]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 | 4 | */ |
c1d7c514 | 5 | |
ec6b910f | 6 | #include <linux/sched.h> |
f361bf4a | 7 | #include <linux/sched/signal.h> |
edbd8d4e | 8 | #include <linux/pagemap.h> |
ec44a35c | 9 | #include <linux/writeback.h> |
21af804c | 10 | #include <linux/blkdev.h> |
b7a9f29f | 11 | #include <linux/sort.h> |
4184ea7f | 12 | #include <linux/rcupdate.h> |
817d52f8 | 13 | #include <linux/kthread.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
dff51cd1 | 15 | #include <linux/ratelimit.h> |
b150a4f1 | 16 | #include <linux/percpu_counter.h> |
69fe2d75 | 17 | #include <linux/lockdep.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
602cbe91 | 19 | #include "misc.h" |
995946dd | 20 | #include "tree-log.h" |
fec577fb CM |
21 | #include "disk-io.h" |
22 | #include "print-tree.h" | |
0b86a832 | 23 | #include "volumes.h" |
53b381b3 | 24 | #include "raid56.h" |
925baedd | 25 | #include "locking.h" |
fa9c0d79 | 26 | #include "free-space-cache.h" |
1e144fb8 | 27 | #include "free-space-tree.h" |
6ab0a202 | 28 | #include "sysfs.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
8719aaae | 31 | #include "space-info.h" |
d12ffdd1 | 32 | #include "block-rsv.h" |
86736342 | 33 | #include "delalloc-space.h" |
aac0023c | 34 | #include "block-group.h" |
fec577fb | 35 | |
709c0486 AJ |
36 | #undef SCRAMBLE_DELAYED_REFS |
37 | ||
9f9b8e8d | 38 | |
5d4f98a2 | 39 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
40 | struct btrfs_delayed_ref_node *node, u64 parent, |
41 | u64 root_objectid, u64 owner_objectid, | |
42 | u64 owner_offset, int refs_to_drop, | |
43 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
44 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
45 | struct extent_buffer *leaf, | |
46 | struct btrfs_extent_item *ei); | |
47 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
48 | u64 parent, u64 root_objectid, |
49 | u64 flags, u64 owner, u64 offset, | |
50 | struct btrfs_key *ins, int ref_mod); | |
51 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 52 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 53 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
54 | static int find_next_key(struct btrfs_path *path, int level, |
55 | struct btrfs_key *key); | |
6a63209f | 56 | |
0f9dd46c JB |
57 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
58 | { | |
59 | return (cache->flags & bits) == bits; | |
60 | } | |
61 | ||
6f410d1b JB |
62 | int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info, |
63 | u64 start, u64 num_bytes) | |
817d52f8 | 64 | { |
11833d66 | 65 | u64 end = start + num_bytes - 1; |
0b246afa | 66 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 67 | start, end, EXTENT_UPTODATE); |
0b246afa | 68 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 69 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
70 | return 0; |
71 | } | |
817d52f8 | 72 | |
6f410d1b | 73 | void btrfs_free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 74 | { |
9e715da8 | 75 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 76 | u64 start, end; |
817d52f8 | 77 | |
11833d66 YZ |
78 | start = cache->key.objectid; |
79 | end = start + cache->key.offset - 1; | |
80 | ||
0b246afa | 81 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 82 | start, end, EXTENT_UPTODATE); |
0b246afa | 83 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 84 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
85 | } |
86 | ||
78192442 | 87 | static u64 generic_ref_to_space_flags(struct btrfs_ref *ref) |
0d9f824d | 88 | { |
ddf30cf0 QW |
89 | if (ref->type == BTRFS_REF_METADATA) { |
90 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
78192442 | 91 | return BTRFS_BLOCK_GROUP_SYSTEM; |
0d9f824d | 92 | else |
78192442 | 93 | return BTRFS_BLOCK_GROUP_METADATA; |
0d9f824d | 94 | } |
78192442 QW |
95 | return BTRFS_BLOCK_GROUP_DATA; |
96 | } | |
97 | ||
98 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, | |
99 | struct btrfs_ref *ref) | |
100 | { | |
101 | struct btrfs_space_info *space_info; | |
102 | u64 flags = generic_ref_to_space_flags(ref); | |
103 | ||
280c2908 | 104 | space_info = btrfs_find_space_info(fs_info, flags); |
78192442 QW |
105 | ASSERT(space_info); |
106 | percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len, | |
107 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
108 | } | |
109 | ||
110 | static void sub_pinned_bytes(struct btrfs_fs_info *fs_info, | |
111 | struct btrfs_ref *ref) | |
112 | { | |
113 | struct btrfs_space_info *space_info; | |
114 | u64 flags = generic_ref_to_space_flags(ref); | |
0d9f824d | 115 | |
280c2908 | 116 | space_info = btrfs_find_space_info(fs_info, flags); |
55e8196a | 117 | ASSERT(space_info); |
78192442 | 118 | percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len, |
dec59fa3 | 119 | BTRFS_TOTAL_BYTES_PINNED_BATCH); |
0d9f824d OS |
120 | } |
121 | ||
1a4ed8fd | 122 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 123 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
124 | { |
125 | int ret; | |
126 | struct btrfs_key key; | |
31840ae1 | 127 | struct btrfs_path *path; |
e02119d5 | 128 | |
31840ae1 | 129 | path = btrfs_alloc_path(); |
d8926bb3 MF |
130 | if (!path) |
131 | return -ENOMEM; | |
132 | ||
e02119d5 CM |
133 | key.objectid = start; |
134 | key.offset = len; | |
3173a18f | 135 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 136 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 137 | btrfs_free_path(path); |
7bb86316 CM |
138 | return ret; |
139 | } | |
140 | ||
a22285a6 | 141 | /* |
3173a18f | 142 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
143 | * |
144 | * the head node for delayed ref is used to store the sum of all the | |
145 | * reference count modifications queued up in the rbtree. the head | |
146 | * node may also store the extent flags to set. This way you can check | |
147 | * to see what the reference count and extent flags would be if all of | |
148 | * the delayed refs are not processed. | |
149 | */ | |
150 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 151 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 152 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
153 | { |
154 | struct btrfs_delayed_ref_head *head; | |
155 | struct btrfs_delayed_ref_root *delayed_refs; | |
156 | struct btrfs_path *path; | |
157 | struct btrfs_extent_item *ei; | |
158 | struct extent_buffer *leaf; | |
159 | struct btrfs_key key; | |
160 | u32 item_size; | |
161 | u64 num_refs; | |
162 | u64 extent_flags; | |
163 | int ret; | |
164 | ||
3173a18f JB |
165 | /* |
166 | * If we don't have skinny metadata, don't bother doing anything | |
167 | * different | |
168 | */ | |
0b246afa JM |
169 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
170 | offset = fs_info->nodesize; | |
3173a18f JB |
171 | metadata = 0; |
172 | } | |
173 | ||
a22285a6 YZ |
174 | path = btrfs_alloc_path(); |
175 | if (!path) | |
176 | return -ENOMEM; | |
177 | ||
a22285a6 YZ |
178 | if (!trans) { |
179 | path->skip_locking = 1; | |
180 | path->search_commit_root = 1; | |
181 | } | |
639eefc8 FDBM |
182 | |
183 | search_again: | |
184 | key.objectid = bytenr; | |
185 | key.offset = offset; | |
186 | if (metadata) | |
187 | key.type = BTRFS_METADATA_ITEM_KEY; | |
188 | else | |
189 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
190 | ||
0b246afa | 191 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
192 | if (ret < 0) |
193 | goto out_free; | |
194 | ||
3173a18f | 195 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
196 | if (path->slots[0]) { |
197 | path->slots[0]--; | |
198 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
199 | path->slots[0]); | |
200 | if (key.objectid == bytenr && | |
201 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 202 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
203 | ret = 0; |
204 | } | |
3173a18f JB |
205 | } |
206 | ||
a22285a6 YZ |
207 | if (ret == 0) { |
208 | leaf = path->nodes[0]; | |
209 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
210 | if (item_size >= sizeof(*ei)) { | |
211 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
212 | struct btrfs_extent_item); | |
213 | num_refs = btrfs_extent_refs(leaf, ei); | |
214 | extent_flags = btrfs_extent_flags(leaf, ei); | |
215 | } else { | |
ba3c2b19 NB |
216 | ret = -EINVAL; |
217 | btrfs_print_v0_err(fs_info); | |
218 | if (trans) | |
219 | btrfs_abort_transaction(trans, ret); | |
220 | else | |
221 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
222 | ||
223 | goto out_free; | |
a22285a6 | 224 | } |
ba3c2b19 | 225 | |
a22285a6 YZ |
226 | BUG_ON(num_refs == 0); |
227 | } else { | |
228 | num_refs = 0; | |
229 | extent_flags = 0; | |
230 | ret = 0; | |
231 | } | |
232 | ||
233 | if (!trans) | |
234 | goto out; | |
235 | ||
236 | delayed_refs = &trans->transaction->delayed_refs; | |
237 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 238 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
239 | if (head) { |
240 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 241 | refcount_inc(&head->refs); |
a22285a6 YZ |
242 | spin_unlock(&delayed_refs->lock); |
243 | ||
b3b4aa74 | 244 | btrfs_release_path(path); |
a22285a6 | 245 | |
8cc33e5c DS |
246 | /* |
247 | * Mutex was contended, block until it's released and try | |
248 | * again | |
249 | */ | |
a22285a6 YZ |
250 | mutex_lock(&head->mutex); |
251 | mutex_unlock(&head->mutex); | |
d278850e | 252 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 253 | goto search_again; |
a22285a6 | 254 | } |
d7df2c79 | 255 | spin_lock(&head->lock); |
a22285a6 YZ |
256 | if (head->extent_op && head->extent_op->update_flags) |
257 | extent_flags |= head->extent_op->flags_to_set; | |
258 | else | |
259 | BUG_ON(num_refs == 0); | |
260 | ||
d278850e | 261 | num_refs += head->ref_mod; |
d7df2c79 | 262 | spin_unlock(&head->lock); |
a22285a6 YZ |
263 | mutex_unlock(&head->mutex); |
264 | } | |
265 | spin_unlock(&delayed_refs->lock); | |
266 | out: | |
267 | WARN_ON(num_refs == 0); | |
268 | if (refs) | |
269 | *refs = num_refs; | |
270 | if (flags) | |
271 | *flags = extent_flags; | |
272 | out_free: | |
273 | btrfs_free_path(path); | |
274 | return ret; | |
275 | } | |
276 | ||
d8d5f3e1 CM |
277 | /* |
278 | * Back reference rules. Back refs have three main goals: | |
279 | * | |
280 | * 1) differentiate between all holders of references to an extent so that | |
281 | * when a reference is dropped we can make sure it was a valid reference | |
282 | * before freeing the extent. | |
283 | * | |
284 | * 2) Provide enough information to quickly find the holders of an extent | |
285 | * if we notice a given block is corrupted or bad. | |
286 | * | |
287 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
288 | * maintenance. This is actually the same as #2, but with a slightly | |
289 | * different use case. | |
290 | * | |
5d4f98a2 YZ |
291 | * There are two kinds of back refs. The implicit back refs is optimized |
292 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
293 | * back refs of this kind provide information about the block's owner tree | |
294 | * and the pointer's key. These information allow us to find the block by | |
295 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
296 | * referenced by their owner trees. The location of tree block is recorded | |
297 | * in the back refs. Actually the full back refs is generic, and can be | |
298 | * used in all cases the implicit back refs is used. The major shortcoming | |
299 | * of the full back refs is its overhead. Every time a tree block gets | |
300 | * COWed, we have to update back refs entry for all pointers in it. | |
301 | * | |
302 | * For a newly allocated tree block, we use implicit back refs for | |
303 | * pointers in it. This means most tree related operations only involve | |
304 | * implicit back refs. For a tree block created in old transaction, the | |
305 | * only way to drop a reference to it is COW it. So we can detect the | |
306 | * event that tree block loses its owner tree's reference and do the | |
307 | * back refs conversion. | |
308 | * | |
01327610 | 309 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
310 | * |
311 | * The reference count of the block is one and the tree is the block's | |
312 | * owner tree. Nothing to do in this case. | |
313 | * | |
314 | * The reference count of the block is one and the tree is not the | |
315 | * block's owner tree. In this case, full back refs is used for pointers | |
316 | * in the block. Remove these full back refs, add implicit back refs for | |
317 | * every pointers in the new block. | |
318 | * | |
319 | * The reference count of the block is greater than one and the tree is | |
320 | * the block's owner tree. In this case, implicit back refs is used for | |
321 | * pointers in the block. Add full back refs for every pointers in the | |
322 | * block, increase lower level extents' reference counts. The original | |
323 | * implicit back refs are entailed to the new block. | |
324 | * | |
325 | * The reference count of the block is greater than one and the tree is | |
326 | * not the block's owner tree. Add implicit back refs for every pointer in | |
327 | * the new block, increase lower level extents' reference count. | |
328 | * | |
329 | * Back Reference Key composing: | |
330 | * | |
331 | * The key objectid corresponds to the first byte in the extent, | |
332 | * The key type is used to differentiate between types of back refs. | |
333 | * There are different meanings of the key offset for different types | |
334 | * of back refs. | |
335 | * | |
d8d5f3e1 CM |
336 | * File extents can be referenced by: |
337 | * | |
338 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 339 | * - different files inside a single subvolume |
d8d5f3e1 CM |
340 | * - different offsets inside a file (bookend extents in file.c) |
341 | * | |
5d4f98a2 | 342 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
343 | * |
344 | * - Objectid of the subvolume root | |
d8d5f3e1 | 345 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
346 | * - original offset in the file |
347 | * - how many bookend extents | |
d8d5f3e1 | 348 | * |
5d4f98a2 YZ |
349 | * The key offset for the implicit back refs is hash of the first |
350 | * three fields. | |
d8d5f3e1 | 351 | * |
5d4f98a2 | 352 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 353 | * |
5d4f98a2 | 354 | * - number of pointers in the tree leaf |
d8d5f3e1 | 355 | * |
5d4f98a2 YZ |
356 | * The key offset for the implicit back refs is the first byte of |
357 | * the tree leaf | |
d8d5f3e1 | 358 | * |
5d4f98a2 YZ |
359 | * When a file extent is allocated, The implicit back refs is used. |
360 | * the fields are filled in: | |
d8d5f3e1 | 361 | * |
5d4f98a2 | 362 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 363 | * |
5d4f98a2 YZ |
364 | * When a file extent is removed file truncation, we find the |
365 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 366 | * |
5d4f98a2 | 367 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 368 | * |
5d4f98a2 | 369 | * Btree extents can be referenced by: |
d8d5f3e1 | 370 | * |
5d4f98a2 | 371 | * - Different subvolumes |
d8d5f3e1 | 372 | * |
5d4f98a2 YZ |
373 | * Both the implicit back refs and the full back refs for tree blocks |
374 | * only consist of key. The key offset for the implicit back refs is | |
375 | * objectid of block's owner tree. The key offset for the full back refs | |
376 | * is the first byte of parent block. | |
d8d5f3e1 | 377 | * |
5d4f98a2 YZ |
378 | * When implicit back refs is used, information about the lowest key and |
379 | * level of the tree block are required. These information are stored in | |
380 | * tree block info structure. | |
d8d5f3e1 | 381 | */ |
31840ae1 | 382 | |
167ce953 LB |
383 | /* |
384 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 385 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
386 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
387 | */ | |
388 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
389 | struct btrfs_extent_inline_ref *iref, | |
390 | enum btrfs_inline_ref_type is_data) | |
391 | { | |
392 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 393 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
394 | |
395 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
396 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
397 | type == BTRFS_SHARED_DATA_REF_KEY || | |
398 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
399 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 400 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 401 | return type; |
64ecdb64 LB |
402 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
403 | ASSERT(eb->fs_info); | |
404 | /* | |
405 | * Every shared one has parent tree | |
406 | * block, which must be aligned to | |
407 | * nodesize. | |
408 | */ | |
409 | if (offset && | |
410 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
411 | return type; | |
412 | } | |
167ce953 | 413 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 414 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 415 | return type; |
64ecdb64 LB |
416 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
417 | ASSERT(eb->fs_info); | |
418 | /* | |
419 | * Every shared one has parent tree | |
420 | * block, which must be aligned to | |
421 | * nodesize. | |
422 | */ | |
423 | if (offset && | |
424 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
425 | return type; | |
426 | } | |
167ce953 LB |
427 | } else { |
428 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
429 | return type; | |
430 | } | |
431 | } | |
432 | ||
433 | btrfs_print_leaf((struct extent_buffer *)eb); | |
434 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
435 | eb->start, type); | |
436 | WARN_ON(1); | |
437 | ||
438 | return BTRFS_REF_TYPE_INVALID; | |
439 | } | |
440 | ||
0785a9aa | 441 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
442 | { |
443 | u32 high_crc = ~(u32)0; | |
444 | u32 low_crc = ~(u32)0; | |
445 | __le64 lenum; | |
446 | ||
447 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 448 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 449 | lenum = cpu_to_le64(owner); |
65019df8 | 450 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 451 | lenum = cpu_to_le64(offset); |
65019df8 | 452 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
453 | |
454 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
455 | } | |
456 | ||
457 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
458 | struct btrfs_extent_data_ref *ref) | |
459 | { | |
460 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
461 | btrfs_extent_data_ref_objectid(leaf, ref), | |
462 | btrfs_extent_data_ref_offset(leaf, ref)); | |
463 | } | |
464 | ||
465 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
466 | struct btrfs_extent_data_ref *ref, | |
467 | u64 root_objectid, u64 owner, u64 offset) | |
468 | { | |
469 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
470 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
471 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
472 | return 0; | |
473 | return 1; | |
474 | } | |
475 | ||
476 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
477 | struct btrfs_path *path, |
478 | u64 bytenr, u64 parent, | |
479 | u64 root_objectid, | |
480 | u64 owner, u64 offset) | |
481 | { | |
bd1d53ef | 482 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
483 | struct btrfs_key key; |
484 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 485 | struct extent_buffer *leaf; |
5d4f98a2 | 486 | u32 nritems; |
74493f7a | 487 | int ret; |
5d4f98a2 YZ |
488 | int recow; |
489 | int err = -ENOENT; | |
74493f7a | 490 | |
31840ae1 | 491 | key.objectid = bytenr; |
5d4f98a2 YZ |
492 | if (parent) { |
493 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
494 | key.offset = parent; | |
495 | } else { | |
496 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
497 | key.offset = hash_extent_data_ref(root_objectid, | |
498 | owner, offset); | |
499 | } | |
500 | again: | |
501 | recow = 0; | |
502 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
503 | if (ret < 0) { | |
504 | err = ret; | |
505 | goto fail; | |
506 | } | |
31840ae1 | 507 | |
5d4f98a2 YZ |
508 | if (parent) { |
509 | if (!ret) | |
510 | return 0; | |
5d4f98a2 | 511 | goto fail; |
31840ae1 ZY |
512 | } |
513 | ||
514 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
515 | nritems = btrfs_header_nritems(leaf); |
516 | while (1) { | |
517 | if (path->slots[0] >= nritems) { | |
518 | ret = btrfs_next_leaf(root, path); | |
519 | if (ret < 0) | |
520 | err = ret; | |
521 | if (ret) | |
522 | goto fail; | |
523 | ||
524 | leaf = path->nodes[0]; | |
525 | nritems = btrfs_header_nritems(leaf); | |
526 | recow = 1; | |
527 | } | |
528 | ||
529 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
530 | if (key.objectid != bytenr || | |
531 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
532 | goto fail; | |
533 | ||
534 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
535 | struct btrfs_extent_data_ref); | |
536 | ||
537 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
538 | owner, offset)) { | |
539 | if (recow) { | |
b3b4aa74 | 540 | btrfs_release_path(path); |
5d4f98a2 YZ |
541 | goto again; |
542 | } | |
543 | err = 0; | |
544 | break; | |
545 | } | |
546 | path->slots[0]++; | |
31840ae1 | 547 | } |
5d4f98a2 YZ |
548 | fail: |
549 | return err; | |
31840ae1 ZY |
550 | } |
551 | ||
5d4f98a2 | 552 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
553 | struct btrfs_path *path, |
554 | u64 bytenr, u64 parent, | |
555 | u64 root_objectid, u64 owner, | |
556 | u64 offset, int refs_to_add) | |
31840ae1 | 557 | { |
62b895af | 558 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
559 | struct btrfs_key key; |
560 | struct extent_buffer *leaf; | |
5d4f98a2 | 561 | u32 size; |
31840ae1 ZY |
562 | u32 num_refs; |
563 | int ret; | |
74493f7a | 564 | |
74493f7a | 565 | key.objectid = bytenr; |
5d4f98a2 YZ |
566 | if (parent) { |
567 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
568 | key.offset = parent; | |
569 | size = sizeof(struct btrfs_shared_data_ref); | |
570 | } else { | |
571 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
572 | key.offset = hash_extent_data_ref(root_objectid, | |
573 | owner, offset); | |
574 | size = sizeof(struct btrfs_extent_data_ref); | |
575 | } | |
74493f7a | 576 | |
5d4f98a2 YZ |
577 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
578 | if (ret && ret != -EEXIST) | |
579 | goto fail; | |
580 | ||
581 | leaf = path->nodes[0]; | |
582 | if (parent) { | |
583 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 584 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
585 | struct btrfs_shared_data_ref); |
586 | if (ret == 0) { | |
587 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
588 | } else { | |
589 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
590 | num_refs += refs_to_add; | |
591 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 592 | } |
5d4f98a2 YZ |
593 | } else { |
594 | struct btrfs_extent_data_ref *ref; | |
595 | while (ret == -EEXIST) { | |
596 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
597 | struct btrfs_extent_data_ref); | |
598 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
599 | owner, offset)) | |
600 | break; | |
b3b4aa74 | 601 | btrfs_release_path(path); |
5d4f98a2 YZ |
602 | key.offset++; |
603 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
604 | size); | |
605 | if (ret && ret != -EEXIST) | |
606 | goto fail; | |
31840ae1 | 607 | |
5d4f98a2 YZ |
608 | leaf = path->nodes[0]; |
609 | } | |
610 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
611 | struct btrfs_extent_data_ref); | |
612 | if (ret == 0) { | |
613 | btrfs_set_extent_data_ref_root(leaf, ref, | |
614 | root_objectid); | |
615 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
616 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
617 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
618 | } else { | |
619 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
620 | num_refs += refs_to_add; | |
621 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 622 | } |
31840ae1 | 623 | } |
5d4f98a2 YZ |
624 | btrfs_mark_buffer_dirty(leaf); |
625 | ret = 0; | |
626 | fail: | |
b3b4aa74 | 627 | btrfs_release_path(path); |
7bb86316 | 628 | return ret; |
74493f7a CM |
629 | } |
630 | ||
5d4f98a2 | 631 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 632 | struct btrfs_path *path, |
fcebe456 | 633 | int refs_to_drop, int *last_ref) |
31840ae1 | 634 | { |
5d4f98a2 YZ |
635 | struct btrfs_key key; |
636 | struct btrfs_extent_data_ref *ref1 = NULL; | |
637 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 638 | struct extent_buffer *leaf; |
5d4f98a2 | 639 | u32 num_refs = 0; |
31840ae1 ZY |
640 | int ret = 0; |
641 | ||
642 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
643 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
644 | ||
645 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
646 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
647 | struct btrfs_extent_data_ref); | |
648 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
649 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
650 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
651 | struct btrfs_shared_data_ref); | |
652 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 653 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
654 | btrfs_print_v0_err(trans->fs_info); |
655 | btrfs_abort_transaction(trans, -EINVAL); | |
656 | return -EINVAL; | |
5d4f98a2 YZ |
657 | } else { |
658 | BUG(); | |
659 | } | |
660 | ||
56bec294 CM |
661 | BUG_ON(num_refs < refs_to_drop); |
662 | num_refs -= refs_to_drop; | |
5d4f98a2 | 663 | |
31840ae1 | 664 | if (num_refs == 0) { |
e9f6290d | 665 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 666 | *last_ref = 1; |
31840ae1 | 667 | } else { |
5d4f98a2 YZ |
668 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
669 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
670 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
671 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
672 | btrfs_mark_buffer_dirty(leaf); |
673 | } | |
31840ae1 ZY |
674 | return ret; |
675 | } | |
676 | ||
9ed0dea0 | 677 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 678 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 679 | { |
5d4f98a2 YZ |
680 | struct btrfs_key key; |
681 | struct extent_buffer *leaf; | |
682 | struct btrfs_extent_data_ref *ref1; | |
683 | struct btrfs_shared_data_ref *ref2; | |
684 | u32 num_refs = 0; | |
3de28d57 | 685 | int type; |
5d4f98a2 YZ |
686 | |
687 | leaf = path->nodes[0]; | |
688 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
689 | |
690 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 691 | if (iref) { |
3de28d57 LB |
692 | /* |
693 | * If type is invalid, we should have bailed out earlier than | |
694 | * this call. | |
695 | */ | |
696 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
697 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
698 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
699 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
700 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
701 | } else { | |
702 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
703 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
704 | } | |
705 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
706 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
