299 | } |
4300 | ||
3a45bb20 | 4301 | err = btrfs_commit_transaction(trans); |
79787eaa | 4302 | if (err) |
fb2d83ee | 4303 | goto out_unset; |
5d4f98a2 YZ |
4304 | |
4305 | merge_reloc_roots(rc); | |
4306 | ||
4307 | unset_reloc_control(rc); | |
4308 | ||
7a7eaa40 | 4309 | trans = btrfs_join_transaction(rc->extent_root); |
62b99540 | 4310 | if (IS_ERR(trans)) { |
3612b495 | 4311 | err = PTR_ERR(trans); |
6217b0fa | 4312 | goto out_clean; |
62b99540 | 4313 | } |
3a45bb20 | 4314 | err = btrfs_commit_transaction(trans); |
6217b0fa | 4315 | out_clean: |
d2311e69 QW |
4316 | ret = clean_dirty_subvols(rc); |
4317 | if (ret < 0 && !err) | |
4318 | err = ret; | |
fb2d83ee JB |
4319 | out_unset: |
4320 | unset_reloc_control(rc); | |
907d2710 DS |
4321 | out_end: |
4322 | reloc_chunk_end(fs_info); | |
1a0afa0e | 4323 | free_reloc_control(rc); |
3612b495 | 4324 | out: |
a7571232 | 4325 | free_reloc_roots(&reloc_roots); |
aca1bba6 | 4326 | |
5d4f98a2 YZ |
4327 | btrfs_free_path(path); |
4328 | ||
4329 | if (err == 0) { | |
4330 | /* cleanup orphan inode in data relocation tree */ | |
aeb935a4 QW |
4331 | fs_root = btrfs_grab_root(fs_info->data_reloc_root); |
4332 | ASSERT(fs_root); | |
4333 | err = btrfs_orphan_cleanup(fs_root); | |
4334 | btrfs_put_root(fs_root); | |
5d4f98a2 YZ |
4335 | } |
4336 | return err; | |
4337 | } | |
4338 | ||
4339 | /* | |
4340 | * helper to add ordered checksum for data relocation. | |
4341 | * | |
4342 | * cloning checksum properly handles the nodatasum extents. | |
4343 | * it also saves CPU time to re-calculate the checksum. | |
4344 | */ | |
34bfaf15 | 4345 | int btrfs_reloc_clone_csums(struct btrfs_ordered_extent *ordered) |
5d4f98a2 | 4346 | { |
34bfaf15 | 4347 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); |
7bfa9535 | 4348 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
34bfaf15 CH |
4349 | u64 disk_bytenr = ordered->file_offset + inode->index_cnt; |
4350 | struct btrfs_root *csum_root = btrfs_csum_root(fs_info, disk_bytenr); | |
5d4f98a2 | 4351 | LIST_HEAD(list); |
34bfaf15 | 4352 | int ret; |
5d4f98a2 | 4353 | |
97e38239 | 4354 | ret = btrfs_lookup_csums_list(csum_root, disk_bytenr, |
34bfaf15 CH |
4355 | disk_bytenr + ordered->num_bytes - 1, |
4356 | &list, 0, false); | |
79787eaa | 4357 | if (ret) |
34bfaf15 | 4358 | return ret; |
5d4f98a2 YZ |
4359 | |
4360 | while (!list_empty(&list)) { | |
34bfaf15 CH |
4361 | struct btrfs_ordered_sum *sums = |
4362 | list_entry(list.next, struct btrfs_ordered_sum, list); | |
4363 | ||
5d4f98a2 YZ |
4364 | list_del_init(&sums->list); |
4365 | ||
4577b014 JB |
4366 | /* |
4367 | * We need to offset the new_bytenr based on where the csum is. | |
4368 | * We need to do this because we will read in entire prealloc | |
4369 | * extents but we may have written to say the middle of the | |
4370 | * prealloc extent, so we need to make sure the csum goes with | |
4371 | * the right disk offset. | |
4372 | * | |
4373 | * We can do this because the data reloc inode refers strictly | |
4374 | * to the on disk bytes, so we don't have to worry about | |
4375 | * disk_len vs real len like with real inodes since it's all | |
