]>
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" |
b0643e59 | 35 | #include "discard.h" |
fec577fb | 36 | |
709c0486 AJ |
37 | #undef SCRAMBLE_DELAYED_REFS |
38 | ||
9f9b8e8d | 39 | |
5d4f98a2 | 40 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
41 | struct btrfs_delayed_ref_node *node, u64 parent, |
42 | u64 root_objectid, u64 owner_objectid, | |
43 | u64 owner_offset, int refs_to_drop, | |
44 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
45 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
46 | struct extent_buffer *leaf, | |
47 | struct btrfs_extent_item *ei); | |
48 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
49 | u64 parent, u64 root_objectid, |
50 | u64 flags, u64 owner, u64 offset, | |
51 | struct btrfs_key *ins, int ref_mod); | |
52 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 53 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 54 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
55 | static int find_next_key(struct btrfs_path *path, int level, |
56 | struct btrfs_key *key); | |
6a63209f | 57 | |
32da5386 | 58 | static int block_group_bits(struct btrfs_block_group *cache, u64 bits) |
0f9dd46c JB |
59 | { |
60 | return (cache->flags & bits) == bits; | |
61 | } | |
62 | ||
6f410d1b JB |
63 | int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info, |
64 | u64 start, u64 num_bytes) | |
817d52f8 | 65 | { |
11833d66 | 66 | u64 end = start + num_bytes - 1; |
0b246afa | 67 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 68 | start, end, EXTENT_UPTODATE); |
0b246afa | 69 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 70 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
71 | return 0; |
72 | } | |
817d52f8 | 73 | |
32da5386 | 74 | void btrfs_free_excluded_extents(struct btrfs_block_group *cache) |
11833d66 | 75 | { |
9e715da8 | 76 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 77 | u64 start, end; |
817d52f8 | 78 | |
b3470b5d DS |
79 | start = cache->start; |
80 | end = start + cache->length - 1; | |
11833d66 | 81 | |
0b246afa | 82 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 83 | start, end, EXTENT_UPTODATE); |
0b246afa | 84 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 85 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
86 | } |
87 | ||
78192442 | 88 | static u64 generic_ref_to_space_flags(struct btrfs_ref *ref) |
0d9f824d | 89 | { |
ddf30cf0 QW |
90 | if (ref->type == BTRFS_REF_METADATA) { |
91 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
78192442 | 92 | return BTRFS_BLOCK_GROUP_SYSTEM; |
0d9f824d | 93 | else |
78192442 | 94 | return BTRFS_BLOCK_GROUP_METADATA; |
0d9f824d | 95 | } |
78192442 QW |
96 | return BTRFS_BLOCK_GROUP_DATA; |
97 | } | |
98 | ||
99 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, | |
100 | struct btrfs_ref *ref) | |
101 | { | |
102 | struct btrfs_space_info *space_info; | |
103 | u64 flags = generic_ref_to_space_flags(ref); | |
104 | ||
280c2908 | 105 | space_info = btrfs_find_space_info(fs_info, flags); |
78192442 QW |
106 | ASSERT(space_info); |
107 | percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len, | |
108 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
109 | } | |
110 | ||
111 | static void sub_pinned_bytes(struct btrfs_fs_info *fs_info, | |
112 | struct btrfs_ref *ref) | |
113 | { | |
114 | struct btrfs_space_info *space_info; | |
115 | u64 flags = generic_ref_to_space_flags(ref); | |
0d9f824d | 116 | |
280c2908 | 117 | space_info = btrfs_find_space_info(fs_info, flags); |
55e8196a | 118 | ASSERT(space_info); |
78192442 | 119 | percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len, |
dec59fa3 | 120 | BTRFS_TOTAL_BYTES_PINNED_BATCH); |
0d9f824d OS |
121 | } |
122 | ||
1a4ed8fd | 123 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 124 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
125 | { |
126 | int ret; | |
127 | struct btrfs_key key; | |
31840ae1 | 128 | struct btrfs_path *path; |
e02119d5 | 129 | |
31840ae1 | 130 | path = btrfs_alloc_path(); |
d8926bb3 MF |
131 | if (!path) |
132 | return -ENOMEM; | |
133 | ||
e02119d5 CM |
134 | key.objectid = start; |
135 | key.offset = len; | |
3173a18f | 136 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 137 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 138 | btrfs_free_path(path); |
7bb86316 CM |
139 | return ret; |
140 | } | |
141 | ||
a22285a6 | 142 | /* |
3173a18f | 143 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
144 | * |
145 | * the head node for delayed ref is used to store the sum of all the | |
146 | * reference count modifications queued up in the rbtree. the head | |
147 | * node may also store the extent flags to set. This way you can check | |
148 | * to see what the reference count and extent flags would be if all of | |
149 | * the delayed refs are not processed. | |
150 | */ | |
151 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 152 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 153 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
154 | { |
155 | struct btrfs_delayed_ref_head *head; | |
156 | struct btrfs_delayed_ref_root *delayed_refs; | |
157 | struct btrfs_path *path; | |
158 | struct btrfs_extent_item *ei; | |
159 | struct extent_buffer *leaf; | |
160 | struct btrfs_key key; | |
161 | u32 item_size; | |
162 | u64 num_refs; | |
163 | u64 extent_flags; | |
164 | int ret; | |
165 | ||
3173a18f JB |
166 | /* |
167 | * If we don't have skinny metadata, don't bother doing anything | |
168 | * different | |
169 | */ | |
0b246afa JM |
170 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
171 | offset = fs_info->nodesize; | |
3173a18f JB |
172 | metadata = 0; |
173 | } | |
174 | ||
a22285a6 YZ |
175 | path = btrfs_alloc_path(); |
176 | if (!path) | |
177 | return -ENOMEM; | |
178 | ||
a22285a6 YZ |
179 | if (!trans) { |
180 | path->skip_locking = 1; | |
181 | path->search_commit_root = 1; | |
182 | } | |
639eefc8 FDBM |
183 | |
184 | search_again: | |
185 | key.objectid = bytenr; | |
186 | key.offset = offset; | |
187 | if (metadata) | |
188 | key.type = BTRFS_METADATA_ITEM_KEY; | |
189 | else | |
190 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
191 | ||
0b246afa | 192 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
193 | if (ret < 0) |
194 | goto out_free; | |
195 | ||
3173a18f | 196 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
197 | if (path->slots[0]) { |
198 | path->slots[0]--; | |
199 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
200 | path->slots[0]); | |
201 | if (key.objectid == bytenr && | |
202 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 203 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
204 | ret = 0; |
205 | } | |
3173a18f JB |
206 | } |
207 | ||
a22285a6 YZ |
208 | if (ret == 0) { |
209 | leaf = path->nodes[0]; | |
210 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
211 | if (item_size >= sizeof(*ei)) { | |
212 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
213 | struct btrfs_extent_item); | |
214 | num_refs = btrfs_extent_refs(leaf, ei); | |
215 | extent_flags = btrfs_extent_flags(leaf, ei); | |
216 | } else { | |
ba3c2b19 NB |
217 | ret = -EINVAL; |
218 | btrfs_print_v0_err(fs_info); | |
219 | if (trans) | |
220 | btrfs_abort_transaction(trans, ret); | |
221 | else | |
222 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
223 | ||
224 | goto out_free; | |
a22285a6 | 225 | } |
ba3c2b19 | 226 | |
a22285a6 YZ |
227 | BUG_ON(num_refs == 0); |
228 | } else { | |
229 | num_refs = 0; | |
230 | extent_flags = 0; | |
231 | ret = 0; | |
232 | } | |
233 | ||
234 | if (!trans) | |
235 | goto out; | |
236 | ||
237 | delayed_refs = &trans->transaction->delayed_refs; | |
238 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 239 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
240 | if (head) { |
241 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 242 | refcount_inc(&head->refs); |
a22285a6 YZ |
243 | spin_unlock(&delayed_refs->lock); |
244 | ||
b3b4aa74 | 245 | btrfs_release_path(path); |
a22285a6 | 246 | |
8cc33e5c DS |
247 | /* |
248 | * Mutex was contended, block until it's released and try | |
249 | * again | |
250 | */ | |
a22285a6 YZ |
251 | mutex_lock(&head->mutex); |
252 | mutex_unlock(&head->mutex); | |
d278850e | 253 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 254 | goto search_again; |
a22285a6 | 255 | } |
d7df2c79 | 256 | spin_lock(&head->lock); |
a22285a6 YZ |
257 | if (head->extent_op && head->extent_op->update_flags) |
258 | extent_flags |= head->extent_op->flags_to_set; | |
259 | else | |
260 | BUG_ON(num_refs == 0); | |
261 | ||
d278850e | 262 | num_refs += head->ref_mod; |
d7df2c79 | 263 | spin_unlock(&head->lock); |
a22285a6 YZ |
264 | mutex_unlock(&head->mutex); |
265 | } | |
266 | spin_unlock(&delayed_refs->lock); | |
267 | out: | |
268 | WARN_ON(num_refs == 0); | |
269 | if (refs) | |
270 | *refs = num_refs; | |
271 | if (flags) | |
272 | *flags = extent_flags; | |
273 | out_free: | |
274 | btrfs_free_path(path); | |
275 | return ret; | |
276 | } | |
277 | ||
d8d5f3e1 CM |
278 | /* |
279 | * Back reference rules. Back refs have three main goals: | |
280 | * | |
281 | * 1) differentiate between all holders of references to an extent so that | |
282 | * when a reference is dropped we can make sure it was a valid reference | |
283 | * before freeing the extent. | |
284 | * | |
285 | * 2) Provide enough information to quickly find the holders of an extent | |
286 | * if we notice a given block is corrupted or bad. | |
287 | * | |
288 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
289 | * maintenance. This is actually the same as #2, but with a slightly | |
290 | * different use case. | |
291 | * | |
5d4f98a2 YZ |
292 | * There are two kinds of back refs. The implicit back refs is optimized |
293 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
294 | * back refs of this kind provide information about the block's owner tree | |
295 | * and the pointer's key. These information allow us to find the block by | |
296 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
297 | * referenced by their owner trees. The location of tree block is recorded | |
298 | * in the back refs. Actually the full back refs is generic, and can be | |
299 | * used in all cases the implicit back refs is used. The major shortcoming | |
300 | * of the full back refs is its overhead. Every time a tree block gets | |
301 | * COWed, we have to update back refs entry for all pointers in it. | |
302 | * | |
303 | * For a newly allocated tree block, we use implicit back refs for | |
304 | * pointers in it. This means most tree related operations only involve | |
305 | * implicit back refs. For a tree block created in old transaction, the | |
306 | * only way to drop a reference to it is COW it. So we can detect the | |
307 | * event that tree block loses its owner tree's reference and do the | |
308 | * back refs conversion. | |
309 | * | |
01327610 | 310 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
311 | * |
312 | * The reference count of the block is one and the tree is the block's | |
313 | * owner tree. Nothing to do in this case. | |
314 | * | |
315 | * The reference count of the block is one and the tree is not the | |
316 | * block's owner tree. In this case, full back refs is used for pointers | |
317 | * in the block. Remove these full back refs, add implicit back refs for | |
318 | * every pointers in the new block. | |
319 | * | |
320 | * The reference count of the block is greater than one and the tree is | |
321 | * the block's owner tree. In this case, implicit back refs is used for | |
322 | * pointers in the block. Add full back refs for every pointers in the | |
323 | * block, increase lower level extents' reference counts. The original | |
324 | * implicit back refs are entailed to the new block. | |
325 | * | |
326 | * The reference count of the block is greater than one and the tree is | |
327 | * not the block's owner tree. Add implicit back refs for every pointer in | |
328 | * the new block, increase lower level extents' reference count. | |
329 | * | |
330 | * Back Reference Key composing: | |
331 | * | |
332 | * The key objectid corresponds to the first byte in the extent, | |
333 | * The key type is used to differentiate between types of back refs. | |
334 | * There are different meanings of the key offset for different types | |
335 | * of back refs. | |
336 | * | |
d8d5f3e1 CM |
337 | * File extents can be referenced by: |
338 | * | |
339 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 340 | * - different files inside a single subvolume |
d8d5f3e1 CM |
341 | * - different offsets inside a file (bookend extents in file.c) |
342 | * | |
5d4f98a2 | 343 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
344 | * |
345 | * - Objectid of the subvolume root | |
d8d5f3e1 | 346 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
347 | * - original offset in the file |
348 | * - how many bookend extents | |
d8d5f3e1 | 349 | * |
5d4f98a2 YZ |
350 | * The key offset for the implicit back refs is hash of the first |
351 | * three fields. | |
d8d5f3e1 | 352 | * |
5d4f98a2 | 353 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 354 | * |
5d4f98a2 | 355 | * - number of pointers in the tree leaf |
d8d5f3e1 | 356 | * |
5d4f98a2 YZ |
357 | * The key offset for the implicit back refs is the first byte of |
358 | * the tree leaf | |
d8d5f3e1 | 359 | * |
5d4f98a2 YZ |
360 | * When a file extent is allocated, The implicit back refs is used. |
361 | * the fields are filled in: | |
d8d5f3e1 | 362 | * |
5d4f98a2 | 363 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 364 | * |
5d4f98a2 YZ |
365 | * When a file extent is removed file truncation, we find the |
366 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 367 | * |
5d4f98a2 | 368 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 369 | * |
5d4f98a2 | 370 | * Btree extents can be referenced by: |
d8d5f3e1 | 371 | * |
5d4f98a2 | 372 | * - Different subvolumes |
d8d5f3e1 | 373 | * |
5d4f98a2 YZ |
374 | * Both the implicit back refs and the full back refs for tree blocks |
375 | * only consist of key. The key offset for the implicit back refs is | |
376 | * objectid of block's owner tree. The key offset for the full back refs | |
377 | * is the first byte of parent block. | |
d8d5f3e1 | 378 | * |
5d4f98a2 YZ |
379 | * When implicit back refs is used, information about the lowest key and |
380 | * level of the tree block are required. These information are stored in | |
381 | * tree block info structure. | |
d8d5f3e1 | 382 | */ |
31840ae1 | 383 | |
167ce953 LB |
384 | /* |
385 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 386 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
387 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
388 | */ | |
389 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
390 | struct btrfs_extent_inline_ref *iref, | |
391 | enum btrfs_inline_ref_type is_data) | |
392 | { | |
393 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 394 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
395 | |
396 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
397 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
398 | type == BTRFS_SHARED_DATA_REF_KEY || | |
399 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
400 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 401 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 402 | return type; |
64ecdb64 LB |
403 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
404 | ASSERT(eb->fs_info); | |
405 | /* | |
406 | * Every shared one has parent tree | |
407 | * block, which must be aligned to | |
408 | * nodesize. | |
409 | */ | |
410 | if (offset && | |
411 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
412 | return type; | |
413 | } | |
167ce953 | 414 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 415 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 416 | return type; |
64ecdb64 LB |
417 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
418 | ASSERT(eb->fs_info); | |
419 | /* | |
420 | * Every shared one has parent tree | |
421 | * block, which must be aligned to | |
422 | * nodesize. | |
423 | */ | |
424 | if (offset && | |
425 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
426 | return type; | |
427 | } | |
167ce953 LB |
428 | } else { |
429 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
430 | return type; | |
431 | } | |
432 | } | |
433 | ||
434 | btrfs_print_leaf((struct extent_buffer *)eb); | |
435 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
436 | eb->start, type); | |
437 | WARN_ON(1); | |
438 | ||
439 | return BTRFS_REF_TYPE_INVALID; | |
440 | } | |
441 | ||
0785a9aa | 442 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
443 | { |
444 | u32 high_crc = ~(u32)0; | |
445 | u32 low_crc = ~(u32)0; | |
446 | __le64 lenum; | |
447 | ||
448 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 449 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 450 | lenum = cpu_to_le64(owner); |
65019df8 | 451 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 452 | lenum = cpu_to_le64(offset); |
65019df8 | 453 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
454 | |
455 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
456 | } | |
457 | ||
458 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
459 | struct btrfs_extent_data_ref *ref) | |
460 | { | |
461 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
462 | btrfs_extent_data_ref_objectid(leaf, ref), | |
463 | btrfs_extent_data_ref_offset(leaf, ref)); | |
464 | } | |
465 | ||
466 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
467 | struct btrfs_extent_data_ref *ref, | |
468 | u64 root_objectid, u64 owner, u64 offset) | |
469 | { | |
470 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
471 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
472 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
473 | return 0; | |
474 | return 1; | |
475 | } | |
476 | ||
477 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
478 | struct btrfs_path *path, |
479 | u64 bytenr, u64 parent, | |
480 | u64 root_objectid, | |
481 | u64 owner, u64 offset) | |
482 | { | |
bd1d53ef | 483 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
484 | struct btrfs_key key; |
485 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 486 | struct extent_buffer *leaf; |
5d4f98a2 | 487 | u32 nritems; |
74493f7a | 488 | int ret; |
5d4f98a2 YZ |
489 | int recow; |
490 | int err = -ENOENT; | |
74493f7a | 491 | |
31840ae1 | 492 | key.objectid = bytenr; |
5d4f98a2 YZ |
493 | if (parent) { |
494 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
495 | key.offset = parent; | |
496 | } else { | |
497 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
498 | key.offset = hash_extent_data_ref(root_objectid, | |
499 | owner, offset); | |
500 | } | |
501 | again: | |
502 | recow = 0; | |
503 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
504 | if (ret < 0) { | |
505 | err = ret; | |
506 | goto fail; | |
507 | } | |
31840ae1 | 508 | |
5d4f98a2 YZ |
509 | if (parent) { |
510 | if (!ret) | |
511 | return 0; | |
5d4f98a2 | 512 | goto fail; |
31840ae1 ZY |
513 | } |
514 | ||
515 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
516 | nritems = btrfs_header_nritems(leaf); |
517 | while (1) { | |
518 | if (path->slots[0] >= nritems) { | |
519 | ret = btrfs_next_leaf(root, path); | |
520 | if (ret < 0) | |
521 | err = ret; | |
522 | if (ret) | |
523 | goto fail; | |
524 | ||
525 | leaf = path->nodes[0]; | |
526 | nritems = btrfs_header_nritems(leaf); | |
527 | recow = 1; | |
528 | } | |
529 | ||
530 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
531 | if (key.objectid != bytenr || | |
532 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
533 | goto fail; | |
534 | ||
535 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
536 | struct btrfs_extent_data_ref); | |
537 | ||
538 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
539 | owner, offset)) { | |
540 | if (recow) { | |
b3b4aa74 | 541 | btrfs_release_path(path); |
5d4f98a2 YZ |
542 | goto again; |
543 | } | |
544 | err = 0; | |
545 | break; | |
546 | } | |
547 | path->slots[0]++; | |
31840ae1 | 548 | } |
5d4f98a2 YZ |
549 | fail: |
550 | return err; | |
31840ae1 ZY |
551 | } |
552 | ||
5d4f98a2 | 553 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
554 | struct btrfs_path *path, |
555 | u64 bytenr, u64 parent, | |