707 | struct btrfs_extent_data_ref); | |
708 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
709 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
710 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
711 | struct btrfs_shared_data_ref); | |
712 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
713 | } else { |
714 | WARN_ON(1); | |
715 | } | |
716 | return num_refs; | |
717 | } | |
15916de8 | 718 | |
5d4f98a2 | 719 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
720 | struct btrfs_path *path, |
721 | u64 bytenr, u64 parent, | |
722 | u64 root_objectid) | |
1f3c79a2 | 723 | { |
b8582eea | 724 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 725 | struct btrfs_key key; |
1f3c79a2 | 726 | int ret; |
1f3c79a2 | 727 | |
5d4f98a2 YZ |
728 | key.objectid = bytenr; |
729 | if (parent) { | |
730 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
731 | key.offset = parent; | |
732 | } else { | |
733 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
734 | key.offset = root_objectid; | |
1f3c79a2 LH |
735 | } |
736 | ||
5d4f98a2 YZ |
737 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
738 | if (ret > 0) | |
739 | ret = -ENOENT; | |
5d4f98a2 | 740 | return ret; |
1f3c79a2 LH |
741 | } |
742 | ||
5d4f98a2 | 743 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
744 | struct btrfs_path *path, |
745 | u64 bytenr, u64 parent, | |
746 | u64 root_objectid) | |
31840ae1 | 747 | { |
5d4f98a2 | 748 | struct btrfs_key key; |
31840ae1 | 749 | int ret; |
31840ae1 | 750 | |
5d4f98a2 YZ |
751 | key.objectid = bytenr; |
752 | if (parent) { | |
753 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
754 | key.offset = parent; | |
755 | } else { | |
756 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
757 | key.offset = root_objectid; | |
758 | } | |
759 | ||
10728404 | 760 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 761 | path, &key, 0); |
b3b4aa74 | 762 | btrfs_release_path(path); |
31840ae1 ZY |
763 | return ret; |
764 | } | |
765 | ||
5d4f98a2 | 766 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 767 | { |
5d4f98a2 YZ |
768 | int type; |
769 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
770 | if (parent > 0) | |
771 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
772 | else | |
773 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
774 | } else { | |
775 | if (parent > 0) | |
776 | type = BTRFS_SHARED_DATA_REF_KEY; | |
777 | else | |
778 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
779 | } | |
780 | return type; | |
31840ae1 | 781 | } |
56bec294 | 782 | |
2c47e605 YZ |
783 | static int find_next_key(struct btrfs_path *path, int level, |
784 | struct btrfs_key *key) | |
56bec294 | 785 | |
02217ed2 | 786 | { |
2c47e605 | 787 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
788 | if (!path->nodes[level]) |
789 | break; | |
5d4f98a2 YZ |
790 | if (path->slots[level] + 1 >= |
791 | btrfs_header_nritems(path->nodes[level])) | |
792 | continue; | |
793 | if (level == 0) | |
794 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
795 | path->slots[level] + 1); | |
796 | else | |
797 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
798 | path->slots[level] + 1); | |
799 | return 0; | |
800 | } | |
801 | return 1; | |
802 | } | |
037e6390 | 803 | |
5d4f98a2 YZ |
804 | /* |
805 | * look for inline back ref. if back ref is found, *ref_ret is set | |
806 | * to the address of inline back ref, and 0 is returned. | |
807 | * | |
808 | * if back ref isn't found, *ref_ret is set to the address where it | |
809 | * should be inserted, and -ENOENT is returned. | |
810 | * | |
811 | * if insert is true and there are too many inline back refs, the path | |
812 | * points to the extent item, and -EAGAIN is returned. | |
813 | * | |
814 | * NOTE: inline back refs are ordered in the same way that back ref | |
815 | * items in the tree are ordered. | |
816 | */ | |
817 | static noinline_for_stack | |
818 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
819 | struct btrfs_path *path, |
820 | struct btrfs_extent_inline_ref **ref_ret, | |
821 | u64 bytenr, u64 num_bytes, | |
822 | u64 parent, u64 root_objectid, | |
823 | u64 owner, u64 offset, int insert) | |
824 | { | |
867cc1fb | 825 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 826 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
827 | struct btrfs_key key; |
828 | struct extent_buffer *leaf; | |
829 | struct btrfs_extent_item *ei; | |
830 | struct btrfs_extent_inline_ref *iref; | |
831 | u64 flags; | |
832 | u64 item_size; | |
833 | unsigned long ptr; | |
834 | unsigned long end; | |
835 | int extra_size; | |
836 | int type; | |
837 | int want; | |
838 | int ret; | |
839 | int err = 0; | |
0b246afa | 840 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 841 | int needed; |
26b8003f | 842 | |
db94535d | 843 | key.objectid = bytenr; |
31840ae1 | 844 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 845 | key.offset = num_bytes; |
31840ae1 | 846 | |
5d4f98a2 YZ |
847 | want = extent_ref_type(parent, owner); |
848 | if (insert) { | |
849 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 850 | path->keep_locks = 1; |
5d4f98a2 YZ |
851 | } else |
852 | extra_size = -1; | |
3173a18f JB |
853 | |
854 | /* | |
16d1c062 NB |
855 | * Owner is our level, so we can just add one to get the level for the |
856 | * block we are interested in. | |
3173a18f JB |
857 | */ |
858 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
859 | key.type = BTRFS_METADATA_ITEM_KEY; | |
860 | key.offset = owner; | |
861 | } | |
862 | ||
863 | again: | |
5d4f98a2 | 864 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 865 | if (ret < 0) { |
5d4f98a2 YZ |
866 | err = ret; |
867 | goto out; | |
868 | } | |
3173a18f JB |
869 | |
870 | /* | |
871 | * We may be a newly converted file system which still has the old fat | |
872 | * extent entries for metadata, so try and see if we have one of those. | |
873 | */ | |
874 | if (ret > 0 && skinny_metadata) { | |
875 | skinny_metadata = false; | |
876 | if (path->slots[0]) { | |
877 | path->slots[0]--; | |
878 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
879 | path->slots[0]); | |
880 | if (key.objectid == bytenr && | |
881 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
882 | key.offset == num_bytes) | |
883 | ret = 0; | |
884 | } | |
885 | if (ret) { | |
9ce49a0b | 886 | key.objectid = bytenr; |
3173a18f JB |
887 | key.type = BTRFS_EXTENT_ITEM_KEY; |
888 | key.offset = num_bytes; | |
889 | btrfs_release_path(path); | |
890 | goto again; | |
891 | } | |
892 | } | |
893 | ||
79787eaa JM |
894 | if (ret && !insert) { |
895 | err = -ENOENT; | |
896 | goto out; | |
fae7f21c | 897 | } else if (WARN_ON(ret)) { |
492104c8 | 898 | err = -EIO; |
492104c8 | 899 | goto out; |
79787eaa | 900 | } |
5d4f98a2 YZ |
901 | |
902 | leaf = path->nodes[0]; | |
903 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 904 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
905 | err = -EINVAL; |
906 | btrfs_print_v0_err(fs_info); | |
907 | btrfs_abort_transaction(trans, err); | |
908 | goto out; | |
909 | } | |
5d4f98a2 | 910 | |
5d4f98a2 YZ |
911 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
912 | flags = btrfs_extent_flags(leaf, ei); | |
913 | ||
914 | ptr = (unsigned long)(ei + 1); | |
915 | end = (unsigned long)ei + item_size; | |
916 | ||
3173a18f | 917 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
918 | ptr += sizeof(struct btrfs_tree_block_info); |
919 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
920 | } |
921 | ||
3de28d57 LB |
922 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
923 | needed = BTRFS_REF_TYPE_DATA; | |
924 | else | |
925 | needed = BTRFS_REF_TYPE_BLOCK; | |
926 | ||
5d4f98a2 YZ |
927 | err = -ENOENT; |
928 | while (1) { | |
929 | if (ptr >= end) { | |
930 | WARN_ON(ptr > end); | |
931 | break; | |
932 | } | |
933 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
934 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
935 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 936 | err = -EUCLEAN; |
3de28d57 LB |
937 | goto out; |
938 | } | |
939 | ||
5d4f98a2 YZ |
940 | if (want < type) |
941 | break; | |
942 | if (want > type) { | |
943 | ptr += btrfs_extent_inline_ref_size(type); | |
944 | continue; | |
945 | } | |
946 | ||
947 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
948 | struct btrfs_extent_data_ref *dref; | |
949 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
950 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
951 | owner, offset)) { | |
952 | err = 0; | |
953 | break; | |
954 | } | |
955 | if (hash_extent_data_ref_item(leaf, dref) < | |
956 | hash_extent_data_ref(root_objectid, owner, offset)) | |
957 | break; | |
958 | } else { | |
959 | u64 ref_offset; | |
960 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
961 | if (parent > 0) { | |
962 | if (parent == ref_offset) { | |
963 | err = 0; | |
964 | break; | |
965 | } | |
966 | if (ref_offset < parent) | |
967 | break; | |
968 | } else { | |
969 | if (root_objectid == ref_offset) { | |
970 | err = 0; | |
971 | break; | |
972 | } | |
973 | if (ref_offset < root_objectid) | |
974 | break; | |
975 | } | |
976 | } | |
977 | ptr += btrfs_extent_inline_ref_size(type); | |
978 | } | |
979 | if (err == -ENOENT && insert) { | |
980 | if (item_size + extra_size >= | |
981 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
982 | err = -EAGAIN; | |
983 | goto out; | |
984 | } | |
985 | /* | |
986 | * To add new inline back ref, we have to make sure | |
987 | * there is no corresponding back ref item. | |
988 | * For simplicity, we just do not add new inline back | |
989 | * ref if there is any kind of item for this block | |
990 | */ | |
2c47e605 YZ |
991 | if (find_next_key(path, 0, &key) == 0 && |
992 | key.objectid == bytenr && | |
85d4198e | 993 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
994 | err = -EAGAIN; |
995 | goto out; | |
996 | } | |
997 | } | |
998 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
999 | out: | |
85d4198e | 1000 | if (insert) { |
5d4f98a2 YZ |
1001 | path->keep_locks = 0; |
1002 | btrfs_unlock_up_safe(path, 1); | |
1003 | } | |
1004 | return err; | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * helper to add new inline back ref | |
1009 | */ | |
1010 | static noinline_for_stack | |
87bde3cd | 1011 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1012 | struct btrfs_path *path, |
1013 | struct btrfs_extent_inline_ref *iref, | |
1014 | u64 parent, u64 root_objectid, | |
1015 | u64 owner, u64 offset, int refs_to_add, | |
1016 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1017 | { |
1018 | struct extent_buffer *leaf; | |
1019 | struct btrfs_extent_item *ei; | |
1020 | unsigned long ptr; | |
1021 | unsigned long end; | |
1022 | unsigned long item_offset; | |
1023 | u64 refs; | |
1024 | int size; | |
1025 | int type; | |
5d4f98a2 YZ |
1026 | |
1027 | leaf = path->nodes[0]; | |
1028 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1029 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1030 | ||
1031 | type = extent_ref_type(parent, owner); | |
1032 | size = btrfs_extent_inline_ref_size(type); | |
1033 | ||
c71dd880 | 1034 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1035 | |
1036 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1037 | refs = btrfs_extent_refs(leaf, ei); | |
1038 | refs += refs_to_add; | |
1039 | btrfs_set_extent_refs(leaf, ei, refs); | |
1040 | if (extent_op) | |
1041 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1042 | ||
1043 | ptr = (unsigned long)ei + item_offset; | |
1044 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1045 | if (ptr < end - size) | |
1046 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1047 | end - size - ptr); | |
1048 | ||
1049 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1050 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1051 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1052 | struct btrfs_extent_data_ref *dref; | |
1053 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1054 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1055 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1056 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1057 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1058 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1059 | struct btrfs_shared_data_ref *sref; | |
1060 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1061 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1062 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1063 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1064 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1065 | } else { | |
1066 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1067 | } | |
1068 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1069 | } |
1070 | ||
1071 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1072 | struct btrfs_path *path, |
1073 | struct btrfs_extent_inline_ref **ref_ret, | |
1074 | u64 bytenr, u64 num_bytes, u64 parent, | |
1075 | u64 root_objectid, u64 owner, u64 offset) | |
1076 | { | |
1077 | int ret; | |
1078 | ||
867cc1fb NB |
1079 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1080 | num_bytes, parent, root_objectid, | |
1081 | owner, offset, 0); | |
5d4f98a2 | 1082 | if (ret != -ENOENT) |
54aa1f4d | 1083 | return ret; |
5d4f98a2 | 1084 | |
b3b4aa74 | 1085 | btrfs_release_path(path); |
5d4f98a2 YZ |
1086 | *ref_ret = NULL; |
1087 | ||
1088 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1089 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1090 | root_objectid); | |
5d4f98a2 | 1091 | } else { |
bd1d53ef NB |
1092 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1093 | root_objectid, owner, offset); | |
b9473439 | 1094 | } |
5d4f98a2 YZ |
1095 | return ret; |
1096 | } | |
31840ae1 | 1097 | |
5d4f98a2 YZ |
1098 | /* |
1099 | * helper to update/remove inline back ref | |
1100 | */ | |
1101 | static noinline_for_stack | |
61a18f1c | 1102 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1103 | struct btrfs_extent_inline_ref *iref, |
1104 | int refs_to_mod, | |
fcebe456 JB |
1105 | struct btrfs_delayed_extent_op *extent_op, |
1106 | int *last_ref) | |
5d4f98a2 | 1107 | { |
61a18f1c | 1108 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1109 | struct btrfs_extent_item *ei; |
1110 | struct btrfs_extent_data_ref *dref = NULL; | |
1111 | struct btrfs_shared_data_ref *sref = NULL; | |
1112 | unsigned long ptr; | |
1113 | unsigned long end; | |
1114 | u32 item_size; | |
1115 | int size; | |
1116 | int type; | |
5d4f98a2 YZ |
1117 | u64 refs; |
1118 | ||
5d4f98a2 YZ |
1119 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1120 | refs = btrfs_extent_refs(leaf, ei); | |
1121 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1122 | refs += refs_to_mod; | |
1123 | btrfs_set_extent_refs(leaf, ei, refs); | |
1124 | if (extent_op) | |
1125 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1126 | ||
3de28d57 LB |
1127 | /* |
1128 | * If type is invalid, we should have bailed out after | |
1129 | * lookup_inline_extent_backref(). | |
1130 | */ | |
1131 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1132 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1133 | |
1134 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1135 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1136 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1137 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1138 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1139 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1140 | } else { | |
1141 | refs = 1; | |
1142 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1143 | } |
31840ae1 | 1144 | |
5d4f98a2 YZ |
1145 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1146 | refs += refs_to_mod; | |
1147 | ||
1148 | if (refs > 0) { | |
1149 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1150 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1151 | else | |
1152 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1153 | } else { | |
fcebe456 | 1154 | *last_ref = 1; |
5d4f98a2 YZ |
1155 | size = btrfs_extent_inline_ref_size(type); |
1156 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1157 | ptr = (unsigned long)iref; | |
1158 | end = (unsigned long)ei + item_size; | |
1159 | if (ptr + size < end) | |
1160 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1161 | end - ptr - size); | |
1162 | item_size -= size; | |
78ac4f9e | 1163 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1164 | } |
1165 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1166 | } |
1167 | ||
1168 | static noinline_for_stack | |
1169 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1170 | struct btrfs_path *path, |
1171 | u64 bytenr, u64 num_bytes, u64 parent, | |
1172 | u64 root_objectid, u64 owner, | |
1173 | u64 offset, int refs_to_add, | |
1174 | struct btrfs_delayed_extent_op *extent_op) | |
1175 | { | |
1176 | struct btrfs_extent_inline_ref *iref; | |
1177 | int ret; | |
1178 | ||
867cc1fb NB |
1179 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1180 | num_bytes, parent, root_objectid, | |
1181 | owner, offset, 1); | |
5d4f98a2 YZ |
1182 | if (ret == 0) { |
1183 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1184 | update_inline_extent_backref(path, iref, refs_to_add, |
1185 | extent_op, NULL); | |
5d4f98a2 | 1186 | } else if (ret == -ENOENT) { |
a639cdeb | 1187 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1188 | root_objectid, owner, offset, |
1189 | refs_to_add, extent_op); | |
1190 | ret = 0; | |
771ed689 | 1191 | } |
5d4f98a2 YZ |
1192 | return ret; |
1193 | } | |
31840ae1 | 1194 | |
5d4f98a2 | 1195 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1196 | struct btrfs_path *path, |
1197 | u64 bytenr, u64 parent, u64 root_objectid, | |
1198 | u64 owner, u64 offset, int refs_to_add) | |
1199 | { | |
1200 | int ret; | |
1201 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1202 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1203 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1204 | root_objectid); | |
5d4f98a2 | 1205 | } else { |
62b895af NB |
1206 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1207 | root_objectid, owner, offset, | |
1208 | refs_to_add); | |
5d4f98a2 YZ |
1209 | } |
1210 | return ret; | |
1211 | } | |
56bec294 | 1212 | |
5d4f98a2 | 1213 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1214 | struct btrfs_path *path, |
1215 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1216 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1217 | { |
143bede5 | 1218 | int ret = 0; |
b9473439 | 1219 | |
5d4f98a2 YZ |
1220 | BUG_ON(!is_data && refs_to_drop != 1); |
1221 | if (iref) { | |
61a18f1c NB |
1222 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1223 | last_ref); | |
5d4f98a2 | 1224 | } else if (is_data) { |
e9f6290d | 1225 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1226 | last_ref); |
5d4f98a2 | 1227 | } else { |
fcebe456 | 1228 | *last_ref = 1; |
87cc7a8a | 1229 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1230 | } |
1231 | return ret; | |
1232 | } | |
1233 | ||
d04c6b88 JM |
1234 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1235 | u64 *discarded_bytes) | |
5d4f98a2 | 1236 | { |
86557861 JM |
1237 | int j, ret = 0; |
1238 | u64 bytes_left, end; | |
4d89d377 | 1239 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1240 | |
4d89d377 JM |
1241 | if (WARN_ON(start != aligned_start)) { |
1242 | len -= aligned_start - start; | |
1243 | len = round_down(len, 1 << 9); | |
1244 | start = aligned_start; | |
1245 | } | |
d04c6b88 | 1246 | |
4d89d377 | 1247 | *discarded_bytes = 0; |
86557861 JM |
1248 | |
1249 | if (!len) | |
1250 | return 0; | |
1251 | ||
1252 | end = start + len; | |
1253 | bytes_left = len; | |
1254 | ||
1255 | /* Skip any superblocks on this device. */ | |
1256 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1257 | u64 sb_start = btrfs_sb_offset(j); | |
1258 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1259 | u64 size = sb_start - start; | |
1260 | ||
1261 | if (!in_range(sb_start, start, bytes_left) && | |
1262 | !in_range(sb_end, start, bytes_left) && | |
1263 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1264 | continue; | |
1265 | ||
1266 | /* | |
1267 | * Superblock spans beginning of range. Adjust start and | |
1268 | * try again. | |
1269 | */ | |
1270 | if (sb_start <= start) { | |
1271 | start += sb_end - start; | |
1272 | if (start > end) { | |
1273 | bytes_left = 0; | |
1274 | break; | |
1275 | } | |
1276 | bytes_left = end - start; | |
1277 | continue; | |
1278 | } | |
1279 | ||
1280 | if (size) { | |
1281 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1282 | GFP_NOFS, 0); | |
1283 | if (!ret) | |
1284 | *discarded_bytes += size; | |
1285 | else if (ret != -EOPNOTSUPP) | |
1286 | return ret; | |
1287 | } | |
1288 | ||
1289 | start = sb_end; | |
1290 | if (start > end) { | |
1291 | bytes_left = 0; | |
1292 | break; | |
1293 | } | |
1294 | bytes_left = end - start; | |
1295 | } | |
1296 | ||
1297 | if (bytes_left) { | |
1298 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1299 | GFP_NOFS, 0); |
1300 | if (!ret) | |
86557861 | 1301 | *discarded_bytes += bytes_left; |
4d89d377 | 1302 | } |
d04c6b88 | 1303 | return ret; |
5d4f98a2 | 1304 | } |
5d4f98a2 | 1305 | |
2ff7e61e | 1306 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1307 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1308 | { |
5d4f98a2 | 1309 | int ret; |
5378e607 | 1310 | u64 discarded_bytes = 0; |
a1d3c478 | 1311 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1312 | |
e244a0ae | 1313 | |
2999241d FM |
1314 | /* |
1315 | * Avoid races with device replace and make sure our bbio has devices | |
1316 | * associated to its stripes that don't go away while we are discarding. | |
1317 | */ | |
0b246afa | 1318 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 1319 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
1320 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
1321 | &bbio, 0); | |
79787eaa | 1322 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1323 | if (!ret) { |
a1d3c478 | 1324 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1325 | int i; |
1326 | ||
5d4f98a2 | 1327 | |
a1d3c478 | 1328 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1329 | u64 bytes; |
38b5f68e AJ |
1330 | struct request_queue *req_q; |
1331 | ||
627e0873 FM |
1332 | if (!stripe->dev->bdev) { |
1333 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1334 | continue; | |
1335 | } | |
38b5f68e AJ |
1336 | req_q = bdev_get_queue(stripe->dev->bdev); |
1337 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1338 | continue; |
1339 | ||
5378e607 LD |
1340 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1341 | stripe->physical, | |
d04c6b88 JM |
1342 | stripe->length, |
1343 | &bytes); | |
5378e607 | 1344 | if (!ret) |
d04c6b88 | 1345 | discarded_bytes += bytes; |
5378e607 | 1346 | else if (ret != -EOPNOTSUPP) |
79787eaa | 1347 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1348 | |
1349 | /* | |
1350 | * Just in case we get back EOPNOTSUPP for some reason, | |
1351 | * just ignore the return value so we don't screw up | |
1352 | * people calling discard_extent. | |
1353 | */ | |
1354 | ret = 0; | |
5d4f98a2 | 1355 | } |
6e9606d2 | 1356 | btrfs_put_bbio(bbio); |
5d4f98a2 | 1357 | } |
0b246afa | 1358 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1359 | |
1360 | if (actual_bytes) | |
1361 | *actual_bytes = discarded_bytes; | |
1362 | ||
5d4f98a2 | 1363 | |
53b381b3 DW |
1364 | if (ret == -EOPNOTSUPP) |
1365 | ret = 0; | |
5d4f98a2 | 1366 | return ret; |
5d4f98a2 YZ |
1367 | } |
1368 | ||
79787eaa | 1369 | /* Can return -ENOMEM */ |
5d4f98a2 | 1370 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1371 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1372 | { |
82fa113f | 1373 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1374 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1375 | int ret; |
66d7e7f0 | 1376 | |
82fa113f QW |
1377 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1378 | generic_ref->action); | |
1379 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1380 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1381 | |
82fa113f QW |
1382 | if (generic_ref->type == BTRFS_REF_METADATA) |
1383 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1384 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1385 | else |
1386 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1387 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1388 | |
82fa113f | 1389 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1390 | |
ddf30cf0 | 1391 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1392 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 1393 | |
5d4f98a2 YZ |
1394 | return ret; |
1395 | } | |
1396 | ||
bd3c685e NB |
1397 | /* |
1398 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1399 | * | |
1400 | * @trans: Handle of transaction | |
1401 | * | |
1402 | * @node: The delayed ref node used to get the bytenr/length for | |
1403 | * extent whose references are incremented. | |
1404 | * | |
1405 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1406 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1407 | * bytenr of the parent block. Since new extents are always | |
1408 | * created with indirect references, this will only be the case | |
1409 | * when relocating a shared extent. In that case, root_objectid | |
1410 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1411 | * be 0 | |
1412 | * | |
1413 | * @root_objectid: The id of the root where this modification has originated, | |
1414 | * this can be either one of the well-known metadata trees or | |
1415 | * the subvolume id which references this extent. | |
1416 | * | |
1417 | * @owner: For data extents it is the inode number of the owning file. | |
1418 | * For metadata extents this parameter holds the level in the | |
1419 | * tree of the extent. | |
1420 | * | |
1421 | * @offset: For metadata extents the offset is ignored and is currently | |
1422 | * always passed as 0. For data extents it is the fileoffset | |
1423 | * this extent belongs to. | |
1424 | * | |
1425 | * @refs_to_add Number of references to add | |
1426 | * | |
1427 | * @extent_op Pointer to a structure, holding information necessary when | |
1428 | * updating a tree block's flags | |
1429 | * | |
1430 | */ | |
5d4f98a2 | 1431 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1432 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1433 | u64 parent, u64 root_objectid, |
1434 | u64 owner, u64 offset, int refs_to_add, | |
1435 | struct btrfs_delayed_extent_op *extent_op) | |
1436 | { | |
1437 | struct btrfs_path *path; | |
1438 | struct extent_buffer *leaf; | |
1439 | struct btrfs_extent_item *item; | |
fcebe456 | 1440 | struct btrfs_key key; |
c682f9b3 QW |
1441 | u64 bytenr = node->bytenr; |
1442 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1443 | u64 refs; |
1444 | int ret; | |
5d4f98a2 YZ |
1445 | |
1446 | path = btrfs_alloc_path(); | |
1447 | if (!path) | |
1448 | return -ENOMEM; | |
1449 | ||
e4058b54 | 1450 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
1451 | path->leave_spinning = 1; |
1452 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
1453 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1454 | parent, root_objectid, owner, | |
1455 | offset, refs_to_add, extent_op); | |