4376 | * disk length. | |
4377 | */ | |
34bfaf15 | 4378 | sums->logical = ordered->disk_bytenr + sums->logical - disk_bytenr; |
f9756261 | 4379 | btrfs_add_ordered_sum(ordered, sums); |
5d4f98a2 | 4380 | } |
34bfaf15 CH |
4381 | |
4382 | return 0; | |
5d4f98a2 | 4383 | } |
3fd0a558 | 4384 | |
83d4cfd4 JB |
4385 | int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, |
4386 | struct btrfs_root *root, struct extent_buffer *buf, | |
4387 | struct extent_buffer *cow) | |
3fd0a558 | 4388 | { |
0b246afa | 4389 | struct btrfs_fs_info *fs_info = root->fs_info; |
3fd0a558 | 4390 | struct reloc_control *rc; |
a26195a5 | 4391 | struct btrfs_backref_node *node; |
3fd0a558 YZ |
4392 | int first_cow = 0; |
4393 | int level; | |
83d4cfd4 | 4394 | int ret = 0; |
3fd0a558 | 4395 | |
0b246afa | 4396 | rc = fs_info->reloc_ctl; |
3fd0a558 | 4397 | if (!rc) |
83d4cfd4 | 4398 | return 0; |
3fd0a558 | 4399 | |
37f00a6d | 4400 | BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root)); |
3fd0a558 YZ |
4401 | |
4402 | level = btrfs_header_level(buf); | |
4403 | if (btrfs_header_generation(buf) <= | |
4404 | btrfs_root_last_snapshot(&root->root_item)) | |
4405 | first_cow = 1; | |
4406 | ||
4407 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && | |
4408 | rc->create_reloc_tree) { | |
4409 | WARN_ON(!first_cow && level == 0); | |
4410 | ||
4411 | node = rc->backref_cache.path[level]; | |
4412 | BUG_ON(node->bytenr != buf->start && | |
4413 | node->new_bytenr != buf->start); | |
4414 | ||
b0fe7078 | 4415 | btrfs_backref_drop_node_buffer(node); |
67439dad | 4416 | atomic_inc(&cow->refs); |
3fd0a558 YZ |
4417 | node->eb = cow; |
4418 | node->new_bytenr = cow->start; | |
4419 | ||
4420 | if (!node->pending) { | |
4421 | list_move_tail(&node->list, | |
4422 | &rc->backref_cache.pending[level]); | |
4423 | node->pending = 1; | |
4424 | } | |
4425 | ||
4426 | if (first_cow) | |
9569cc20 | 4427 | mark_block_processed(rc, node); |
3fd0a558 YZ |
4428 | |
4429 | if (first_cow && level > 0) | |
4430 | rc->nodes_relocated += buf->len; | |
4431 | } | |
4432 | ||
83d4cfd4 | 4433 | if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) |
3fd0a558 | 4434 | ret = replace_file_extents(trans, rc, root, cow); |
83d4cfd4 | 4435 | return ret; |
3fd0a558 YZ |
4436 | } |
4437 | ||
4438 | /* | |
4439 | * called before creating snapshot. it calculates metadata reservation | |
01327610 | 4440 | * required for relocating tree blocks in the snapshot |
3fd0a558 | 4441 | */ |
147d256e | 4442 | void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, |
3fd0a558 YZ |
4443 | u64 *bytes_to_reserve) |
4444 | { | |
10995c04 QW |
4445 | struct btrfs_root *root = pending->root; |
4446 | struct reloc_control *rc = root->fs_info->reloc_ctl; | |
3fd0a558 | 4447 | |
6282675e | 4448 | if (!rc || !have_reloc_root(root)) |
3fd0a558 YZ |
4449 | return; |
4450 | ||
3fd0a558 YZ |
4451 | if (!rc->merge_reloc_tree) |
4452 | return; | |
4453 | ||
4454 | root = root->reloc_root; | |
4455 | BUG_ON(btrfs_root_refs(&root->root_item) == 0); | |
4456 | /* | |
4457 | * relocation is in the stage of merging trees. the space | |