556 | u64 root_objectid, u64 owner, | |
557 | u64 offset, int refs_to_add) | |
31840ae1 | 558 | { |
62b895af | 559 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
560 | struct btrfs_key key; |
561 | struct extent_buffer *leaf; | |
5d4f98a2 | 562 | u32 size; |
31840ae1 ZY |
563 | u32 num_refs; |
564 | int ret; | |
74493f7a | 565 | |
74493f7a | 566 | key.objectid = bytenr; |
5d4f98a2 YZ |
567 | if (parent) { |
568 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
569 | key.offset = parent; | |
570 | size = sizeof(struct btrfs_shared_data_ref); | |
571 | } else { | |
572 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
573 | key.offset = hash_extent_data_ref(root_objectid, | |
574 | owner, offset); | |
575 | size = sizeof(struct btrfs_extent_data_ref); | |
576 | } | |
74493f7a | 577 | |
5d4f98a2 YZ |
578 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
579 | if (ret && ret != -EEXIST) | |
580 | goto fail; | |
581 | ||
582 | leaf = path->nodes[0]; | |
583 | if (parent) { | |
584 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 585 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
586 | struct btrfs_shared_data_ref); |
587 | if (ret == 0) { | |
588 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
589 | } else { | |
590 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
591 | num_refs += refs_to_add; | |
592 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 593 | } |
5d4f98a2 YZ |
594 | } else { |
595 | struct btrfs_extent_data_ref *ref; | |
596 | while (ret == -EEXIST) { | |
597 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
598 | struct btrfs_extent_data_ref); | |
599 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
600 | owner, offset)) | |
601 | break; | |
b3b4aa74 | 602 | btrfs_release_path(path); |
5d4f98a2 YZ |
603 | key.offset++; |
604 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
605 | size); | |
606 | if (ret && ret != -EEXIST) | |
607 | goto fail; | |
31840ae1 | 608 | |
5d4f98a2 YZ |
609 | leaf = path->nodes[0]; |
610 | } | |
611 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
612 | struct btrfs_extent_data_ref); | |
613 | if (ret == 0) { | |
614 | btrfs_set_extent_data_ref_root(leaf, ref, | |
615 | root_objectid); | |
616 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
617 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
618 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
619 | } else { | |
620 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
621 | num_refs += refs_to_add; | |
622 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 623 | } |
31840ae1 | 624 | } |
5d4f98a2 YZ |
625 | btrfs_mark_buffer_dirty(leaf); |
626 | ret = 0; | |
627 | fail: | |
b3b4aa74 | 628 | btrfs_release_path(path); |
7bb86316 | 629 | return ret; |
74493f7a CM |
630 | } |
631 | ||
5d4f98a2 | 632 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 633 | struct btrfs_path *path, |
fcebe456 | 634 | int refs_to_drop, int *last_ref) |
31840ae1 | 635 | { |
5d4f98a2 YZ |
636 | struct btrfs_key key; |
637 | struct btrfs_extent_data_ref *ref1 = NULL; | |
638 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 639 | struct extent_buffer *leaf; |
5d4f98a2 | 640 | u32 num_refs = 0; |
31840ae1 ZY |
641 | int ret = 0; |
642 | ||
643 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
644 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
645 | ||
646 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
647 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
648 | struct btrfs_extent_data_ref); | |
649 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
650 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
651 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
652 | struct btrfs_shared_data_ref); | |
653 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 654 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
655 | btrfs_print_v0_err(trans->fs_info); |
656 | btrfs_abort_transaction(trans, -EINVAL); | |
657 | return -EINVAL; | |
5d4f98a2 YZ |
658 | } else { |
659 | BUG(); | |
660 | } | |
661 | ||
56bec294 CM |
662 | BUG_ON(num_refs < refs_to_drop); |
663 | num_refs -= refs_to_drop; | |
5d4f98a2 | 664 | |
31840ae1 | 665 | if (num_refs == 0) { |
e9f6290d | 666 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 667 | *last_ref = 1; |
31840ae1 | 668 | } else { |
5d4f98a2 YZ |
669 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
670 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
671 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
672 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
673 | btrfs_mark_buffer_dirty(leaf); |
674 | } | |
31840ae1 ZY |
675 | return ret; |
676 | } | |
677 | ||
9ed0dea0 | 678 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 679 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 680 | { |
5d4f98a2 YZ |
681 | struct btrfs_key key; |
682 | struct extent_buffer *leaf; | |
683 | struct btrfs_extent_data_ref *ref1; | |
684 | struct btrfs_shared_data_ref *ref2; | |
685 | u32 num_refs = 0; | |
3de28d57 | 686 | int type; |
5d4f98a2 YZ |
687 | |
688 | leaf = path->nodes[0]; | |
689 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
690 | |
691 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 692 | if (iref) { |
3de28d57 LB |
693 | /* |
694 | * If type is invalid, we should have bailed out earlier than | |
695 | * this call. | |
696 | */ | |
697 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
698 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
699 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
700 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
701 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
702 | } else { | |
703 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
704 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
705 | } | |
706 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
707 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
708 | struct btrfs_extent_data_ref); | |
709 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
710 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
711 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
712 | struct btrfs_shared_data_ref); | |
713 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
714 | } else { |
715 | WARN_ON(1); | |
716 | } | |
717 | return num_refs; | |
718 | } | |
15916de8 | 719 | |
5d4f98a2 | 720 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
721 | struct btrfs_path *path, |
722 | u64 bytenr, u64 parent, | |
723 | u64 root_objectid) | |
1f3c79a2 | 724 | { |
b8582eea | 725 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 726 | struct btrfs_key key; |
1f3c79a2 | 727 | int ret; |
1f3c79a2 | 728 | |
5d4f98a2 YZ |
729 | key.objectid = bytenr; |
730 | if (parent) { | |
731 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
732 | key.offset = parent; | |
733 | } else { | |
734 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
735 | key.offset = root_objectid; | |
1f3c79a2 LH |
736 | } |
737 | ||
5d4f98a2 YZ |
738 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
739 | if (ret > 0) | |
740 | ret = -ENOENT; | |
5d4f98a2 | 741 | return ret; |
1f3c79a2 LH |
742 | } |
743 | ||
5d4f98a2 | 744 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
745 | struct btrfs_path *path, |
746 | u64 bytenr, u64 parent, | |
747 | u64 root_objectid) | |
31840ae1 | 748 | { |
5d4f98a2 | 749 | struct btrfs_key key; |
31840ae1 | 750 | int ret; |
31840ae1 | 751 | |
5d4f98a2 YZ |
752 | key.objectid = bytenr; |
753 | if (parent) { | |
754 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
755 | key.offset = parent; | |
756 | } else { | |
757 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
758 | key.offset = root_objectid; | |
759 | } | |
760 | ||
10728404 | 761 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 762 | path, &key, 0); |
b3b4aa74 | 763 | btrfs_release_path(path); |
31840ae1 ZY |
764 | return ret; |
765 | } | |
766 | ||
5d4f98a2 | 767 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 768 | { |
5d4f98a2 YZ |
769 | int type; |
770 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
771 | if (parent > 0) | |
772 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
773 | else | |
774 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
775 | } else { | |
776 | if (parent > 0) | |
777 | type = BTRFS_SHARED_DATA_REF_KEY; | |
778 | else | |
779 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
780 | } | |
781 | return type; | |
31840ae1 | 782 | } |
56bec294 | 783 | |
2c47e605 YZ |
784 | static int find_next_key(struct btrfs_path *path, int level, |
785 | struct btrfs_key *key) | |
56bec294 | 786 | |
02217ed2 | 787 | { |
2c47e605 | 788 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
789 | if (!path->nodes[level]) |
790 | break; | |
5d4f98a2 YZ |
791 | if (path->slots[level] + 1 >= |
792 | btrfs_header_nritems(path->nodes[level])) | |
793 | continue; | |
794 | if (level == 0) | |
795 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
796 | path->slots[level] + 1); | |
797 | else | |
798 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
799 | path->slots[level] + 1); | |
800 | return 0; | |
801 | } | |
802 | return 1; | |
803 | } | |
037e6390 | 804 | |
5d4f98a2 YZ |
805 | /* |
806 | * look for inline back ref. if back ref is found, *ref_ret is set | |
807 | * to the address of inline back ref, and 0 is returned. | |
808 | * | |
809 | * if back ref isn't found, *ref_ret is set to the address where it | |
810 | * should be inserted, and -ENOENT is returned. | |
811 | * | |
812 | * if insert is true and there are too many inline back refs, the path | |
813 | * points to the extent item, and -EAGAIN is returned. | |
814 | * | |
815 | * NOTE: inline back refs are ordered in the same way that back ref | |
816 | * items in the tree are ordered. | |
817 | */ | |
818 | static noinline_for_stack | |
819 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
820 | struct btrfs_path *path, |
821 | struct btrfs_extent_inline_ref **ref_ret, | |
822 | u64 bytenr, u64 num_bytes, | |
823 | u64 parent, u64 root_objectid, | |
824 | u64 owner, u64 offset, int insert) | |
825 | { | |
867cc1fb | 826 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 827 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
828 | struct btrfs_key key; |
829 | struct extent_buffer *leaf; | |
830 | struct btrfs_extent_item *ei; | |
831 | struct btrfs_extent_inline_ref *iref; | |
832 | u64 flags; | |
833 | u64 item_size; | |
834 | unsigned long ptr; | |
835 | unsigned long end; | |
836 | int extra_size; | |
837 | int type; | |
838 | int want; | |
839 | int ret; | |
840 | int err = 0; | |
0b246afa | 841 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 842 | int needed; |
26b8003f | 843 | |
db94535d | 844 | key.objectid = bytenr; |
31840ae1 | 845 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 846 | key.offset = num_bytes; |
31840ae1 | 847 | |
5d4f98a2 YZ |
848 | want = extent_ref_type(parent, owner); |
849 | if (insert) { | |
850 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 851 | path->keep_locks = 1; |
5d4f98a2 YZ |
852 | } else |
853 | extra_size = -1; | |
3173a18f JB |
854 | |
855 | /* | |
16d1c062 NB |
856 | * Owner is our level, so we can just add one to get the level for the |
857 | * block we are interested in. | |
3173a18f JB |
858 | */ |
859 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
860 | key.type = BTRFS_METADATA_ITEM_KEY; | |
861 | key.offset = owner; | |
862 | } | |
863 | ||
864 | again: | |
5d4f98a2 | 865 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 866 | if (ret < 0) { |
5d4f98a2 YZ |
867 | err = ret; |
868 | goto out; | |
869 | } | |
3173a18f JB |
870 | |
871 | /* | |
872 | * We may be a newly converted file system which still has the old fat | |
873 | * extent entries for metadata, so try and see if we have one of those. | |
874 | */ | |
875 | if (ret > 0 && skinny_metadata) { | |
876 | skinny_metadata = false; | |
877 | if (path->slots[0]) { | |
878 | path->slots[0]--; | |
879 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
880 | path->slots[0]); | |
881 | if (key.objectid == bytenr && | |
882 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
883 | key.offset == num_bytes) | |
884 | ret = 0; | |
885 | } | |
886 | if (ret) { | |
9ce49a0b | 887 | key.objectid = bytenr; |
3173a18f JB |
888 | key.type = BTRFS_EXTENT_ITEM_KEY; |
889 | key.offset = num_bytes; | |
890 | btrfs_release_path(path); | |
891 | goto again; | |
892 | } | |
893 | } | |
894 | ||
79787eaa JM |
895 | if (ret && !insert) { |
896 | err = -ENOENT; | |
897 | goto out; | |
fae7f21c | 898 | } else if (WARN_ON(ret)) { |
492104c8 | 899 | err = -EIO; |
492104c8 | 900 | goto out; |
79787eaa | 901 | } |
5d4f98a2 YZ |
902 | |
903 | leaf = path->nodes[0]; | |
904 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 905 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
906 | err = -EINVAL; |
907 | btrfs_print_v0_err(fs_info); | |
908 | btrfs_abort_transaction(trans, err); | |
909 | goto out; | |
910 | } | |
5d4f98a2 | 911 | |
5d4f98a2 YZ |
912 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
913 | flags = btrfs_extent_flags(leaf, ei); | |
914 | ||
915 | ptr = (unsigned long)(ei + 1); | |
916 | end = (unsigned long)ei + item_size; | |
917 | ||
3173a18f | 918 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
919 | ptr += sizeof(struct btrfs_tree_block_info); |
920 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
921 | } |
922 | ||
3de28d57 LB |
923 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
924 | needed = BTRFS_REF_TYPE_DATA; | |
925 | else | |
926 | needed = BTRFS_REF_TYPE_BLOCK; | |
927 | ||
5d4f98a2 YZ |
928 | err = -ENOENT; |
929 | while (1) { | |
930 | if (ptr >= end) { | |
931 | WARN_ON(ptr > end); | |
932 | break; | |
933 | } | |
934 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
935 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
936 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 937 | err = -EUCLEAN; |
3de28d57 LB |
938 | goto out; |
939 | } | |
940 | ||
5d4f98a2 YZ |
941 | if (want < type) |
942 | break; | |
943 | if (want > type) { | |
944 | ptr += btrfs_extent_inline_ref_size(type); | |
945 | continue; | |
946 | } | |
947 | ||
948 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
949 | struct btrfs_extent_data_ref *dref; | |
950 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
951 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
952 | owner, offset)) { | |
953 | err = 0; | |
954 | break; | |
955 | } | |
956 | if (hash_extent_data_ref_item(leaf, dref) < | |
957 | hash_extent_data_ref(root_objectid, owner, offset)) | |
958 | break; | |
959 | } else { | |
960 | u64 ref_offset; | |
961 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
962 | if (parent > 0) { | |
963 | if (parent == ref_offset) { | |
964 | err = 0; | |
965 | break; | |
966 | } | |
967 | if (ref_offset < parent) | |
968 | break; | |
969 | } else { | |
970 | if (root_objectid == ref_offset) { | |
971 | err = 0; | |
972 | break; | |
973 | } | |
974 | if (ref_offset < root_objectid) | |
975 | break; | |
976 | } | |
977 | } | |
978 | ptr += btrfs_extent_inline_ref_size(type); | |
979 | } | |
980 | if (err == -ENOENT && insert) { | |
981 | if (item_size + extra_size >= | |
982 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
983 | err = -EAGAIN; | |
984 | goto out; | |
985 | } | |
986 | /* | |
987 | * To add new inline back ref, we have to make sure | |
988 | * there is no corresponding back ref item. | |
989 | * For simplicity, we just do not add new inline back | |
990 | * ref if there is any kind of item for this block | |
991 | */ | |
2c47e605 YZ |
992 | if (find_next_key(path, 0, &key) == 0 && |
993 | key.objectid == bytenr && | |
85d4198e | 994 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
995 | err = -EAGAIN; |
996 | goto out; | |
997 | } | |
998 | } | |
999 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1000 | out: | |
85d4198e | 1001 | if (insert) { |
5d4f98a2 YZ |
1002 | path->keep_locks = 0; |
1003 | btrfs_unlock_up_safe(path, 1); | |
1004 | } | |
1005 | return err; | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * helper to add new inline back ref | |
1010 | */ | |
1011 | static noinline_for_stack | |
87bde3cd | 1012 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1013 | struct btrfs_path *path, |
1014 | struct btrfs_extent_inline_ref *iref, | |
1015 | u64 parent, u64 root_objectid, | |
1016 | u64 owner, u64 offset, int refs_to_add, | |
1017 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1018 | { |
1019 | struct extent_buffer *leaf; | |
1020 | struct btrfs_extent_item *ei; | |
1021 | unsigned long ptr; | |
1022 | unsigned long end; | |
1023 | unsigned long item_offset; | |
1024 | u64 refs; | |
1025 | int size; | |
1026 | int type; | |
5d4f98a2 YZ |
1027 | |
1028 | leaf = path->nodes[0]; | |
1029 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1030 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1031 | ||
1032 | type = extent_ref_type(parent, owner); | |
1033 | size = btrfs_extent_inline_ref_size(type); | |
1034 | ||
c71dd880 | 1035 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1036 | |
1037 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1038 | refs = btrfs_extent_refs(leaf, ei); | |
1039 | refs += refs_to_add; | |
1040 | btrfs_set_extent_refs(leaf, ei, refs); | |
1041 | if (extent_op) | |
1042 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1043 | ||
1044 | ptr = (unsigned long)ei + item_offset; | |
1045 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1046 | if (ptr < end - size) | |
1047 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1048 | end - size - ptr); | |
1049 | ||
1050 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1051 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1052 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1053 | struct btrfs_extent_data_ref *dref; | |
1054 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1055 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1056 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1057 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1058 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1059 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1060 | struct btrfs_shared_data_ref *sref; | |
1061 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1062 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1063 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1064 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1065 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1066 | } else { | |
1067 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1068 | } | |
1069 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1070 | } |
1071 | ||
1072 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1073 | struct btrfs_path *path, |
1074 | struct btrfs_extent_inline_ref **ref_ret, | |
1075 | u64 bytenr, u64 num_bytes, u64 parent, | |
1076 | u64 root_objectid, u64 owner, u64 offset) | |
1077 | { | |
1078 | int ret; | |
1079 | ||
867cc1fb NB |
1080 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1081 | num_bytes, parent, root_objectid, | |
1082 | owner, offset, 0); | |
5d4f98a2 | 1083 | if (ret != -ENOENT) |
54aa1f4d | 1084 | return ret; |
5d4f98a2 | 1085 | |
b3b4aa74 | 1086 | btrfs_release_path(path); |
5d4f98a2 YZ |
1087 | *ref_ret = NULL; |
1088 | ||
1089 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1090 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1091 | root_objectid); | |
5d4f98a2 | 1092 | } else { |
bd1d53ef NB |
1093 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1094 | root_objectid, owner, offset); | |
b9473439 | 1095 | } |
5d4f98a2 YZ |
1096 | return ret; |
1097 | } | |
31840ae1 | 1098 | |
5d4f98a2 YZ |
1099 | /* |
1100 | * helper to update/remove inline back ref | |
1101 | */ | |
1102 | static noinline_for_stack | |
61a18f1c | 1103 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1104 | struct btrfs_extent_inline_ref *iref, |
1105 | int refs_to_mod, | |
fcebe456 JB |
1106 | struct btrfs_delayed_extent_op *extent_op, |
1107 | int *last_ref) | |
5d4f98a2 | 1108 | { |
61a18f1c | 1109 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1110 | struct btrfs_extent_item *ei; |
1111 | struct btrfs_extent_data_ref *dref = NULL; | |
1112 | struct btrfs_shared_data_ref *sref = NULL; | |
1113 | unsigned long ptr; | |
1114 | unsigned long end; | |
1115 | u32 item_size; | |