0ed4792a | 1456 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1457 | goto out; |
fcebe456 JB |
1458 | |
1459 | /* | |
1460 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1461 | * inline extent ref, so just update the reference count and add a | |
1462 | * normal backref. | |
1463 | */ | |
5d4f98a2 | 1464 | leaf = path->nodes[0]; |
fcebe456 | 1465 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1466 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1467 | refs = btrfs_extent_refs(leaf, item); | |
1468 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1469 | if (extent_op) | |
1470 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1471 | |
5d4f98a2 | 1472 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1473 | btrfs_release_path(path); |
56bec294 | 1474 | |
e4058b54 | 1475 | path->reada = READA_FORWARD; |
b9473439 | 1476 | path->leave_spinning = 1; |
56bec294 | 1477 | /* now insert the actual backref */ |
37593410 NB |
1478 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
1479 | owner, offset, refs_to_add); | |
79787eaa | 1480 | if (ret) |
66642832 | 1481 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1482 | out: |
56bec294 | 1483 | btrfs_free_path(path); |
30d133fc | 1484 | return ret; |
56bec294 CM |
1485 | } |
1486 | ||
5d4f98a2 | 1487 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1488 | struct btrfs_delayed_ref_node *node, |
1489 | struct btrfs_delayed_extent_op *extent_op, | |
1490 | int insert_reserved) | |
56bec294 | 1491 | { |
5d4f98a2 YZ |
1492 | int ret = 0; |
1493 | struct btrfs_delayed_data_ref *ref; | |
1494 | struct btrfs_key ins; | |
1495 | u64 parent = 0; | |
1496 | u64 ref_root = 0; | |
1497 | u64 flags = 0; | |
1498 | ||
1499 | ins.objectid = node->bytenr; | |
1500 | ins.offset = node->num_bytes; | |
1501 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1502 | ||
1503 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1504 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1505 | |
5d4f98a2 YZ |
1506 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1507 | parent = ref->parent; | |
fcebe456 | 1508 | ref_root = ref->root; |
5d4f98a2 YZ |
1509 | |
1510 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1511 | if (extent_op) |
5d4f98a2 | 1512 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1513 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1514 | flags, ref->objectid, | |
1515 | ref->offset, &ins, | |
1516 | node->ref_mod); | |
5d4f98a2 | 1517 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1518 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1519 | ref->objectid, ref->offset, | |
1520 | node->ref_mod, extent_op); | |
5d4f98a2 | 1521 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1522 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1523 | ref_root, ref->objectid, |
1524 | ref->offset, node->ref_mod, | |
c682f9b3 | 1525 | extent_op); |
5d4f98a2 YZ |
1526 | } else { |
1527 | BUG(); | |
1528 | } | |
1529 | return ret; | |
1530 | } | |
1531 | ||
1532 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1533 | struct extent_buffer *leaf, | |
1534 | struct btrfs_extent_item *ei) | |
1535 | { | |
1536 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1537 | if (extent_op->update_flags) { | |
1538 | flags |= extent_op->flags_to_set; | |
1539 | btrfs_set_extent_flags(leaf, ei, flags); | |
1540 | } | |
1541 | ||
1542 | if (extent_op->update_key) { | |
1543 | struct btrfs_tree_block_info *bi; | |
1544 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1545 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1546 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1547 | } | |
1548 | } | |
1549 | ||
1550 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1551 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1552 | struct btrfs_delayed_extent_op *extent_op) |
1553 | { | |
20b9a2d6 | 1554 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1555 | struct btrfs_key key; |
1556 | struct btrfs_path *path; | |
1557 | struct btrfs_extent_item *ei; | |
1558 | struct extent_buffer *leaf; | |
1559 | u32 item_size; | |
56bec294 | 1560 | int ret; |
5d4f98a2 | 1561 | int err = 0; |
b1c79e09 | 1562 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1563 | |
79787eaa JM |
1564 | if (trans->aborted) |
1565 | return 0; | |
1566 | ||
0b246afa | 1567 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1568 | metadata = 0; |
1569 | ||
5d4f98a2 YZ |
1570 | path = btrfs_alloc_path(); |
1571 | if (!path) | |
1572 | return -ENOMEM; | |
1573 | ||
d278850e | 1574 | key.objectid = head->bytenr; |
5d4f98a2 | 1575 | |
3173a18f | 1576 | if (metadata) { |
3173a18f | 1577 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1578 | key.offset = extent_op->level; |
3173a18f JB |
1579 | } else { |
1580 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1581 | key.offset = head->num_bytes; |
3173a18f JB |
1582 | } |
1583 | ||
1584 | again: | |
e4058b54 | 1585 | path->reada = READA_FORWARD; |
5d4f98a2 | 1586 | path->leave_spinning = 1; |
0b246afa | 1587 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1588 | if (ret < 0) { |
1589 | err = ret; | |
1590 | goto out; | |
1591 | } | |
1592 | if (ret > 0) { | |
3173a18f | 1593 | if (metadata) { |
55994887 FDBM |
1594 | if (path->slots[0] > 0) { |
1595 | path->slots[0]--; | |
1596 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1597 | path->slots[0]); | |
d278850e | 1598 | if (key.objectid == head->bytenr && |
55994887 | 1599 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1600 | key.offset == head->num_bytes) |
55994887 FDBM |
1601 | ret = 0; |
1602 | } | |
1603 | if (ret > 0) { | |
1604 | btrfs_release_path(path); | |
1605 | metadata = 0; | |
3173a18f | 1606 | |
d278850e JB |
1607 | key.objectid = head->bytenr; |
1608 | key.offset = head->num_bytes; | |
55994887 FDBM |
1609 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1610 | goto again; | |
1611 | } | |
1612 | } else { | |
1613 | err = -EIO; | |
1614 | goto out; | |
3173a18f | 1615 | } |
5d4f98a2 YZ |
1616 | } |
1617 | ||
1618 | leaf = path->nodes[0]; | |
1619 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1620 | |
6d8ff4e4 | 1621 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1622 | err = -EINVAL; |
1623 | btrfs_print_v0_err(fs_info); | |
1624 | btrfs_abort_transaction(trans, err); | |
1625 | goto out; | |
1626 | } | |
1627 | ||
5d4f98a2 YZ |
1628 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1629 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1630 | |
5d4f98a2 YZ |
1631 | btrfs_mark_buffer_dirty(leaf); |
1632 | out: | |
1633 | btrfs_free_path(path); | |
1634 | return err; | |
56bec294 CM |
1635 | } |
1636 | ||
5d4f98a2 | 1637 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1638 | struct btrfs_delayed_ref_node *node, |
1639 | struct btrfs_delayed_extent_op *extent_op, | |
1640 | int insert_reserved) | |
56bec294 CM |
1641 | { |
1642 | int ret = 0; | |
5d4f98a2 | 1643 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1644 | u64 parent = 0; |
1645 | u64 ref_root = 0; | |
56bec294 | 1646 | |
5d4f98a2 | 1647 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1648 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1649 | |
5d4f98a2 YZ |
1650 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1651 | parent = ref->parent; | |
fcebe456 | 1652 | ref_root = ref->root; |
5d4f98a2 | 1653 | |
02794222 | 1654 | if (node->ref_mod != 1) { |
f97806f2 | 1655 | btrfs_err(trans->fs_info, |
02794222 LB |
1656 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1657 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1658 | parent); | |
1659 | return -EIO; | |
1660 | } | |
5d4f98a2 | 1661 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1662 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1663 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1664 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1665 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1666 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1667 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1668 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1669 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1670 | } else { |
1671 | BUG(); | |
1672 | } | |
56bec294 CM |
1673 | return ret; |
1674 | } | |
1675 | ||
1676 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1677 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1678 | struct btrfs_delayed_ref_node *node, |
1679 | struct btrfs_delayed_extent_op *extent_op, | |
1680 | int insert_reserved) | |
56bec294 | 1681 | { |
79787eaa JM |
1682 | int ret = 0; |
1683 | ||
857cc2fc JB |
1684 | if (trans->aborted) { |
1685 | if (insert_reserved) | |
5fac7f9e | 1686 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 1687 | node->num_bytes, 1); |
79787eaa | 1688 | return 0; |
857cc2fc | 1689 | } |
79787eaa | 1690 | |
5d4f98a2 YZ |
1691 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1692 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1693 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1694 | insert_reserved); |
1695 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1696 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1697 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1698 | insert_reserved); |
1699 | else | |
1700 | BUG(); | |
80ee54bf JB |
1701 | if (ret && insert_reserved) |
1702 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
1703 | node->num_bytes, 1); | |
5d4f98a2 | 1704 | return ret; |
56bec294 CM |
1705 | } |
1706 | ||
c6fc2454 | 1707 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1708 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1709 | { | |
cffc3374 FM |
1710 | struct btrfs_delayed_ref_node *ref; |
1711 | ||
e3d03965 | 1712 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1713 | return NULL; |
d7df2c79 | 1714 | |
cffc3374 FM |
1715 | /* |
1716 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1717 | * This is to prevent a ref count from going down to zero, which deletes | |
1718 | * the extent item from the extent tree, when there still are references | |
1719 | * to add, which would fail because they would not find the extent item. | |
1720 | */ | |
1d57ee94 WX |
1721 | if (!list_empty(&head->ref_add_list)) |
1722 | return list_first_entry(&head->ref_add_list, | |
1723 | struct btrfs_delayed_ref_node, add_list); | |
1724 | ||
e3d03965 | 1725 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1726 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1727 | ASSERT(list_empty(&ref->add_list)); |
1728 | return ref; | |
56bec294 CM |
1729 | } |
1730 | ||
2eadaa22 JB |
1731 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1732 | struct btrfs_delayed_ref_head *head) | |
1733 | { | |
1734 | spin_lock(&delayed_refs->lock); | |
1735 | head->processing = 0; | |
1736 | delayed_refs->num_heads_ready++; | |
1737 | spin_unlock(&delayed_refs->lock); | |
1738 | btrfs_delayed_ref_unlock(head); | |
1739 | } | |
1740 | ||
bedc6617 JB |
1741 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1742 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1743 | { |
1744 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1745 | |
1746 | if (!extent_op) | |
bedc6617 JB |
1747 | return NULL; |
1748 | ||
b00e6250 | 1749 | if (head->must_insert_reserved) { |
bedc6617 | 1750 | head->extent_op = NULL; |
b00e6250 | 1751 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1752 | return NULL; |
b00e6250 | 1753 | } |
bedc6617 JB |
1754 | return extent_op; |
1755 | } | |
1756 | ||
1757 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1758 | struct btrfs_delayed_ref_head *head) | |
1759 | { | |
1760 | struct btrfs_delayed_extent_op *extent_op; | |
1761 | int ret; | |
1762 | ||
1763 | extent_op = cleanup_extent_op(head); | |
1764 | if (!extent_op) | |
1765 | return 0; | |
1766 | head->extent_op = NULL; | |
b00e6250 | 1767 | spin_unlock(&head->lock); |
20b9a2d6 | 1768 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1769 | btrfs_free_delayed_extent_op(extent_op); |
1770 | return ret ? ret : 1; | |
1771 | } | |
1772 | ||
31890da0 JB |
1773 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1774 | struct btrfs_delayed_ref_root *delayed_refs, | |
1775 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1776 | { |
ba2c4d4e | 1777 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
1778 | |
1779 | if (head->total_ref_mod < 0) { | |
1780 | struct btrfs_space_info *space_info; | |
1781 | u64 flags; | |
1782 | ||
1783 | if (head->is_data) | |
1784 | flags = BTRFS_BLOCK_GROUP_DATA; | |
1785 | else if (head->is_system) | |
1786 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
1787 | else | |
1788 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 1789 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
1790 | ASSERT(space_info); |
1791 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1792 | -head->num_bytes, | |
1793 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1794 | ||
ba2c4d4e JB |
1795 | /* |
1796 | * We had csum deletions accounted for in our delayed refs rsv, | |
1797 | * we need to drop the csum leaves for this update from our | |
1798 | * delayed_refs_rsv. | |
1799 | */ | |
07c47775 JB |
1800 | if (head->is_data) { |
1801 | spin_lock(&delayed_refs->lock); | |
1802 | delayed_refs->pending_csums -= head->num_bytes; | |
1803 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
1804 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
1805 | head->num_bytes); | |
07c47775 JB |
1806 | } |
1807 | } | |
1808 | ||
ba2c4d4e | 1809 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1810 | } |
1811 | ||
194ab0bc | 1812 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1813 | struct btrfs_delayed_ref_head *head) |
1814 | { | |
f9871edd NB |
1815 | |
1816 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1817 | struct btrfs_delayed_ref_root *delayed_refs; |
1818 | int ret; | |
1819 | ||
1820 | delayed_refs = &trans->transaction->delayed_refs; | |
1821 | ||
bedc6617 | 1822 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1823 | if (ret < 0) { |
1824 | unselect_delayed_ref_head(delayed_refs, head); | |
1825 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1826 | return ret; | |
1827 | } else if (ret) { | |
1828 | return ret; | |
1829 | } | |
1830 | ||
1831 | /* | |
1832 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1833 | * and then re-check to make sure nobody got added. | |
1834 | */ | |
1835 | spin_unlock(&head->lock); | |
1836 | spin_lock(&delayed_refs->lock); | |
1837 | spin_lock(&head->lock); | |
e3d03965 | 1838 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1839 | spin_unlock(&head->lock); |
1840 | spin_unlock(&delayed_refs->lock); | |
1841 | return 1; | |
1842 | } | |
d7baffda | 1843 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1844 | spin_unlock(&head->lock); |
1e7a1421 | 1845 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1846 | |
c1103f7a | 1847 | if (head->must_insert_reserved) { |
d278850e JB |
1848 | btrfs_pin_extent(fs_info, head->bytenr, |
1849 | head->num_bytes, 1); | |
c1103f7a | 1850 | if (head->is_data) { |
d278850e JB |
1851 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
1852 | head->num_bytes); | |
c1103f7a JB |
1853 | } |
1854 | } | |
1855 | ||
31890da0 | 1856 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1857 | |
1858 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1859 | btrfs_delayed_ref_unlock(head); |
d278850e | 1860 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1861 | return 0; |
1862 | } | |
1863 | ||
b1cdbcb5 NB |
1864 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1865 | struct btrfs_trans_handle *trans) | |
1866 | { | |
1867 | struct btrfs_delayed_ref_root *delayed_refs = | |
1868 | &trans->transaction->delayed_refs; | |
1869 | struct btrfs_delayed_ref_head *head = NULL; | |
1870 | int ret; | |
1871 | ||
1872 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1873 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1874 | if (!head) { |
1875 | spin_unlock(&delayed_refs->lock); | |
1876 | return head; | |
1877 | } | |
1878 | ||
1879 | /* | |
1880 | * Grab the lock that says we are going to process all the refs for | |
1881 | * this head | |
1882 | */ | |
9e920a6f | 1883 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1884 | spin_unlock(&delayed_refs->lock); |
1885 | ||
1886 | /* | |
1887 | * We may have dropped the spin lock to get the head mutex lock, and | |
1888 | * that might have given someone else time to free the head. If that's | |
1889 | * true, it has been removed from our list and we can move on. | |
1890 | */ | |
1891 | if (ret == -EAGAIN) | |
1892 | head = ERR_PTR(-EAGAIN); | |
1893 | ||
1894 | return head; | |
1895 | } | |
1896 | ||
e7261386 NB |
1897 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1898 | struct btrfs_delayed_ref_head *locked_ref, | |
1899 | unsigned long *run_refs) | |
1900 | { | |
1901 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1902 | struct btrfs_delayed_ref_root *delayed_refs; | |
1903 | struct btrfs_delayed_extent_op *extent_op; | |
1904 | struct btrfs_delayed_ref_node *ref; | |
1905 | int must_insert_reserved = 0; | |
1906 | int ret; | |
1907 | ||
1908 | delayed_refs = &trans->transaction->delayed_refs; | |
1909 | ||
0110a4c4 NB |
1910 | lockdep_assert_held(&locked_ref->mutex); |
1911 | lockdep_assert_held(&locked_ref->lock); | |
1912 | ||
e7261386 NB |
1913 | while ((ref = select_delayed_ref(locked_ref))) { |
1914 | if (ref->seq && | |
1915 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1916 | spin_unlock(&locked_ref->lock); | |
1917 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1918 | return -EAGAIN; | |
1919 | } | |
1920 | ||
1921 | (*run_refs)++; | |
1922 | ref->in_tree = 0; | |
1923 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1924 | RB_CLEAR_NODE(&ref->ref_node); | |
1925 | if (!list_empty(&ref->add_list)) | |
1926 | list_del(&ref->add_list); | |
1927 | /* | |
1928 | * When we play the delayed ref, also correct the ref_mod on | |
1929 | * head | |
1930 | */ | |
1931 | switch (ref->action) { | |
1932 | case BTRFS_ADD_DELAYED_REF: | |
1933 | case BTRFS_ADD_DELAYED_EXTENT: | |
1934 | locked_ref->ref_mod -= ref->ref_mod; | |
1935 | break; | |
1936 | case BTRFS_DROP_DELAYED_REF: | |
1937 | locked_ref->ref_mod += ref->ref_mod; | |
1938 | break; | |
1939 | default: | |
1940 | WARN_ON(1); | |
1941 | } | |
1942 | atomic_dec(&delayed_refs->num_entries); | |
1943 | ||
1944 | /* | |
1945 | * Record the must_insert_reserved flag before we drop the | |
1946 | * spin lock. | |
1947 | */ | |
1948 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1949 | locked_ref->must_insert_reserved = 0; | |
1950 | ||
1951 | extent_op = locked_ref->extent_op; | |
1952 | locked_ref->extent_op = NULL; | |
1953 | spin_unlock(&locked_ref->lock); | |
1954 | ||
1955 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1956 | must_insert_reserved); | |
1957 | ||
1958 | btrfs_free_delayed_extent_op(extent_op); | |
1959 | if (ret) { | |
1960 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1961 | btrfs_put_delayed_ref(ref); | |
1962 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1963 | ret); | |
1964 | return ret; | |
1965 | } | |
1966 | ||
1967 | btrfs_put_delayed_ref(ref); | |
1968 | cond_resched(); | |
1969 | ||
1970 | spin_lock(&locked_ref->lock); | |
1971 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1972 | } | |
1973 | ||
1974 | return 0; | |
1975 | } | |
1976 | ||
79787eaa JM |
1977 | /* |
1978 | * Returns 0 on success or if called with an already aborted transaction. | |
1979 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1980 | */ | |
d7df2c79 | 1981 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 1982 | unsigned long nr) |
56bec294 | 1983 | { |
0a1e458a | 1984 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 1985 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 1986 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 1987 | ktime_t start = ktime_get(); |
56bec294 | 1988 | int ret; |
d7df2c79 | 1989 | unsigned long count = 0; |
0a2b2a84 | 1990 | unsigned long actual_count = 0; |
56bec294 CM |
1991 | |
1992 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 1993 | do { |
56bec294 | 1994 | if (!locked_ref) { |
b1cdbcb5 | 1995 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
1996 | if (IS_ERR_OR_NULL(locked_ref)) { |
1997 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
1998 | continue; | |
1999 | } else { | |
2000 | break; | |
2001 | } | |
56bec294 | 2002 | } |
0110a4c4 | 2003 | count++; |
56bec294 | 2004 | } |
2c3cf7d5 FM |
2005 | /* |
2006 | * We need to try and merge add/drops of the same ref since we | |
2007 | * can run into issues with relocate dropping the implicit ref | |
2008 | * and then it being added back again before the drop can | |
2009 | * finish. If we merged anything we need to re-loop so we can | |
2010 | * get a good ref. | |
2011 | * Or we can get node references of the same type that weren't | |
2012 | * merged when created due to bumps in the tree mod seq, and | |
2013 | * we need to merge them to prevent adding an inline extent | |
2014 | * backref before dropping it (triggering a BUG_ON at | |
2015 | * insert_inline_extent_backref()). | |
2016 | */ | |
d7df2c79 | 2017 | spin_lock(&locked_ref->lock); |
be97f133 | 2018 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2019 | |
0110a4c4 NB |
2020 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2021 | &actual_count); | |
2022 | if (ret < 0 && ret != -EAGAIN) { | |
2023 | /* | |
2024 | * Error, btrfs_run_delayed_refs_for_head already | |
2025 | * unlocked everything so just bail out | |
2026 | */ | |
2027 | return ret; | |
2028 | } else if (!ret) { | |
2029 | /* | |
2030 | * Success, perform the usual cleanup of a processed | |
2031 | * head | |
2032 | */ | |
f9871edd | 2033 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2034 | if (ret > 0 ) { |
b00e6250 JB |
2035 | /* We dropped our lock, we need to loop. */ |
2036 | ret = 0; | |
d7df2c79 | 2037 | continue; |
194ab0bc JB |
2038 | } else if (ret) { |
2039 | return ret; | |
5d4f98a2 | 2040 | } |
22cd2e7d | 2041 | } |
1ce7a5ec | 2042 | |
b00e6250 | 2043 | /* |
0110a4c4 NB |
2044 | * Either success case or btrfs_run_delayed_refs_for_head |
2045 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2046 | */ |
b00e6250 | 2047 | |
0110a4c4 | 2048 | locked_ref = NULL; |
c3e69d58 | 2049 | cond_resched(); |
0110a4c4 | 2050 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2051 | |
2052 | /* | |
2053 | * We don't want to include ref heads since we can have empty ref heads | |
2054 | * and those will drastically skew our runtime down since we just do | |
2055 | * accounting, no actual extent tree updates. | |
2056 | */ | |
2057 | if (actual_count > 0) { | |
2058 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2059 | u64 avg; | |
2060 | ||
2061 | /* | |
2062 | * We weigh the current average higher than our current runtime | |
2063 | * to avoid large swings in the average. | |
2064 | */ | |
2065 | spin_lock(&delayed_refs->lock); | |
2066 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2067 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2068 | spin_unlock(&delayed_refs->lock); |
2069 | } | |
d7df2c79 | 2070 | return 0; |
c3e69d58 CM |
2071 | } |
2072 | ||
709c0486 AJ |
2073 | #ifdef SCRAMBLE_DELAYED_REFS |
2074 | /* | |
2075 | * Normally delayed refs get processed in ascending bytenr order. This | |
2076 | * correlates in most cases to the order added. To expose dependencies on this | |
2077 | * order, we start to process the tree in the middle instead of the beginning | |
2078 | */ | |
2079 | static u64 find_middle(struct rb_root *root) | |
2080 | { | |
2081 | struct rb_node *n = root->rb_node; | |
2082 | struct btrfs_delayed_ref_node *entry; | |
2083 | int alt = 1; | |
2084 | u64 middle; | |
2085 | u64 first = 0, last = 0; | |
2086 | ||
2087 | n = rb_first(root); | |
2088 | if (n) { | |
2089 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2090 | first = entry->bytenr; | |
2091 | } | |
2092 | n = rb_last(root); | |
2093 | if (n) { | |
2094 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2095 | last = entry->bytenr; | |
2096 | } | |
2097 | n = root->rb_node; | |
2098 | ||
2099 | while (n) { | |
2100 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2101 | WARN_ON(!entry->in_tree); | |
2102 | ||
2103 | middle = entry->bytenr; | |
2104 | ||
2105 | if (alt) | |
2106 | n = n->rb_left; | |
2107 | else | |
2108 | n = n->rb_right; | |
2109 | ||
2110 | alt = 1 - alt; | |
2111 | } | |
2112 | return middle; | |
2113 | } | |
2114 | #endif | |
2115 | ||
2ff7e61e | 2116 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2117 | { |
2118 | u64 num_bytes; | |
2119 | ||
2120 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2121 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2122 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2123 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2124 | ||
2125 | /* | |
2126 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2127 | * closer to what we're really going to want to use. |
1be41b78 | 2128 | */ |
0b246afa | 2129 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2130 | } |
2131 | ||
1262133b JB |
2132 | /* |
2133 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2134 | * would require to store the csums for that many bytes. | |
2135 | */ | |
2ff7e61e | 2136 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2137 | { |
2138 | u64 csum_size; | |
2139 | u64 num_csums_per_leaf; | |
2140 | u64 num_csums; | |
2141 | ||
0b246afa | 2142 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2143 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2144 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2145 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2146 | num_csums += num_csums_per_leaf - 1; |
2147 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2148 | return num_csums; | |
2149 | } | |
2150 | ||
c3e69d58 CM |
2151 | /* |
2152 | * this starts processing the delayed reference count updates and | |
2153 | * extent insertions we have queued up so far. count can be | |
2154 | * 0, which means to process everything in the tree at the start | |
2155 | * of the run (but not newly added entries), or it can be some target | |
2156 | * number you'd like to process. | |
79787eaa JM |
2157 | * |
2158 | * Returns 0 on success or if called with an aborted transaction | |
2159 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2160 | */ |
2161 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2162 | unsigned long count) |
c3e69d58 | 2163 | { |
c79a70b1 | 2164 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2165 | struct rb_node *node; |
2166 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2167 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2168 | int ret; |
2169 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2170 | |
79787eaa JM |
2171 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2172 | if (trans->aborted) | |
2173 | return 0; | |
2174 | ||
0b246afa | 2175 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2176 | return 0; |
2177 | ||
c3e69d58 | 2178 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2179 | if (count == 0) |
d7df2c79 | 2180 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2181 | |
c3e69d58 | 2182 | again: |
709c0486 AJ |
2183 | #ifdef SCRAMBLE_DELAYED_REFS |
2184 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2185 | #endif | |
0a1e458a | 2186 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2187 | if (ret < 0) { |