4458 | * used by merging a reloc tree is twice the size of | |
4459 | * relocated tree nodes in the worst case. half for cowing | |
4460 | * the reloc tree, half for cowing the fs tree. the space | |
4461 | * used by cowing the reloc tree will be freed after the | |
4462 | * tree is dropped. if we create snapshot, cowing the fs | |
4463 | * tree may use more space than it frees. so we need | |
4464 | * reserve extra space. | |
4465 | */ | |
4466 | *bytes_to_reserve += rc->nodes_relocated; | |
4467 | } | |
4468 | ||
4469 | /* | |
4470 | * called after snapshot is created. migrate block reservation | |
4471 | * and create reloc root for the newly created snapshot | |
f44deb74 JB |
4472 | * |
4473 | * This is similar to btrfs_init_reloc_root(), we come out of here with two | |
4474 | * references held on the reloc_root, one for root->reloc_root and one for | |
4475 | * rc->reloc_roots. | |
3fd0a558 | 4476 | */ |
49b25e05 | 4477 | int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, |
3fd0a558 YZ |
4478 | struct btrfs_pending_snapshot *pending) |
4479 | { | |
4480 | struct btrfs_root *root = pending->root; | |
4481 | struct btrfs_root *reloc_root; | |
4482 | struct btrfs_root *new_root; | |
10995c04 | 4483 | struct reloc_control *rc = root->fs_info->reloc_ctl; |
3fd0a558 YZ |
4484 | int ret; |
4485 | ||
6282675e | 4486 | if (!rc || !have_reloc_root(root)) |
49b25e05 | 4487 | return 0; |
3fd0a558 YZ |
4488 | |
4489 | rc = root->fs_info->reloc_ctl; | |
4490 | rc->merging_rsv_size += rc->nodes_relocated; | |
4491 | ||
4492 | if (rc->merge_reloc_tree) { | |
4493 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
4494 | rc->block_rsv, | |
3a584174 | 4495 | rc->nodes_relocated, true); |
49b25e05 JM |
4496 | if (ret) |
4497 | return ret; | |
3fd0a558 YZ |
4498 | } |
4499 | ||
4500 | new_root = pending->snap; | |
4501 | reloc_root = create_reloc_root(trans, root->reloc_root, | |
4502 | new_root->root_key.objectid); | |
49b25e05 JM |
4503 | if (IS_ERR(reloc_root)) |
4504 | return PTR_ERR(reloc_root); | |
3fd0a558 | 4505 | |
ffd7b339 | 4506 | ret = __add_reloc_root(reloc_root); |
57a304cf | 4507 | ASSERT(ret != -EEXIST); |
3c925863 JB |
4508 | if (ret) { |
4509 | /* Pairs with create_reloc_root */ | |
4510 | btrfs_put_root(reloc_root); | |
4511 | return ret; | |
4512 | } | |
f44deb74 | 4513 | new_root->reloc_root = btrfs_grab_root(reloc_root); |
3fd0a558 | 4514 | |
49b25e05 | 4515 | if (rc->create_reloc_tree) |
3fd0a558 | 4516 | ret = clone_backref_node(trans, rc, root, reloc_root); |
49b25e05 | 4517 | return ret; |
3fd0a558 | 4518 | } |
b9a9a850 QW |
4519 | |
4520 | /* | |
4521 | * Get the current bytenr for the block group which is being relocated. | |
4522 | * | |
4523 | * Return U64_MAX if no running relocation. | |
4524 | */ | |
4525 | u64 btrfs_get_reloc_bg_bytenr(struct btrfs_fs_info *fs_info) | |
4526 | { | |
4527 | u64 logical = U64_MAX; | |
4528 | ||
4529 | lockdep_assert_held(&fs_info->reloc_mutex); | |
4530 | ||
4531 | if (fs_info->reloc_ctl && fs_info->reloc_ctl->block_group) | |
4532 | logical = fs_info->reloc_ctl->block_group->start; | |
4533 | return logical; | |
4534 | } |