1116 | int size; | |
1117 | int type; | |
5d4f98a2 YZ |
1118 | u64 refs; |
1119 | ||
5d4f98a2 YZ |
1120 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1121 | refs = btrfs_extent_refs(leaf, ei); | |
1122 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1123 | refs += refs_to_mod; | |
1124 | btrfs_set_extent_refs(leaf, ei, refs); | |
1125 | if (extent_op) | |
1126 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1127 | ||
3de28d57 LB |
1128 | /* |
1129 | * If type is invalid, we should have bailed out after | |
1130 | * lookup_inline_extent_backref(). | |
1131 | */ | |
1132 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1133 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1134 | |
1135 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1136 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1137 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1138 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1139 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1140 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1141 | } else { | |
1142 | refs = 1; | |
1143 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1144 | } |
31840ae1 | 1145 | |
5d4f98a2 YZ |
1146 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1147 | refs += refs_to_mod; | |
1148 | ||
1149 | if (refs > 0) { | |
1150 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1151 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1152 | else | |
1153 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1154 | } else { | |
fcebe456 | 1155 | *last_ref = 1; |
5d4f98a2 YZ |
1156 | size = btrfs_extent_inline_ref_size(type); |
1157 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1158 | ptr = (unsigned long)iref; | |
1159 | end = (unsigned long)ei + item_size; | |
1160 | if (ptr + size < end) | |
1161 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1162 | end - ptr - size); | |
1163 | item_size -= size; | |
78ac4f9e | 1164 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1165 | } |
1166 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1167 | } |
1168 | ||
1169 | static noinline_for_stack | |
1170 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1171 | struct btrfs_path *path, |
1172 | u64 bytenr, u64 num_bytes, u64 parent, | |
1173 | u64 root_objectid, u64 owner, | |
1174 | u64 offset, int refs_to_add, | |
1175 | struct btrfs_delayed_extent_op *extent_op) | |
1176 | { | |
1177 | struct btrfs_extent_inline_ref *iref; | |
1178 | int ret; | |
1179 | ||
867cc1fb NB |
1180 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1181 | num_bytes, parent, root_objectid, | |
1182 | owner, offset, 1); | |
5d4f98a2 YZ |
1183 | if (ret == 0) { |
1184 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1185 | update_inline_extent_backref(path, iref, refs_to_add, |
1186 | extent_op, NULL); | |
5d4f98a2 | 1187 | } else if (ret == -ENOENT) { |
a639cdeb | 1188 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1189 | root_objectid, owner, offset, |
1190 | refs_to_add, extent_op); | |
1191 | ret = 0; | |
771ed689 | 1192 | } |
5d4f98a2 YZ |
1193 | return ret; |
1194 | } | |
31840ae1 | 1195 | |
5d4f98a2 | 1196 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1197 | struct btrfs_path *path, |
1198 | u64 bytenr, u64 parent, u64 root_objectid, | |
1199 | u64 owner, u64 offset, int refs_to_add) | |
1200 | { | |
1201 | int ret; | |
1202 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1203 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1204 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1205 | root_objectid); | |
5d4f98a2 | 1206 | } else { |
62b895af NB |
1207 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1208 | root_objectid, owner, offset, | |
1209 | refs_to_add); | |
5d4f98a2 YZ |
1210 | } |
1211 | return ret; | |
1212 | } | |
56bec294 | 1213 | |
5d4f98a2 | 1214 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1215 | struct btrfs_path *path, |
1216 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1217 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1218 | { |
143bede5 | 1219 | int ret = 0; |
b9473439 | 1220 | |
5d4f98a2 YZ |
1221 | BUG_ON(!is_data && refs_to_drop != 1); |
1222 | if (iref) { | |
61a18f1c NB |
1223 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1224 | last_ref); | |
5d4f98a2 | 1225 | } else if (is_data) { |
e9f6290d | 1226 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1227 | last_ref); |
5d4f98a2 | 1228 | } else { |
fcebe456 | 1229 | *last_ref = 1; |
87cc7a8a | 1230 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1231 | } |
1232 | return ret; | |
1233 | } | |
1234 | ||
d04c6b88 JM |
1235 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1236 | u64 *discarded_bytes) | |
5d4f98a2 | 1237 | { |
86557861 JM |
1238 | int j, ret = 0; |
1239 | u64 bytes_left, end; | |
4d89d377 | 1240 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1241 | |
4d89d377 JM |
1242 | if (WARN_ON(start != aligned_start)) { |
1243 | len -= aligned_start - start; | |
1244 | len = round_down(len, 1 << 9); | |
1245 | start = aligned_start; | |
1246 | } | |
d04c6b88 | 1247 | |
4d89d377 | 1248 | *discarded_bytes = 0; |
86557861 JM |
1249 | |
1250 | if (!len) | |
1251 | return 0; | |
1252 | ||
1253 | end = start + len; | |
1254 | bytes_left = len; | |
1255 | ||
1256 | /* Skip any superblocks on this device. */ | |
1257 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1258 | u64 sb_start = btrfs_sb_offset(j); | |
1259 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1260 | u64 size = sb_start - start; | |
1261 | ||
1262 | if (!in_range(sb_start, start, bytes_left) && | |
1263 | !in_range(sb_end, start, bytes_left) && | |
1264 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1265 | continue; | |
1266 | ||
1267 | /* | |
1268 | * Superblock spans beginning of range. Adjust start and | |
1269 | * try again. | |
1270 | */ | |
1271 | if (sb_start <= start) { | |
1272 | start += sb_end - start; | |
1273 | if (start > end) { | |
1274 | bytes_left = 0; | |
1275 | break; | |
1276 | } | |
1277 | bytes_left = end - start; | |
1278 | continue; | |
1279 | } | |
1280 | ||
1281 | if (size) { | |
1282 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1283 | GFP_NOFS, 0); | |
1284 | if (!ret) | |
1285 | *discarded_bytes += size; | |
1286 | else if (ret != -EOPNOTSUPP) | |
1287 | return ret; | |
1288 | } | |
1289 | ||
1290 | start = sb_end; | |
1291 | if (start > end) { | |
1292 | bytes_left = 0; | |
1293 | break; | |
1294 | } | |
1295 | bytes_left = end - start; | |
1296 | } | |
1297 | ||
1298 | if (bytes_left) { | |
1299 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1300 | GFP_NOFS, 0); |
1301 | if (!ret) | |
86557861 | 1302 | *discarded_bytes += bytes_left; |
4d89d377 | 1303 | } |
d04c6b88 | 1304 | return ret; |
5d4f98a2 | 1305 | } |
5d4f98a2 | 1306 | |
2ff7e61e | 1307 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1308 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1309 | { |
6b7faadd | 1310 | int ret = 0; |
5378e607 | 1311 | u64 discarded_bytes = 0; |
6b7faadd QW |
1312 | u64 end = bytenr + num_bytes; |
1313 | u64 cur = bytenr; | |
a1d3c478 | 1314 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1315 | |
e244a0ae | 1316 | |
2999241d FM |
1317 | /* |
1318 | * Avoid races with device replace and make sure our bbio has devices | |
1319 | * associated to its stripes that don't go away while we are discarding. | |
1320 | */ | |
0b246afa | 1321 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd QW |
1322 | while (cur < end) { |
1323 | struct btrfs_bio_stripe *stripe; | |
5d4f98a2 YZ |
1324 | int i; |
1325 | ||
6b7faadd QW |
1326 | num_bytes = end - cur; |
1327 | /* Tell the block device(s) that the sectors can be discarded */ | |
1328 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
1329 | &num_bytes, &bbio, 0); | |
1330 | /* | |
1331 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1332 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1333 | * thus we can't continue anyway. | |
1334 | */ | |
1335 | if (ret < 0) | |
1336 | goto out; | |
5d4f98a2 | 1337 | |
6b7faadd | 1338 | stripe = bbio->stripes; |
a1d3c478 | 1339 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1340 | u64 bytes; |
38b5f68e AJ |
1341 | struct request_queue *req_q; |
1342 | ||
627e0873 FM |
1343 | if (!stripe->dev->bdev) { |
1344 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1345 | continue; | |
1346 | } | |
38b5f68e AJ |
1347 | req_q = bdev_get_queue(stripe->dev->bdev); |
1348 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1349 | continue; |
1350 | ||
5378e607 LD |
1351 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1352 | stripe->physical, | |
d04c6b88 JM |
1353 | stripe->length, |
1354 | &bytes); | |
6b7faadd | 1355 | if (!ret) { |
d04c6b88 | 1356 | discarded_bytes += bytes; |
6b7faadd QW |
1357 | } else if (ret != -EOPNOTSUPP) { |
1358 | /* | |
1359 | * Logic errors or -ENOMEM, or -EIO, but | |
1360 | * unlikely to happen. | |
1361 | * | |
1362 | * And since there are two loops, explicitly | |
1363 | * go to out to avoid confusion. | |
1364 | */ | |
1365 | btrfs_put_bbio(bbio); | |
1366 | goto out; | |
1367 | } | |
d5e2003c JB |
1368 | |
1369 | /* | |
1370 | * Just in case we get back EOPNOTSUPP for some reason, | |
1371 | * just ignore the return value so we don't screw up | |
1372 | * people calling discard_extent. | |
1373 | */ | |
1374 | ret = 0; | |
5d4f98a2 | 1375 | } |
6e9606d2 | 1376 | btrfs_put_bbio(bbio); |
6b7faadd | 1377 | cur += num_bytes; |
5d4f98a2 | 1378 | } |
6b7faadd | 1379 | out: |
0b246afa | 1380 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1381 | |
1382 | if (actual_bytes) | |
1383 | *actual_bytes = discarded_bytes; | |
1384 | ||
5d4f98a2 | 1385 | |
53b381b3 DW |
1386 | if (ret == -EOPNOTSUPP) |
1387 | ret = 0; | |
5d4f98a2 | 1388 | return ret; |
5d4f98a2 YZ |
1389 | } |
1390 | ||
79787eaa | 1391 | /* Can return -ENOMEM */ |
5d4f98a2 | 1392 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1393 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1394 | { |
82fa113f | 1395 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1396 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1397 | int ret; |
66d7e7f0 | 1398 | |
82fa113f QW |
1399 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1400 | generic_ref->action); | |
1401 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1402 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1403 | |
82fa113f QW |
1404 | if (generic_ref->type == BTRFS_REF_METADATA) |
1405 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1406 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1407 | else |
1408 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1409 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1410 | |
82fa113f | 1411 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1412 | |
ddf30cf0 | 1413 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1414 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 1415 | |
5d4f98a2 YZ |
1416 | return ret; |
1417 | } | |
1418 | ||
bd3c685e NB |
1419 | /* |
1420 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1421 | * | |
1422 | * @trans: Handle of transaction | |
1423 | * | |
1424 | * @node: The delayed ref node used to get the bytenr/length for | |
1425 | * extent whose references are incremented. | |
1426 | * | |
1427 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1428 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1429 | * bytenr of the parent block. Since new extents are always | |
1430 | * created with indirect references, this will only be the case | |
1431 | * when relocating a shared extent. In that case, root_objectid | |
1432 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1433 | * be 0 | |
1434 | * | |
1435 | * @root_objectid: The id of the root where this modification has originated, | |
1436 | * this can be either one of the well-known metadata trees or | |
1437 | * the subvolume id which references this extent. | |
1438 | * | |
1439 | * @owner: For data extents it is the inode number of the owning file. | |
1440 | * For metadata extents this parameter holds the level in the | |
1441 | * tree of the extent. | |
1442 | * | |
1443 | * @offset: For metadata extents the offset is ignored and is currently | |
1444 | * always passed as 0. For data extents it is the fileoffset | |
1445 | * this extent belongs to. | |
1446 | * | |
1447 | * @refs_to_add Number of references to add | |
1448 | * | |
1449 | * @extent_op Pointer to a structure, holding information necessary when | |
1450 | * updating a tree block's flags | |
1451 | * | |
1452 | */ | |
5d4f98a2 | 1453 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1454 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1455 | u64 parent, u64 root_objectid, |
1456 | u64 owner, u64 offset, int refs_to_add, | |
1457 | struct btrfs_delayed_extent_op *extent_op) | |
1458 | { | |
1459 | struct btrfs_path *path; | |
1460 | struct extent_buffer *leaf; | |
1461 | struct btrfs_extent_item *item; | |
fcebe456 | 1462 | struct btrfs_key key; |
c682f9b3 QW |
1463 | u64 bytenr = node->bytenr; |
1464 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1465 | u64 refs; |
1466 | int ret; | |
5d4f98a2 YZ |
1467 | |
1468 | path = btrfs_alloc_path(); | |
1469 | if (!path) | |
1470 | return -ENOMEM; | |
1471 | ||
e4058b54 | 1472 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
1473 | path->leave_spinning = 1; |
1474 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
1475 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1476 | parent, root_objectid, owner, | |
1477 | offset, refs_to_add, extent_op); | |
0ed4792a | 1478 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1479 | goto out; |
fcebe456 JB |
1480 | |
1481 | /* | |
1482 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1483 | * inline extent ref, so just update the reference count and add a | |
1484 | * normal backref. | |
1485 | */ | |
5d4f98a2 | 1486 | leaf = path->nodes[0]; |
fcebe456 | 1487 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1488 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1489 | refs = btrfs_extent_refs(leaf, item); | |
1490 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1491 | if (extent_op) | |
1492 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1493 | |
5d4f98a2 | 1494 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1495 | btrfs_release_path(path); |
56bec294 | 1496 | |
e4058b54 | 1497 | path->reada = READA_FORWARD; |
b9473439 | 1498 | path->leave_spinning = 1; |
56bec294 | 1499 | /* now insert the actual backref */ |
37593410 NB |
1500 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
1501 | owner, offset, refs_to_add); | |
79787eaa | 1502 | if (ret) |
66642832 | 1503 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1504 | out: |
56bec294 | 1505 | btrfs_free_path(path); |
30d133fc | 1506 | return ret; |
56bec294 CM |
1507 | } |
1508 | ||
5d4f98a2 | 1509 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1510 | struct btrfs_delayed_ref_node *node, |
1511 | struct btrfs_delayed_extent_op *extent_op, | |
1512 | int insert_reserved) | |
56bec294 | 1513 | { |
5d4f98a2 YZ |
1514 | int ret = 0; |
1515 | struct btrfs_delayed_data_ref *ref; | |
1516 | struct btrfs_key ins; | |
1517 | u64 parent = 0; | |
1518 | u64 ref_root = 0; | |
1519 | u64 flags = 0; | |
1520 | ||
1521 | ins.objectid = node->bytenr; | |
1522 | ins.offset = node->num_bytes; | |
1523 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1524 | ||
1525 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1526 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1527 | |
5d4f98a2 YZ |
1528 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1529 | parent = ref->parent; | |
fcebe456 | 1530 | ref_root = ref->root; |
5d4f98a2 YZ |
1531 | |
1532 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1533 | if (extent_op) |
5d4f98a2 | 1534 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1535 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1536 | flags, ref->objectid, | |
1537 | ref->offset, &ins, | |
1538 | node->ref_mod); | |
5d4f98a2 | 1539 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1540 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1541 | ref->objectid, ref->offset, | |
1542 | node->ref_mod, extent_op); | |
5d4f98a2 | 1543 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1544 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1545 | ref_root, ref->objectid, |
1546 | ref->offset, node->ref_mod, | |
c682f9b3 | 1547 | extent_op); |
5d4f98a2 YZ |
1548 | } else { |
1549 | BUG(); | |
1550 | } | |
1551 | return ret; | |
1552 | } | |
1553 | ||
1554 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1555 | struct extent_buffer *leaf, | |
1556 | struct btrfs_extent_item *ei) | |
1557 | { | |
1558 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1559 | if (extent_op->update_flags) { | |
1560 | flags |= extent_op->flags_to_set; | |
1561 | btrfs_set_extent_flags(leaf, ei, flags); | |
1562 | } | |
1563 | ||
1564 | if (extent_op->update_key) { | |
1565 | struct btrfs_tree_block_info *bi; | |
1566 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1567 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1568 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1569 | } | |
1570 | } | |
1571 | ||
1572 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1573 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1574 | struct btrfs_delayed_extent_op *extent_op) |
1575 | { | |
20b9a2d6 | 1576 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1577 | struct btrfs_key key; |
1578 | struct btrfs_path *path; | |
1579 | struct btrfs_extent_item *ei; | |
1580 | struct extent_buffer *leaf; | |
1581 | u32 item_size; | |
56bec294 | 1582 | int ret; |
5d4f98a2 | 1583 | int err = 0; |
b1c79e09 | 1584 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1585 | |
bf31f87f | 1586 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1587 | return 0; |
1588 | ||
0b246afa | 1589 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1590 | metadata = 0; |
1591 | ||
5d4f98a2 YZ |
1592 | path = btrfs_alloc_path(); |
1593 | if (!path) | |
1594 | return -ENOMEM; | |
1595 | ||
d278850e | 1596 | key.objectid = head->bytenr; |
5d4f98a2 | 1597 | |
3173a18f | 1598 | if (metadata) { |
3173a18f | 1599 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1600 | key.offset = extent_op->level; |
3173a18f JB |
1601 | } else { |
1602 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1603 | key.offset = head->num_bytes; |
3173a18f JB |
1604 | } |
1605 | ||
1606 | again: | |
e4058b54 | 1607 | path->reada = READA_FORWARD; |
5d4f98a2 | 1608 | path->leave_spinning = 1; |
0b246afa | 1609 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1610 | if (ret < 0) { |
1611 | err = ret; | |
1612 | goto out; | |
1613 | } | |
1614 | if (ret > 0) { | |
3173a18f | 1615 | if (metadata) { |
55994887 FDBM |
1616 | if (path->slots[0] > 0) { |
1617 | path->slots[0]--; | |
1618 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1619 | path->slots[0]); | |
d278850e | 1620 | if (key.objectid == head->bytenr && |
55994887 | 1621 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1622 | key.offset == head->num_bytes) |
55994887 FDBM |
1623 | ret = 0; |
1624 | } | |
1625 | if (ret > 0) { | |
1626 | btrfs_release_path(path); | |
1627 | metadata = 0; | |
3173a18f | 1628 | |
d278850e JB |
1629 | key.objectid = head->bytenr; |
1630 | key.offset = head->num_bytes; | |
55994887 FDBM |
1631 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1632 | goto again; | |
1633 | } | |
1634 | } else { | |
1635 | err = -EIO; | |
1636 | goto out; | |
3173a18f | 1637 | } |
5d4f98a2 YZ |
1638 | } |
1639 | ||
1640 | leaf = path->nodes[0]; | |
1641 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1642 | |
6d8ff4e4 | 1643 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1644 | err = -EINVAL; |
1645 | btrfs_print_v0_err(fs_info); | |
1646 | btrfs_abort_transaction(trans, err); | |
1647 | goto out; | |
1648 | } | |
1649 | ||
5d4f98a2 YZ |
1650 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1651 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1652 | |
5d4f98a2 YZ |
1653 | btrfs_mark_buffer_dirty(leaf); |
1654 | out: | |
1655 | btrfs_free_path(path); | |
1656 | return err; | |
56bec294 CM |
1657 | } |
1658 | ||
5d4f98a2 | 1659 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1660 | struct btrfs_delayed_ref_node *node, |
1661 | struct btrfs_delayed_extent_op *extent_op, | |