66642832 | 2188 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2189 | return ret; |
eb099670 | 2190 | } |
c3e69d58 | 2191 | |
56bec294 | 2192 | if (run_all) { |
119e80df | 2193 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2194 | |
d7df2c79 | 2195 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2196 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2197 | if (!node) { |
2198 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2199 | goto out; |
d7df2c79 | 2200 | } |
d278850e JB |
2201 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2202 | href_node); | |
2203 | refcount_inc(&head->refs); | |
2204 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2205 | |
d278850e JB |
2206 | /* Mutex was contended, block until it's released and retry. */ |
2207 | mutex_lock(&head->mutex); | |
2208 | mutex_unlock(&head->mutex); | |
56bec294 | 2209 | |
d278850e | 2210 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2211 | cond_resched(); |
56bec294 | 2212 | goto again; |
5f39d397 | 2213 | } |
54aa1f4d | 2214 | out: |
a28ec197 CM |
2215 | return 0; |
2216 | } | |
2217 | ||
5d4f98a2 | 2218 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
5d4f98a2 | 2219 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2220 | int level, int is_data) |
5d4f98a2 YZ |
2221 | { |
2222 | struct btrfs_delayed_extent_op *extent_op; | |
2223 | int ret; | |
2224 | ||
78a6184a | 2225 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2226 | if (!extent_op) |
2227 | return -ENOMEM; | |
2228 | ||
2229 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2230 | extent_op->update_flags = true; |
2231 | extent_op->update_key = false; | |
2232 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2233 | extent_op->level = level; |
5d4f98a2 | 2234 | |
c6e340bc | 2235 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); |
5d4f98a2 | 2236 | if (ret) |
78a6184a | 2237 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2238 | return ret; |
2239 | } | |
2240 | ||
e4c3b2dc | 2241 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2242 | struct btrfs_path *path, |
2243 | u64 objectid, u64 offset, u64 bytenr) | |
2244 | { | |
2245 | struct btrfs_delayed_ref_head *head; | |
2246 | struct btrfs_delayed_ref_node *ref; | |
2247 | struct btrfs_delayed_data_ref *data_ref; | |
2248 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2249 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2250 | struct rb_node *node; |
5d4f98a2 YZ |
2251 | int ret = 0; |
2252 | ||
998ac6d2 | 2253 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2254 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2255 | if (cur_trans) |
2256 | refcount_inc(&cur_trans->use_count); | |
2257 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2258 | if (!cur_trans) |
2259 | return 0; | |
2260 | ||
2261 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2262 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2263 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2264 | if (!head) { |
2265 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2266 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2267 | return 0; |
2268 | } | |
5d4f98a2 YZ |
2269 | |
2270 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2271 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2272 | spin_unlock(&delayed_refs->lock); |
2273 | ||
b3b4aa74 | 2274 | btrfs_release_path(path); |
5d4f98a2 | 2275 | |
8cc33e5c DS |
2276 | /* |
2277 | * Mutex was contended, block until it's released and let | |
2278 | * caller try again | |
2279 | */ | |
5d4f98a2 YZ |
2280 | mutex_lock(&head->mutex); |
2281 | mutex_unlock(&head->mutex); | |
d278850e | 2282 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2283 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2284 | return -EAGAIN; |
2285 | } | |
d7df2c79 | 2286 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2287 | |
d7df2c79 | 2288 | spin_lock(&head->lock); |
0e0adbcf JB |
2289 | /* |
2290 | * XXX: We should replace this with a proper search function in the | |
2291 | * future. | |
2292 | */ | |
e3d03965 LB |
2293 | for (node = rb_first_cached(&head->ref_tree); node; |
2294 | node = rb_next(node)) { | |
0e0adbcf | 2295 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2296 | /* If it's a shared ref we know a cross reference exists */ |
2297 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2298 | ret = 1; | |
2299 | break; | |
2300 | } | |
5d4f98a2 | 2301 | |
d7df2c79 | 2302 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2303 | |
d7df2c79 JB |
2304 | /* |
2305 | * If our ref doesn't match the one we're currently looking at | |
2306 | * then we have a cross reference. | |
2307 | */ | |
2308 | if (data_ref->root != root->root_key.objectid || | |
2309 | data_ref->objectid != objectid || | |
2310 | data_ref->offset != offset) { | |
2311 | ret = 1; | |
2312 | break; | |
2313 | } | |
5d4f98a2 | 2314 | } |
d7df2c79 | 2315 | spin_unlock(&head->lock); |
5d4f98a2 | 2316 | mutex_unlock(&head->mutex); |
998ac6d2 | 2317 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2318 | return ret; |
2319 | } | |
2320 | ||
e4c3b2dc | 2321 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2322 | struct btrfs_path *path, |
2323 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 2324 | { |
0b246afa JM |
2325 | struct btrfs_fs_info *fs_info = root->fs_info; |
2326 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2327 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2328 | struct btrfs_extent_data_ref *ref; |
2329 | struct btrfs_extent_inline_ref *iref; | |
2330 | struct btrfs_extent_item *ei; | |
f321e491 | 2331 | struct btrfs_key key; |
5d4f98a2 | 2332 | u32 item_size; |
3de28d57 | 2333 | int type; |
be20aa9d | 2334 | int ret; |
925baedd | 2335 | |
be20aa9d | 2336 | key.objectid = bytenr; |
31840ae1 | 2337 | key.offset = (u64)-1; |
f321e491 | 2338 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2339 | |
be20aa9d CM |
2340 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2341 | if (ret < 0) | |
2342 | goto out; | |
79787eaa | 2343 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2344 | |
2345 | ret = -ENOENT; | |
2346 | if (path->slots[0] == 0) | |
31840ae1 | 2347 | goto out; |
be20aa9d | 2348 | |
31840ae1 | 2349 | path->slots[0]--; |
f321e491 | 2350 | leaf = path->nodes[0]; |
5d4f98a2 | 2351 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2352 | |
5d4f98a2 | 2353 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2354 | goto out; |
f321e491 | 2355 | |
5d4f98a2 YZ |
2356 | ret = 1; |
2357 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2358 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2359 | |
a6bd9cd1 | 2360 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2361 | if (item_size != sizeof(*ei) + |
2362 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2363 | goto out; | |
be20aa9d | 2364 | |
a6bd9cd1 | 2365 | /* If extent created before last snapshot => it's definitely shared */ |
5d4f98a2 YZ |
2366 | if (btrfs_extent_generation(leaf, ei) <= |
2367 | btrfs_root_last_snapshot(&root->root_item)) | |
2368 | goto out; | |
2369 | ||
2370 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2371 | |
a6bd9cd1 | 2372 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2373 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2374 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2375 | goto out; |
2376 | ||
2377 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2378 | if (btrfs_extent_refs(leaf, ei) != | |
2379 | btrfs_extent_data_ref_count(leaf, ref) || | |
2380 | btrfs_extent_data_ref_root(leaf, ref) != | |
2381 | root->root_key.objectid || | |
2382 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2383 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2384 | goto out; | |
2385 | ||
2386 | ret = 0; | |
2387 | out: | |
2388 | return ret; | |
2389 | } | |
2390 | ||
e4c3b2dc LB |
2391 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
2392 | u64 bytenr) | |
5d4f98a2 YZ |
2393 | { |
2394 | struct btrfs_path *path; | |
2395 | int ret; | |
5d4f98a2 YZ |
2396 | |
2397 | path = btrfs_alloc_path(); | |
2398 | if (!path) | |
9132c4ff | 2399 | return -ENOMEM; |
5d4f98a2 YZ |
2400 | |
2401 | do { | |
e4c3b2dc | 2402 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
2403 | offset, bytenr); |
2404 | if (ret && ret != -ENOENT) | |
f321e491 | 2405 | goto out; |
80ff3856 | 2406 | |
380fd066 MT |
2407 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2408 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2409 | |
be20aa9d | 2410 | out: |
80ff3856 | 2411 | btrfs_free_path(path); |
f0486c68 YZ |
2412 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2413 | WARN_ON(ret > 0); | |
f321e491 | 2414 | return ret; |
be20aa9d | 2415 | } |
c5739bba | 2416 | |
5d4f98a2 | 2417 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2418 | struct btrfs_root *root, |
5d4f98a2 | 2419 | struct extent_buffer *buf, |
e339a6b0 | 2420 | int full_backref, int inc) |
31840ae1 | 2421 | { |
0b246afa | 2422 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2423 | u64 bytenr; |
5d4f98a2 YZ |
2424 | u64 num_bytes; |
2425 | u64 parent; | |
31840ae1 | 2426 | u64 ref_root; |
31840ae1 | 2427 | u32 nritems; |
31840ae1 ZY |
2428 | struct btrfs_key key; |
2429 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2430 | struct btrfs_ref generic_ref = { 0 }; |
2431 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2432 | int i; |
82fa113f | 2433 | int action; |
31840ae1 ZY |
2434 | int level; |
2435 | int ret = 0; | |
fccb84c9 | 2436 | |
0b246afa | 2437 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2438 | return 0; |
fccb84c9 | 2439 | |
31840ae1 | 2440 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2441 | nritems = btrfs_header_nritems(buf); |
2442 | level = btrfs_header_level(buf); | |
2443 | ||
27cdeb70 | 2444 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 2445 | return 0; |
31840ae1 | 2446 | |
5d4f98a2 YZ |
2447 | if (full_backref) |
2448 | parent = buf->start; | |
2449 | else | |
2450 | parent = 0; | |
82fa113f QW |
2451 | if (inc) |
2452 | action = BTRFS_ADD_DELAYED_REF; | |
2453 | else | |
2454 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2455 | |
2456 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2457 | if (level == 0) { |
5d4f98a2 | 2458 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2459 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2460 | continue; |
5d4f98a2 | 2461 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2462 | struct btrfs_file_extent_item); |
2463 | if (btrfs_file_extent_type(buf, fi) == | |
2464 | BTRFS_FILE_EXTENT_INLINE) | |
2465 | continue; | |
2466 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2467 | if (bytenr == 0) | |
2468 | continue; | |
5d4f98a2 YZ |
2469 | |
2470 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2471 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2472 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2473 | num_bytes, parent); | |
2474 | generic_ref.real_root = root->root_key.objectid; | |
2475 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2476 | key.offset); | |
2477 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2478 | if (inc) |
82fa113f | 2479 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2480 | else |
ffd4bb2a | 2481 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2482 | if (ret) |
2483 | goto fail; | |
2484 | } else { | |
5d4f98a2 | 2485 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2486 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2487 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2488 | num_bytes, parent); | |
2489 | generic_ref.real_root = root->root_key.objectid; | |
2490 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2491 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2492 | if (inc) |
82fa113f | 2493 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2494 | else |
ffd4bb2a | 2495 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2496 | if (ret) |
2497 | goto fail; | |
2498 | } | |
2499 | } | |
2500 | return 0; | |
2501 | fail: | |
5d4f98a2 YZ |
2502 | return ret; |
2503 | } | |
2504 | ||
2505 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2506 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2507 | { |
e339a6b0 | 2508 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2509 | } |
2510 | ||
2511 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2512 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2513 | { |
e339a6b0 | 2514 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2515 | } |
2516 | ||
2ff7e61e | 2517 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
2518 | { |
2519 | struct btrfs_block_group_cache *block_group; | |
2520 | int readonly = 0; | |
2521 | ||
0b246afa | 2522 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2523 | if (!block_group || block_group->ro) |
2524 | readonly = 1; | |
2525 | if (block_group) | |
fa9c0d79 | 2526 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2527 | return readonly; |
2528 | } | |
2529 | ||
1b86826d | 2530 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2531 | { |
0b246afa | 2532 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2533 | u64 flags; |
53b381b3 | 2534 | u64 ret; |
9ed74f2d | 2535 | |
b742bb82 YZ |
2536 | if (data) |
2537 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2538 | else if (root == fs_info->chunk_root) |
b742bb82 | 2539 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2540 | else |
b742bb82 | 2541 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2542 | |
878d7b67 | 2543 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2544 | return ret; |
6a63209f | 2545 | } |
9ed74f2d | 2546 | |
2ff7e61e | 2547 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2548 | { |
0f9dd46c | 2549 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 2550 | u64 bytenr; |
0f9dd46c | 2551 | |
0b246afa JM |
2552 | spin_lock(&fs_info->block_group_cache_lock); |
2553 | bytenr = fs_info->first_logical_byte; | |
2554 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2555 | |
2556 | if (bytenr < (u64)-1) | |
2557 | return bytenr; | |
2558 | ||
0b246afa | 2559 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2560 | if (!cache) |
a061fc8d | 2561 | return 0; |
0f9dd46c | 2562 | |
d2fb3437 | 2563 | bytenr = cache->key.objectid; |
fa9c0d79 | 2564 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2565 | |
2566 | return bytenr; | |
a061fc8d CM |
2567 | } |
2568 | ||
fdf08605 | 2569 | static int pin_down_extent(struct btrfs_block_group_cache *cache, |
f0486c68 | 2570 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2571 | { |
fdf08605 DS |
2572 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2573 | ||
11833d66 YZ |
2574 | spin_lock(&cache->space_info->lock); |
2575 | spin_lock(&cache->lock); | |
2576 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2577 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2578 | num_bytes); | |
11833d66 YZ |
2579 | if (reserved) { |
2580 | cache->reserved -= num_bytes; | |
2581 | cache->space_info->bytes_reserved -= num_bytes; | |
2582 | } | |
2583 | spin_unlock(&cache->lock); | |
2584 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2585 | |
dec59fa3 EL |
2586 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
2587 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 2588 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
2589 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2590 | return 0; | |
2591 | } | |
68b38550 | 2592 | |
f0486c68 YZ |
2593 | /* |
2594 | * this function must be called within transaction | |
2595 | */ | |
2ff7e61e | 2596 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
2597 | u64 bytenr, u64 num_bytes, int reserved) |
2598 | { | |
2599 | struct btrfs_block_group_cache *cache; | |
68b38550 | 2600 | |
0b246afa | 2601 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 2602 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2603 | |
fdf08605 | 2604 | pin_down_extent(cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2605 | |
2606 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2607 | return 0; |
2608 | } | |
2609 | ||
f0486c68 | 2610 | /* |
e688b725 CM |
2611 | * this function must be called within transaction |
2612 | */ | |
2ff7e61e | 2613 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
2614 | u64 bytenr, u64 num_bytes) |
2615 | { | |
2616 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 2617 | int ret; |
e688b725 | 2618 | |
0b246afa | 2619 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
2620 | if (!cache) |
2621 | return -EINVAL; | |
e688b725 CM |
2622 | |
2623 | /* | |
2624 | * pull in the free space cache (if any) so that our pin | |
2625 | * removes the free space from the cache. We have load_only set | |
2626 | * to one because the slow code to read in the free extents does check | |
2627 | * the pinned extents. | |
2628 | */ | |
676f1f75 | 2629 | btrfs_cache_block_group(cache, 1); |
e688b725 | 2630 | |
fdf08605 | 2631 | pin_down_extent(cache, bytenr, num_bytes, 0); |
e688b725 CM |
2632 | |
2633 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2634 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 2635 | btrfs_put_block_group(cache); |
b50c6e25 | 2636 | return ret; |
e688b725 CM |
2637 | } |
2638 | ||
2ff7e61e JM |
2639 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2640 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2641 | { |
2642 | int ret; | |
2643 | struct btrfs_block_group_cache *block_group; | |
2644 | struct btrfs_caching_control *caching_ctl; | |
2645 | ||
0b246afa | 2646 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2647 | if (!block_group) |
2648 | return -EINVAL; | |
2649 | ||
676f1f75 | 2650 | btrfs_cache_block_group(block_group, 0); |
e3cb339f | 2651 | caching_ctl = btrfs_get_caching_control(block_group); |
8c2a1a30 JB |
2652 | |
2653 | if (!caching_ctl) { | |
2654 | /* Logic error */ | |
676f1f75 | 2655 | BUG_ON(!btrfs_block_group_cache_done(block_group)); |
8c2a1a30 JB |
2656 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2657 | } else { | |
2658 | mutex_lock(&caching_ctl->mutex); | |
2659 | ||
2660 | if (start >= caching_ctl->progress) { | |
6f410d1b JB |
2661 | ret = btrfs_add_excluded_extent(fs_info, start, |
2662 | num_bytes); | |
8c2a1a30 JB |
2663 | } else if (start + num_bytes <= caching_ctl->progress) { |
2664 | ret = btrfs_remove_free_space(block_group, | |
2665 | start, num_bytes); | |
2666 | } else { | |
2667 | num_bytes = caching_ctl->progress - start; | |
2668 | ret = btrfs_remove_free_space(block_group, | |
2669 | start, num_bytes); | |
2670 | if (ret) | |
2671 | goto out_lock; | |
2672 | ||
2673 | num_bytes = (start + num_bytes) - | |
2674 | caching_ctl->progress; | |
2675 | start = caching_ctl->progress; | |
6f410d1b JB |
2676 | ret = btrfs_add_excluded_extent(fs_info, start, |
2677 | num_bytes); | |
8c2a1a30 JB |
2678 | } |
2679 | out_lock: | |
2680 | mutex_unlock(&caching_ctl->mutex); | |
e3cb339f | 2681 | btrfs_put_caching_control(caching_ctl); |
8c2a1a30 JB |
2682 | } |
2683 | btrfs_put_block_group(block_group); | |
2684 | return ret; | |
2685 | } | |
2686 | ||
bcdc428c | 2687 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2688 | { |
bcdc428c | 2689 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2690 | struct btrfs_file_extent_item *item; |
2691 | struct btrfs_key key; | |
2692 | int found_type; | |
2693 | int i; | |
b89311ef | 2694 | int ret = 0; |
8c2a1a30 | 2695 | |
2ff7e61e | 2696 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2697 | return 0; |
2698 | ||
2699 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2700 | btrfs_item_key_to_cpu(eb, &key, i); | |
2701 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2702 | continue; | |
2703 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2704 | found_type = btrfs_file_extent_type(eb, item); | |
2705 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2706 | continue; | |
2707 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2708 | continue; | |
2709 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2710 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2711 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2712 | if (ret) | |
2713 | break; | |
8c2a1a30 JB |
2714 | } |
2715 | ||
b89311ef | 2716 | return ret; |
8c2a1a30 JB |
2717 | } |
2718 | ||
9cfa3e34 FM |
2719 | static void |
2720 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
2721 | { | |
2722 | atomic_inc(&bg->reservations); | |
2723 | } | |
2724 | ||
8b74c03e | 2725 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 2726 | { |
11833d66 YZ |
2727 | struct btrfs_caching_control *next; |
2728 | struct btrfs_caching_control *caching_ctl; | |
2729 | struct btrfs_block_group_cache *cache; | |
e8569813 | 2730 | |
9e351cc8 | 2731 | down_write(&fs_info->commit_root_sem); |
25179201 | 2732 | |
11833d66 YZ |
2733 | list_for_each_entry_safe(caching_ctl, next, |
2734 | &fs_info->caching_block_groups, list) { | |
2735 | cache = caching_ctl->block_group; | |
676f1f75 | 2736 | if (btrfs_block_group_cache_done(cache)) { |
11833d66 YZ |
2737 | cache->last_byte_to_unpin = (u64)-1; |
2738 | list_del_init(&caching_ctl->list); | |
e3cb339f | 2739 | btrfs_put_caching_control(caching_ctl); |
e8569813 | 2740 | } else { |
11833d66 | 2741 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 2742 | } |
e8569813 | 2743 | } |
11833d66 YZ |
2744 | |
2745 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
2746 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
2747 | else | |
2748 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
2749 | ||
9e351cc8 | 2750 | up_write(&fs_info->commit_root_sem); |
8929ecfa | 2751 | |
67f9c220 | 2752 | btrfs_update_global_block_rsv(fs_info); |
e8569813 ZY |
2753 | } |
2754 | ||
c759c4e1 JB |
2755 | /* |
2756 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2757 | * what it should be based on the mount options. | |
2758 | */ | |
2759 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2760 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2761 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2762 | { |
2763 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2764 | |
2765 | *empty_cluster = 0; | |
2766 | if (btrfs_mixed_space_info(space_info)) | |
2767 | return ret; | |
2768 | ||
c759c4e1 | 2769 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2770 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2771 | if (btrfs_test_opt(fs_info, SSD)) |
2772 | *empty_cluster = SZ_2M; | |
2773 | else | |
ee22184b | 2774 | *empty_cluster = SZ_64K; |
583b7231 HK |
2775 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2776 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2777 | *empty_cluster = SZ_2M; | |
0b246afa | 2778 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2779 | } |
2780 | ||
2781 | return ret; | |
2782 | } | |
2783 | ||
2ff7e61e JM |
2784 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2785 | u64 start, u64 end, | |
678886bd | 2786 | const bool return_free_space) |
ccd467d6 | 2787 | { |
11833d66 | 2788 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
2789 | struct btrfs_space_info *space_info; |
2790 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2791 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2792 | u64 len; |
c759c4e1 JB |
2793 | u64 total_unpinned = 0; |
2794 | u64 empty_cluster = 0; | |
7b398f8e | 2795 | bool readonly; |
ccd467d6 | 2796 | |
11833d66 | 2797 | while (start <= end) { |
7b398f8e | 2798 | readonly = false; |
11833d66 YZ |
2799 | if (!cache || |
2800 | start >= cache->key.objectid + cache->key.offset) { | |
2801 | if (cache) | |
2802 | btrfs_put_block_group(cache); | |
c759c4e1 | 2803 | total_unpinned = 0; |
11833d66 | 2804 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2805 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2806 | |
2ff7e61e | 2807 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2808 | cache->space_info, |
2809 | &empty_cluster); | |
2810 | empty_cluster <<= 1; | |
11833d66 YZ |
2811 | } |
2812 | ||
2813 | len = cache->key.objectid + cache->key.offset - start; | |
2814 | len = min(len, end + 1 - start); | |
2815 | ||
2816 | if (start < cache->last_byte_to_unpin) { | |
2817 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
2818 | if (return_free_space) |
2819 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
2820 | } |
2821 | ||
f0486c68 | 2822 | start += len; |
c759c4e1 | 2823 | total_unpinned += len; |
7b398f8e | 2824 | space_info = cache->space_info; |
f0486c68 | 2825 | |
c759c4e1 JB |
2826 | /* |
2827 | * If this space cluster has been marked as fragmented and we've | |
2828 | * unpinned enough in this block group to potentially allow a | |
2829 | * cluster to be created inside of it go ahead and clear the | |
2830 | * fragmented check. | |
2831 | */ | |
2832 | if (cluster && cluster->fragmented && | |
2833 | total_unpinned > empty_cluster) { | |
2834 | spin_lock(&cluster->lock); | |
2835 | cluster->fragmented = 0; | |
2836 | spin_unlock(&cluster->lock); | |
2837 | } | |
2838 | ||
7b398f8e | 2839 | spin_lock(&space_info->lock); |
11833d66 YZ |
2840 | spin_lock(&cache->lock); |
2841 | cache->pinned -= len; | |
bb96c4e5 | 2842 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2843 | space_info->max_extent_size = 0; |
dec59fa3 EL |
2844 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2845 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
2846 | if (cache->ro) { |
2847 | space_info->bytes_readonly += len; | |
2848 | readonly = true; | |
2849 | } | |
11833d66 | 2850 | spin_unlock(&cache->lock); |
957780eb JB |
2851 | if (!readonly && return_free_space && |
2852 | global_rsv->space_info == space_info) { | |
2853 | u64 to_add = len; | |
92ac58ec | 2854 | |
7b398f8e JB |
2855 | spin_lock(&global_rsv->lock); |
2856 | if (!global_rsv->full) { | |
957780eb JB |
2857 | to_add = min(len, global_rsv->size - |
2858 | global_rsv->reserved); | |
2859 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2860 | btrfs_space_info_update_bytes_may_use(fs_info, |
2861 | space_info, to_add); | |
7b398f8e JB |
2862 | if (global_rsv->reserved >= global_rsv->size) |
2863 | global_rsv->full = 1; | |
957780eb | 2864 | len -= to_add; |
7b398f8e JB |
2865 | } |
2866 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
2867 | /* Add to any tickets we may have */ |
2868 | if (len) | |
d44b72aa JB |
2869 | btrfs_space_info_add_new_bytes(fs_info, |
2870 | space_info, len); | |
7b398f8e JB |
2871 | } |
2872 | spin_unlock(&space_info->lock); | |
ccd467d6 | 2873 | } |
11833d66 YZ |
2874 | |
2875 | if (cache) | |
2876 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2877 | return 0; |
2878 | } | |
2879 | ||
5ead2dd0 | 2880 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2881 | { |
5ead2dd0 | 2882 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
2883 | struct btrfs_block_group_cache *block_group, *tmp; |
2884 | struct list_head *deleted_bgs; | |
11833d66 | 2885 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2886 | u64 start; |
2887 | u64 end; | |
a28ec197 | 2888 | int ret; |
a28ec197 | 2889 | |
11833d66 YZ |
2890 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
2891 | unpin = &fs_info->freed_extents[1]; | |
2892 | else | |
2893 | unpin = &fs_info->freed_extents[0]; | |
2894 | ||
e33e17ee | 2895 | while (!trans->aborted) { |
0e6ec385 FM |
2896 | struct extent_state *cached_state = NULL; |
2897 | ||
d4b450cd | 2898 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2899 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2900 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2901 | if (ret) { |
2902 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2903 | break; |
d4b450cd | 2904 | } |
1f3c79a2 | 2905 | |
0b246afa | 2906 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 2907 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2908 | end + 1 - start, NULL); |
1f3c79a2 | 2909 | |
0e6ec385 | 2910 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2911 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2912 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2913 | free_extent_state(cached_state); |
b9473439 | 2914 | cond_resched(); |
a28ec197 | 2915 | } |
817d52f8 | 2916 | |
e33e17ee JM |
2917 | /* |
2918 | * Transaction is finished. We don't need the lock anymore. We | |
2919 | * do need to clean up the block groups in case of a transaction | |
2920 | * abort. | |
2921 | */ | |
2922 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2923 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2924 | u64 trimmed = 0; | |
2925 | ||
2926 | ret = -EROFS; | |
2927 | if (!trans->aborted) | |
2ff7e61e | 2928 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
2929 | block_group->key.objectid, |
2930 | block_group->key.offset, | |
2931 | &trimmed); | |
2932 | ||
2933 | list_del_init(&block_group->bg_list); | |
2934 | btrfs_put_block_group_trimming(block_group); | |
2935 | btrfs_put_block_group(block_group); | |
2936 | ||
2937 | if (ret) { | |
2938 | const char *errstr = btrfs_decode_error(ret); | |
2939 | btrfs_warn(fs_info, | |
913e1535 | 2940 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2941 | ret, errstr); |
2942 | } | |
2943 | } | |
2944 | ||
e20d96d6 CM |
2945 | return 0; |
2946 | } | |
2947 | ||
5d4f98a2 | 2948 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2949 | struct btrfs_delayed_ref_node *node, u64 parent, |
2950 | u64 root_objectid, u64 owner_objectid, | |
2951 | u64 owner_offset, int refs_to_drop, | |
2952 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2953 | { |
e72cb923 | 2954 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2955 | struct btrfs_key key; |
5d4f98a2 | 2956 | struct btrfs_path *path; |
1261ec42 | 2957 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2958 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2959 | struct btrfs_extent_item *ei; |
2960 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2961 | int ret; |
5d4f98a2 | 2962 | int is_data; |
952fccac CM |
2963 | int extent_slot = 0; |
2964 | int found_extent = 0; | |
2965 | int num_to_del = 1; | |
5d4f98a2 YZ |
2966 | u32 item_size; |
2967 | u64 refs; | |
c682f9b3 QW |
2968 | u64 bytenr = node->bytenr; |
2969 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2970 | int last_ref = 0; |
0b246afa | 2971 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 2972 | |
5caf2a00 | 2973 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2974 | if (!path) |
2975 | return -ENOMEM; | |
5f26f772 | 2976 | |
e4058b54 | 2977 | path->reada = READA_FORWARD; |
b9473439 | 2978 | path->leave_spinning = 1; |
5d4f98a2 YZ |
2979 | |
2980 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
2981 | BUG_ON(!is_data && refs_to_drop != 1); | |
2982 | ||
3173a18f | 2983 | if (is_data) |
897ca819 | 2984 | skinny_metadata = false; |
3173a18f | 2985 | |
fbe4801b NB |
2986 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
2987 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 2988 | owner_offset); |
7bb86316 | 2989 | if (ret == 0) { |
952fccac | 2990 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
2991 | while (extent_slot >= 0) { |
2992 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 2993 | extent_slot); |
5d4f98a2 | 2994 | if (key.objectid != bytenr) |
952fccac | 2995 | break; |
5d4f98a2 YZ |
2996 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
2997 | key.offset == num_bytes) { | |
952fccac CM |
2998 | found_extent = 1; |
2999 | break; | |
3000 | } | |
3173a18f JB |
3001 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3002 | key.offset == owner_objectid) { | |
3003 | found_extent = 1; | |
3004 | break; | |
3005 | } | |
952fccac CM |
3006 | if (path->slots[0] - extent_slot > 5) |
3007 | break; | |
5d4f98a2 | 3008 | extent_slot--; |
952fccac | 3009 | } |
a79865c6 | 3010 | |
31840ae1 | 3011 | if (!found_extent) { |
5d4f98a2 | 3012 | BUG_ON(iref); |
87cc7a8a | 3013 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3014 | refs_to_drop, |
fcebe456 | 3015 | is_data, &last_ref); |
005d6427 | 3016 | if (ret) { |
66642832 | 3017 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3018 | goto out; |
3019 | } | |
b3b4aa74 | 3020 | btrfs_release_path(path); |
b9473439 | 3021 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3022 | |
3023 | key.objectid = bytenr; | |
3024 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3025 | key.offset = num_bytes; | |
3026 | ||
3173a18f JB |
3027 | if (!is_data && skinny_metadata) { |
3028 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3029 | key.offset = owner_objectid; | |
3030 | } | |
3031 | ||
31840ae1 ZY |
3032 | ret = btrfs_search_slot(trans, extent_root, |
3033 | &key, path, -1, 1); | |
3173a18f JB |
3034 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3035 | /* | |
3036 | * Couldn't find our skinny metadata item, | |
3037 | * see if we have ye olde extent item. | |
3038 | */ | |
3039 | path->slots[0]--; | |
3040 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3041 | path->slots[0]); | |
3042 | if (key.objectid == bytenr && | |
3043 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3044 | key.offset == num_bytes) | |
3045 | ret = 0; | |
3046 | } | |
3047 | ||
3048 | if (ret > 0 && skinny_metadata) { | |
3049 | skinny_metadata = false; | |
9ce49a0b | 3050 | key.objectid = bytenr; |
3173a18f JB |
3051 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3052 | key.offset = num_bytes; | |
3053 | btrfs_release_path(path); | |
3054 | ret = btrfs_search_slot(trans, extent_root, | |
3055 | &key, path, -1, 1); | |
3056 | } | |
3057 | ||
f3465ca4 | 3058 | if (ret) { |
5d163e0e JM |
3059 | btrfs_err(info, |
3060 | "umm, got %d back from search, was looking for %llu", | |
3061 | ret, bytenr); | |
b783e62d | 3062 | if (ret > 0) |
a4f78750 | 3063 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3064 | } |
005d6427 | 3065 | if (ret < 0) { |
66642832 | 3066 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3067 | goto out; |
3068 | } | |
31840ae1 ZY |
3069 | extent_slot = path->slots[0]; |
3070 | } | |
fae7f21c | 3071 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3072 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3073 | btrfs_err(info, |
3074 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3075 | bytenr, parent, root_objectid, owner_objectid, |
3076 | owner_offset); | |
66642832 | 3077 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3078 | goto out; |
79787eaa | 3079 | } else { |
66642832 | 3080 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3081 | goto out; |
7bb86316 | 3082 | } |
5f39d397 CM |
3083 | |
3084 | leaf = path->nodes[0]; | |
5d4f98a2 | 3085 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3086 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3087 | ret = -EINVAL; |
3088 | btrfs_print_v0_err(info); | |
3089 | btrfs_abort_transaction(trans, ret); | |
3090 | goto out; | |
3091 | } | |
952fccac | 3092 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3093 | struct btrfs_extent_item); |
3173a18f JB |
3094 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3095 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
3096 | struct btrfs_tree_block_info *bi; |
3097 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
3098 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3099 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3100 | } | |
56bec294 | 3101 | |
5d4f98a2 | 3102 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3103 | if (refs < refs_to_drop) { |
5d163e0e JM |
3104 | btrfs_err(info, |
3105 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
3106 | refs_to_drop, refs, bytenr); | |
32b02538 | 3107 | ret = -EINVAL; |
66642832 | 3108 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
3109 | goto out; |
3110 | } | |
56bec294 | 3111 | refs -= refs_to_drop; |
5f39d397 | 3112 | |
5d4f98a2 YZ |
3113 | if (refs > 0) { |
3114 | if (extent_op) | |
3115 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3116 | /* | |
3117 | * In the case of inline back ref, reference count will | |
3118 | * be updated by remove_extent_backref | |
952fccac | 3119 | */ |
5d4f98a2 YZ |
3120 | if (iref) { |
3121 | BUG_ON(!found_extent); | |
3122 | } else { | |
3123 | btrfs_set_extent_refs(leaf, ei, refs); | |
3124 | btrfs_mark_buffer_dirty(leaf); | |
3125 | } | |
3126 | if (found_extent) { | |
87cc7a8a NB |
3127 | ret = remove_extent_backref(trans, path, iref, |
3128 | refs_to_drop, is_data, | |
3129 | &last_ref); | |
005d6427 | 3130 | if (ret) { |
66642832 | 3131 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3132 | goto out; |
3133 | } | |
952fccac | 3134 | } |
5d4f98a2 | 3135 | } else { |
5d4f98a2 YZ |
3136 | if (found_extent) { |
3137 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 3138 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
3139 | if (iref) { |
3140 | BUG_ON(path->slots[0] != extent_slot); | |
3141 | } else { | |
3142 | BUG_ON(path->slots[0] != extent_slot + 1); | |
3143 | path->slots[0] = extent_slot; | |
3144 | num_to_del = 2; | |
3145 | } | |
78fae27e | 3146 | } |
b9473439 | 3147 | |
fcebe456 | 3148 | last_ref = 1; |
952fccac CM |
3149 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3150 | num_to_del); | |
005d6427 | 3151 | if (ret) { |
66642832 | 3152 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3153 | goto out; |
3154 | } | |
b3b4aa74 | 3155 | btrfs_release_path(path); |
21af804c | 3156 | |
5d4f98a2 | 3157 | if (is_data) { |
5b4aacef | 3158 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 3159 | if (ret) { |
66642832 | 3160 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3161 | goto out; |
3162 | } | |
459931ec CM |
3163 | } |
3164 | ||
e7355e50 | 3165 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3166 | if (ret) { |
66642832 | 3167 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3168 | goto out; |
3169 | } | |
3170 | ||
ade4b516 | 3171 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3172 | if (ret) { |
66642832 | 3173 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3174 | goto out; |
3175 | } | |
a28ec197 | 3176 | } |
fcebe456 JB |
3177 | btrfs_release_path(path); |
3178 | ||
79787eaa | 3179 | out: |
5caf2a00 | 3180 | btrfs_free_path(path); |
a28ec197 CM |
3181 | return ret; |
3182 | } | |
3183 | ||
1887be66 | 3184 | /* |
f0486c68 | 3185 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3186 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3187 | * a given extent, and if there are no other delayed refs to be processed, it | |
3188 | * removes it from the tree. | |
3189 | */ | |
3190 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3191 | u64 bytenr) |
1887be66 CM |
3192 | { |
3193 | struct btrfs_delayed_ref_head *head; | |
3194 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3195 | int ret = 0; |
1887be66 CM |
3196 | |
3197 | delayed_refs = &trans->transaction->delayed_refs; | |
3198 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3199 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3200 | if (!head) |
cf93da7b | 3201 | goto out_delayed_unlock; |
1887be66 | 3202 | |
d7df2c79 | 3203 | spin_lock(&head->lock); |
e3d03965 | 3204 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3205 | goto out; |
3206 | ||
bedc6617 JB |
3207 | if (cleanup_extent_op(head) != NULL) |
3208 | goto out; | |
5d4f98a2 | 3209 | |
1887be66 CM |
3210 | /* |
3211 | * waiting for the lock here would deadlock. If someone else has it | |
3212 | * locked they are already in the process of dropping it anyway | |
3213 | */ | |
3214 | if (!mutex_trylock(&head->mutex)) | |
3215 | goto out; | |
3216 | ||
d7baffda | 3217 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3218 | head->processing = 0; |
d7baffda | 3219 | |
d7df2c79 | 3220 | spin_unlock(&head->lock); |
1887be66 CM |
3221 | spin_unlock(&delayed_refs->lock); |
3222 | ||
f0486c68 YZ |
3223 | BUG_ON(head->extent_op); |
3224 | if (head->must_insert_reserved) | |
3225 | ret = 1; | |
3226 | ||
31890da0 | 3227 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3228 | mutex_unlock(&head->mutex); |
d278850e | 3229 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3230 | return ret; |
1887be66 | 3231 | out: |
d7df2c79 | 3232 | spin_unlock(&head->lock); |
cf93da7b CM |
3233 | |
3234 | out_delayed_unlock: | |
1887be66 CM |
3235 | spin_unlock(&delayed_refs->lock); |
3236 | return 0; | |
3237 | } | |
3238 | ||
f0486c68 YZ |
3239 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3240 | struct btrfs_root *root, | |
3241 | struct extent_buffer *buf, | |
5581a51a | 3242 | u64 parent, int last_ref) |
f0486c68 | 3243 | { |
0b246afa | 3244 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3245 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 3246 | int pin = 1; |
f0486c68 YZ |
3247 | int ret; |
3248 | ||
ed4f255b QW |
3249 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3250 | buf->start, buf->len, parent); | |
3251 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3252 | root->root_key.objectid); | |
3253 | ||
f0486c68 | 3254 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
3255 | int old_ref_mod, new_ref_mod; |
3256 | ||
8a5040f7 | 3257 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 3258 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 3259 | &old_ref_mod, &new_ref_mod); |
79787eaa | 3260 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 3261 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
3262 | } |
3263 | ||
0a16c7d7 | 3264 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
3265 | struct btrfs_block_group_cache *cache; |
3266 | ||
f0486c68 | 3267 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3268 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 3269 | if (!ret) |
37be25bc | 3270 | goto out; |
f0486c68 YZ |
3271 | } |
3272 | ||
4da8b76d | 3273 | pin = 0; |
0b246afa | 3274 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3275 | |
f0486c68 | 3276 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
fdf08605 | 3277 | pin_down_extent(cache, buf->start, buf->len, 1); |
6219872d | 3278 | btrfs_put_block_group(cache); |
37be25bc | 3279 | goto out; |
f0486c68 YZ |
3280 | } |
3281 | ||
3282 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3283 | ||
3284 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3285 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3286 | btrfs_put_block_group(cache); |
71ff6437 | 3287 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3288 | } |
3289 | out: | |
b150a4f1 | 3290 | if (pin) |
78192442 | 3291 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 3292 | |
0a16c7d7 OS |
3293 | if (last_ref) { |
3294 | /* | |
3295 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3296 | * matter anymore. | |
3297 | */ | |
3298 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3299 | } | |
f0486c68 YZ |
3300 | } |
3301 | ||
79787eaa | 3302 | /* Can return -ENOMEM */ |
ffd4bb2a | 3303 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3304 | { |
ffd4bb2a | 3305 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 3306 | int old_ref_mod, new_ref_mod; |
925baedd CM |
3307 | int ret; |
3308 | ||
f5ee5c9a | 3309 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3310 | return 0; |
fccb84c9 | 3311 | |
56bec294 CM |
3312 | /* |
3313 | * tree log blocks never actually go into the extent allocation | |
3314 | * tree, just update pinning info and exit early. | |
56bec294 | 3315 | */ |
ffd4bb2a QW |
3316 | if ((ref->type == BTRFS_REF_METADATA && |
3317 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3318 | (ref->type == BTRFS_REF_DATA && | |
3319 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3320 | /* unlocks the pinned mutex */ |
ffd4bb2a | 3321 | btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1); |
d7eae340 | 3322 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 3323 | ret = 0; |
ffd4bb2a QW |
3324 | } else if (ref->type == BTRFS_REF_METADATA) { |
3325 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 3326 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 3327 | } else { |
ffd4bb2a | 3328 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 3329 | &old_ref_mod, &new_ref_mod); |
56bec294 | 3330 | } |
d7eae340 | 3331 | |
ffd4bb2a QW |
3332 | if (!((ref->type == BTRFS_REF_METADATA && |
3333 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3334 | (ref->type == BTRFS_REF_DATA && | |
3335 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3336 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3337 | |
ddf30cf0 | 3338 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 3339 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 3340 | |
925baedd CM |
3341 | return ret; |
3342 | } | |
3343 | ||
817d52f8 | 3344 | enum btrfs_loop_type { |
f262fa8d DS |
3345 | LOOP_CACHING_NOWAIT, |
3346 | LOOP_CACHING_WAIT, | |
3347 | LOOP_ALLOC_CHUNK, | |
3348 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3349 | }; |
3350 | ||
e570fd27 MX |
3351 | static inline void |
3352 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
3353 | int delalloc) | |
3354 | { | |
3355 | if (delalloc) | |
3356 | down_read(&cache->data_rwsem); | |
3357 | } | |
3358 | ||
3359 | static inline void | |
3360 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
3361 | int delalloc) | |
3362 | { | |
3363 | btrfs_get_block_group(cache); | |
3364 | if (delalloc) | |
3365 | down_read(&cache->data_rwsem); | |
3366 | } | |
3367 | ||
3368 | static struct btrfs_block_group_cache * | |
3369 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
3370 | struct btrfs_free_cluster *cluster, | |
3371 | int delalloc) | |
3372 | { | |
89771cc9 | 3373 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 3374 | |
e570fd27 | 3375 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3376 | while (1) { |
3377 | used_bg = cluster->block_group; | |
3378 | if (!used_bg) | |
3379 | return NULL; | |
3380 | ||
3381 | if (used_bg == block_group) | |
e570fd27 MX |
3382 | return used_bg; |
3383 | ||
6719afdc | 3384 | btrfs_get_block_group(used_bg); |
e570fd27 | 3385 | |
6719afdc GU |
3386 | if (!delalloc) |
3387 | return used_bg; | |
e570fd27 | 3388 | |
6719afdc GU |
3389 | if (down_read_trylock(&used_bg->data_rwsem)) |
3390 | return used_bg; | |
e570fd27 | 3391 | |
6719afdc | 3392 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3393 | |
e321f8a8 LB |
3394 | /* We should only have one-level nested. */ |
3395 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3396 | |
6719afdc GU |
3397 | spin_lock(&cluster->refill_lock); |
3398 | if (used_bg == cluster->block_group) | |
3399 | return used_bg; | |
e570fd27 | 3400 | |
6719afdc GU |
3401 | up_read(&used_bg->data_rwsem); |
3402 | btrfs_put_block_group(used_bg); | |
3403 | } | |
e570fd27 MX |
3404 | } |
3405 | ||
3406 | static inline void | |
3407 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
3408 | int delalloc) | |
3409 | { | |
3410 | if (delalloc) | |
3411 | up_read(&cache->data_rwsem); | |
3412 | btrfs_put_block_group(cache); | |
3413 | } | |
3414 | ||
b4bd745d QW |
3415 | /* |
3416 | * Structure used internally for find_free_extent() function. Wraps needed | |
3417 | * parameters. | |
3418 | */ | |
3419 | struct find_free_extent_ctl { | |
3420 | /* Basic allocation info */ | |
3421 | u64 ram_bytes; | |
3422 | u64 num_bytes; | |
3423 | u64 empty_size; | |
3424 | u64 flags; | |
3425 | int delalloc; | |
3426 | ||
3427 | /* Where to start the search inside the bg */ | |
3428 | u64 search_start; | |
3429 | ||
3430 | /* For clustered allocation */ | |
3431 | u64 empty_cluster; | |
3432 | ||
3433 | bool have_caching_bg; | |
3434 | bool orig_have_caching_bg; | |
3435 | ||
3436 | /* RAID index, converted from flags */ | |
3437 | int index; | |
3438 | ||
e72d79d6 QW |
3439 | /* |
3440 | * Current loop number, check find_free_extent_update_loop() for details | |
3441 | */ | |
b4bd745d QW |
3442 | int loop; |
3443 | ||
3444 | /* | |
3445 | * Whether we're refilling a cluster, if true we need to re-search | |
3446 | * current block group but don't try to refill the cluster again. | |
3447 | */ | |
3448 | bool retry_clustered; | |
3449 | ||
3450 | /* | |
3451 | * Whether we're updating free space cache, if true we need to re-search | |
3452 | * current block group but don't try updating free space cache again. | |
3453 | */ | |
3454 | bool retry_unclustered; | |
3455 | ||
3456 | /* If current block group is cached */ | |
3457 | int cached; | |
3458 | ||
3459 | /* Max contiguous hole found */ | |
3460 | u64 max_extent_size; | |
3461 | ||
3462 | /* Total free space from free space cache, not always contiguous */ | |
3463 | u64 total_free_space; | |
3464 | ||
3465 | /* Found result */ | |
3466 | u64 found_offset; | |
3467 | }; | |
3468 | ||
d06e3bb6 QW |
3469 | |
3470 | /* | |
3471 | * Helper function for find_free_extent(). | |
3472 | * | |
3473 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3474 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3475 | * Return >0 to inform caller that we find nothing | |
3476 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3477 | */ | |
3478 | static int find_free_extent_clustered(struct btrfs_block_group_cache *bg, | |
3479 | struct btrfs_free_cluster *last_ptr, | |
3480 | struct find_free_extent_ctl *ffe_ctl, | |
3481 | struct btrfs_block_group_cache **cluster_bg_ret) | |
3482 | { | |
d06e3bb6 QW |
3483 | struct btrfs_block_group_cache *cluster_bg; |
3484 | u64 aligned_cluster; | |
3485 | u64 offset; | |
3486 | int ret; | |
3487 | ||
3488 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3489 | if (!cluster_bg) | |
3490 | goto refill_cluster; | |
3491 | if (cluster_bg != bg && (cluster_bg->ro || | |
3492 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3493 | goto release_cluster; | |
3494 | ||
3495 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
3496 | ffe_ctl->num_bytes, cluster_bg->key.objectid, | |
3497 | &ffe_ctl->max_extent_size); | |
3498 | if (offset) { | |
3499 | /* We have a block, we're done */ | |
3500 | spin_unlock(&last_ptr->refill_lock); | |
3501 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3502 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3503 | *cluster_bg_ret = cluster_bg; | |
3504 | ffe_ctl->found_offset = offset; | |
3505 | return 0; | |
3506 | } | |
3507 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3508 | ||
3509 | release_cluster: | |
3510 | /* | |
3511 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3512 | * lets just skip it and let the allocator find whatever block it can | |
3513 | * find. If we reach this point, we will have tried the cluster | |
3514 | * allocator plenty of times and not have found anything, so we are | |
3515 | * likely way too fragmented for the clustering stuff to find anything. | |
3516 | * | |
3517 | * However, if the cluster is taken from the current block group, | |
3518 | * release the cluster first, so that we stand a better chance of | |
3519 | * succeeding in the unclustered allocation. | |
3520 | */ | |
3521 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3522 | spin_unlock(&last_ptr->refill_lock); | |
3523 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3524 | return -ENOENT; | |
3525 | } | |
3526 | ||
3527 | /* This cluster didn't work out, free it and start over */ | |
3528 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3529 | ||
3530 | if (cluster_bg != bg) | |
3531 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3532 | ||
3533 | refill_cluster: | |
3534 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3535 | spin_unlock(&last_ptr->refill_lock); | |
3536 | return -ENOENT; | |
3537 | } | |
3538 | ||
3539 | aligned_cluster = max_t(u64, | |
3540 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3541 | bg->full_stripe_len); | |
2ceeae2e DS |
3542 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3543 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3544 | if (ret == 0) { |
3545 | /* Now pull our allocation out of this cluster */ | |
3546 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3547 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3548 | &ffe_ctl->max_extent_size); | |
3549 | if (offset) { | |
3550 | /* We found one, proceed */ | |
3551 | spin_unlock(&last_ptr->refill_lock); | |
3552 | trace_btrfs_reserve_extent_cluster(bg, | |
3553 | ffe_ctl->search_start, | |
3554 | ffe_ctl->num_bytes); | |
3555 | ffe_ctl->found_offset = offset; | |
3556 | return 0; | |
3557 | } | |
3558 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3559 | !ffe_ctl->retry_clustered) { | |
3560 | spin_unlock(&last_ptr->refill_lock); | |
3561 | ||
3562 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3563 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3564 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3565 | return -EAGAIN; | |
3566 | } | |
3567 | /* | |
3568 | * At this point we either didn't find a cluster or we weren't able to | |
3569 | * allocate a block from our cluster. Free the cluster we've been | |
3570 | * trying to use, and go to the next block group. | |
3571 | */ | |
3572 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3573 | spin_unlock(&last_ptr->refill_lock); | |
3574 | return 1; | |
3575 | } | |
3576 | ||
e1a41848 QW |
3577 | /* |
3578 | * Return >0 to inform caller that we find nothing | |
3579 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3580 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3581 | */ | |
3582 | static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg, | |
3583 | struct btrfs_free_cluster *last_ptr, | |
3584 | struct find_free_extent_ctl *ffe_ctl) | |
3585 | { | |
3586 | u64 offset; | |
3587 | ||
3588 | /* | |
3589 | * We are doing an unclustered allocation, set the fragmented flag so | |
3590 | * we don't bother trying to setup a cluster again until we get more | |
3591 | * space. | |
3592 | */ | |
3593 | if (unlikely(last_ptr)) { | |
3594 | spin_lock(&last_ptr->lock); | |
3595 | last_ptr->fragmented = 1; | |
3596 | spin_unlock(&last_ptr->lock); | |
3597 | } | |
3598 | if (ffe_ctl->cached) { | |
3599 | struct btrfs_free_space_ctl *free_space_ctl; | |
3600 | ||
3601 | free_space_ctl = bg->free_space_ctl; | |
3602 | spin_lock(&free_space_ctl->tree_lock); | |
3603 | if (free_space_ctl->free_space < | |
3604 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3605 | ffe_ctl->empty_size) { | |
3606 | ffe_ctl->total_free_space = max_t(u64, | |
3607 | ffe_ctl->total_free_space, | |
3608 | free_space_ctl->free_space); | |
3609 | spin_unlock(&free_space_ctl->tree_lock); | |
3610 | return 1; | |
3611 | } | |
3612 | spin_unlock(&free_space_ctl->tree_lock); | |
3613 | } | |
3614 | ||
3615 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3616 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3617 | &ffe_ctl->max_extent_size); | |
3618 | ||
3619 | /* | |
3620 | * If we didn't find a chunk, and we haven't failed on this block group | |
3621 | * before, and this block group is in the middle of caching and we are | |
3622 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3623 | * set @retry_unclustered to true. | |
3624 | * | |
3625 | * If @retry_unclustered is true then we've already waited on this | |