1662 | int insert_reserved) | |
56bec294 CM |
1663 | { |
1664 | int ret = 0; | |
5d4f98a2 | 1665 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1666 | u64 parent = 0; |
1667 | u64 ref_root = 0; | |
56bec294 | 1668 | |
5d4f98a2 | 1669 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1670 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1671 | |
5d4f98a2 YZ |
1672 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1673 | parent = ref->parent; | |
fcebe456 | 1674 | ref_root = ref->root; |
5d4f98a2 | 1675 | |
02794222 | 1676 | if (node->ref_mod != 1) { |
f97806f2 | 1677 | btrfs_err(trans->fs_info, |
02794222 LB |
1678 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1679 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1680 | parent); | |
1681 | return -EIO; | |
1682 | } | |
5d4f98a2 | 1683 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1684 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1685 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1686 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1687 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1688 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1689 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1690 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1691 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1692 | } else { |
1693 | BUG(); | |
1694 | } | |
56bec294 CM |
1695 | return ret; |
1696 | } | |
1697 | ||
1698 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1699 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1700 | struct btrfs_delayed_ref_node *node, |
1701 | struct btrfs_delayed_extent_op *extent_op, | |
1702 | int insert_reserved) | |
56bec294 | 1703 | { |
79787eaa JM |
1704 | int ret = 0; |
1705 | ||
bf31f87f | 1706 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1707 | if (insert_reserved) |
5fac7f9e | 1708 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 1709 | node->num_bytes, 1); |
79787eaa | 1710 | return 0; |
857cc2fc | 1711 | } |
79787eaa | 1712 | |
5d4f98a2 YZ |
1713 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1714 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1715 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1716 | insert_reserved); |
1717 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1718 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1719 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1720 | insert_reserved); |
1721 | else | |
1722 | BUG(); | |
80ee54bf JB |
1723 | if (ret && insert_reserved) |
1724 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
1725 | node->num_bytes, 1); | |
5d4f98a2 | 1726 | return ret; |
56bec294 CM |
1727 | } |
1728 | ||
c6fc2454 | 1729 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1730 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1731 | { | |
cffc3374 FM |
1732 | struct btrfs_delayed_ref_node *ref; |
1733 | ||
e3d03965 | 1734 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1735 | return NULL; |
d7df2c79 | 1736 | |
cffc3374 FM |
1737 | /* |
1738 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1739 | * This is to prevent a ref count from going down to zero, which deletes | |
1740 | * the extent item from the extent tree, when there still are references | |
1741 | * to add, which would fail because they would not find the extent item. | |
1742 | */ | |
1d57ee94 WX |
1743 | if (!list_empty(&head->ref_add_list)) |
1744 | return list_first_entry(&head->ref_add_list, | |
1745 | struct btrfs_delayed_ref_node, add_list); | |
1746 | ||
e3d03965 | 1747 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1748 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1749 | ASSERT(list_empty(&ref->add_list)); |
1750 | return ref; | |
56bec294 CM |
1751 | } |
1752 | ||
2eadaa22 JB |
1753 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1754 | struct btrfs_delayed_ref_head *head) | |
1755 | { | |
1756 | spin_lock(&delayed_refs->lock); | |
1757 | head->processing = 0; | |
1758 | delayed_refs->num_heads_ready++; | |
1759 | spin_unlock(&delayed_refs->lock); | |
1760 | btrfs_delayed_ref_unlock(head); | |
1761 | } | |
1762 | ||
bedc6617 JB |
1763 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1764 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1765 | { |
1766 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1767 | |
1768 | if (!extent_op) | |
bedc6617 JB |
1769 | return NULL; |
1770 | ||
b00e6250 | 1771 | if (head->must_insert_reserved) { |
bedc6617 | 1772 | head->extent_op = NULL; |
b00e6250 | 1773 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1774 | return NULL; |
b00e6250 | 1775 | } |
bedc6617 JB |
1776 | return extent_op; |
1777 | } | |
1778 | ||
1779 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1780 | struct btrfs_delayed_ref_head *head) | |
1781 | { | |
1782 | struct btrfs_delayed_extent_op *extent_op; | |
1783 | int ret; | |
1784 | ||
1785 | extent_op = cleanup_extent_op(head); | |
1786 | if (!extent_op) | |
1787 | return 0; | |
1788 | head->extent_op = NULL; | |
b00e6250 | 1789 | spin_unlock(&head->lock); |
20b9a2d6 | 1790 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1791 | btrfs_free_delayed_extent_op(extent_op); |
1792 | return ret ? ret : 1; | |
1793 | } | |
1794 | ||
31890da0 JB |
1795 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1796 | struct btrfs_delayed_ref_root *delayed_refs, | |
1797 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1798 | { |
ba2c4d4e | 1799 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
1800 | |
1801 | if (head->total_ref_mod < 0) { | |
1802 | struct btrfs_space_info *space_info; | |
1803 | u64 flags; | |
1804 | ||
1805 | if (head->is_data) | |
1806 | flags = BTRFS_BLOCK_GROUP_DATA; | |
1807 | else if (head->is_system) | |
1808 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
1809 | else | |
1810 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 1811 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
1812 | ASSERT(space_info); |
1813 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1814 | -head->num_bytes, | |
1815 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1816 | ||
ba2c4d4e JB |
1817 | /* |
1818 | * We had csum deletions accounted for in our delayed refs rsv, | |
1819 | * we need to drop the csum leaves for this update from our | |
1820 | * delayed_refs_rsv. | |
1821 | */ | |
07c47775 JB |
1822 | if (head->is_data) { |
1823 | spin_lock(&delayed_refs->lock); | |
1824 | delayed_refs->pending_csums -= head->num_bytes; | |
1825 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
1826 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
1827 | head->num_bytes); | |
07c47775 JB |
1828 | } |
1829 | } | |
1830 | ||
ba2c4d4e | 1831 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1832 | } |
1833 | ||
194ab0bc | 1834 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1835 | struct btrfs_delayed_ref_head *head) |
1836 | { | |
f9871edd NB |
1837 | |
1838 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1839 | struct btrfs_delayed_ref_root *delayed_refs; |
1840 | int ret; | |
1841 | ||
1842 | delayed_refs = &trans->transaction->delayed_refs; | |
1843 | ||
bedc6617 | 1844 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1845 | if (ret < 0) { |
1846 | unselect_delayed_ref_head(delayed_refs, head); | |
1847 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1848 | return ret; | |
1849 | } else if (ret) { | |
1850 | return ret; | |
1851 | } | |
1852 | ||
1853 | /* | |
1854 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1855 | * and then re-check to make sure nobody got added. | |
1856 | */ | |
1857 | spin_unlock(&head->lock); | |
1858 | spin_lock(&delayed_refs->lock); | |
1859 | spin_lock(&head->lock); | |
e3d03965 | 1860 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1861 | spin_unlock(&head->lock); |
1862 | spin_unlock(&delayed_refs->lock); | |
1863 | return 1; | |
1864 | } | |
d7baffda | 1865 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1866 | spin_unlock(&head->lock); |
1e7a1421 | 1867 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1868 | |
c1103f7a | 1869 | if (head->must_insert_reserved) { |
d278850e JB |
1870 | btrfs_pin_extent(fs_info, head->bytenr, |
1871 | head->num_bytes, 1); | |
c1103f7a | 1872 | if (head->is_data) { |
40e046ac FM |
1873 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1874 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1875 | } |
1876 | } | |
1877 | ||
31890da0 | 1878 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1879 | |
1880 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1881 | btrfs_delayed_ref_unlock(head); |
d278850e | 1882 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1883 | return 0; |
1884 | } | |
1885 | ||
b1cdbcb5 NB |
1886 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1887 | struct btrfs_trans_handle *trans) | |
1888 | { | |
1889 | struct btrfs_delayed_ref_root *delayed_refs = | |
1890 | &trans->transaction->delayed_refs; | |
1891 | struct btrfs_delayed_ref_head *head = NULL; | |
1892 | int ret; | |
1893 | ||
1894 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1895 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1896 | if (!head) { |
1897 | spin_unlock(&delayed_refs->lock); | |
1898 | return head; | |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * Grab the lock that says we are going to process all the refs for | |
1903 | * this head | |
1904 | */ | |
9e920a6f | 1905 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1906 | spin_unlock(&delayed_refs->lock); |
1907 | ||
1908 | /* | |
1909 | * We may have dropped the spin lock to get the head mutex lock, and | |
1910 | * that might have given someone else time to free the head. If that's | |
1911 | * true, it has been removed from our list and we can move on. | |
1912 | */ | |
1913 | if (ret == -EAGAIN) | |
1914 | head = ERR_PTR(-EAGAIN); | |
1915 | ||
1916 | return head; | |
1917 | } | |
1918 | ||
e7261386 NB |
1919 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1920 | struct btrfs_delayed_ref_head *locked_ref, | |
1921 | unsigned long *run_refs) | |
1922 | { | |
1923 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1924 | struct btrfs_delayed_ref_root *delayed_refs; | |
1925 | struct btrfs_delayed_extent_op *extent_op; | |
1926 | struct btrfs_delayed_ref_node *ref; | |
1927 | int must_insert_reserved = 0; | |
1928 | int ret; | |
1929 | ||
1930 | delayed_refs = &trans->transaction->delayed_refs; | |
1931 | ||
0110a4c4 NB |
1932 | lockdep_assert_held(&locked_ref->mutex); |
1933 | lockdep_assert_held(&locked_ref->lock); | |
1934 | ||
e7261386 NB |
1935 | while ((ref = select_delayed_ref(locked_ref))) { |
1936 | if (ref->seq && | |
1937 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1938 | spin_unlock(&locked_ref->lock); | |
1939 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1940 | return -EAGAIN; | |
1941 | } | |
1942 | ||
1943 | (*run_refs)++; | |
1944 | ref->in_tree = 0; | |
1945 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1946 | RB_CLEAR_NODE(&ref->ref_node); | |
1947 | if (!list_empty(&ref->add_list)) | |
1948 | list_del(&ref->add_list); | |
1949 | /* | |
1950 | * When we play the delayed ref, also correct the ref_mod on | |
1951 | * head | |
1952 | */ | |
1953 | switch (ref->action) { | |
1954 | case BTRFS_ADD_DELAYED_REF: | |
1955 | case BTRFS_ADD_DELAYED_EXTENT: | |
1956 | locked_ref->ref_mod -= ref->ref_mod; | |
1957 | break; | |
1958 | case BTRFS_DROP_DELAYED_REF: | |
1959 | locked_ref->ref_mod += ref->ref_mod; | |
1960 | break; | |
1961 | default: | |
1962 | WARN_ON(1); | |
1963 | } | |
1964 | atomic_dec(&delayed_refs->num_entries); | |
1965 | ||
1966 | /* | |
1967 | * Record the must_insert_reserved flag before we drop the | |
1968 | * spin lock. | |
1969 | */ | |
1970 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1971 | locked_ref->must_insert_reserved = 0; | |
1972 | ||
1973 | extent_op = locked_ref->extent_op; | |
1974 | locked_ref->extent_op = NULL; | |
1975 | spin_unlock(&locked_ref->lock); | |
1976 | ||
1977 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1978 | must_insert_reserved); | |
1979 | ||
1980 | btrfs_free_delayed_extent_op(extent_op); | |
1981 | if (ret) { | |
1982 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1983 | btrfs_put_delayed_ref(ref); | |
1984 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1985 | ret); | |
1986 | return ret; | |
1987 | } | |
1988 | ||
1989 | btrfs_put_delayed_ref(ref); | |
1990 | cond_resched(); | |
1991 | ||
1992 | spin_lock(&locked_ref->lock); | |
1993 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1994 | } | |
1995 | ||
1996 | return 0; | |
1997 | } | |
1998 | ||
79787eaa JM |
1999 | /* |
2000 | * Returns 0 on success or if called with an already aborted transaction. | |
2001 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2002 | */ | |
d7df2c79 | 2003 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2004 | unsigned long nr) |
56bec294 | 2005 | { |
0a1e458a | 2006 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2007 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2008 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2009 | ktime_t start = ktime_get(); |
56bec294 | 2010 | int ret; |
d7df2c79 | 2011 | unsigned long count = 0; |
0a2b2a84 | 2012 | unsigned long actual_count = 0; |
56bec294 CM |
2013 | |
2014 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2015 | do { |
56bec294 | 2016 | if (!locked_ref) { |
b1cdbcb5 | 2017 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2018 | if (IS_ERR_OR_NULL(locked_ref)) { |
2019 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2020 | continue; | |
2021 | } else { | |
2022 | break; | |
2023 | } | |
56bec294 | 2024 | } |
0110a4c4 | 2025 | count++; |
56bec294 | 2026 | } |
2c3cf7d5 FM |
2027 | /* |
2028 | * We need to try and merge add/drops of the same ref since we | |
2029 | * can run into issues with relocate dropping the implicit ref | |
2030 | * and then it being added back again before the drop can | |
2031 | * finish. If we merged anything we need to re-loop so we can | |
2032 | * get a good ref. | |
2033 | * Or we can get node references of the same type that weren't | |
2034 | * merged when created due to bumps in the tree mod seq, and | |
2035 | * we need to merge them to prevent adding an inline extent | |
2036 | * backref before dropping it (triggering a BUG_ON at | |
2037 | * insert_inline_extent_backref()). | |
2038 | */ | |
d7df2c79 | 2039 | spin_lock(&locked_ref->lock); |
be97f133 | 2040 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2041 | |
0110a4c4 NB |
2042 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2043 | &actual_count); | |
2044 | if (ret < 0 && ret != -EAGAIN) { | |
2045 | /* | |
2046 | * Error, btrfs_run_delayed_refs_for_head already | |
2047 | * unlocked everything so just bail out | |
2048 | */ | |
2049 | return ret; | |
2050 | } else if (!ret) { | |
2051 | /* | |
2052 | * Success, perform the usual cleanup of a processed | |
2053 | * head | |
2054 | */ | |
f9871edd | 2055 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2056 | if (ret > 0 ) { |
b00e6250 JB |
2057 | /* We dropped our lock, we need to loop. */ |
2058 | ret = 0; | |
d7df2c79 | 2059 | continue; |
194ab0bc JB |
2060 | } else if (ret) { |
2061 | return ret; | |
5d4f98a2 | 2062 | } |
22cd2e7d | 2063 | } |
1ce7a5ec | 2064 | |
b00e6250 | 2065 | /* |
0110a4c4 NB |
2066 | * Either success case or btrfs_run_delayed_refs_for_head |
2067 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2068 | */ |
b00e6250 | 2069 | |
0110a4c4 | 2070 | locked_ref = NULL; |
c3e69d58 | 2071 | cond_resched(); |
0110a4c4 | 2072 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2073 | |
2074 | /* | |
2075 | * We don't want to include ref heads since we can have empty ref heads | |
2076 | * and those will drastically skew our runtime down since we just do | |
2077 | * accounting, no actual extent tree updates. | |
2078 | */ | |
2079 | if (actual_count > 0) { | |
2080 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2081 | u64 avg; | |
2082 | ||
2083 | /* | |
2084 | * We weigh the current average higher than our current runtime | |
2085 | * to avoid large swings in the average. | |
2086 | */ | |
2087 | spin_lock(&delayed_refs->lock); | |
2088 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2089 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2090 | spin_unlock(&delayed_refs->lock); |
2091 | } | |
d7df2c79 | 2092 | return 0; |
c3e69d58 CM |
2093 | } |
2094 | ||
709c0486 AJ |
2095 | #ifdef SCRAMBLE_DELAYED_REFS |
2096 | /* | |
2097 | * Normally delayed refs get processed in ascending bytenr order. This | |
2098 | * correlates in most cases to the order added. To expose dependencies on this | |
2099 | * order, we start to process the tree in the middle instead of the beginning | |
2100 | */ | |
2101 | static u64 find_middle(struct rb_root *root) | |
2102 | { | |
2103 | struct rb_node *n = root->rb_node; | |
2104 | struct btrfs_delayed_ref_node *entry; | |
2105 | int alt = 1; | |
2106 | u64 middle; | |
2107 | u64 first = 0, last = 0; | |
2108 | ||
2109 | n = rb_first(root); | |
2110 | if (n) { | |
2111 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2112 | first = entry->bytenr; | |
2113 | } | |
2114 | n = rb_last(root); | |
2115 | if (n) { | |
2116 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2117 | last = entry->bytenr; | |
2118 | } | |
2119 | n = root->rb_node; | |
2120 | ||
2121 | while (n) { | |
2122 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2123 | WARN_ON(!entry->in_tree); | |
2124 | ||
2125 | middle = entry->bytenr; | |
2126 | ||
2127 | if (alt) | |
2128 | n = n->rb_left; | |
2129 | else | |
2130 | n = n->rb_right; | |
2131 | ||
2132 | alt = 1 - alt; | |
2133 | } | |
2134 | return middle; | |
2135 | } | |
2136 | #endif | |
2137 | ||
2ff7e61e | 2138 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2139 | { |
2140 | u64 num_bytes; | |
2141 | ||
2142 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2143 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2144 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2145 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2146 | ||
2147 | /* | |
2148 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2149 | * closer to what we're really going to want to use. |
1be41b78 | 2150 | */ |
0b246afa | 2151 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2152 | } |
2153 | ||
1262133b JB |
2154 | /* |
2155 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2156 | * would require to store the csums for that many bytes. | |
2157 | */ | |
2ff7e61e | 2158 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2159 | { |
2160 | u64 csum_size; | |
2161 | u64 num_csums_per_leaf; | |
2162 | u64 num_csums; | |
2163 | ||
0b246afa | 2164 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2165 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2166 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2167 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2168 | num_csums += num_csums_per_leaf - 1; |
2169 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2170 | return num_csums; | |
2171 | } | |
2172 | ||
c3e69d58 CM |
2173 | /* |
2174 | * this starts processing the delayed reference count updates and | |
2175 | * extent insertions we have queued up so far. count can be | |
2176 | * 0, which means to process everything in the tree at the start | |
2177 | * of the run (but not newly added entries), or it can be some target | |
2178 | * number you'd like to process. | |
79787eaa JM |
2179 | * |
2180 | * Returns 0 on success or if called with an aborted transaction | |
2181 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2182 | */ |
2183 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2184 | unsigned long count) |
c3e69d58 | 2185 | { |
c79a70b1 | 2186 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2187 | struct rb_node *node; |
2188 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2189 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2190 | int ret; |
2191 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2192 | |
79787eaa | 2193 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2194 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2195 | return 0; |
2196 | ||
0b246afa | 2197 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2198 | return 0; |
2199 | ||
c3e69d58 | 2200 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2201 | if (count == 0) |