3626 | * block group once and should move on to the next block group. | |
3627 | */ | |
3628 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3629 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3630 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3631 | ffe_ctl->empty_size); | |
e1a41848 QW |
3632 | ffe_ctl->retry_unclustered = true; |
3633 | return -EAGAIN; | |
3634 | } else if (!offset) { | |
3635 | return 1; | |
3636 | } | |
3637 | ffe_ctl->found_offset = offset; | |
3638 | return 0; | |
3639 | } | |
3640 | ||
e72d79d6 QW |
3641 | /* |
3642 | * Return >0 means caller needs to re-search for free extent | |
3643 | * Return 0 means we have the needed free extent. | |
3644 | * Return <0 means we failed to locate any free extent. | |
3645 | */ | |
3646 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
3647 | struct btrfs_free_cluster *last_ptr, | |
3648 | struct btrfs_key *ins, | |
3649 | struct find_free_extent_ctl *ffe_ctl, | |
3650 | int full_search, bool use_cluster) | |
3651 | { | |
3652 | struct btrfs_root *root = fs_info->extent_root; | |
3653 | int ret; | |
3654 | ||
3655 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3656 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3657 | ffe_ctl->orig_have_caching_bg = true; | |
3658 | ||
3659 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3660 | ffe_ctl->have_caching_bg) | |
3661 | return 1; | |
3662 | ||
3663 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3664 | return 1; | |
3665 | ||
3666 | if (ins->objectid) { | |
3667 | if (!use_cluster && last_ptr) { | |
3668 | spin_lock(&last_ptr->lock); | |
3669 | last_ptr->window_start = ins->objectid; | |
3670 | spin_unlock(&last_ptr->lock); | |
3671 | } | |
3672 | return 0; | |
3673 | } | |
3674 | ||
3675 | /* | |
3676 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3677 | * caching kthreads as we move along | |
3678 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3679 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3680 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3681 | * again | |
3682 | */ | |
3683 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3684 | ffe_ctl->index = 0; | |
3685 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3686 | /* | |
3687 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3688 | * don't have any uncached bgs and we've already done a | |
3689 | * full search through. | |
3690 | */ | |
3691 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3692 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3693 | else | |
3694 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3695 | } else { | |
3696 | ffe_ctl->loop++; | |
3697 | } | |
3698 | ||
3699 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3700 | struct btrfs_trans_handle *trans; | |
3701 | int exist = 0; | |
3702 | ||
3703 | trans = current->journal_info; | |
3704 | if (trans) | |
3705 | exist = 1; | |
3706 | else | |
3707 | trans = btrfs_join_transaction(root); | |
3708 | ||
3709 | if (IS_ERR(trans)) { | |
3710 | ret = PTR_ERR(trans); | |
3711 | return ret; | |
3712 | } | |
3713 | ||
fc471cb0 JB |
3714 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3715 | CHUNK_ALLOC_FORCE); | |
e72d79d6 QW |
3716 | |
3717 | /* | |
3718 | * If we can't allocate a new chunk we've already looped | |
3719 | * through at least once, move on to the NO_EMPTY_SIZE | |
3720 | * case. | |
3721 | */ | |
3722 | if (ret == -ENOSPC) | |
3723 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3724 | ||
3725 | /* Do not bail out on ENOSPC since we can do more. */ | |
3726 | if (ret < 0 && ret != -ENOSPC) | |
3727 | btrfs_abort_transaction(trans, ret); | |
3728 | else | |
3729 | ret = 0; | |
3730 | if (!exist) | |
3731 | btrfs_end_transaction(trans); | |
3732 | if (ret) | |
3733 | return ret; | |
3734 | } | |
3735 | ||
3736 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
3737 | /* | |
3738 | * Don't loop again if we already have no empty_size and | |
3739 | * no empty_cluster. | |
3740 | */ | |
3741 | if (ffe_ctl->empty_size == 0 && | |
3742 | ffe_ctl->empty_cluster == 0) | |
3743 | return -ENOSPC; | |
3744 | ffe_ctl->empty_size = 0; | |
3745 | ffe_ctl->empty_cluster = 0; | |
3746 | } | |
3747 | return 1; | |
3748 | } | |
3749 | return -ENOSPC; | |
3750 | } | |
3751 | ||
fec577fb CM |
3752 | /* |
3753 | * walks the btree of allocated extents and find a hole of a given size. | |
3754 | * The key ins is changed to record the hole: | |
a4820398 | 3755 | * ins->objectid == start position |
62e2749e | 3756 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 3757 | * ins->offset == the size of the hole. |
fec577fb | 3758 | * Any available blocks before search_start are skipped. |
a4820398 MX |
3759 | * |
3760 | * If there is no suitable free space, we will record the max size of | |
3761 | * the free space extent currently. | |
e72d79d6 QW |
3762 | * |
3763 | * The overall logic and call chain: | |
3764 | * | |
3765 | * find_free_extent() | |
3766 | * |- Iterate through all block groups | |
3767 | * | |- Get a valid block group | |
3768 | * | |- Try to do clustered allocation in that block group | |
3769 | * | |- Try to do unclustered allocation in that block group | |
3770 | * | |- Check if the result is valid | |
3771 | * | | |- If valid, then exit | |
3772 | * | |- Jump to next block group | |
3773 | * | | |
3774 | * |- Push harder to find free extents | |
3775 | * |- If not found, re-iterate all block groups | |
fec577fb | 3776 | */ |
87bde3cd | 3777 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
3778 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
3779 | u64 hint_byte, struct btrfs_key *ins, | |
3780 | u64 flags, int delalloc) | |
fec577fb | 3781 | { |
80eb234a | 3782 | int ret = 0; |
fa9c0d79 | 3783 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 3784 | struct btrfs_block_group_cache *block_group = NULL; |
b4bd745d | 3785 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 3786 | struct btrfs_space_info *space_info; |
67377734 | 3787 | bool use_cluster = true; |
a5e681d9 | 3788 | bool full_search = false; |
fec577fb | 3789 | |
0b246afa | 3790 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d QW |
3791 | |
3792 | ffe_ctl.ram_bytes = ram_bytes; | |
3793 | ffe_ctl.num_bytes = num_bytes; | |
3794 | ffe_ctl.empty_size = empty_size; | |
3795 | ffe_ctl.flags = flags; | |
3796 | ffe_ctl.search_start = 0; | |
3797 | ffe_ctl.retry_clustered = false; | |
3798 | ffe_ctl.retry_unclustered = false; | |
3799 | ffe_ctl.delalloc = delalloc; | |
3800 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
3801 | ffe_ctl.have_caching_bg = false; | |
3802 | ffe_ctl.orig_have_caching_bg = false; | |
3803 | ffe_ctl.found_offset = 0; | |
3804 | ||
962a298f | 3805 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
3806 | ins->objectid = 0; |
3807 | ins->offset = 0; | |
b1a4d965 | 3808 | |
71ff6437 | 3809 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 3810 | |
280c2908 | 3811 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 3812 | if (!space_info) { |
0b246afa | 3813 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
3814 | return -ENOSPC; |
3815 | } | |
2552d17e | 3816 | |
67377734 | 3817 | /* |
4f4db217 JB |
3818 | * If our free space is heavily fragmented we may not be able to make |
3819 | * big contiguous allocations, so instead of doing the expensive search | |
3820 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3821 | * can go ahead and search for a more manageable chunk. | |
3822 | * | |
3823 | * If our max_extent_size is large enough for our allocation simply | |
3824 | * disable clustering since we will likely not be able to find enough | |
3825 | * space to create a cluster and induce latency trying. | |
67377734 | 3826 | */ |
4f4db217 JB |
3827 | if (unlikely(space_info->max_extent_size)) { |
3828 | spin_lock(&space_info->lock); | |
3829 | if (space_info->max_extent_size && | |
3830 | num_bytes > space_info->max_extent_size) { | |
3831 | ins->offset = space_info->max_extent_size; | |
3832 | spin_unlock(&space_info->lock); | |
3833 | return -ENOSPC; | |
3834 | } else if (space_info->max_extent_size) { | |
3835 | use_cluster = false; | |
3836 | } | |
3837 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 3838 | } |
0f9dd46c | 3839 | |
b4bd745d QW |
3840 | last_ptr = fetch_cluster_info(fs_info, space_info, |
3841 | &ffe_ctl.empty_cluster); | |
239b14b3 | 3842 | if (last_ptr) { |
fa9c0d79 CM |
3843 | spin_lock(&last_ptr->lock); |
3844 | if (last_ptr->block_group) | |
3845 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
3846 | if (last_ptr->fragmented) { |
3847 | /* | |
3848 | * We still set window_start so we can keep track of the | |
3849 | * last place we found an allocation to try and save | |
3850 | * some time. | |
3851 | */ | |
3852 | hint_byte = last_ptr->window_start; | |
3853 | use_cluster = false; | |
3854 | } | |
fa9c0d79 | 3855 | spin_unlock(&last_ptr->lock); |
239b14b3 | 3856 | } |
fa9c0d79 | 3857 | |
b4bd745d QW |
3858 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
3859 | first_logical_byte(fs_info, 0)); | |
3860 | ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte); | |
3861 | if (ffe_ctl.search_start == hint_byte) { | |
3862 | block_group = btrfs_lookup_block_group(fs_info, | |
3863 | ffe_ctl.search_start); | |
817d52f8 JB |
3864 | /* |
3865 | * we don't want to use the block group if it doesn't match our | |
3866 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
3867 | * |
3868 | * However if we are re-searching with an ideal block group | |
3869 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 3870 | */ |
b6919a58 | 3871 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 3872 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 3873 | down_read(&space_info->groups_sem); |
44fb5511 CM |
3874 | if (list_empty(&block_group->list) || |
3875 | block_group->ro) { | |
3876 | /* | |
3877 | * someone is removing this block group, | |
3878 | * we can't jump into the have_block_group | |
3879 | * target because our list pointers are not | |
3880 | * valid | |
3881 | */ | |
3882 | btrfs_put_block_group(block_group); | |
3883 | up_read(&space_info->groups_sem); | |
ccf0e725 | 3884 | } else { |
b4bd745d | 3885 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 3886 | block_group->flags); |
e570fd27 | 3887 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 3888 | goto have_block_group; |
ccf0e725 | 3889 | } |
2552d17e | 3890 | } else if (block_group) { |
fa9c0d79 | 3891 | btrfs_put_block_group(block_group); |
2552d17e | 3892 | } |
42e70e7a | 3893 | } |
2552d17e | 3894 | search: |
b4bd745d QW |
3895 | ffe_ctl.have_caching_bg = false; |
3896 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
3897 | ffe_ctl.index == 0) | |
a5e681d9 | 3898 | full_search = true; |
80eb234a | 3899 | down_read(&space_info->groups_sem); |
b4bd745d QW |
3900 | list_for_each_entry(block_group, |
3901 | &space_info->block_groups[ffe_ctl.index], list) { | |
14443937 JM |
3902 | /* If the block group is read-only, we can skip it entirely. */ |
3903 | if (unlikely(block_group->ro)) | |
3904 | continue; | |
3905 | ||
e570fd27 | 3906 | btrfs_grab_block_group(block_group, delalloc); |
b4bd745d | 3907 | ffe_ctl.search_start = block_group->key.objectid; |
42e70e7a | 3908 | |
83a50de9 CM |
3909 | /* |
3910 | * this can happen if we end up cycling through all the | |
3911 | * raid types, but we want to make sure we only allocate | |
3912 | * for the proper type. | |
3913 | */ | |
b6919a58 | 3914 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 3915 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 3916 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 3917 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
3918 | BTRFS_BLOCK_GROUP_RAID10; |
3919 | ||
3920 | /* | |
3921 | * if they asked for extra copies and this block group | |
3922 | * doesn't provide them, bail. This does allow us to | |
3923 | * fill raid0 from raid1. | |
3924 | */ | |
b6919a58 | 3925 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 3926 | goto loop; |
2a28468e QW |
3927 | |
3928 | /* | |
3929 | * This block group has different flags than we want. | |
3930 | * It's possible that we have MIXED_GROUP flag but no | |
3931 | * block group is mixed. Just skip such block group. | |
3932 | */ | |
3933 | btrfs_release_block_group(block_group, delalloc); | |
3934 | continue; | |
83a50de9 CM |
3935 | } |
3936 | ||
2552d17e | 3937 | have_block_group: |
676f1f75 | 3938 | ffe_ctl.cached = btrfs_block_group_cache_done(block_group); |
b4bd745d QW |
3939 | if (unlikely(!ffe_ctl.cached)) { |
3940 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 3941 | ret = btrfs_cache_block_group(block_group, 0); |
1d4284bd CM |
3942 | BUG_ON(ret < 0); |
3943 | ret = 0; | |
817d52f8 JB |
3944 | } |
3945 | ||
36cce922 JB |
3946 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
3947 | goto loop; | |
0f9dd46c | 3948 | |
0a24325e | 3949 | /* |
062c05c4 AO |
3950 | * Ok we want to try and use the cluster allocator, so |
3951 | * lets look there | |
0a24325e | 3952 | */ |
c759c4e1 | 3953 | if (last_ptr && use_cluster) { |
d06e3bb6 | 3954 | struct btrfs_block_group_cache *cluster_bg = NULL; |
fa9c0d79 | 3955 | |
d06e3bb6 QW |
3956 | ret = find_free_extent_clustered(block_group, last_ptr, |
3957 | &ffe_ctl, &cluster_bg); | |
062c05c4 | 3958 | |
fa9c0d79 | 3959 | if (ret == 0) { |
d06e3bb6 QW |
3960 | if (cluster_bg && cluster_bg != block_group) { |
3961 | btrfs_release_block_group(block_group, | |
3962 | delalloc); | |
3963 | block_group = cluster_bg; | |
fa9c0d79 | 3964 | } |
d06e3bb6 QW |
3965 | goto checks; |
3966 | } else if (ret == -EAGAIN) { | |
817d52f8 | 3967 | goto have_block_group; |
d06e3bb6 QW |
3968 | } else if (ret > 0) { |
3969 | goto loop; | |
fa9c0d79 | 3970 | } |
d06e3bb6 | 3971 | /* ret == -ENOENT case falls through */ |
fa9c0d79 CM |
3972 | } |
3973 | ||
e1a41848 QW |
3974 | ret = find_free_extent_unclustered(block_group, last_ptr, |
3975 | &ffe_ctl); | |
3976 | if (ret == -EAGAIN) | |
817d52f8 | 3977 | goto have_block_group; |
e1a41848 | 3978 | else if (ret > 0) |
1cdda9b8 | 3979 | goto loop; |
e1a41848 | 3980 | /* ret == 0 case falls through */ |
fa9c0d79 | 3981 | checks: |
b4bd745d QW |
3982 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
3983 | fs_info->stripesize); | |
25179201 | 3984 | |
2552d17e | 3985 | /* move on to the next group */ |
b4bd745d | 3986 | if (ffe_ctl.search_start + num_bytes > |
215a63d1 | 3987 | block_group->key.objectid + block_group->key.offset) { |
b4bd745d QW |
3988 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
3989 | num_bytes); | |
2552d17e | 3990 | goto loop; |
6226cb0a | 3991 | } |
f5a31e16 | 3992 | |
b4bd745d QW |
3993 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
3994 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
3995 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 3996 | |
18513091 WX |
3997 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
3998 | num_bytes, delalloc); | |
f0486c68 | 3999 | if (ret == -EAGAIN) { |
b4bd745d QW |
4000 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4001 | num_bytes); | |
2552d17e | 4002 | goto loop; |
0f9dd46c | 4003 | } |
9cfa3e34 | 4004 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4005 | |
f0486c68 | 4006 | /* we are all good, lets return */ |
b4bd745d | 4007 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4008 | ins->offset = num_bytes; |
d2fb3437 | 4009 | |
b4bd745d QW |
4010 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4011 | num_bytes); | |
e570fd27 | 4012 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4013 | break; |
4014 | loop: | |
b4bd745d QW |
4015 | ffe_ctl.retry_clustered = false; |
4016 | ffe_ctl.retry_unclustered = false; | |
3e72ee88 | 4017 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
b4bd745d | 4018 | ffe_ctl.index); |
e570fd27 | 4019 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 4020 | cond_resched(); |
2552d17e JB |
4021 | } |
4022 | up_read(&space_info->groups_sem); | |
4023 | ||
e72d79d6 QW |
4024 | ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl, |
4025 | full_search, use_cluster); | |
4026 | if (ret > 0) | |
b742bb82 YZ |
4027 | goto search; |
4028 | ||
4f4db217 | 4029 | if (ret == -ENOSPC) { |
b4bd745d QW |
4030 | /* |
4031 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4032 | * any contiguous hole. | |
4033 | */ | |
4034 | if (!ffe_ctl.max_extent_size) | |
4035 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4036 | spin_lock(&space_info->lock); |
b4bd745d | 4037 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4038 | spin_unlock(&space_info->lock); |
b4bd745d | 4039 | ins->offset = ffe_ctl.max_extent_size; |
4f4db217 | 4040 | } |
0f70abe2 | 4041 | return ret; |
fec577fb | 4042 | } |
ec44a35c | 4043 | |
6f47c706 NB |
4044 | /* |
4045 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4046 | * hole that is at least as big as @num_bytes. | |
4047 | * | |
4048 | * @root - The root that will contain this extent | |
4049 | * | |
4050 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4051 | * is used for accounting purposes. This value differs | |
4052 | * from @num_bytes only in the case of compressed extents. | |
4053 | * | |
4054 | * @num_bytes - Number of bytes to allocate on-disk. | |
4055 | * | |
4056 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4057 | * allocator should try to satisfy. In some cases | |
4058 | * @num_bytes may be larger than what is required and if | |
4059 | * the filesystem is fragmented then allocation fails. | |
4060 | * However, the presence of @min_alloc_size gives a | |
4061 | * chance to try and satisfy the smaller allocation. | |
4062 | * | |
4063 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4064 | * size in bytes the allocator should try to find free | |
4065 | * next to the block it returns. This is just a hint and | |
4066 | * may be ignored by the allocator. | |
4067 | * | |
4068 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4069 | * address passed. It might be ignored. | |
4070 | * | |
4071 | * @ins - This key is modified to record the found hole. It will | |
4072 | * have the following values: | |
4073 | * ins->objectid == start position | |
4074 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4075 | * ins->offset == the size of the hole. | |
4076 | * | |
4077 | * @is_data - Boolean flag indicating whether an extent is | |
4078 | * allocated for data (true) or metadata (false) | |
4079 | * | |
4080 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4081 | * delalloc or not. If 'true' data_rwsem of block groups | |
4082 | * is going to be acquired. | |
4083 | * | |
4084 | * | |
4085 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4086 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4087 | * largest available hole the allocator managed to find. | |
4088 | */ | |
18513091 | 4089 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4090 | u64 num_bytes, u64 min_alloc_size, |
4091 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4092 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4093 | { |
ab8d0fc4 | 4094 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4095 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4096 | u64 flags; |
fec577fb | 4097 | int ret; |
925baedd | 4098 | |
1b86826d | 4099 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4100 | again: |
0b246afa | 4101 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 4102 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 4103 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4104 | if (!ret && !is_data) { |
ab8d0fc4 | 4105 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4106 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4107 | if (!final_tried && ins->offset) { |
4108 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4109 | num_bytes = round_down(num_bytes, |
0b246afa | 4110 | fs_info->sectorsize); |
9e622d6b | 4111 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4112 | ram_bytes = num_bytes; |
9e622d6b MX |
4113 | if (num_bytes == min_alloc_size) |
4114 | final_tried = true; | |
4115 | goto again; | |
ab8d0fc4 | 4116 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4117 | struct btrfs_space_info *sinfo; |
4118 | ||
280c2908 | 4119 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4120 | btrfs_err(fs_info, |
5d163e0e JM |
4121 | "allocation failed flags %llu, wanted %llu", |
4122 | flags, num_bytes); | |
53804280 | 4123 | if (sinfo) |
5da6afeb JB |
4124 | btrfs_dump_space_info(fs_info, sinfo, |
4125 | num_bytes, 1); | |
9e622d6b | 4126 | } |
925baedd | 4127 | } |
0f9dd46c JB |
4128 | |
4129 | return ret; | |
e6dcd2dc CM |
4130 | } |
4131 | ||
2ff7e61e | 4132 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
4133 | u64 start, u64 len, |
4134 | int pin, int delalloc) | |
65b51a00 | 4135 | { |
0f9dd46c | 4136 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 4137 | int ret = 0; |
0f9dd46c | 4138 | |
0b246afa | 4139 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4140 | if (!cache) { |
0b246afa JM |
4141 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4142 | start); | |
0f9dd46c JB |
4143 | return -ENOSPC; |
4144 | } | |
1f3c79a2 | 4145 | |
e688b725 | 4146 | if (pin) |
fdf08605 | 4147 | pin_down_extent(cache, start, len, 1); |
e688b725 | 4148 | else { |
0b246afa | 4149 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 4150 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 4151 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 4152 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 4153 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 4154 | } |
31193213 | 4155 | |
fa9c0d79 | 4156 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
4157 | return ret; |
4158 | } | |
4159 | ||
2ff7e61e | 4160 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 4161 | u64 start, u64 len, int delalloc) |
e688b725 | 4162 | { |
2ff7e61e | 4163 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
4164 | } |
4165 | ||
2ff7e61e | 4166 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
4167 | u64 start, u64 len) |
4168 | { | |
2ff7e61e | 4169 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
4170 | } |
4171 | ||
5d4f98a2 | 4172 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4173 | u64 parent, u64 root_objectid, |
4174 | u64 flags, u64 owner, u64 offset, | |
4175 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4176 | { |
ef89b824 | 4177 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4178 | int ret; |
e6dcd2dc | 4179 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4180 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4181 | struct btrfs_path *path; |
5d4f98a2 YZ |
4182 | struct extent_buffer *leaf; |
4183 | int type; | |
4184 | u32 size; | |
26b8003f | 4185 | |
5d4f98a2 YZ |
4186 | if (parent > 0) |
4187 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4188 | else | |
4189 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4190 | |
5d4f98a2 | 4191 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4192 | |
4193 | path = btrfs_alloc_path(); | |
db5b493a TI |
4194 | if (!path) |
4195 | return -ENOMEM; | |
47e4bb98 | 4196 | |
b9473439 | 4197 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4198 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4199 | ins, size); | |
79787eaa JM |
4200 | if (ret) { |
4201 | btrfs_free_path(path); | |
4202 | return ret; | |
4203 | } | |
0f9dd46c | 4204 | |
5d4f98a2 YZ |
4205 | leaf = path->nodes[0]; |
4206 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4207 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4208 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4209 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4210 | btrfs_set_extent_flags(leaf, extent_item, | |
4211 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4212 | ||
4213 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4214 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4215 | if (parent > 0) { | |
4216 | struct btrfs_shared_data_ref *ref; | |
4217 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4218 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4219 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4220 | } else { | |
4221 | struct btrfs_extent_data_ref *ref; | |
4222 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4223 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4224 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4225 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4226 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4227 | } | |
47e4bb98 CM |
4228 | |
4229 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4230 | btrfs_free_path(path); |
f510cfec | 4231 | |
25a356d3 | 4232 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4233 | if (ret) |
4234 | return ret; | |
4235 | ||
ade4b516 | 4236 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4237 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4238 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4239 | ins->objectid, ins->offset); |
f5947066 CM |
4240 | BUG(); |
4241 | } | |
71ff6437 | 4242 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4243 | return ret; |
4244 | } | |
4245 | ||
5d4f98a2 | 4246 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4247 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4248 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4249 | { |
9dcdbe01 | 4250 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4251 | int ret; |
5d4f98a2 | 4252 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4253 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4254 | struct btrfs_tree_block_info *block_info; |
4255 | struct btrfs_extent_inline_ref *iref; | |
4256 | struct btrfs_path *path; | |
4257 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4258 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4259 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4260 | u64 num_bytes; |
21ebfbe7 | 4261 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4262 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4263 | |
4e6bd4e0 NB |
4264 | ref = btrfs_delayed_node_to_tree_ref(node); |
4265 | ||
4e6bd4e0 NB |
4266 | extent_key.objectid = node->bytenr; |
4267 | if (skinny_metadata) { | |
4268 | extent_key.offset = ref->level; | |
4269 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4270 | num_bytes = fs_info->nodesize; | |
4271 | } else { | |
4272 | extent_key.offset = node->num_bytes; | |
4273 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4274 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4275 | num_bytes = node->num_bytes; |
4276 | } | |
1c2308f8 | 4277 | |
5d4f98a2 | 4278 | path = btrfs_alloc_path(); |
80ee54bf | 4279 | if (!path) |
d8926bb3 | 4280 | return -ENOMEM; |
56bec294 | 4281 | |
5d4f98a2 YZ |
4282 | path->leave_spinning = 1; |
4283 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 4284 | &extent_key, size); |
79787eaa | 4285 | if (ret) { |
dd825259 | 4286 | btrfs_free_path(path); |
79787eaa JM |
4287 | return ret; |
4288 | } | |
5d4f98a2 YZ |
4289 | |
4290 | leaf = path->nodes[0]; | |
4291 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4292 | struct btrfs_extent_item); | |
4293 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4294 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4295 | btrfs_set_extent_flags(leaf, extent_item, | |
4296 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4297 | |
3173a18f JB |
4298 | if (skinny_metadata) { |
4299 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4300 | } else { | |