d7df2c79 | 2202 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2203 | |
c3e69d58 | 2204 | again: |
709c0486 AJ |
2205 | #ifdef SCRAMBLE_DELAYED_REFS |
2206 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2207 | #endif | |
0a1e458a | 2208 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2209 | if (ret < 0) { |
66642832 | 2210 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2211 | return ret; |
eb099670 | 2212 | } |
c3e69d58 | 2213 | |
56bec294 | 2214 | if (run_all) { |
119e80df | 2215 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2216 | |
d7df2c79 | 2217 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2218 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2219 | if (!node) { |
2220 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2221 | goto out; |
d7df2c79 | 2222 | } |
d278850e JB |
2223 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2224 | href_node); | |
2225 | refcount_inc(&head->refs); | |
2226 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2227 | |
d278850e JB |
2228 | /* Mutex was contended, block until it's released and retry. */ |
2229 | mutex_lock(&head->mutex); | |
2230 | mutex_unlock(&head->mutex); | |
56bec294 | 2231 | |
d278850e | 2232 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2233 | cond_resched(); |
56bec294 | 2234 | goto again; |
5f39d397 | 2235 | } |
54aa1f4d | 2236 | out: |
a28ec197 CM |
2237 | return 0; |
2238 | } | |
2239 | ||
5d4f98a2 | 2240 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
5d4f98a2 | 2241 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2242 | int level, int is_data) |
5d4f98a2 YZ |
2243 | { |
2244 | struct btrfs_delayed_extent_op *extent_op; | |
2245 | int ret; | |
2246 | ||
78a6184a | 2247 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2248 | if (!extent_op) |
2249 | return -ENOMEM; | |
2250 | ||
2251 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2252 | extent_op->update_flags = true; |
2253 | extent_op->update_key = false; | |
2254 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2255 | extent_op->level = level; |
5d4f98a2 | 2256 | |
c6e340bc | 2257 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); |
5d4f98a2 | 2258 | if (ret) |
78a6184a | 2259 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2260 | return ret; |
2261 | } | |
2262 | ||
e4c3b2dc | 2263 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2264 | struct btrfs_path *path, |
2265 | u64 objectid, u64 offset, u64 bytenr) | |
2266 | { | |
2267 | struct btrfs_delayed_ref_head *head; | |
2268 | struct btrfs_delayed_ref_node *ref; | |
2269 | struct btrfs_delayed_data_ref *data_ref; | |
2270 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2271 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2272 | struct rb_node *node; |
5d4f98a2 YZ |
2273 | int ret = 0; |
2274 | ||
998ac6d2 | 2275 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2276 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2277 | if (cur_trans) |
2278 | refcount_inc(&cur_trans->use_count); | |
2279 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2280 | if (!cur_trans) |
2281 | return 0; | |
2282 | ||
2283 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2284 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2285 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2286 | if (!head) { |
2287 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2288 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2289 | return 0; |
2290 | } | |
5d4f98a2 YZ |
2291 | |
2292 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2293 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2294 | spin_unlock(&delayed_refs->lock); |
2295 | ||
b3b4aa74 | 2296 | btrfs_release_path(path); |
5d4f98a2 | 2297 | |
8cc33e5c DS |
2298 | /* |
2299 | * Mutex was contended, block until it's released and let | |
2300 | * caller try again | |
2301 | */ | |
5d4f98a2 YZ |
2302 | mutex_lock(&head->mutex); |
2303 | mutex_unlock(&head->mutex); | |
d278850e | 2304 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2305 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2306 | return -EAGAIN; |
2307 | } | |
d7df2c79 | 2308 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2309 | |
d7df2c79 | 2310 | spin_lock(&head->lock); |
0e0adbcf JB |
2311 | /* |
2312 | * XXX: We should replace this with a proper search function in the | |
2313 | * future. | |
2314 | */ | |
e3d03965 LB |
2315 | for (node = rb_first_cached(&head->ref_tree); node; |
2316 | node = rb_next(node)) { | |
0e0adbcf | 2317 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2318 | /* If it's a shared ref we know a cross reference exists */ |
2319 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2320 | ret = 1; | |
2321 | break; | |
2322 | } | |
5d4f98a2 | 2323 | |
d7df2c79 | 2324 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2325 | |
d7df2c79 JB |
2326 | /* |
2327 | * If our ref doesn't match the one we're currently looking at | |
2328 | * then we have a cross reference. | |
2329 | */ | |
2330 | if (data_ref->root != root->root_key.objectid || | |
2331 | data_ref->objectid != objectid || | |
2332 | data_ref->offset != offset) { | |
2333 | ret = 1; | |
2334 | break; | |
2335 | } | |
5d4f98a2 | 2336 | } |
d7df2c79 | 2337 | spin_unlock(&head->lock); |
5d4f98a2 | 2338 | mutex_unlock(&head->mutex); |
998ac6d2 | 2339 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2340 | return ret; |
2341 | } | |
2342 | ||
e4c3b2dc | 2343 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2344 | struct btrfs_path *path, |
2345 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 2346 | { |
0b246afa JM |
2347 | struct btrfs_fs_info *fs_info = root->fs_info; |
2348 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2349 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2350 | struct btrfs_extent_data_ref *ref; |
2351 | struct btrfs_extent_inline_ref *iref; | |
2352 | struct btrfs_extent_item *ei; | |
f321e491 | 2353 | struct btrfs_key key; |
5d4f98a2 | 2354 | u32 item_size; |
3de28d57 | 2355 | int type; |
be20aa9d | 2356 | int ret; |
925baedd | 2357 | |
be20aa9d | 2358 | key.objectid = bytenr; |
31840ae1 | 2359 | key.offset = (u64)-1; |
f321e491 | 2360 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2361 | |
be20aa9d CM |
2362 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2363 | if (ret < 0) | |
2364 | goto out; | |
79787eaa | 2365 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2366 | |
2367 | ret = -ENOENT; | |
2368 | if (path->slots[0] == 0) | |
31840ae1 | 2369 | goto out; |
be20aa9d | 2370 | |
31840ae1 | 2371 | path->slots[0]--; |
f321e491 | 2372 | leaf = path->nodes[0]; |
5d4f98a2 | 2373 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2374 | |
5d4f98a2 | 2375 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2376 | goto out; |
f321e491 | 2377 | |
5d4f98a2 YZ |
2378 | ret = 1; |
2379 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2380 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2381 | |
a6bd9cd1 | 2382 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2383 | if (item_size != sizeof(*ei) + |
2384 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2385 | goto out; | |
be20aa9d | 2386 | |
a6bd9cd1 | 2387 | /* If extent created before last snapshot => it's definitely shared */ |
5d4f98a2 YZ |
2388 | if (btrfs_extent_generation(leaf, ei) <= |
2389 | btrfs_root_last_snapshot(&root->root_item)) | |
2390 | goto out; | |
2391 | ||
2392 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2393 | |
a6bd9cd1 | 2394 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2395 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2396 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2397 | goto out; |
2398 | ||
2399 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2400 | if (btrfs_extent_refs(leaf, ei) != | |
2401 | btrfs_extent_data_ref_count(leaf, ref) || | |
2402 | btrfs_extent_data_ref_root(leaf, ref) != | |
2403 | root->root_key.objectid || | |
2404 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2405 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2406 | goto out; | |
2407 | ||
2408 | ret = 0; | |
2409 | out: | |
2410 | return ret; | |
2411 | } | |
2412 | ||
e4c3b2dc LB |
2413 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
2414 | u64 bytenr) | |
5d4f98a2 YZ |
2415 | { |
2416 | struct btrfs_path *path; | |
2417 | int ret; | |
5d4f98a2 YZ |
2418 | |
2419 | path = btrfs_alloc_path(); | |
2420 | if (!path) | |
9132c4ff | 2421 | return -ENOMEM; |
5d4f98a2 YZ |
2422 | |
2423 | do { | |
e4c3b2dc | 2424 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
2425 | offset, bytenr); |
2426 | if (ret && ret != -ENOENT) | |
f321e491 | 2427 | goto out; |
80ff3856 | 2428 | |
380fd066 MT |
2429 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2430 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2431 | |
be20aa9d | 2432 | out: |
80ff3856 | 2433 | btrfs_free_path(path); |
f0486c68 YZ |
2434 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2435 | WARN_ON(ret > 0); | |
f321e491 | 2436 | return ret; |
be20aa9d | 2437 | } |
c5739bba | 2438 | |
5d4f98a2 | 2439 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2440 | struct btrfs_root *root, |
5d4f98a2 | 2441 | struct extent_buffer *buf, |
e339a6b0 | 2442 | int full_backref, int inc) |
31840ae1 | 2443 | { |
0b246afa | 2444 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2445 | u64 bytenr; |
5d4f98a2 YZ |
2446 | u64 num_bytes; |
2447 | u64 parent; | |
31840ae1 | 2448 | u64 ref_root; |
31840ae1 | 2449 | u32 nritems; |
31840ae1 ZY |
2450 | struct btrfs_key key; |
2451 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2452 | struct btrfs_ref generic_ref = { 0 }; |
2453 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2454 | int i; |
82fa113f | 2455 | int action; |
31840ae1 ZY |
2456 | int level; |
2457 | int ret = 0; | |
fccb84c9 | 2458 | |
0b246afa | 2459 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2460 | return 0; |
fccb84c9 | 2461 | |
31840ae1 | 2462 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2463 | nritems = btrfs_header_nritems(buf); |
2464 | level = btrfs_header_level(buf); | |
2465 | ||
27cdeb70 | 2466 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 2467 | return 0; |
31840ae1 | 2468 | |
5d4f98a2 YZ |
2469 | if (full_backref) |
2470 | parent = buf->start; | |
2471 | else | |
2472 | parent = 0; | |
82fa113f QW |
2473 | if (inc) |
2474 | action = BTRFS_ADD_DELAYED_REF; | |
2475 | else | |
2476 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2477 | |
2478 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2479 | if (level == 0) { |
5d4f98a2 | 2480 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2481 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2482 | continue; |
5d4f98a2 | 2483 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2484 | struct btrfs_file_extent_item); |
2485 | if (btrfs_file_extent_type(buf, fi) == | |
2486 | BTRFS_FILE_EXTENT_INLINE) | |
2487 | continue; | |
2488 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2489 | if (bytenr == 0) | |
2490 | continue; | |
5d4f98a2 YZ |
2491 | |
2492 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2493 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2494 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2495 | num_bytes, parent); | |
2496 | generic_ref.real_root = root->root_key.objectid; | |
2497 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2498 | key.offset); | |
2499 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2500 | if (inc) |
82fa113f | 2501 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2502 | else |
ffd4bb2a | 2503 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2504 | if (ret) |
2505 | goto fail; | |
2506 | } else { | |
5d4f98a2 | 2507 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2508 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2509 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2510 | num_bytes, parent); | |
2511 | generic_ref.real_root = root->root_key.objectid; | |
2512 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2513 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2514 | if (inc) |
82fa113f | 2515 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2516 | else |
ffd4bb2a | 2517 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2518 | if (ret) |
2519 | goto fail; | |
2520 | } | |
2521 | } | |
2522 | return 0; | |
2523 | fail: | |
5d4f98a2 YZ |
2524 | return ret; |
2525 | } | |
2526 | ||
2527 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2528 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2529 | { |
e339a6b0 | 2530 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2531 | } |
2532 | ||
2533 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2534 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2535 | { |
e339a6b0 | 2536 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2537 | } |
2538 | ||
2ff7e61e | 2539 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 | 2540 | { |
32da5386 | 2541 | struct btrfs_block_group *block_group; |
d2fb3437 YZ |
2542 | int readonly = 0; |
2543 | ||
0b246afa | 2544 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2545 | if (!block_group || block_group->ro) |
2546 | readonly = 1; | |
2547 | if (block_group) | |
fa9c0d79 | 2548 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2549 | return readonly; |
2550 | } | |
2551 | ||
1b86826d | 2552 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2553 | { |
0b246afa | 2554 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2555 | u64 flags; |
53b381b3 | 2556 | u64 ret; |
9ed74f2d | 2557 | |
b742bb82 YZ |
2558 | if (data) |
2559 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2560 | else if (root == fs_info->chunk_root) |
b742bb82 | 2561 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2562 | else |
b742bb82 | 2563 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2564 | |
878d7b67 | 2565 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2566 | return ret; |
6a63209f | 2567 | } |
9ed74f2d | 2568 | |
2ff7e61e | 2569 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2570 | { |
32da5386 | 2571 | struct btrfs_block_group *cache; |
d2fb3437 | 2572 | u64 bytenr; |
0f9dd46c | 2573 | |
0b246afa JM |
2574 | spin_lock(&fs_info->block_group_cache_lock); |
2575 | bytenr = fs_info->first_logical_byte; | |
2576 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2577 | |
2578 | if (bytenr < (u64)-1) | |
2579 | return bytenr; | |
2580 | ||
0b246afa | 2581 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2582 | if (!cache) |
a061fc8d | 2583 | return 0; |
0f9dd46c | 2584 | |
b3470b5d | 2585 | bytenr = cache->start; |
fa9c0d79 | 2586 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2587 | |
2588 | return bytenr; | |
a061fc8d CM |
2589 | } |
2590 | ||
32da5386 | 2591 | static int pin_down_extent(struct btrfs_block_group *cache, |
f0486c68 | 2592 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2593 | { |
fdf08605 DS |
2594 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2595 | ||
11833d66 YZ |
2596 | spin_lock(&cache->space_info->lock); |
2597 | spin_lock(&cache->lock); | |
2598 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2599 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2600 | num_bytes); | |
11833d66 YZ |
2601 | if (reserved) { |
2602 | cache->reserved -= num_bytes; | |
2603 | cache->space_info->bytes_reserved -= num_bytes; | |
2604 | } | |
2605 | spin_unlock(&cache->lock); | |
2606 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2607 | |
dec59fa3 EL |
2608 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
2609 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 2610 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
2611 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2612 | return 0; | |
2613 | } | |
68b38550 | 2614 | |
2ff7e61e | 2615 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
2616 | u64 bytenr, u64 num_bytes, int reserved) |
2617 | { | |
32da5386 | 2618 | struct btrfs_block_group *cache; |
68b38550 | 2619 | |
ce6d3eb6 NB |
2620 | ASSERT(fs_info->running_transaction); |
2621 | ||
0b246afa | 2622 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 2623 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2624 | |
fdf08605 | 2625 | pin_down_extent(cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2626 | |
2627 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2628 | return 0; |
2629 | } | |
2630 | ||
f0486c68 | 2631 | /* |
e688b725 CM |
2632 | * this function must be called within transaction |
2633 | */ | |
2ff7e61e | 2634 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
2635 | u64 bytenr, u64 num_bytes) |
2636 | { | |
32da5386 | 2637 | struct btrfs_block_group *cache; |
b50c6e25 | 2638 | int ret; |
e688b725 | 2639 | |
0b246afa | 2640 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
2641 | if (!cache) |
2642 | return -EINVAL; | |
e688b725 CM |
2643 | |
2644 | /* | |
2645 | * pull in the free space cache (if any) so that our pin | |
2646 | * removes the free space from the cache. We have load_only set | |
2647 | * to one because the slow code to read in the free extents does check | |
2648 | * the pinned extents. | |
2649 | */ | |
676f1f75 | 2650 | btrfs_cache_block_group(cache, 1); |
e688b725 | 2651 | |
fdf08605 | 2652 | pin_down_extent(cache, bytenr, num_bytes, 0); |
e688b725 CM |
2653 | |
2654 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2655 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 2656 | btrfs_put_block_group(cache); |
b50c6e25 | 2657 | return ret; |
e688b725 CM |
2658 | } |
2659 | ||
2ff7e61e JM |
2660 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2661 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2662 | { |
2663 | int ret; | |
32da5386 | 2664 | struct btrfs_block_group *block_group; |
8c2a1a30 JB |
2665 | struct btrfs_caching_control *caching_ctl; |
2666 | ||
0b246afa | 2667 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2668 | if (!block_group) |
2669 | return -EINVAL; | |
2670 | ||
676f1f75 | 2671 | btrfs_cache_block_group(block_group, 0); |
e3cb339f | 2672 | caching_ctl = btrfs_get_caching_control(block_group); |
8c2a1a30 JB |
2673 | |
2674 | if (!caching_ctl) { | |
2675 | /* Logic error */ | |
32da5386 | 2676 | BUG_ON(!btrfs_block_group_done(block_group)); |
8c2a1a30 JB |
2677 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2678 | } else { | |
2679 | mutex_lock(&caching_ctl->mutex); | |
2680 | ||
2681 | if (start >= caching_ctl->progress) { | |
6f410d1b JB |
2682 | ret = btrfs_add_excluded_extent(fs_info, start, |
2683 | num_bytes); | |
8c2a1a30 JB |
2684 | } else if (start + num_bytes <= caching_ctl->progress) { |
2685 | ret = btrfs_remove_free_space(block_group, | |
2686 | start, num_bytes); | |
2687 | } else { | |
2688 | num_bytes = caching_ctl->progress - start; | |
2689 | ret = btrfs_remove_free_space(block_group, | |
2690 | start, num_bytes); | |
2691 | if (ret) | |
2692 | goto out_lock; | |
2693 | ||
2694 | num_bytes = (start + num_bytes) - | |
2695 | caching_ctl->progress; | |
2696 | start = caching_ctl->progress; | |
6f410d1b JB |
2697 | ret = btrfs_add_excluded_extent(fs_info, start, |
2698 | num_bytes); | |
8c2a1a30 JB |
2699 | } |
2700 | out_lock: | |
2701 | mutex_unlock(&caching_ctl->mutex); | |
e3cb339f | 2702 | btrfs_put_caching_control(caching_ctl); |
8c2a1a30 JB |
2703 | } |
2704 | btrfs_put_block_group(block_group); | |
2705 | return ret; | |
2706 | } | |
2707 | ||
bcdc428c | 2708 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2709 | { |
bcdc428c | 2710 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2711 | struct btrfs_file_extent_item *item; |
2712 | struct btrfs_key key; | |
2713 | int found_type; | |
2714 | int i; | |
b89311ef | 2715 | int ret = 0; |
8c2a1a30 | 2716 | |
2ff7e61e | 2717 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2718 | return 0; |
2719 | ||
2720 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2721 | btrfs_item_key_to_cpu(eb, &key, i); | |
2722 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2723 | continue; | |
2724 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2725 | found_type = btrfs_file_extent_type(eb, item); | |
2726 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2727 | continue; | |
2728 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2729 | continue; | |
2730 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2731 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2732 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2733 | if (ret) | |