4301 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4302 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4303 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4304 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4305 | } | |
5d4f98a2 | 4306 | |
d4b20733 | 4307 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4308 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4309 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4310 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4311 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4312 | } else { |
4313 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4314 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4315 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4316 | } |
4317 | ||
4318 | btrfs_mark_buffer_dirty(leaf); | |
4319 | btrfs_free_path(path); | |
4320 | ||
4e6bd4e0 NB |
4321 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4322 | num_bytes); | |
1e144fb8 OS |
4323 | if (ret) |
4324 | return ret; | |
4325 | ||
ade4b516 JB |
4326 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4327 | fs_info->nodesize, 1); | |
79787eaa | 4328 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4329 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4330 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4331 | BUG(); |
4332 | } | |
0be5dc67 | 4333 | |
4e6bd4e0 | 4334 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4335 | fs_info->nodesize); |
5d4f98a2 YZ |
4336 | return ret; |
4337 | } | |
4338 | ||
4339 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4340 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4341 | u64 offset, u64 ram_bytes, |
4342 | struct btrfs_key *ins) | |
5d4f98a2 | 4343 | { |
76675593 | 4344 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
4345 | int ret; |
4346 | ||
84f7d8e6 | 4347 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4348 | |
76675593 QW |
4349 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4350 | ins->objectid, ins->offset, 0); | |
4351 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4352 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
4353 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
4354 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
4355 | return ret; |
4356 | } | |
e02119d5 CM |
4357 | |
4358 | /* | |
4359 | * this is used by the tree logging recovery code. It records that | |
4360 | * an extent has been allocated and makes sure to clear the free | |
4361 | * space cache bits as well | |
4362 | */ | |
5d4f98a2 | 4363 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4364 | u64 root_objectid, u64 owner, u64 offset, |
4365 | struct btrfs_key *ins) | |
e02119d5 | 4366 | { |
61da2abf | 4367 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
4368 | int ret; |
4369 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 4370 | struct btrfs_space_info *space_info; |
11833d66 | 4371 | |
8c2a1a30 JB |
4372 | /* |
4373 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4374 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4375 | */ |
0b246afa | 4376 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4377 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4378 | ins->offset); | |
b50c6e25 | 4379 | if (ret) |
8c2a1a30 | 4380 | return ret; |
11833d66 YZ |
4381 | } |
4382 | ||
0b246afa | 4383 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4384 | if (!block_group) |
4385 | return -EINVAL; | |
4386 | ||
ed7a6948 WX |
4387 | space_info = block_group->space_info; |
4388 | spin_lock(&space_info->lock); | |
4389 | spin_lock(&block_group->lock); | |
4390 | space_info->bytes_reserved += ins->offset; | |
4391 | block_group->reserved += ins->offset; | |
4392 | spin_unlock(&block_group->lock); | |
4393 | spin_unlock(&space_info->lock); | |
4394 | ||
ef89b824 NB |
4395 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4396 | offset, ins, 1); | |
b50c6e25 | 4397 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4398 | return ret; |
4399 | } | |
4400 | ||
48a3b636 ES |
4401 | static struct extent_buffer * |
4402 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 4403 | u64 bytenr, int level, u64 owner) |
65b51a00 | 4404 | { |
0b246afa | 4405 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4406 | struct extent_buffer *buf; |
4407 | ||
2ff7e61e | 4408 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4409 | if (IS_ERR(buf)) |
4410 | return buf; | |
4411 | ||
b72c3aba QW |
4412 | /* |
4413 | * Extra safety check in case the extent tree is corrupted and extent | |
4414 | * allocator chooses to use a tree block which is already used and | |
4415 | * locked. | |
4416 | */ | |
4417 | if (buf->lock_owner == current->pid) { | |
4418 | btrfs_err_rl(fs_info, | |
4419 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4420 | buf->start, btrfs_header_owner(buf), current->pid); | |
4421 | free_extent_buffer(buf); | |
4422 | return ERR_PTR(-EUCLEAN); | |
4423 | } | |
4424 | ||
85d4e461 | 4425 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 4426 | btrfs_tree_lock(buf); |
6a884d7d | 4427 | btrfs_clean_tree_block(buf); |
3083ee2e | 4428 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 4429 | |
8bead258 | 4430 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 4431 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4432 | |
bc877d28 NB |
4433 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4434 | btrfs_set_header_level(buf, level); | |
4435 | btrfs_set_header_bytenr(buf, buf->start); | |
4436 | btrfs_set_header_generation(buf, trans->transid); | |
4437 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4438 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4439 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4440 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4441 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4442 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4443 | /* |
4444 | * we allow two log transactions at a time, use different | |
52042d8e | 4445 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4446 | */ |
656f30db | 4447 | if (buf->log_index == 0) |
8cef4e16 YZ |
4448 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4449 | buf->start + buf->len - 1, GFP_NOFS); | |
4450 | else | |
4451 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4452 | buf->start + buf->len - 1); |
d0c803c4 | 4453 | } else { |
656f30db | 4454 | buf->log_index = -1; |
d0c803c4 | 4455 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4456 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4457 | } |
64c12921 | 4458 | trans->dirty = true; |
b4ce94de | 4459 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4460 | return buf; |
4461 | } | |
4462 | ||
fec577fb | 4463 | /* |
f0486c68 | 4464 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4465 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4466 | */ |
4d75f8a9 | 4467 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4468 | struct btrfs_root *root, |
4469 | u64 parent, u64 root_objectid, | |
4470 | const struct btrfs_disk_key *key, | |
4471 | int level, u64 hint, | |
4472 | u64 empty_size) | |
fec577fb | 4473 | { |
0b246afa | 4474 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4475 | struct btrfs_key ins; |
f0486c68 | 4476 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4477 | struct extent_buffer *buf; |
67b7859e | 4478 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4479 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4480 | u64 flags = 0; |
4481 | int ret; | |
0b246afa JM |
4482 | u32 blocksize = fs_info->nodesize; |
4483 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4484 | |
05653ef3 | 4485 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4486 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4487 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 4488 | level, root_objectid); |
faa2dbf0 JB |
4489 | if (!IS_ERR(buf)) |
4490 | root->alloc_bytenr += blocksize; | |
4491 | return buf; | |
4492 | } | |
05653ef3 | 4493 | #endif |
fccb84c9 | 4494 | |
67f9c220 | 4495 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4496 | if (IS_ERR(block_rsv)) |
4497 | return ERR_CAST(block_rsv); | |
4498 | ||
18513091 | 4499 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4500 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4501 | if (ret) |
4502 | goto out_unuse; | |
55c69072 | 4503 | |
bc877d28 NB |
4504 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
4505 | root_objectid); | |
67b7859e OS |
4506 | if (IS_ERR(buf)) { |
4507 | ret = PTR_ERR(buf); | |
4508 | goto out_free_reserved; | |
4509 | } | |
f0486c68 YZ |
4510 | |
4511 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4512 | if (parent == 0) | |
4513 | parent = ins.objectid; | |
4514 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4515 | } else | |
4516 | BUG_ON(parent > 0); | |
4517 | ||
4518 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4519 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4520 | if (!extent_op) { |
4521 | ret = -ENOMEM; | |
4522 | goto out_free_buf; | |
4523 | } | |
f0486c68 YZ |
4524 | if (key) |
4525 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4526 | else | |
4527 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4528 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4529 | extent_op->update_key = skinny_metadata ? false : true; |
4530 | extent_op->update_flags = true; | |
4531 | extent_op->is_data = false; | |
b1c79e09 | 4532 | extent_op->level = level; |
f0486c68 | 4533 | |
ed4f255b QW |
4534 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4535 | ins.objectid, ins.offset, parent); | |
4536 | generic_ref.real_root = root->root_key.objectid; | |
4537 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4538 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 4539 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 4540 | extent_op, NULL, NULL); |
67b7859e OS |
4541 | if (ret) |
4542 | goto out_free_delayed; | |
f0486c68 | 4543 | } |
fec577fb | 4544 | return buf; |
67b7859e OS |
4545 | |
4546 | out_free_delayed: | |
4547 | btrfs_free_delayed_extent_op(extent_op); | |
4548 | out_free_buf: | |
4549 | free_extent_buffer(buf); | |
4550 | out_free_reserved: | |
2ff7e61e | 4551 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4552 | out_unuse: |
67f9c220 | 4553 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4554 | return ERR_PTR(ret); |
fec577fb | 4555 | } |
a28ec197 | 4556 | |
2c47e605 YZ |
4557 | struct walk_control { |
4558 | u64 refs[BTRFS_MAX_LEVEL]; | |
4559 | u64 flags[BTRFS_MAX_LEVEL]; | |
4560 | struct btrfs_key update_progress; | |
aea6f028 JB |
4561 | struct btrfs_key drop_progress; |
4562 | int drop_level; | |
2c47e605 YZ |
4563 | int stage; |
4564 | int level; | |
4565 | int shared_level; | |
4566 | int update_ref; | |
4567 | int keep_locks; | |
1c4850e2 YZ |
4568 | int reada_slot; |
4569 | int reada_count; | |
78c52d9e | 4570 | int restarted; |
2c47e605 YZ |
4571 | }; |
4572 | ||
4573 | #define DROP_REFERENCE 1 | |
4574 | #define UPDATE_BACKREF 2 | |
4575 | ||
1c4850e2 YZ |
4576 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4577 | struct btrfs_root *root, | |
4578 | struct walk_control *wc, | |
4579 | struct btrfs_path *path) | |
6407bf6d | 4580 | { |
0b246afa | 4581 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4582 | u64 bytenr; |
4583 | u64 generation; | |
4584 | u64 refs; | |
94fcca9f | 4585 | u64 flags; |
5d4f98a2 | 4586 | u32 nritems; |
1c4850e2 YZ |
4587 | struct btrfs_key key; |
4588 | struct extent_buffer *eb; | |
6407bf6d | 4589 | int ret; |
1c4850e2 YZ |
4590 | int slot; |
4591 | int nread = 0; | |
6407bf6d | 4592 | |
1c4850e2 YZ |
4593 | if (path->slots[wc->level] < wc->reada_slot) { |
4594 | wc->reada_count = wc->reada_count * 2 / 3; | |
4595 | wc->reada_count = max(wc->reada_count, 2); | |
4596 | } else { | |
4597 | wc->reada_count = wc->reada_count * 3 / 2; | |
4598 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4599 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4600 | } |
7bb86316 | 4601 | |
1c4850e2 YZ |
4602 | eb = path->nodes[wc->level]; |
4603 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4604 | |
1c4850e2 YZ |
4605 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4606 | if (nread >= wc->reada_count) | |
4607 | break; | |
bd56b302 | 4608 | |
2dd3e67b | 4609 | cond_resched(); |
1c4850e2 YZ |
4610 | bytenr = btrfs_node_blockptr(eb, slot); |
4611 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4612 | |
1c4850e2 YZ |
4613 | if (slot == path->slots[wc->level]) |
4614 | goto reada; | |
5d4f98a2 | 4615 | |
1c4850e2 YZ |
4616 | if (wc->stage == UPDATE_BACKREF && |
4617 | generation <= root->root_key.offset) | |
bd56b302 CM |
4618 | continue; |
4619 | ||
94fcca9f | 4620 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4621 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4622 | wc->level - 1, 1, &refs, |
4623 | &flags); | |
79787eaa JM |
4624 | /* We don't care about errors in readahead. */ |
4625 | if (ret < 0) | |
4626 | continue; | |
94fcca9f YZ |
4627 | BUG_ON(refs == 0); |
4628 | ||
1c4850e2 | 4629 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4630 | if (refs == 1) |
4631 | goto reada; | |
bd56b302 | 4632 | |
94fcca9f YZ |
4633 | if (wc->level == 1 && |
4634 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4635 | continue; | |
1c4850e2 YZ |
4636 | if (!wc->update_ref || |
4637 | generation <= root->root_key.offset) | |
4638 | continue; | |
4639 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4640 | ret = btrfs_comp_cpu_keys(&key, | |
4641 | &wc->update_progress); | |
4642 | if (ret < 0) | |
4643 | continue; | |
94fcca9f YZ |
4644 | } else { |
4645 | if (wc->level == 1 && | |
4646 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4647 | continue; | |
6407bf6d | 4648 | } |
1c4850e2 | 4649 | reada: |
2ff7e61e | 4650 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 4651 | nread++; |
20524f02 | 4652 | } |
1c4850e2 | 4653 | wc->reada_slot = slot; |
20524f02 | 4654 | } |
2c47e605 | 4655 | |
f82d02d9 | 4656 | /* |
2c016dc2 | 4657 | * helper to process tree block while walking down the tree. |
2c47e605 | 4658 | * |
2c47e605 YZ |
4659 | * when wc->stage == UPDATE_BACKREF, this function updates |
4660 | * back refs for pointers in the block. | |
4661 | * | |
4662 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4663 | */ |
2c47e605 | 4664 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4665 | struct btrfs_root *root, |
2c47e605 | 4666 | struct btrfs_path *path, |
94fcca9f | 4667 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4668 | { |
2ff7e61e | 4669 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4670 | int level = wc->level; |
4671 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4672 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4673 | int ret; |
4674 | ||
2c47e605 YZ |
4675 | if (wc->stage == UPDATE_BACKREF && |
4676 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4677 | return 1; | |
f82d02d9 | 4678 | |
2c47e605 YZ |
4679 | /* |
4680 | * when reference count of tree block is 1, it won't increase | |
4681 | * again. once full backref flag is set, we never clear it. | |
4682 | */ | |
94fcca9f YZ |
4683 | if (lookup_info && |
4684 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4685 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4686 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4687 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4688 | eb->start, level, 1, |
2c47e605 YZ |
4689 | &wc->refs[level], |
4690 | &wc->flags[level]); | |
79787eaa JM |
4691 | BUG_ON(ret == -ENOMEM); |
4692 | if (ret) | |
4693 | return ret; | |
2c47e605 YZ |
4694 | BUG_ON(wc->refs[level] == 0); |
4695 | } | |
5d4f98a2 | 4696 | |
2c47e605 YZ |
4697 | if (wc->stage == DROP_REFERENCE) { |
4698 | if (wc->refs[level] > 1) | |
4699 | return 1; | |
f82d02d9 | 4700 | |
2c47e605 | 4701 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4702 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4703 | path->locks[level] = 0; |
4704 | } | |
4705 | return 0; | |
4706 | } | |
f82d02d9 | 4707 | |
2c47e605 YZ |
4708 | /* wc->stage == UPDATE_BACKREF */ |
4709 | if (!(wc->flags[level] & flag)) { | |
4710 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4711 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4712 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4713 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4714 | BUG_ON(ret); /* -ENOMEM */ |
f5c8daa5 | 4715 | ret = btrfs_set_disk_extent_flags(trans, eb->start, |
b1c79e09 JB |
4716 | eb->len, flag, |
4717 | btrfs_header_level(eb), 0); | |
79787eaa | 4718 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4719 | wc->flags[level] |= flag; |
4720 | } | |
4721 | ||
4722 | /* | |
4723 | * the block is shared by multiple trees, so it's not good to | |
4724 | * keep the tree lock | |
4725 | */ | |
4726 | if (path->locks[level] && level > 0) { | |
bd681513 | 4727 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4728 | path->locks[level] = 0; |
4729 | } | |
4730 | return 0; | |
4731 | } | |
4732 | ||
78c52d9e JB |
4733 | /* |
4734 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4735 | * would have a drop_progress key that hadn't been updated properly. | |
4736 | */ | |
4737 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4738 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4739 | int level) | |
4740 | { | |
4741 | struct btrfs_path *path; | |
4742 | struct btrfs_extent_inline_ref *iref; | |
4743 | int ret; | |
4744 | ||
4745 | path = btrfs_alloc_path(); | |
4746 | if (!path) | |
4747 | return -ENOMEM; | |
4748 | ||
4749 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
4750 | root->fs_info->nodesize, parent, | |
4751 | root->root_key.objectid, level, 0); | |
4752 | btrfs_free_path(path); | |
4753 | if (ret == -ENOENT) | |
4754 | return 0; | |
4755 | if (ret < 0) | |
4756 | return ret; | |
4757 | return 1; | |
4758 | } | |
4759 | ||
1c4850e2 | 4760 | /* |
2c016dc2 | 4761 | * helper to process tree block pointer. |
1c4850e2 YZ |
4762 | * |
4763 | * when wc->stage == DROP_REFERENCE, this function checks | |
4764 | * reference count of the block pointed to. if the block | |
4765 | * is shared and we need update back refs for the subtree | |
4766 | * rooted at the block, this function changes wc->stage to | |
4767 | * UPDATE_BACKREF. if the block is shared and there is no | |
4768 | * need to update back, this function drops the reference | |
4769 | * to the block. | |
4770 | * | |
4771 | * NOTE: return value 1 means we should stop walking down. | |
4772 | */ | |
4773 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
4774 | struct btrfs_root *root, | |
4775 | struct btrfs_path *path, | |
94fcca9f | 4776 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 4777 | { |
0b246afa | 4778 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4779 | u64 bytenr; |
4780 | u64 generation; | |
4781 | u64 parent; | |
1c4850e2 | 4782 | struct btrfs_key key; |
581c1760 | 4783 | struct btrfs_key first_key; |
ffd4bb2a | 4784 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
4785 | struct extent_buffer *next; |
4786 | int level = wc->level; | |
4787 | int reada = 0; | |
4788 | int ret = 0; | |
1152651a | 4789 | bool need_account = false; |
1c4850e2 YZ |
4790 | |
4791 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
4792 | path->slots[level]); | |
4793 | /* | |
4794 | * if the lower level block was created before the snapshot | |
4795 | * was created, we know there is no need to update back refs | |
4796 | * for the subtree | |
4797 | */ | |
4798 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
4799 | generation <= root->root_key.offset) { |
4800 | *lookup_info = 1; | |
1c4850e2 | 4801 | return 1; |
94fcca9f | 4802 | } |
1c4850e2 YZ |
4803 | |
4804 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
4805 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
4806 | path->slots[level]); | |
1c4850e2 | 4807 | |
0b246afa | 4808 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 4809 | if (!next) { |
2ff7e61e | 4810 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4811 | if (IS_ERR(next)) |
4812 | return PTR_ERR(next); | |
4813 | ||
b2aaaa3b JB |
4814 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
4815 | level - 1); | |
1c4850e2 YZ |
4816 | reada = 1; |
4817 | } | |
4818 | btrfs_tree_lock(next); | |
8bead258 | 4819 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 4820 | |
2ff7e61e | 4821 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
4822 | &wc->refs[level - 1], |
4823 | &wc->flags[level - 1]); | |
4867268c JB |
4824 | if (ret < 0) |
4825 | goto out_unlock; | |
79787eaa | 4826 | |
c2cf52eb | 4827 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 4828 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
4829 | ret = -EIO; |
4830 | goto out_unlock; | |
c2cf52eb | 4831 | } |
94fcca9f | 4832 | *lookup_info = 0; |
1c4850e2 | 4833 | |
94fcca9f | 4834 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 4835 | if (wc->refs[level - 1] > 1) { |
1152651a | 4836 | need_account = true; |
94fcca9f YZ |
4837 | if (level == 1 && |
4838 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4839 | goto skip; | |
4840 | ||
1c4850e2 YZ |
4841 | if (!wc->update_ref || |
4842 | generation <= root->root_key.offset) | |
4843 | goto skip; | |
4844 | ||
4845 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
4846 | path->slots[level]); | |
4847 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
4848 | if (ret < 0) | |
4849 | goto skip; | |
4850 | ||
4851 | wc->stage = UPDATE_BACKREF; | |
4852 | wc->shared_level = level - 1; | |
4853 | } | |
94fcca9f YZ |
4854 | } else { |
4855 | if (level == 1 && | |
4856 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4857 | goto skip; | |
1c4850e2 YZ |
4858 | } |
4859 | ||
b9fab919 | 4860 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
4861 | btrfs_tree_unlock(next); |
4862 | free_extent_buffer(next); | |
4863 | next = NULL; | |
94fcca9f | 4864 | *lookup_info = 1; |
1c4850e2 YZ |
4865 | } |
4866 | ||
4867 | if (!next) { | |
4868 | if (reada && level == 1) | |
4869 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
4870 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
4871 | &first_key); | |
64c043de LB |
4872 | if (IS_ERR(next)) { |
4873 | return PTR_ERR(next); | |
4874 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 4875 | free_extent_buffer(next); |
97d9a8a4 | 4876 | return -EIO; |
416bc658 | 4877 | } |
1c4850e2 | 4878 | btrfs_tree_lock(next); |
8bead258 | 4879 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
4880 | } |
4881 | ||
4882 | level--; | |
4867268c JB |
4883 | ASSERT(level == btrfs_header_level(next)); |
4884 | if (level != btrfs_header_level(next)) { | |
4885 | btrfs_err(root->fs_info, "mismatched level"); | |
4886 | ret = -EIO; | |
4887 | goto out_unlock; | |
4888 | } | |
1c4850e2 YZ |
4889 | path->nodes[level] = next; |
4890 | path->slots[level] = 0; | |
bd681513 | 4891 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
4892 | wc->level = level; |
4893 | if (wc->level == 1) | |
4894 | wc->reada_slot = 0; | |
4895 | return 0; | |
4896 | skip: | |
4897 | wc->refs[level - 1] = 0; | |
4898 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
4899 | if (wc->stage == DROP_REFERENCE) { |
4900 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
4901 | parent = path->nodes[level]->start; | |
4902 | } else { | |
4867268c | 4903 | ASSERT(root->root_key.objectid == |
94fcca9f | 4904 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
4905 | if (root->root_key.objectid != |
4906 | btrfs_header_owner(path->nodes[level])) { | |
4907 | btrfs_err(root->fs_info, | |
4908 | "mismatched block owner"); | |
4909 | ret = -EIO; | |
4910 | goto out_unlock; | |
4911 | } | |
94fcca9f YZ |
4912 | parent = 0; |
4913 | } | |
1c4850e2 | 4914 | |
78c52d9e JB |
4915 | /* |
4916 | * If we had a drop_progress we need to verify the refs are set | |
4917 | * as expected. If we find our ref then we know that from here | |
4918 | * on out everything should be correct, and we can clear the | |
4919 | * ->restarted flag. | |
4920 | */ | |
4921 | if (wc->restarted) { | |
4922 | ret = check_ref_exists(trans, root, bytenr, parent, | |
4923 | level - 1); | |
4924 | if (ret < 0) | |
4925 | goto out_unlock; | |
4926 | if (ret == 0) | |
4927 | goto no_delete; | |
4928 | ret = 0; | |
4929 | wc->restarted = 0; | |
4930 | } | |
4931 | ||
2cd86d30 QW |
4932 | /* |
4933 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
4934 | * already accounted them at merge time (replace_path), | |
4935 | * thus we could skip expensive subtree trace here. | |
4936 | */ | |
4937 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
4938 | need_account) { | |
deb40627 | 4939 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 4940 | generation, level - 1); |
1152651a | 4941 | if (ret) { |
0b246afa | 4942 | btrfs_err_rl(fs_info, |
5d163e0e JM |
4943 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
4944 | ret); | |
1152651a MF |
4945 | } |
4946 | } | |
aea6f028 JB |
4947 | |
4948 | /* | |
4949 | * We need to update the next key in our walk control so we can | |
4950 | * update the drop_progress key accordingly. We don't care if | |
4951 | * find_next_key doesn't find a key because that means we're at | |
4952 | * the end and are going to clean up now. | |
4953 | */ | |
4954 | wc->drop_level = level; | |
4955 | find_next_key(path, level, &wc->drop_progress); | |
4956 | ||
ffd4bb2a QW |
4957 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
4958 | fs_info->nodesize, parent); | |
4959 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
4960 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
4961 | if (ret) |
4962 | goto out_unlock; | |
1c4850e2 | 4963 | } |
78c52d9e | 4964 | no_delete: |
4867268c JB |
4965 | *lookup_info = 1; |
4966 | ret = 1; | |
4967 | ||
4968 | out_unlock: | |
1c4850e2 YZ |
4969 | btrfs_tree_unlock(next); |
4970 | free_extent_buffer(next); | |
4867268c JB |
4971 | |
4972 | return ret; | |
1c4850e2 YZ |
4973 | } |
4974 | ||
2c47e605 | 4975 | /* |
2c016dc2 | 4976 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
4977 | * |
4978 | * when wc->stage == DROP_REFERENCE, this function drops | |
4979 | * reference count on the block. | |
4980 | * | |
4981 | * when wc->stage == UPDATE_BACKREF, this function changes | |
4982 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
4983 | * to UPDATE_BACKREF previously while processing the block. | |
4984 | * | |
4985 | * NOTE: return value 1 means we should stop walking up. | |
4986 | */ | |
4987 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
4988 | struct btrfs_root *root, | |
4989 | struct btrfs_path *path, | |
4990 | struct walk_control *wc) | |
4991 | { | |
0b246afa | 4992 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 4993 | int ret; |
2c47e605 YZ |
4994 | int level = wc->level; |
4995 | struct extent_buffer *eb = path->nodes[level]; | |
4996 | u64 parent = 0; | |
4997 | ||
4998 | if (wc->stage == UPDATE_BACKREF) { | |
4999 | BUG_ON(wc->shared_level < level); | |
5000 | if (level < wc->shared_level) | |
5001 | goto out; | |
5002 | ||
2c47e605 YZ |
5003 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5004 | if (ret > 0) | |
5005 | wc->update_ref = 0; | |
5006 | ||
5007 | wc->stage = DROP_REFERENCE; | |
5008 | wc->shared_level = -1; | |
5009 | path->slots[level] = 0; | |
5010 | ||
5011 | /* | |
5012 | * check reference count again if the block isn't locked. | |
5013 | * we should start walking down the tree again if reference | |