2734 | break; | |
8c2a1a30 JB |
2735 | } |
2736 | ||
b89311ef | 2737 | return ret; |
8c2a1a30 JB |
2738 | } |
2739 | ||
9cfa3e34 | 2740 | static void |
32da5386 | 2741 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2742 | { |
2743 | atomic_inc(&bg->reservations); | |
2744 | } | |
2745 | ||
8b74c03e | 2746 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 2747 | { |
11833d66 YZ |
2748 | struct btrfs_caching_control *next; |
2749 | struct btrfs_caching_control *caching_ctl; | |
32da5386 | 2750 | struct btrfs_block_group *cache; |
e8569813 | 2751 | |
9e351cc8 | 2752 | down_write(&fs_info->commit_root_sem); |
25179201 | 2753 | |
11833d66 YZ |
2754 | list_for_each_entry_safe(caching_ctl, next, |
2755 | &fs_info->caching_block_groups, list) { | |
2756 | cache = caching_ctl->block_group; | |
32da5386 | 2757 | if (btrfs_block_group_done(cache)) { |
11833d66 YZ |
2758 | cache->last_byte_to_unpin = (u64)-1; |
2759 | list_del_init(&caching_ctl->list); | |
e3cb339f | 2760 | btrfs_put_caching_control(caching_ctl); |
e8569813 | 2761 | } else { |
11833d66 | 2762 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 2763 | } |
e8569813 | 2764 | } |
11833d66 YZ |
2765 | |
2766 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
2767 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
2768 | else | |
2769 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
2770 | ||
9e351cc8 | 2771 | up_write(&fs_info->commit_root_sem); |
8929ecfa | 2772 | |
67f9c220 | 2773 | btrfs_update_global_block_rsv(fs_info); |
e8569813 ZY |
2774 | } |
2775 | ||
c759c4e1 JB |
2776 | /* |
2777 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2778 | * what it should be based on the mount options. | |
2779 | */ | |
2780 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2781 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2782 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2783 | { |
2784 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2785 | |
2786 | *empty_cluster = 0; | |
2787 | if (btrfs_mixed_space_info(space_info)) | |
2788 | return ret; | |
2789 | ||
c759c4e1 | 2790 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2791 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2792 | if (btrfs_test_opt(fs_info, SSD)) |
2793 | *empty_cluster = SZ_2M; | |
2794 | else | |
ee22184b | 2795 | *empty_cluster = SZ_64K; |
583b7231 HK |
2796 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2797 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2798 | *empty_cluster = SZ_2M; | |
0b246afa | 2799 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2800 | } |
2801 | ||
2802 | return ret; | |
2803 | } | |
2804 | ||
2ff7e61e JM |
2805 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2806 | u64 start, u64 end, | |
678886bd | 2807 | const bool return_free_space) |
ccd467d6 | 2808 | { |
32da5386 | 2809 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2810 | struct btrfs_space_info *space_info; |
2811 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2812 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2813 | u64 len; |
c759c4e1 JB |
2814 | u64 total_unpinned = 0; |
2815 | u64 empty_cluster = 0; | |
7b398f8e | 2816 | bool readonly; |
ccd467d6 | 2817 | |
11833d66 | 2818 | while (start <= end) { |
7b398f8e | 2819 | readonly = false; |
11833d66 | 2820 | if (!cache || |
b3470b5d | 2821 | start >= cache->start + cache->length) { |
11833d66 YZ |
2822 | if (cache) |
2823 | btrfs_put_block_group(cache); | |
c759c4e1 | 2824 | total_unpinned = 0; |
11833d66 | 2825 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2826 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2827 | |
2ff7e61e | 2828 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2829 | cache->space_info, |
2830 | &empty_cluster); | |
2831 | empty_cluster <<= 1; | |
11833d66 YZ |
2832 | } |
2833 | ||
b3470b5d | 2834 | len = cache->start + cache->length - start; |
11833d66 YZ |
2835 | len = min(len, end + 1 - start); |
2836 | ||
2837 | if (start < cache->last_byte_to_unpin) { | |
2838 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
2839 | if (return_free_space) |
2840 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
2841 | } |
2842 | ||
f0486c68 | 2843 | start += len; |
c759c4e1 | 2844 | total_unpinned += len; |
7b398f8e | 2845 | space_info = cache->space_info; |
f0486c68 | 2846 | |
c759c4e1 JB |
2847 | /* |
2848 | * If this space cluster has been marked as fragmented and we've | |
2849 | * unpinned enough in this block group to potentially allow a | |
2850 | * cluster to be created inside of it go ahead and clear the | |
2851 | * fragmented check. | |
2852 | */ | |
2853 | if (cluster && cluster->fragmented && | |
2854 | total_unpinned > empty_cluster) { | |
2855 | spin_lock(&cluster->lock); | |
2856 | cluster->fragmented = 0; | |
2857 | spin_unlock(&cluster->lock); | |
2858 | } | |
2859 | ||
7b398f8e | 2860 | spin_lock(&space_info->lock); |
11833d66 YZ |
2861 | spin_lock(&cache->lock); |
2862 | cache->pinned -= len; | |
bb96c4e5 | 2863 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2864 | space_info->max_extent_size = 0; |
dec59fa3 EL |
2865 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2866 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
2867 | if (cache->ro) { |
2868 | space_info->bytes_readonly += len; | |
2869 | readonly = true; | |
2870 | } | |
11833d66 | 2871 | spin_unlock(&cache->lock); |
957780eb JB |
2872 | if (!readonly && return_free_space && |
2873 | global_rsv->space_info == space_info) { | |
2874 | u64 to_add = len; | |
92ac58ec | 2875 | |
7b398f8e JB |
2876 | spin_lock(&global_rsv->lock); |
2877 | if (!global_rsv->full) { | |
957780eb JB |
2878 | to_add = min(len, global_rsv->size - |
2879 | global_rsv->reserved); | |
2880 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2881 | btrfs_space_info_update_bytes_may_use(fs_info, |
2882 | space_info, to_add); | |
7b398f8e JB |
2883 | if (global_rsv->reserved >= global_rsv->size) |
2884 | global_rsv->full = 1; | |
957780eb | 2885 | len -= to_add; |
7b398f8e JB |
2886 | } |
2887 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
2888 | /* Add to any tickets we may have */ |
2889 | if (len) | |
18fa2284 JB |
2890 | btrfs_try_granting_tickets(fs_info, |
2891 | space_info); | |
7b398f8e JB |
2892 | } |
2893 | spin_unlock(&space_info->lock); | |
ccd467d6 | 2894 | } |
11833d66 YZ |
2895 | |
2896 | if (cache) | |
2897 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2898 | return 0; |
2899 | } | |
2900 | ||
5ead2dd0 | 2901 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2902 | { |
5ead2dd0 | 2903 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2904 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2905 | struct list_head *deleted_bgs; |
11833d66 | 2906 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2907 | u64 start; |
2908 | u64 end; | |
a28ec197 | 2909 | int ret; |
a28ec197 | 2910 | |
11833d66 YZ |
2911 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
2912 | unpin = &fs_info->freed_extents[1]; | |
2913 | else | |
2914 | unpin = &fs_info->freed_extents[0]; | |
2915 | ||
bf31f87f | 2916 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2917 | struct extent_state *cached_state = NULL; |
2918 | ||
d4b450cd | 2919 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2920 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2921 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2922 | if (ret) { |
2923 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2924 | break; |
d4b450cd | 2925 | } |
1f3c79a2 | 2926 | |
46b27f50 | 2927 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2928 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2929 | end + 1 - start, NULL); |
1f3c79a2 | 2930 | |
0e6ec385 | 2931 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2932 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2933 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2934 | free_extent_state(cached_state); |
b9473439 | 2935 | cond_resched(); |
a28ec197 | 2936 | } |
817d52f8 | 2937 | |
a2309300 DZ |
2938 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2939 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2940 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2941 | } |
b0643e59 | 2942 | |
e33e17ee JM |
2943 | /* |
2944 | * Transaction is finished. We don't need the lock anymore. We | |
2945 | * do need to clean up the block groups in case of a transaction | |
2946 | * abort. | |
2947 | */ | |
2948 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2949 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2950 | u64 trimmed = 0; | |
2951 | ||
2952 | ret = -EROFS; | |
bf31f87f | 2953 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2954 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2955 | block_group->start, |
2956 | block_group->length, | |
e33e17ee JM |
2957 | &trimmed); |
2958 | ||
2959 | list_del_init(&block_group->bg_list); | |
2960 | btrfs_put_block_group_trimming(block_group); | |
2961 | btrfs_put_block_group(block_group); | |
2962 | ||
2963 | if (ret) { | |
2964 | const char *errstr = btrfs_decode_error(ret); | |
2965 | btrfs_warn(fs_info, | |
913e1535 | 2966 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2967 | ret, errstr); |
2968 | } | |
2969 | } | |
2970 | ||
e20d96d6 CM |
2971 | return 0; |
2972 | } | |
2973 | ||
5d4f98a2 | 2974 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2975 | struct btrfs_delayed_ref_node *node, u64 parent, |
2976 | u64 root_objectid, u64 owner_objectid, | |
2977 | u64 owner_offset, int refs_to_drop, | |
2978 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2979 | { |
e72cb923 | 2980 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2981 | struct btrfs_key key; |
5d4f98a2 | 2982 | struct btrfs_path *path; |
1261ec42 | 2983 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2984 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2985 | struct btrfs_extent_item *ei; |
2986 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2987 | int ret; |
5d4f98a2 | 2988 | int is_data; |
952fccac CM |
2989 | int extent_slot = 0; |
2990 | int found_extent = 0; | |
2991 | int num_to_del = 1; | |
5d4f98a2 YZ |
2992 | u32 item_size; |
2993 | u64 refs; | |
c682f9b3 QW |
2994 | u64 bytenr = node->bytenr; |
2995 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2996 | int last_ref = 0; |
0b246afa | 2997 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 2998 | |
5caf2a00 | 2999 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3000 | if (!path) |
3001 | return -ENOMEM; | |
5f26f772 | 3002 | |
e4058b54 | 3003 | path->reada = READA_FORWARD; |
b9473439 | 3004 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3005 | |
3006 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
3007 | BUG_ON(!is_data && refs_to_drop != 1); | |
3008 | ||
3173a18f | 3009 | if (is_data) |
897ca819 | 3010 | skinny_metadata = false; |
3173a18f | 3011 | |
fbe4801b NB |
3012 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
3013 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 3014 | owner_offset); |
7bb86316 | 3015 | if (ret == 0) { |
952fccac | 3016 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3017 | while (extent_slot >= 0) { |
3018 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3019 | extent_slot); |
5d4f98a2 | 3020 | if (key.objectid != bytenr) |
952fccac | 3021 | break; |
5d4f98a2 YZ |
3022 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3023 | key.offset == num_bytes) { | |
952fccac CM |
3024 | found_extent = 1; |
3025 | break; | |
3026 | } | |
3173a18f JB |
3027 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3028 | key.offset == owner_objectid) { | |
3029 | found_extent = 1; | |
3030 | break; | |
3031 | } | |
952fccac CM |
3032 | if (path->slots[0] - extent_slot > 5) |
3033 | break; | |
5d4f98a2 | 3034 | extent_slot--; |
952fccac | 3035 | } |
a79865c6 | 3036 | |
31840ae1 | 3037 | if (!found_extent) { |
5d4f98a2 | 3038 | BUG_ON(iref); |
87cc7a8a | 3039 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3040 | refs_to_drop, |
fcebe456 | 3041 | is_data, &last_ref); |
005d6427 | 3042 | if (ret) { |
66642832 | 3043 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3044 | goto out; |
3045 | } | |
b3b4aa74 | 3046 | btrfs_release_path(path); |
b9473439 | 3047 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3048 | |
3049 | key.objectid = bytenr; | |
3050 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3051 | key.offset = num_bytes; | |
3052 | ||
3173a18f JB |
3053 | if (!is_data && skinny_metadata) { |
3054 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3055 | key.offset = owner_objectid; | |
3056 | } | |
3057 | ||
31840ae1 ZY |
3058 | ret = btrfs_search_slot(trans, extent_root, |
3059 | &key, path, -1, 1); | |
3173a18f JB |
3060 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3061 | /* | |
3062 | * Couldn't find our skinny metadata item, | |
3063 | * see if we have ye olde extent item. | |
3064 | */ | |
3065 | path->slots[0]--; | |
3066 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3067 | path->slots[0]); | |
3068 | if (key.objectid == bytenr && | |
3069 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3070 | key.offset == num_bytes) | |
3071 | ret = 0; | |
3072 | } | |
3073 | ||
3074 | if (ret > 0 && skinny_metadata) { | |
3075 | skinny_metadata = false; | |
9ce49a0b | 3076 | key.objectid = bytenr; |
3173a18f JB |
3077 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3078 | key.offset = num_bytes; | |
3079 | btrfs_release_path(path); | |
3080 | ret = btrfs_search_slot(trans, extent_root, | |
3081 | &key, path, -1, 1); | |
3082 | } | |
3083 | ||
f3465ca4 | 3084 | if (ret) { |
5d163e0e JM |
3085 | btrfs_err(info, |
3086 | "umm, got %d back from search, was looking for %llu", | |
3087 | ret, bytenr); | |
b783e62d | 3088 | if (ret > 0) |
a4f78750 | 3089 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3090 | } |
005d6427 | 3091 | if (ret < 0) { |
66642832 | 3092 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3093 | goto out; |
3094 | } | |
31840ae1 ZY |
3095 | extent_slot = path->slots[0]; |
3096 | } | |
fae7f21c | 3097 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3098 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3099 | btrfs_err(info, |
3100 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3101 | bytenr, parent, root_objectid, owner_objectid, |
3102 | owner_offset); | |
66642832 | 3103 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3104 | goto out; |
79787eaa | 3105 | } else { |
66642832 | 3106 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3107 | goto out; |
7bb86316 | 3108 | } |
5f39d397 CM |
3109 | |
3110 | leaf = path->nodes[0]; | |
5d4f98a2 | 3111 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3112 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3113 | ret = -EINVAL; |
3114 | btrfs_print_v0_err(info); | |
3115 | btrfs_abort_transaction(trans, ret); | |
3116 | goto out; | |
3117 | } | |
952fccac | 3118 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3119 | struct btrfs_extent_item); |
3173a18f JB |
3120 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3121 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
3122 | struct btrfs_tree_block_info *bi; |
3123 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
3124 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3125 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3126 | } | |
56bec294 | 3127 | |
5d4f98a2 | 3128 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3129 | if (refs < refs_to_drop) { |
5d163e0e JM |
3130 | btrfs_err(info, |
3131 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
3132 | refs_to_drop, refs, bytenr); | |
32b02538 | 3133 | ret = -EINVAL; |
66642832 | 3134 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
3135 | goto out; |
3136 | } | |
56bec294 | 3137 | refs -= refs_to_drop; |
5f39d397 | 3138 | |
5d4f98a2 YZ |
3139 | if (refs > 0) { |
3140 | if (extent_op) | |
3141 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3142 | /* | |
3143 | * In the case of inline back ref, reference count will | |
3144 | * be updated by remove_extent_backref | |
952fccac | 3145 | */ |
5d4f98a2 YZ |
3146 | if (iref) { |
3147 | BUG_ON(!found_extent); | |
3148 | } else { | |
3149 | btrfs_set_extent_refs(leaf, ei, refs); | |
3150 | btrfs_mark_buffer_dirty(leaf); | |
3151 | } | |
3152 | if (found_extent) { | |
87cc7a8a NB |
3153 | ret = remove_extent_backref(trans, path, iref, |
3154 | refs_to_drop, is_data, | |
3155 | &last_ref); | |
005d6427 | 3156 | if (ret) { |
66642832 | 3157 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3158 | goto out; |
3159 | } | |
952fccac | 3160 | } |
5d4f98a2 | 3161 | } else { |
5d4f98a2 YZ |
3162 | if (found_extent) { |
3163 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 3164 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
3165 | if (iref) { |
3166 | BUG_ON(path->slots[0] != extent_slot); | |
3167 | } else { | |
3168 | BUG_ON(path->slots[0] != extent_slot + 1); | |
3169 | path->slots[0] = extent_slot; | |
3170 | num_to_del = 2; | |
3171 | } | |
78fae27e | 3172 | } |
b9473439 | 3173 | |
fcebe456 | 3174 | last_ref = 1; |
952fccac CM |
3175 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3176 | num_to_del); | |
005d6427 | 3177 | if (ret) { |
66642832 | 3178 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3179 | goto out; |
3180 | } | |
b3b4aa74 | 3181 | btrfs_release_path(path); |
21af804c | 3182 | |
5d4f98a2 | 3183 | if (is_data) { |
40e046ac FM |
3184 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3185 | num_bytes); | |
005d6427 | 3186 | if (ret) { |
66642832 | 3187 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3188 | goto out; |
3189 | } | |
459931ec CM |
3190 | } |
3191 | ||
e7355e50 | 3192 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3193 | if (ret) { |
66642832 | 3194 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3195 | goto out; |
3196 | } | |
3197 | ||
ade4b516 | 3198 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3199 | if (ret) { |
66642832 | 3200 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3201 | goto out; |
3202 | } | |
a28ec197 | 3203 | } |
fcebe456 JB |
3204 | btrfs_release_path(path); |
3205 | ||
79787eaa | 3206 | out: |
5caf2a00 | 3207 | btrfs_free_path(path); |
a28ec197 CM |
3208 | return ret; |
3209 | } | |
3210 | ||
1887be66 | 3211 | /* |
f0486c68 | 3212 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3213 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3214 | * a given extent, and if there are no other delayed refs to be processed, it | |
3215 | * removes it from the tree. | |
3216 | */ | |
3217 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3218 | u64 bytenr) |
1887be66 CM |
3219 | { |
3220 | struct btrfs_delayed_ref_head *head; | |
3221 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3222 | int ret = 0; |
1887be66 CM |
3223 | |
3224 | delayed_refs = &trans->transaction->delayed_refs; | |
3225 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3226 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3227 | if (!head) |
cf93da7b | 3228 | goto out_delayed_unlock; |
1887be66 | 3229 | |
d7df2c79 | 3230 | spin_lock(&head->lock); |
e3d03965 | 3231 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3232 | goto out; |
3233 | ||
bedc6617 JB |
3234 | if (cleanup_extent_op(head) != NULL) |
3235 | goto out; | |
5d4f98a2 | 3236 | |
1887be66 CM |
3237 | /* |
3238 | * waiting for the lock here would deadlock. If someone else has it | |
3239 | * locked they are already in the process of dropping it anyway | |
3240 | */ | |
3241 | if (!mutex_trylock(&head->mutex)) | |
3242 | goto out; | |
3243 | ||
d7baffda | 3244 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3245 | head->processing = 0; |
d7baffda | 3246 | |
d7df2c79 | 3247 | spin_unlock(&head->lock); |
1887be66 CM |
3248 | spin_unlock(&delayed_refs->lock); |
3249 | ||
f0486c68 YZ |
3250 | BUG_ON(head->extent_op); |
3251 | if (head->must_insert_reserved) | |
3252 | ret = 1; | |
3253 | ||
31890da0 | 3254 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3255 | mutex_unlock(&head->mutex); |
d278850e | 3256 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3257 | return ret; |
1887be66 | 3258 | out: |
d7df2c79 | 3259 | spin_unlock(&head->lock); |
cf93da7b CM |
3260 | |
3261 | out_delayed_unlock: | |
1887be66 CM |
3262 | spin_unlock(&delayed_refs->lock); |
3263 | return 0; | |
3264 | } | |
3265 | ||
f0486c68 YZ |
3266 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3267 | struct btrfs_root *root, | |
3268 | struct extent_buffer *buf, | |
5581a51a | 3269 | u64 parent, int last_ref) |
f0486c68 | 3270 | { |
0b246afa | 3271 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3272 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 3273 | int pin = 1; |
f0486c68 YZ |
3274 | int ret; |
3275 | ||
ed4f255b QW |
3276 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3277 | buf->start, buf->len, parent); | |
3278 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3279 | root->root_key.objectid); | |
3280 | ||
f0486c68 | 3281 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
3282 | int old_ref_mod, new_ref_mod; |
3283 | ||
8a5040f7 | 3284 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 3285 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 3286 | &old_ref_mod, &new_ref_mod); |
79787eaa | 3287 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 3288 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
3289 | } |
3290 | ||
0a16c7d7 | 3291 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3292 | struct btrfs_block_group *cache; |
6219872d | 3293 | |
f0486c68 | 3294 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3295 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 3296 | if (!ret) |
37be25bc | 3297 | goto out; |
f0486c68 YZ |
3298 | } |
3299 | ||
4da8b76d | 3300 | pin = 0; |
0b246afa | 3301 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3302 | |
f0486c68 | 3303 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
fdf08605 | 3304 | pin_down_extent(cache, buf->start, buf->len, 1); |
6219872d | 3305 | btrfs_put_block_group(cache); |
37be25bc | 3306 | goto out; |
f0486c68 YZ |
3307 | } |
3308 | ||
3309 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3310 | ||
3311 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3312 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3313 | btrfs_put_block_group(cache); |
71ff6437 | 3314 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3315 | } |
3316 | out: | |
b150a4f1 | 3317 | if (pin) |
78192442 | 3318 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 3319 | |
0a16c7d7 OS |
3320 | if (last_ref) { |
3321 | /* | |
3322 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3323 | * matter anymore. | |
3324 | */ | |
3325 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3326 | } | |
f0486c68 YZ |
3327 | } |
3328 | ||
79787eaa | 3329 | /* Can return -ENOMEM */ |
ffd4bb2a | 3330 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3331 | { |
ffd4bb2a | 3332 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 3333 | int old_ref_mod, new_ref_mod; |
925baedd CM |
3334 | int ret; |
3335 | ||
f5ee5c9a | 3336 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3337 | return 0; |
fccb84c9 | 3338 | |
56bec294 CM |
3339 | /* |
3340 | * tree log blocks never actually go into the extent allocation | |
3341 | * tree, just update pinning info and exit early. | |
56bec294 | 3342 | */ |
ffd4bb2a QW |
3343 | if ((ref->type == BTRFS_REF_METADATA && |
3344 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3345 | (ref->type == BTRFS_REF_DATA && | |
3346 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3347 | /* unlocks the pinned mutex */ |
ffd4bb2a | 3348 | btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1); |
d7eae340 | 3349 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 3350 | ret = 0; |
ffd4bb2a QW |
3351 | } else if (ref->type == BTRFS_REF_METADATA) { |
3352 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 3353 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 3354 | } else { |
ffd4bb2a | 3355 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 3356 | &old_ref_mod, &new_ref_mod); |
56bec294 | 3357 | } |
d7eae340 | 3358 | |
ffd4bb2a QW |
3359 | if (!((ref->type == BTRFS_REF_METADATA && |
3360 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3361 | (ref->type == BTRFS_REF_DATA && | |
3362 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3363 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3364 | |
ddf30cf0 | 3365 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 3366 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 3367 | |
925baedd CM |
3368 | return ret; |
3369 | } | |
3370 | ||
817d52f8 | 3371 | enum btrfs_loop_type { |
f262fa8d DS |
3372 | LOOP_CACHING_NOWAIT, |
3373 | LOOP_CACHING_WAIT, | |
3374 | LOOP_ALLOC_CHUNK, | |
3375 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3376 | }; |
3377 | ||
e570fd27 | 3378 | static inline void |
32da5386 | 3379 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3380 | int delalloc) |
3381 | { | |
3382 | if (delalloc) | |
3383 | down_read(&cache->data_rwsem); | |
3384 | } | |
3385 | ||
32da5386 | 3386 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3387 | int delalloc) |
3388 | { | |
3389 | btrfs_get_block_group(cache); | |
3390 | if (delalloc) | |
3391 | down_read(&cache->data_rwsem); | |
3392 | } | |
3393 | ||
32da5386 DS |
3394 | static struct btrfs_block_group *btrfs_lock_cluster( |
3395 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3396 | struct btrfs_free_cluster *cluster, |
3397 | int delalloc) | |
3398 | { | |
32da5386 | 3399 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3400 | |
e570fd27 | 3401 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3402 | while (1) { |
3403 | used_bg = cluster->block_group; | |
3404 | if (!used_bg) | |
3405 | return NULL; | |
3406 | ||
3407 | if (used_bg == block_group) | |
e570fd27 MX |
3408 | return used_bg; |
3409 | ||
6719afdc | 3410 | btrfs_get_block_group(used_bg); |
e570fd27 | 3411 | |
6719afdc GU |
3412 | if (!delalloc) |
3413 | return used_bg; | |
e570fd27 | 3414 | |
6719afdc GU |
3415 | if (down_read_trylock(&used_bg->data_rwsem)) |
3416 | return used_bg; | |
e570fd27 | 3417 | |
6719afdc | 3418 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3419 | |
e321f8a8 LB |
3420 | /* We should only have one-level nested. */ |
3421 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3422 | |
6719afdc GU |
3423 | spin_lock(&cluster->refill_lock); |
3424 | if (used_bg == cluster->block_group) | |
3425 | return used_bg; | |
e570fd27 | 3426 | |
6719afdc GU |
3427 | up_read(&used_bg->data_rwsem); |
3428 | btrfs_put_block_group(used_bg); | |
3429 | } | |
e570fd27 MX |
3430 | } |
3431 | ||
3432 | static inline void | |
32da5386 | 3433 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3434 | int delalloc) |
3435 | { | |
3436 | if (delalloc) | |
3437 | up_read(&cache->data_rwsem); | |
3438 | btrfs_put_block_group(cache); | |
3439 | } | |
3440 | ||
b4bd745d QW |
3441 | /* |
3442 | * Structure used internally for find_free_extent() function. Wraps needed | |
3443 | * parameters. | |
3444 | */ | |
3445 | struct find_free_extent_ctl { | |
3446 | /* Basic allocation info */ | |
b4bd745d QW |
3447 | u64 num_bytes; |
3448 | u64 empty_size; | |
3449 | u64 flags; | |
3450 | int delalloc; | |
3451 | ||
3452 | /* Where to start the search inside the bg */ | |
3453 | u64 search_start; | |
3454 | ||
3455 | /* For clustered allocation */ | |
3456 | u64 empty_cluster; | |
3457 | ||
3458 | bool have_caching_bg; | |
3459 | bool orig_have_caching_bg; | |
3460 | ||
3461 | /* RAID index, converted from flags */ | |
3462 | int index; | |
3463 | ||
e72d79d6 QW |
3464 | /* |
3465 | * Current loop number, check find_free_extent_update_loop() for details | |
3466 | */ | |
b4bd745d QW |
3467 | int loop; |
3468 | ||
3469 | /* | |
3470 | * Whether we're refilling a cluster, if true we need to re-search | |
3471 | * current block group but don't try to refill the cluster again. | |
3472 | */ | |
3473 | bool retry_clustered; | |
3474 | ||
3475 | /* | |
3476 | * Whether we're updating free space cache, if true we need to re-search | |
3477 | * current block group but don't try updating free space cache again. | |
3478 | */ | |
3479 | bool retry_unclustered; | |
3480 | ||
3481 | /* If current block group is cached */ | |
3482 | int cached; | |
3483 | ||
3484 | /* Max contiguous hole found */ | |
3485 | u64 max_extent_size; | |
3486 | ||
3487 | /* Total free space from free space cache, not always contiguous */ | |
3488 | u64 total_free_space; | |
3489 | ||
3490 | /* Found result */ | |
3491 | u64 found_offset; | |
3492 | }; | |
3493 | ||
d06e3bb6 QW |
3494 | |
3495 | /* | |
3496 | * Helper function for find_free_extent(). | |
3497 | * | |
3498 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3499 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3500 | * Return >0 to inform caller that we find nothing | |
3501 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3502 | */ | |
32da5386 | 3503 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
d06e3bb6 QW |
3504 | struct btrfs_free_cluster *last_ptr, |
3505 | struct find_free_extent_ctl *ffe_ctl, | |
32da5386 | 3506 | struct btrfs_block_group **cluster_bg_ret) |
d06e3bb6 | 3507 | { |
32da5386 | 3508 | struct btrfs_block_group *cluster_bg; |
d06e3bb6 QW |
3509 | u64 aligned_cluster; |
3510 | u64 offset; | |
3511 | int ret; | |
3512 | ||
3513 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3514 | if (!cluster_bg) | |
3515 | goto refill_cluster; | |
3516 | if (cluster_bg != bg && (cluster_bg->ro || | |
3517 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3518 | goto release_cluster; | |
3519 | ||
3520 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3521 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3522 | &ffe_ctl->max_extent_size); |
3523 | if (offset) { | |
3524 | /* We have a block, we're done */ | |
3525 | spin_unlock(&last_ptr->refill_lock); | |
3526 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3527 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3528 | *cluster_bg_ret = cluster_bg; | |
3529 | ffe_ctl->found_offset = offset; | |
3530 | return 0; | |
3531 | } | |
3532 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3533 | ||
3534 | release_cluster: | |
3535 | /* | |
3536 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3537 | * lets just skip it and let the allocator find whatever block it can | |
3538 | * find. If we reach this point, we will have tried the cluster | |
3539 | * allocator plenty of times and not have found anything, so we are | |
3540 | * likely way too fragmented for the clustering stuff to find anything. | |
3541 | * | |
3542 | * However, if the cluster is taken from the current block group, | |
3543 | * release the cluster first, so that we stand a better chance of | |
3544 | * succeeding in the unclustered allocation. | |
3545 | */ | |
3546 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3547 | spin_unlock(&last_ptr->refill_lock); | |
3548 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3549 | return -ENOENT; | |
3550 | } | |
3551 | ||
3552 | /* This cluster didn't work out, free it and start over */ | |
3553 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3554 | ||
3555 | if (cluster_bg != bg) | |
3556 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3557 | ||
3558 | refill_cluster: | |
3559 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3560 | spin_unlock(&last_ptr->refill_lock); | |
3561 | return -ENOENT; | |
3562 | } | |
3563 | ||
3564 | aligned_cluster = max_t(u64, | |
3565 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3566 | bg->full_stripe_len); | |
2ceeae2e DS |
3567 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3568 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3569 | if (ret == 0) { |
3570 | /* Now pull our allocation out of this cluster */ | |
3571 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3572 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3573 | &ffe_ctl->max_extent_size); | |
3574 | if (offset) { | |
3575 | /* We found one, proceed */ | |
3576 | spin_unlock(&last_ptr->refill_lock); | |
3577 | trace_btrfs_reserve_extent_cluster(bg, | |
3578 | ffe_ctl->search_start, | |
3579 | ffe_ctl->num_bytes); | |
3580 | ffe_ctl->found_offset = offset; | |
3581 | return 0; | |
3582 | } | |
3583 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3584 | !ffe_ctl->retry_clustered) { | |
3585 | spin_unlock(&last_ptr->refill_lock); | |
3586 | ||
3587 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3588 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3589 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3590 | return -EAGAIN; | |
3591 | } | |
3592 | /* | |
3593 | * At this point we either didn't find a cluster or we weren't able to | |
3594 | * allocate a block from our cluster. Free the cluster we've been | |
3595 | * trying to use, and go to the next block group. | |
3596 | */ | |
3597 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3598 | spin_unlock(&last_ptr->refill_lock); | |
3599 | return 1; | |
3600 | } | |
3601 | ||
e1a41848 QW |
3602 | /* |
3603 | * Return >0 to inform caller that we find nothing | |
3604 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3605 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3606 | */ | |
32da5386 | 3607 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
e1a41848 QW |
3608 | struct btrfs_free_cluster *last_ptr, |
3609 | struct find_free_extent_ctl *ffe_ctl) | |
3610 | { | |
3611 | u64 offset; | |
3612 | ||
3613 | /* | |
3614 | * We are doing an unclustered allocation, set the fragmented flag so | |
3615 | * we don't bother trying to setup a cluster again until we get more | |
3616 | * space. | |
3617 | */ | |
3618 | if (unlikely(last_ptr)) { | |
3619 | spin_lock(&last_ptr->lock); | |
3620 | last_ptr->fragmented = 1; | |
3621 | spin_unlock(&last_ptr->lock); | |
3622 | } | |
3623 | if (ffe_ctl->cached) { | |
3624 | struct btrfs_free_space_ctl *free_space_ctl; | |
3625 | ||
3626 | free_space_ctl = bg->free_space_ctl; | |
3627 | spin_lock(&free_space_ctl->tree_lock); | |
3628 | if (free_space_ctl->free_space < | |
3629 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3630 | ffe_ctl->empty_size) { | |
3631 | ffe_ctl->total_free_space = max_t(u64, | |
3632 | ffe_ctl->total_free_space, | |
3633 | free_space_ctl->free_space); | |
3634 | spin_unlock(&free_space_ctl->tree_lock); | |
3635 | return 1; | |
3636 | } | |
3637 | spin_unlock(&free_space_ctl->tree_lock); | |
3638 | } | |
3639 | ||
3640 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3641 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3642 | &ffe_ctl->max_extent_size); | |
3643 | ||
3644 | /* | |
3645 | * If we didn't find a chunk, and we haven't failed on this block group | |
3646 | * before, and this block group is in the middle of caching and we are | |
3647 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3648 | * set @retry_unclustered to true. | |
3649 | * | |
3650 | * If @retry_unclustered is true then we've already waited on this | |
3651 | * block group once and should move on to the next block group. | |
3652 | */ | |
3653 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3654 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3655 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3656 | ffe_ctl->empty_size); | |
e1a41848 QW |
3657 | ffe_ctl->retry_unclustered = true; |
3658 | return -EAGAIN; | |
3659 | } else if (!offset) { | |
3660 | return 1; | |
3661 | } | |
3662 | ffe_ctl->found_offset = offset; | |
3663 | return 0; | |
3664 | } | |
3665 | ||
e72d79d6 QW |
3666 | /* |
3667 | * Return >0 means caller needs to re-search for free extent | |
3668 | * Return 0 means we have the needed free extent. | |
3669 | * Return <0 means we failed to locate any free extent. | |
3670 | */ | |
3671 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
3672 | struct btrfs_free_cluster *last_ptr, | |
3673 | struct btrfs_key *ins, | |
3674 | struct find_free_extent_ctl *ffe_ctl, | |
3675 | int full_search, bool use_cluster) | |
3676 | { | |
3677 | struct btrfs_root *root = fs_info->extent_root; | |
3678 | int ret; | |
3679 | ||
3680 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3681 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3682 | ffe_ctl->orig_have_caching_bg = true; | |
3683 | ||
3684 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3685 | ffe_ctl->have_caching_bg) | |
3686 | return 1; | |
3687 | ||
3688 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3689 | return 1; | |
3690 | ||
3691 | if (ins->objectid) { | |
3692 | if (!use_cluster && last_ptr) { | |
3693 | spin_lock(&last_ptr->lock); | |
3694 | last_ptr->window_start = ins->objectid; | |
3695 | spin_unlock(&last_ptr->lock); | |
3696 | } | |
3697 | return 0; | |
3698 | } | |
3699 | ||
3700 | /* | |
3701 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3702 | * caching kthreads as we move along | |
3703 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3704 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3705 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3706 | * again | |
3707 | */ | |
3708 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3709 | ffe_ctl->index = 0; | |
3710 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3711 | /* | |
3712 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3713 | * don't have any uncached bgs and we've already done a | |
3714 | * full search through. | |
3715 | */ | |
3716 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3717 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3718 | else | |
3719 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3720 | } else { | |
3721 | ffe_ctl->loop++; | |
3722 | } | |
3723 | ||
3724 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3725 | struct btrfs_trans_handle *trans; | |
3726 | int exist = 0; | |
3727 | ||
3728 | trans = current->journal_info; | |
3729 | if (trans) | |
3730 | exist = 1; | |
3731 | else | |
3732 | trans = btrfs_join_transaction(root); | |
3733 | ||
3734 | if (IS_ERR(trans)) { | |
3735 | ret = PTR_ERR(trans); | |
3736 | return ret; | |
3737 | } | |
3738 | ||
fc471cb0 JB |
3739 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3740 | CHUNK_ALLOC_FORCE); | |
e72d79d6 QW |
3741 | |
3742 | /* | |
3743 | * If we can't allocate a new chunk we've already looped | |
3744 | * through at least once, move on to the NO_EMPTY_SIZE | |
3745 | * case. | |
3746 | */ | |
3747 | if (ret == -ENOSPC) | |
3748 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3749 | ||
3750 | /* Do not bail out on ENOSPC since we can do more. */ | |
3751 | if (ret < 0 && ret != -ENOSPC) | |
3752 | btrfs_abort_transaction(trans, ret); | |
3753 | else | |
3754 | ret = 0; | |
3755 | if (!exist) | |
3756 | btrfs_end_transaction(trans); | |
3757 | if (ret) | |
3758 | return ret; | |
3759 | } | |
3760 | ||
3761 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
3762 | /* | |
3763 | * Don't loop again if we already have no empty_size and | |
3764 | * no empty_cluster. | |
3765 | */ | |
3766 | if (ffe_ctl->empty_size == 0 && | |
3767 | ffe_ctl->empty_cluster == 0) | |
3768 | return -ENOSPC; | |
3769 | ffe_ctl->empty_size = 0; | |
3770 | ffe_ctl->empty_cluster = 0; | |
3771 | } | |
3772 | return 1; | |
3773 | } | |
3774 | return -ENOSPC; | |
3775 | } | |
3776 | ||
fec577fb CM |
3777 | /* |
3778 | * walks the btree of allocated extents and find a hole of a given size. | |
3779 | * The key ins is changed to record the hole: | |
a4820398 | 3780 | * ins->objectid == start position |
62e2749e | 3781 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 3782 | * ins->offset == the size of the hole. |
fec577fb | 3783 | * Any available blocks before search_start are skipped. |
a4820398 MX |
3784 | * |
3785 | * If there is no suitable free space, we will record the max size of | |
3786 | * the free space extent currently. | |
e72d79d6 QW |
3787 | * |
3788 | * The overall logic and call chain: | |
3789 | * | |
3790 | * find_free_extent() | |
3791 | * |- Iterate through all block groups | |
3792 | * | |- Get a valid block group | |
3793 | * | |- Try to do clustered allocation in that block group | |
3794 | * | |- Try to do unclustered allocation in that block group | |
3795 | * | |- Check if the result is valid | |
3796 | * | | |- If valid, then exit | |
3797 | * | |- Jump to next block group | |
3798 | * | | |
3799 | * |- Push harder to find free extents | |
3800 | * |- If not found, re-iterate all block groups | |
fec577fb | 3801 | */ |
87bde3cd | 3802 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
3803 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
3804 | u64 hint_byte, struct btrfs_key *ins, | |