5014 | * count is one. | |
5015 | */ | |
5016 | if (!path->locks[level]) { | |
5017 | BUG_ON(level == 0); | |
5018 | btrfs_tree_lock(eb); | |
8bead258 | 5019 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5020 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5021 | |
2ff7e61e | 5022 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5023 | eb->start, level, 1, |
2c47e605 YZ |
5024 | &wc->refs[level], |
5025 | &wc->flags[level]); | |
79787eaa JM |
5026 | if (ret < 0) { |
5027 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5028 | path->locks[level] = 0; |
79787eaa JM |
5029 | return ret; |
5030 | } | |
2c47e605 YZ |
5031 | BUG_ON(wc->refs[level] == 0); |
5032 | if (wc->refs[level] == 1) { | |
bd681513 | 5033 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5034 | path->locks[level] = 0; |
2c47e605 YZ |
5035 | return 1; |
5036 | } | |
f82d02d9 | 5037 | } |
2c47e605 | 5038 | } |
f82d02d9 | 5039 | |
2c47e605 YZ |
5040 | /* wc->stage == DROP_REFERENCE */ |
5041 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5042 | |
2c47e605 YZ |
5043 | if (wc->refs[level] == 1) { |
5044 | if (level == 0) { | |
5045 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5046 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5047 | else |
e339a6b0 | 5048 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5049 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5050 | if (is_fstree(root->root_key.objectid)) { |
5051 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5052 | if (ret) { | |
5053 | btrfs_err_rl(fs_info, | |
5054 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5055 | ret); |
c4140cbf | 5056 | } |
1152651a | 5057 | } |
2c47e605 | 5058 | } |
6a884d7d | 5059 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5060 | if (!path->locks[level] && |
5061 | btrfs_header_generation(eb) == trans->transid) { | |
5062 | btrfs_tree_lock(eb); | |
8bead258 | 5063 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5064 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5065 | } |
6a884d7d | 5066 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5067 | } |
5068 | ||
5069 | if (eb == root->node) { | |
5070 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5071 | parent = eb->start; | |
65c6e82b QW |
5072 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5073 | goto owner_mismatch; | |
2c47e605 YZ |
5074 | } else { |
5075 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5076 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5077 | else if (root->root_key.objectid != |
5078 | btrfs_header_owner(path->nodes[level + 1])) | |
5079 | goto owner_mismatch; | |
f82d02d9 | 5080 | } |
f82d02d9 | 5081 | |
5581a51a | 5082 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5083 | out: |
5084 | wc->refs[level] = 0; | |
5085 | wc->flags[level] = 0; | |
f0486c68 | 5086 | return 0; |
65c6e82b QW |
5087 | |
5088 | owner_mismatch: | |
5089 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5090 | btrfs_header_owner(eb), root->root_key.objectid); | |
5091 | return -EUCLEAN; | |
2c47e605 YZ |
5092 | } |
5093 | ||
5094 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5095 | struct btrfs_root *root, | |
5096 | struct btrfs_path *path, | |
5097 | struct walk_control *wc) | |
5098 | { | |
2c47e605 | 5099 | int level = wc->level; |
94fcca9f | 5100 | int lookup_info = 1; |
2c47e605 YZ |
5101 | int ret; |
5102 | ||
5103 | while (level >= 0) { | |
94fcca9f | 5104 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5105 | if (ret > 0) |
5106 | break; | |
5107 | ||
5108 | if (level == 0) | |
5109 | break; | |
5110 | ||
7a7965f8 YZ |
5111 | if (path->slots[level] >= |
5112 | btrfs_header_nritems(path->nodes[level])) | |
5113 | break; | |
5114 | ||
94fcca9f | 5115 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5116 | if (ret > 0) { |
5117 | path->slots[level]++; | |
5118 | continue; | |
90d2c51d MX |
5119 | } else if (ret < 0) |
5120 | return ret; | |
1c4850e2 | 5121 | level = wc->level; |
f82d02d9 | 5122 | } |
f82d02d9 YZ |
5123 | return 0; |
5124 | } | |
5125 | ||
d397712b | 5126 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5127 | struct btrfs_root *root, |
f82d02d9 | 5128 | struct btrfs_path *path, |
2c47e605 | 5129 | struct walk_control *wc, int max_level) |
20524f02 | 5130 | { |
2c47e605 | 5131 | int level = wc->level; |
20524f02 | 5132 | int ret; |
9f3a7427 | 5133 | |
2c47e605 YZ |
5134 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5135 | while (level < max_level && path->nodes[level]) { | |
5136 | wc->level = level; | |
5137 | if (path->slots[level] + 1 < | |
5138 | btrfs_header_nritems(path->nodes[level])) { | |
5139 | path->slots[level]++; | |
20524f02 CM |
5140 | return 0; |
5141 | } else { | |
2c47e605 YZ |
5142 | ret = walk_up_proc(trans, root, path, wc); |
5143 | if (ret > 0) | |
5144 | return 0; | |
65c6e82b QW |
5145 | if (ret < 0) |
5146 | return ret; | |
bd56b302 | 5147 | |
2c47e605 | 5148 | if (path->locks[level]) { |
bd681513 CM |
5149 | btrfs_tree_unlock_rw(path->nodes[level], |
5150 | path->locks[level]); | |
2c47e605 | 5151 | path->locks[level] = 0; |
f82d02d9 | 5152 | } |
2c47e605 YZ |
5153 | free_extent_buffer(path->nodes[level]); |
5154 | path->nodes[level] = NULL; | |
5155 | level++; | |
20524f02 CM |
5156 | } |
5157 | } | |
5158 | return 1; | |
5159 | } | |
5160 | ||
9aca1d51 | 5161 | /* |
2c47e605 YZ |
5162 | * drop a subvolume tree. |
5163 | * | |
5164 | * this function traverses the tree freeing any blocks that only | |
5165 | * referenced by the tree. | |
5166 | * | |
5167 | * when a shared tree block is found. this function decreases its | |
5168 | * reference count by one. if update_ref is true, this function | |
5169 | * also make sure backrefs for the shared block and all lower level | |
5170 | * blocks are properly updated. | |
9d1a2a3a DS |
5171 | * |
5172 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5173 | */ |
2c536799 | 5174 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
5175 | struct btrfs_block_rsv *block_rsv, int update_ref, |
5176 | int for_reloc) | |
20524f02 | 5177 | { |
ab8d0fc4 | 5178 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5179 | struct btrfs_path *path; |
2c47e605 | 5180 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5181 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5182 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5183 | struct walk_control *wc; |
5184 | struct btrfs_key key; | |
5185 | int err = 0; | |
5186 | int ret; | |
5187 | int level; | |
d29a9f62 | 5188 | bool root_dropped = false; |
20524f02 | 5189 | |
4fd786e6 | 5190 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5191 | |
5caf2a00 | 5192 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5193 | if (!path) { |
5194 | err = -ENOMEM; | |
5195 | goto out; | |
5196 | } | |
20524f02 | 5197 | |
2c47e605 | 5198 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5199 | if (!wc) { |
5200 | btrfs_free_path(path); | |
cb1b69f4 TI |
5201 | err = -ENOMEM; |
5202 | goto out; | |
38a1a919 | 5203 | } |
2c47e605 | 5204 | |
a22285a6 | 5205 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
5206 | if (IS_ERR(trans)) { |
5207 | err = PTR_ERR(trans); | |
5208 | goto out_free; | |
5209 | } | |
98d5dc13 | 5210 | |
0568e82d JB |
5211 | err = btrfs_run_delayed_items(trans); |
5212 | if (err) | |
5213 | goto out_end_trans; | |
5214 | ||
3fd0a558 YZ |
5215 | if (block_rsv) |
5216 | trans->block_rsv = block_rsv; | |
2c47e605 | 5217 | |
83354f07 JB |
5218 | /* |
5219 | * This will help us catch people modifying the fs tree while we're | |
5220 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5221 | * dropped as we unlock the root node and parent nodes as we walk down | |
5222 | * the tree, assuming nothing will change. If something does change | |
5223 | * then we'll have stale information and drop references to blocks we've | |
5224 | * already dropped. | |
5225 | */ | |
5226 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5227 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5228 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5229 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 5230 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 5231 | path->slots[level] = 0; |
bd681513 | 5232 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5233 | memset(&wc->update_progress, 0, |
5234 | sizeof(wc->update_progress)); | |
9f3a7427 | 5235 | } else { |
9f3a7427 | 5236 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5237 | memcpy(&wc->update_progress, &key, |
5238 | sizeof(wc->update_progress)); | |
5239 | ||
6702ed49 | 5240 | level = root_item->drop_level; |
2c47e605 | 5241 | BUG_ON(level == 0); |
6702ed49 | 5242 | path->lowest_level = level; |
2c47e605 YZ |
5243 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5244 | path->lowest_level = 0; | |
5245 | if (ret < 0) { | |
5246 | err = ret; | |
79787eaa | 5247 | goto out_end_trans; |
9f3a7427 | 5248 | } |
1c4850e2 | 5249 | WARN_ON(ret > 0); |
2c47e605 | 5250 | |
7d9eb12c CM |
5251 | /* |
5252 | * unlock our path, this is safe because only this | |
5253 | * function is allowed to delete this snapshot | |
5254 | */ | |
5d4f98a2 | 5255 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5256 | |
5257 | level = btrfs_header_level(root->node); | |
5258 | while (1) { | |
5259 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 5260 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 5261 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5262 | |
2ff7e61e | 5263 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5264 | path->nodes[level]->start, |
3173a18f | 5265 | level, 1, &wc->refs[level], |
2c47e605 | 5266 | &wc->flags[level]); |
79787eaa JM |
5267 | if (ret < 0) { |
5268 | err = ret; | |
5269 | goto out_end_trans; | |
5270 | } | |
2c47e605 YZ |
5271 | BUG_ON(wc->refs[level] == 0); |
5272 | ||
5273 | if (level == root_item->drop_level) | |
5274 | break; | |
5275 | ||
5276 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5277 | path->locks[level] = 0; |
2c47e605 YZ |
5278 | WARN_ON(wc->refs[level] != 1); |
5279 | level--; | |
5280 | } | |
9f3a7427 | 5281 | } |
2c47e605 | 5282 | |
78c52d9e | 5283 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5284 | wc->level = level; |
5285 | wc->shared_level = -1; | |
5286 | wc->stage = DROP_REFERENCE; | |
5287 | wc->update_ref = update_ref; | |
5288 | wc->keep_locks = 0; | |
0b246afa | 5289 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5290 | |
d397712b | 5291 | while (1) { |
9d1a2a3a | 5292 | |
2c47e605 YZ |
5293 | ret = walk_down_tree(trans, root, path, wc); |
5294 | if (ret < 0) { | |
5295 | err = ret; | |
20524f02 | 5296 | break; |
2c47e605 | 5297 | } |
9aca1d51 | 5298 | |
2c47e605 YZ |
5299 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5300 | if (ret < 0) { | |
5301 | err = ret; | |
20524f02 | 5302 | break; |
2c47e605 YZ |
5303 | } |
5304 | ||
5305 | if (ret > 0) { | |
5306 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5307 | break; |
5308 | } | |
2c47e605 YZ |
5309 | |
5310 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5311 | wc->drop_level = wc->level; |
5312 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5313 | &wc->drop_progress, | |
5314 | path->slots[wc->drop_level]); | |
5315 | } | |
5316 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5317 | &wc->drop_progress); | |
5318 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
5319 | |
5320 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5321 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5322 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5323 | ret = btrfs_update_root(trans, tree_root, |
5324 | &root->root_key, | |
5325 | root_item); | |
79787eaa | 5326 | if (ret) { |
66642832 | 5327 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5328 | err = ret; |
5329 | goto out_end_trans; | |
5330 | } | |
2c47e605 | 5331 | |
3a45bb20 | 5332 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5333 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5334 | btrfs_debug(fs_info, |
5335 | "drop snapshot early exit"); | |
3c8f2422 JB |
5336 | err = -EAGAIN; |
5337 | goto out_free; | |
5338 | } | |
5339 | ||
a22285a6 | 5340 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
5341 | if (IS_ERR(trans)) { |
5342 | err = PTR_ERR(trans); | |
5343 | goto out_free; | |
5344 | } | |
3fd0a558 YZ |
5345 | if (block_rsv) |
5346 | trans->block_rsv = block_rsv; | |
c3e69d58 | 5347 | } |
20524f02 | 5348 | } |
b3b4aa74 | 5349 | btrfs_release_path(path); |
79787eaa JM |
5350 | if (err) |
5351 | goto out_end_trans; | |
2c47e605 | 5352 | |
ab9ce7d4 | 5353 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5354 | if (ret) { |
66642832 | 5355 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5356 | err = ret; |
79787eaa JM |
5357 | goto out_end_trans; |
5358 | } | |
2c47e605 | 5359 | |
76dda93c | 5360 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5361 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5362 | NULL, NULL); | |
79787eaa | 5363 | if (ret < 0) { |
66642832 | 5364 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5365 | err = ret; |
5366 | goto out_end_trans; | |
5367 | } else if (ret > 0) { | |
84cd948c JB |
5368 | /* if we fail to delete the orphan item this time |
5369 | * around, it'll get picked up the next time. | |
5370 | * | |
5371 | * The most common failure here is just -ENOENT. | |
5372 | */ | |
5373 | btrfs_del_orphan_item(trans, tree_root, | |
5374 | root->root_key.objectid); | |
76dda93c YZ |
5375 | } |
5376 | } | |
5377 | ||
27cdeb70 | 5378 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 5379 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
5380 | } else { |
5381 | free_extent_buffer(root->node); | |
5382 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 5383 | btrfs_put_fs_root(root); |
76dda93c | 5384 | } |
d29a9f62 | 5385 | root_dropped = true; |
79787eaa | 5386 | out_end_trans: |
3a45bb20 | 5387 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5388 | out_free: |
2c47e605 | 5389 | kfree(wc); |
5caf2a00 | 5390 | btrfs_free_path(path); |
cb1b69f4 | 5391 | out: |
d29a9f62 JB |
5392 | /* |
5393 | * So if we need to stop dropping the snapshot for whatever reason we | |
5394 | * need to make sure to add it back to the dead root list so that we | |
5395 | * keep trying to do the work later. This also cleans up roots if we | |
5396 | * don't have it in the radix (like when we recover after a power fail | |
5397 | * or unmount) so we don't leak memory. | |
5398 | */ | |
897ca819 | 5399 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5400 | btrfs_add_dead_root(root); |
90515e7f | 5401 | if (err && err != -EAGAIN) |
ab8d0fc4 | 5402 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 5403 | return err; |
20524f02 | 5404 | } |
9078a3e1 | 5405 | |
2c47e605 YZ |
5406 | /* |
5407 | * drop subtree rooted at tree block 'node'. | |
5408 | * | |
5409 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5410 | * only used by relocation code |
2c47e605 | 5411 | */ |
f82d02d9 YZ |
5412 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5413 | struct btrfs_root *root, | |
5414 | struct extent_buffer *node, | |
5415 | struct extent_buffer *parent) | |
5416 | { | |
0b246afa | 5417 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5418 | struct btrfs_path *path; |
2c47e605 | 5419 | struct walk_control *wc; |
f82d02d9 YZ |
5420 | int level; |
5421 | int parent_level; | |
5422 | int ret = 0; | |
5423 | int wret; | |
5424 | ||
2c47e605 YZ |
5425 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5426 | ||
f82d02d9 | 5427 | path = btrfs_alloc_path(); |
db5b493a TI |
5428 | if (!path) |
5429 | return -ENOMEM; | |
f82d02d9 | 5430 | |
2c47e605 | 5431 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5432 | if (!wc) { |
5433 | btrfs_free_path(path); | |
5434 | return -ENOMEM; | |
5435 | } | |
2c47e605 | 5436 | |
b9447ef8 | 5437 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
5438 | parent_level = btrfs_header_level(parent); |
5439 | extent_buffer_get(parent); | |
5440 | path->nodes[parent_level] = parent; | |
5441 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5442 | ||
b9447ef8 | 5443 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5444 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5445 | path->nodes[level] = node; |
5446 | path->slots[level] = 0; | |
bd681513 | 5447 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5448 | |
5449 | wc->refs[parent_level] = 1; | |
5450 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5451 | wc->level = level; | |
5452 | wc->shared_level = -1; | |
5453 | wc->stage = DROP_REFERENCE; | |
5454 | wc->update_ref = 0; | |
5455 | wc->keep_locks = 1; | |
0b246afa | 5456 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5457 | |
5458 | while (1) { | |
2c47e605 YZ |
5459 | wret = walk_down_tree(trans, root, path, wc); |
5460 | if (wret < 0) { | |
f82d02d9 | 5461 | ret = wret; |
f82d02d9 | 5462 | break; |
2c47e605 | 5463 | } |
f82d02d9 | 5464 | |
2c47e605 | 5465 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5466 | if (wret < 0) |
5467 | ret = wret; | |
5468 | if (wret != 0) | |
5469 | break; | |
5470 | } | |
5471 | ||
2c47e605 | 5472 | kfree(wc); |
f82d02d9 YZ |
5473 | btrfs_free_path(path); |
5474 | return ret; | |
5475 | } | |
5476 | ||
6d07bcec MX |
5477 | /* |
5478 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5479 | * space_info. takes mirrors into account. |
6d07bcec | 5480 | */ |
633c0aad | 5481 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
5482 | { |
5483 | struct btrfs_block_group_cache *block_group; | |
5484 | u64 free_bytes = 0; | |
5485 | int factor; | |
5486 | ||
01327610 | 5487 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5488 | if (list_empty(&sinfo->ro_bgs)) |
5489 | return 0; | |
5490 | ||
5491 | spin_lock(&sinfo->lock); | |
5492 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5493 | spin_lock(&block_group->lock); |
5494 | ||
5495 | if (!block_group->ro) { | |
5496 | spin_unlock(&block_group->lock); | |
5497 | continue; | |
5498 | } | |
5499 | ||
46df06b8 | 5500 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
5501 | free_bytes += (block_group->key.offset - |
5502 | btrfs_block_group_used(&block_group->item)) * | |
5503 | factor; | |
5504 | ||
5505 | spin_unlock(&block_group->lock); | |
5506 | } | |
6d07bcec MX |
5507 | spin_unlock(&sinfo->lock); |
5508 | ||
5509 | return free_bytes; | |
5510 | } | |
5511 | ||
2ff7e61e JM |
5512 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5513 | u64 start, u64 end) | |
acce952b | 5514 | { |
2ff7e61e | 5515 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5516 | } |
5517 | ||
499f377f JM |
5518 | /* |
5519 | * It used to be that old block groups would be left around forever. | |
5520 | * Iterating over them would be enough to trim unused space. Since we | |
5521 | * now automatically remove them, we also need to iterate over unallocated | |
5522 | * space. | |
5523 | * | |
5524 | * We don't want a transaction for this since the discard may take a | |
5525 | * substantial amount of time. We don't require that a transaction be | |
5526 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5527 | * to ensure that we're not discarding chunks that were released or |
5528 | * allocated in the current transaction. | |
499f377f JM |
5529 | * |
5530 | * Holding the chunks lock will prevent other threads from allocating | |
5531 | * or releasing chunks, but it won't prevent a running transaction | |
5532 | * from committing and releasing the memory that the pending chunks | |
5533 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5534 | * transaction and hold the commit root sem. We only need to hold |
5535 | * it while performing the free space search since we have already | |
5536 | * held back allocations. | |
499f377f | 5537 | */ |
8103d10b | 5538 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5539 | { |
8103d10b | 5540 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5541 | int ret; |
5542 | ||
5543 | *trimmed = 0; | |
5544 | ||
0be88e36 JM |
5545 | /* Discard not supported = nothing to do. */ |
5546 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5547 | return 0; | |
5548 | ||
52042d8e | 5549 | /* Not writable = nothing to do. */ |
ebbede42 | 5550 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5551 | return 0; |
5552 | ||
5553 | /* No free space = nothing to do. */ | |
5554 | if (device->total_bytes <= device->bytes_used) | |
5555 | return 0; | |
5556 | ||
5557 | ret = 0; | |
5558 | ||
5559 | while (1) { | |
fb456252 | 5560 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5561 | u64 bytes; |
5562 | ||
5563 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5564 | if (ret) | |
fee7acc3 | 5565 | break; |
499f377f | 5566 | |
929be17a NB |
5567 | find_first_clear_extent_bit(&device->alloc_state, start, |
5568 | &start, &end, | |
5569 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 NB |
5570 | |
5571 | /* Ensure we skip the reserved area in the first 1M */ | |
5572 | start = max_t(u64, start, SZ_1M); | |
5573 | ||
929be17a NB |
5574 | /* |
5575 | * If find_first_clear_extent_bit find a range that spans the | |
5576 | * end of the device it will set end to -1, in this case it's up | |
5577 | * to the caller to trim the value to the size of the device. | |
5578 | */ | |
5579 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5580 | |
929be17a | 5581 | len = end - start + 1; |
499f377f | 5582 | |
929be17a NB |
5583 | /* We didn't find any extents */ |
5584 | if (!len) { | |
499f377f | 5585 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5586 | ret = 0; |
499f377f JM |
5587 | break; |
5588 | } | |
5589 | ||
929be17a NB |
5590 | ret = btrfs_issue_discard(device->bdev, start, len, |
5591 | &bytes); | |
5592 | if (!ret) | |
5593 | set_extent_bits(&device->alloc_state, start, | |
5594 | start + bytes - 1, | |
5595 | CHUNK_TRIMMED); | |
499f377f JM |
5596 | mutex_unlock(&fs_info->chunk_mutex); |
5597 | ||
5598 | if (ret) | |
5599 | break; | |
5600 | ||
5601 | start += len; | |
5602 | *trimmed += bytes; | |
5603 | ||
5604 | if (fatal_signal_pending(current)) { | |
5605 | ret = -ERESTARTSYS; | |
5606 | break; | |
5607 | } | |
5608 | ||
5609 | cond_resched(); | |
5610 | } | |
5611 | ||
5612 | return ret; | |
5613 | } | |
5614 | ||
93bba24d QW |
5615 | /* |
5616 | * Trim the whole filesystem by: | |
5617 | * 1) trimming the free space in each block group | |
5618 | * 2) trimming the unallocated space on each device | |
5619 | * | |
5620 | * This will also continue trimming even if a block group or device encounters | |
5621 | * an error. The return value will be the last error, or 0 if nothing bad | |
5622 | * happens. | |
5623 | */ | |
2ff7e61e | 5624 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5625 | { |
f7039b1d | 5626 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
5627 | struct btrfs_device *device; |
5628 | struct list_head *devices; | |
f7039b1d | 5629 | u64 group_trimmed; |
07301df7 | 5630 | u64 range_end = U64_MAX; |
f7039b1d LD |
5631 | u64 start; |
5632 | u64 end; | |
5633 | u64 trimmed = 0; | |
93bba24d QW |
5634 | u64 bg_failed = 0; |
5635 | u64 dev_failed = 0; | |
5636 | int bg_ret = 0; | |
5637 | int dev_ret = 0; | |
f7039b1d LD |
5638 | int ret = 0; |
5639 | ||
07301df7 QW |
5640 | /* |
5641 | * Check range overflow if range->len is set. | |
5642 | * The default range->len is U64_MAX. | |
5643 | */ | |
5644 | if (range->len != U64_MAX && | |
5645 | check_add_overflow(range->start, range->len, &range_end)) | |
5646 | return -EINVAL; | |
5647 | ||
6ba9fc8e | 5648 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5649 | for (; cache; cache = btrfs_next_block_group(cache)) { |
07301df7 | 5650 | if (cache->key.objectid >= range_end) { |
f7039b1d LD |
5651 | btrfs_put_block_group(cache); |
5652 | break; | |
5653 | } | |
5654 | ||
5655 | start = max(range->start, cache->key.objectid); | |
07301df7 | 5656 | end = min(range_end, cache->key.objectid + cache->key.offset); |
f7039b1d LD |
5657 | |
5658 | if (end - start >= range->minlen) { | |
676f1f75 JB |
5659 | if (!btrfs_block_group_cache_done(cache)) { |
5660 | ret = btrfs_cache_block_group(cache, 0); | |
1be41b78 | 5661 | if (ret) { |
93bba24d QW |
5662 | bg_failed++; |
5663 | bg_ret = ret; | |
5664 | continue; | |
1be41b78 | 5665 | } |
676f1f75 | 5666 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5667 | if (ret) { |
93bba24d QW |
5668 | bg_failed++; |
5669 | bg_ret = ret; | |
5670 | continue; | |
1be41b78 | 5671 | } |
f7039b1d LD |
5672 | } |
5673 | ret = btrfs_trim_block_group(cache, | |
5674 | &group_trimmed, | |
5675 | start, | |
5676 | end, | |
5677 | range->minlen); | |
5678 | ||
5679 | trimmed += group_trimmed; | |
5680 | if (ret) { | |
93bba24d QW |
5681 | bg_failed++; |
5682 | bg_ret = ret; | |
5683 | continue; | |
f7039b1d LD |
5684 | } |
5685 | } | |
f7039b1d LD |
5686 | } |
5687 | ||
93bba24d QW |
5688 | if (bg_failed) |
5689 | btrfs_warn(fs_info, | |
5690 | "failed to trim %llu block group(s), last error %d", | |
5691 | bg_failed, bg_ret); | |
0b246afa | 5692 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5693 | devices = &fs_info->fs_devices->devices; |
5694 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5695 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5696 | if (ret) { |
5697 | dev_failed++; | |
5698 | dev_ret = ret; | |
499f377f | 5699 | break; |
93bba24d | 5700 | } |
499f377f JM |
5701 | |
5702 | trimmed += group_trimmed; | |
5703 | } | |
0b246afa | 5704 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5705 | |
93bba24d QW |
5706 | if (dev_failed) |
5707 | btrfs_warn(fs_info, | |
5708 | "failed to trim %llu device(s), last error %d", | |
5709 | dev_failed, dev_ret); | |
f7039b1d | 5710 | range->len = trimmed; |
93bba24d QW |
5711 | if (bg_ret) |
5712 | return bg_ret; | |
5713 | return dev_ret; | |
f7039b1d | 5714 | } |
8257b2dc MX |
5715 | |
5716 | /* | |
ea14b57f | 5717 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
5718 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
5719 | * data into the page cache through nocow before the subvolume is snapshoted, | |
5720 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 5721 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 5722 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 5723 | */ |
ea14b57f | 5724 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
5725 | { |
5726 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 5727 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
5728 | } |
5729 | ||
ea14b57f | 5730 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 5731 | { |
ea14b57f | 5732 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
5733 | return 0; |
5734 | ||
5735 | percpu_counter_inc(&root->subv_writers->counter); | |
5736 | /* | |
5737 | * Make sure counter is updated before we check for snapshot creation. | |
5738 | */ | |
5739 | smp_mb(); | |
ea14b57f DS |
5740 | if (atomic_read(&root->will_be_snapshotted)) { |
5741 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
5742 | return 0; |
5743 | } | |
5744 | return 1; | |
5745 | } | |
0bc19f90 | 5746 | |
0bc19f90 ZL |
5747 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
5748 | { | |
5749 | while (true) { | |
5750 | int ret; | |
5751 | ||
ea14b57f | 5752 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
5753 | if (ret) |
5754 | break; | |
4625956a PZ |
5755 | wait_var_event(&root->will_be_snapshotted, |
5756 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
5757 | } |
5758 | } |