3805 | u64 flags, int delalloc) | |
fec577fb | 3806 | { |
80eb234a | 3807 | int ret = 0; |
db8fe64f | 3808 | int cache_block_group_error = 0; |
fa9c0d79 | 3809 | struct btrfs_free_cluster *last_ptr = NULL; |
32da5386 | 3810 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 3811 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 3812 | struct btrfs_space_info *space_info; |
67377734 | 3813 | bool use_cluster = true; |
a5e681d9 | 3814 | bool full_search = false; |
fec577fb | 3815 | |
0b246afa | 3816 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 3817 | |
b4bd745d QW |
3818 | ffe_ctl.num_bytes = num_bytes; |
3819 | ffe_ctl.empty_size = empty_size; | |
3820 | ffe_ctl.flags = flags; | |
3821 | ffe_ctl.search_start = 0; | |
3822 | ffe_ctl.retry_clustered = false; | |
3823 | ffe_ctl.retry_unclustered = false; | |
3824 | ffe_ctl.delalloc = delalloc; | |
3825 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
3826 | ffe_ctl.have_caching_bg = false; | |
3827 | ffe_ctl.orig_have_caching_bg = false; | |
3828 | ffe_ctl.found_offset = 0; | |
3829 | ||
962a298f | 3830 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
3831 | ins->objectid = 0; |
3832 | ins->offset = 0; | |
b1a4d965 | 3833 | |
71ff6437 | 3834 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 3835 | |
280c2908 | 3836 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 3837 | if (!space_info) { |
0b246afa | 3838 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
3839 | return -ENOSPC; |
3840 | } | |
2552d17e | 3841 | |
67377734 | 3842 | /* |
4f4db217 JB |
3843 | * If our free space is heavily fragmented we may not be able to make |
3844 | * big contiguous allocations, so instead of doing the expensive search | |
3845 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3846 | * can go ahead and search for a more manageable chunk. | |
3847 | * | |
3848 | * If our max_extent_size is large enough for our allocation simply | |
3849 | * disable clustering since we will likely not be able to find enough | |
3850 | * space to create a cluster and induce latency trying. | |
67377734 | 3851 | */ |
4f4db217 JB |
3852 | if (unlikely(space_info->max_extent_size)) { |
3853 | spin_lock(&space_info->lock); | |
3854 | if (space_info->max_extent_size && | |
3855 | num_bytes > space_info->max_extent_size) { | |
3856 | ins->offset = space_info->max_extent_size; | |
3857 | spin_unlock(&space_info->lock); | |
3858 | return -ENOSPC; | |
3859 | } else if (space_info->max_extent_size) { | |
3860 | use_cluster = false; | |
3861 | } | |
3862 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 3863 | } |
0f9dd46c | 3864 | |
b4bd745d QW |
3865 | last_ptr = fetch_cluster_info(fs_info, space_info, |
3866 | &ffe_ctl.empty_cluster); | |
239b14b3 | 3867 | if (last_ptr) { |
fa9c0d79 CM |
3868 | spin_lock(&last_ptr->lock); |
3869 | if (last_ptr->block_group) | |
3870 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
3871 | if (last_ptr->fragmented) { |
3872 | /* | |
3873 | * We still set window_start so we can keep track of the | |
3874 | * last place we found an allocation to try and save | |
3875 | * some time. | |
3876 | */ | |
3877 | hint_byte = last_ptr->window_start; | |
3878 | use_cluster = false; | |
3879 | } | |
fa9c0d79 | 3880 | spin_unlock(&last_ptr->lock); |
239b14b3 | 3881 | } |
fa9c0d79 | 3882 | |
b4bd745d QW |
3883 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
3884 | first_logical_byte(fs_info, 0)); | |
3885 | ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte); | |
3886 | if (ffe_ctl.search_start == hint_byte) { | |
3887 | block_group = btrfs_lookup_block_group(fs_info, | |
3888 | ffe_ctl.search_start); | |
817d52f8 JB |
3889 | /* |
3890 | * we don't want to use the block group if it doesn't match our | |
3891 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
3892 | * |
3893 | * However if we are re-searching with an ideal block group | |
3894 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 3895 | */ |
b6919a58 | 3896 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 3897 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 3898 | down_read(&space_info->groups_sem); |
44fb5511 CM |
3899 | if (list_empty(&block_group->list) || |
3900 | block_group->ro) { | |
3901 | /* | |
3902 | * someone is removing this block group, | |
3903 | * we can't jump into the have_block_group | |
3904 | * target because our list pointers are not | |
3905 | * valid | |
3906 | */ | |
3907 | btrfs_put_block_group(block_group); | |
3908 | up_read(&space_info->groups_sem); | |
ccf0e725 | 3909 | } else { |
b4bd745d | 3910 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 3911 | block_group->flags); |
e570fd27 | 3912 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 3913 | goto have_block_group; |
ccf0e725 | 3914 | } |
2552d17e | 3915 | } else if (block_group) { |
fa9c0d79 | 3916 | btrfs_put_block_group(block_group); |
2552d17e | 3917 | } |
42e70e7a | 3918 | } |
2552d17e | 3919 | search: |
b4bd745d QW |
3920 | ffe_ctl.have_caching_bg = false; |
3921 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
3922 | ffe_ctl.index == 0) | |
a5e681d9 | 3923 | full_search = true; |
80eb234a | 3924 | down_read(&space_info->groups_sem); |
b4bd745d QW |
3925 | list_for_each_entry(block_group, |
3926 | &space_info->block_groups[ffe_ctl.index], list) { | |
14443937 JM |
3927 | /* If the block group is read-only, we can skip it entirely. */ |
3928 | if (unlikely(block_group->ro)) | |
3929 | continue; | |
3930 | ||
e570fd27 | 3931 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 3932 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 3933 | |
83a50de9 CM |
3934 | /* |
3935 | * this can happen if we end up cycling through all the | |
3936 | * raid types, but we want to make sure we only allocate | |
3937 | * for the proper type. | |
3938 | */ | |
b6919a58 | 3939 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 3940 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 3941 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 3942 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
3943 | BTRFS_BLOCK_GROUP_RAID10; |
3944 | ||
3945 | /* | |
3946 | * if they asked for extra copies and this block group | |
3947 | * doesn't provide them, bail. This does allow us to | |
3948 | * fill raid0 from raid1. | |
3949 | */ | |
b6919a58 | 3950 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 3951 | goto loop; |
2a28468e QW |
3952 | |
3953 | /* | |
3954 | * This block group has different flags than we want. | |
3955 | * It's possible that we have MIXED_GROUP flag but no | |
3956 | * block group is mixed. Just skip such block group. | |
3957 | */ | |
3958 | btrfs_release_block_group(block_group, delalloc); | |
3959 | continue; | |
83a50de9 CM |
3960 | } |
3961 | ||
2552d17e | 3962 | have_block_group: |
32da5386 | 3963 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
3964 | if (unlikely(!ffe_ctl.cached)) { |
3965 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 3966 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
3967 | |
3968 | /* | |
3969 | * If we get ENOMEM here or something else we want to | |
3970 | * try other block groups, because it may not be fatal. | |
3971 | * However if we can't find anything else we need to | |
3972 | * save our return here so that we return the actual | |
3973 | * error that caused problems, not ENOSPC. | |
3974 | */ | |
3975 | if (ret < 0) { | |
3976 | if (!cache_block_group_error) | |
3977 | cache_block_group_error = ret; | |
3978 | ret = 0; | |
3979 | goto loop; | |
3980 | } | |
1d4284bd | 3981 | ret = 0; |
817d52f8 JB |
3982 | } |
3983 | ||
36cce922 JB |
3984 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
3985 | goto loop; | |
0f9dd46c | 3986 | |
0a24325e | 3987 | /* |
062c05c4 AO |
3988 | * Ok we want to try and use the cluster allocator, so |
3989 | * lets look there | |
0a24325e | 3990 | */ |
c759c4e1 | 3991 | if (last_ptr && use_cluster) { |
32da5386 | 3992 | struct btrfs_block_group *cluster_bg = NULL; |
fa9c0d79 | 3993 | |
d06e3bb6 QW |
3994 | ret = find_free_extent_clustered(block_group, last_ptr, |
3995 | &ffe_ctl, &cluster_bg); | |
062c05c4 | 3996 | |
fa9c0d79 | 3997 | if (ret == 0) { |
d06e3bb6 QW |
3998 | if (cluster_bg && cluster_bg != block_group) { |
3999 | btrfs_release_block_group(block_group, | |
4000 | delalloc); | |
4001 | block_group = cluster_bg; | |
fa9c0d79 | 4002 | } |
d06e3bb6 QW |
4003 | goto checks; |
4004 | } else if (ret == -EAGAIN) { | |
817d52f8 | 4005 | goto have_block_group; |
d06e3bb6 QW |
4006 | } else if (ret > 0) { |
4007 | goto loop; | |
fa9c0d79 | 4008 | } |
d06e3bb6 | 4009 | /* ret == -ENOENT case falls through */ |
fa9c0d79 CM |
4010 | } |
4011 | ||
e1a41848 QW |
4012 | ret = find_free_extent_unclustered(block_group, last_ptr, |
4013 | &ffe_ctl); | |
4014 | if (ret == -EAGAIN) | |
817d52f8 | 4015 | goto have_block_group; |
e1a41848 | 4016 | else if (ret > 0) |
1cdda9b8 | 4017 | goto loop; |
e1a41848 | 4018 | /* ret == 0 case falls through */ |
fa9c0d79 | 4019 | checks: |
b4bd745d QW |
4020 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4021 | fs_info->stripesize); | |
25179201 | 4022 | |
2552d17e | 4023 | /* move on to the next group */ |
b4bd745d | 4024 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4025 | block_group->start + block_group->length) { |
b4bd745d QW |
4026 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4027 | num_bytes); | |
2552d17e | 4028 | goto loop; |
6226cb0a | 4029 | } |
f5a31e16 | 4030 | |
b4bd745d QW |
4031 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
4032 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
4033 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4034 | |
18513091 WX |
4035 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4036 | num_bytes, delalloc); | |
f0486c68 | 4037 | if (ret == -EAGAIN) { |
b4bd745d QW |
4038 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4039 | num_bytes); | |
2552d17e | 4040 | goto loop; |
0f9dd46c | 4041 | } |
9cfa3e34 | 4042 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4043 | |
f0486c68 | 4044 | /* we are all good, lets return */ |
b4bd745d | 4045 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4046 | ins->offset = num_bytes; |
d2fb3437 | 4047 | |
b4bd745d QW |
4048 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4049 | num_bytes); | |
e570fd27 | 4050 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4051 | break; |
4052 | loop: | |
b4bd745d QW |
4053 | ffe_ctl.retry_clustered = false; |
4054 | ffe_ctl.retry_unclustered = false; | |
3e72ee88 | 4055 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
b4bd745d | 4056 | ffe_ctl.index); |
e570fd27 | 4057 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 4058 | cond_resched(); |
2552d17e JB |
4059 | } |
4060 | up_read(&space_info->groups_sem); | |
4061 | ||
e72d79d6 QW |
4062 | ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl, |
4063 | full_search, use_cluster); | |
4064 | if (ret > 0) | |
b742bb82 YZ |
4065 | goto search; |
4066 | ||
db8fe64f | 4067 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4068 | /* |
4069 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4070 | * any contiguous hole. | |
4071 | */ | |
4072 | if (!ffe_ctl.max_extent_size) | |
4073 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4074 | spin_lock(&space_info->lock); |
b4bd745d | 4075 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4076 | spin_unlock(&space_info->lock); |
b4bd745d | 4077 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4078 | } else if (ret == -ENOSPC) { |
4079 | ret = cache_block_group_error; | |
4f4db217 | 4080 | } |
0f70abe2 | 4081 | return ret; |
fec577fb | 4082 | } |
ec44a35c | 4083 | |
6f47c706 NB |
4084 | /* |
4085 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4086 | * hole that is at least as big as @num_bytes. | |
4087 | * | |
4088 | * @root - The root that will contain this extent | |
4089 | * | |
4090 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4091 | * is used for accounting purposes. This value differs | |
4092 | * from @num_bytes only in the case of compressed extents. | |
4093 | * | |
4094 | * @num_bytes - Number of bytes to allocate on-disk. | |
4095 | * | |
4096 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4097 | * allocator should try to satisfy. In some cases | |
4098 | * @num_bytes may be larger than what is required and if | |
4099 | * the filesystem is fragmented then allocation fails. | |
4100 | * However, the presence of @min_alloc_size gives a | |
4101 | * chance to try and satisfy the smaller allocation. | |
4102 | * | |
4103 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4104 | * size in bytes the allocator should try to find free | |
4105 | * next to the block it returns. This is just a hint and | |
4106 | * may be ignored by the allocator. | |
4107 | * | |
4108 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4109 | * address passed. It might be ignored. | |
4110 | * | |
4111 | * @ins - This key is modified to record the found hole. It will | |
4112 | * have the following values: | |
4113 | * ins->objectid == start position | |
4114 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4115 | * ins->offset == the size of the hole. | |
4116 | * | |
4117 | * @is_data - Boolean flag indicating whether an extent is | |
4118 | * allocated for data (true) or metadata (false) | |
4119 | * | |
4120 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4121 | * delalloc or not. If 'true' data_rwsem of block groups | |
4122 | * is going to be acquired. | |
4123 | * | |
4124 | * | |
4125 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4126 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4127 | * largest available hole the allocator managed to find. | |
4128 | */ | |
18513091 | 4129 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4130 | u64 num_bytes, u64 min_alloc_size, |
4131 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4132 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4133 | { |
ab8d0fc4 | 4134 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4135 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4136 | u64 flags; |
fec577fb | 4137 | int ret; |
925baedd | 4138 | |
1b86826d | 4139 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4140 | again: |
0b246afa | 4141 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 4142 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 4143 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4144 | if (!ret && !is_data) { |
ab8d0fc4 | 4145 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4146 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4147 | if (!final_tried && ins->offset) { |
4148 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4149 | num_bytes = round_down(num_bytes, |
0b246afa | 4150 | fs_info->sectorsize); |
9e622d6b | 4151 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4152 | ram_bytes = num_bytes; |
9e622d6b MX |
4153 | if (num_bytes == min_alloc_size) |
4154 | final_tried = true; | |
4155 | goto again; | |
ab8d0fc4 | 4156 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4157 | struct btrfs_space_info *sinfo; |
4158 | ||
280c2908 | 4159 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4160 | btrfs_err(fs_info, |
5d163e0e JM |
4161 | "allocation failed flags %llu, wanted %llu", |
4162 | flags, num_bytes); | |
53804280 | 4163 | if (sinfo) |
5da6afeb JB |
4164 | btrfs_dump_space_info(fs_info, sinfo, |
4165 | num_bytes, 1); | |
9e622d6b | 4166 | } |
925baedd | 4167 | } |
0f9dd46c JB |
4168 | |
4169 | return ret; | |
e6dcd2dc CM |
4170 | } |
4171 | ||
a0fbf736 NB |
4172 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4173 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4174 | { |
32da5386 | 4175 | struct btrfs_block_group *cache; |
0f9dd46c | 4176 | |
0b246afa | 4177 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4178 | if (!cache) { |
0b246afa JM |
4179 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4180 | start); | |
0f9dd46c JB |
4181 | return -ENOSPC; |
4182 | } | |
1f3c79a2 | 4183 | |
a0fbf736 NB |
4184 | btrfs_add_free_space(cache, start, len); |
4185 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4186 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4187 | ||
fa9c0d79 | 4188 | btrfs_put_block_group(cache); |
a0fbf736 | 4189 | return 0; |
e6dcd2dc CM |
4190 | } |
4191 | ||
a0fbf736 | 4192 | int btrfs_pin_reserved_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e688b725 | 4193 | { |
7ef54d54 | 4194 | struct btrfs_block_group *cache; |
a0fbf736 | 4195 | int ret = 0; |
7ef54d54 NB |
4196 | |
4197 | cache = btrfs_lookup_block_group(fs_info, start); | |
4198 | if (!cache) { | |
4199 | btrfs_err(fs_info, "unable to find block group for %llu", start); | |
4200 | return -ENOSPC; | |
4201 | } | |
4202 | ||
a0fbf736 | 4203 | ret = pin_down_extent(cache, start, len, 1); |
7ef54d54 | 4204 | btrfs_put_block_group(cache); |
a0fbf736 | 4205 | return ret; |
e688b725 CM |
4206 | } |
4207 | ||
5d4f98a2 | 4208 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4209 | u64 parent, u64 root_objectid, |
4210 | u64 flags, u64 owner, u64 offset, | |
4211 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4212 | { |
ef89b824 | 4213 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4214 | int ret; |
e6dcd2dc | 4215 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4216 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4217 | struct btrfs_path *path; |
5d4f98a2 YZ |
4218 | struct extent_buffer *leaf; |
4219 | int type; | |
4220 | u32 size; | |
26b8003f | 4221 | |
5d4f98a2 YZ |
4222 | if (parent > 0) |
4223 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4224 | else | |
4225 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4226 | |
5d4f98a2 | 4227 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4228 | |
4229 | path = btrfs_alloc_path(); | |
db5b493a TI |
4230 | if (!path) |
4231 | return -ENOMEM; | |
47e4bb98 | 4232 | |
b9473439 | 4233 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4234 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4235 | ins, size); | |
79787eaa JM |
4236 | if (ret) { |
4237 | btrfs_free_path(path); | |
4238 | return ret; | |
4239 | } | |
0f9dd46c | 4240 | |
5d4f98a2 YZ |
4241 | leaf = path->nodes[0]; |
4242 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4243 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4244 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4245 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4246 | btrfs_set_extent_flags(leaf, extent_item, | |
4247 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4248 | ||
4249 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4250 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4251 | if (parent > 0) { | |
4252 | struct btrfs_shared_data_ref *ref; | |
4253 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4254 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4255 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4256 | } else { | |
4257 | struct btrfs_extent_data_ref *ref; | |
4258 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4259 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4260 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4261 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4262 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4263 | } | |
47e4bb98 CM |
4264 | |
4265 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4266 | btrfs_free_path(path); |
f510cfec | 4267 | |
25a356d3 | 4268 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4269 | if (ret) |
4270 | return ret; | |
4271 | ||
ade4b516 | 4272 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4273 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4274 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4275 | ins->objectid, ins->offset); |
f5947066 CM |
4276 | BUG(); |
4277 | } | |
71ff6437 | 4278 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4279 | return ret; |
4280 | } | |
4281 | ||
5d4f98a2 | 4282 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4283 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4284 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4285 | { |
9dcdbe01 | 4286 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4287 | int ret; |
5d4f98a2 | 4288 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4289 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4290 | struct btrfs_tree_block_info *block_info; |
4291 | struct btrfs_extent_inline_ref *iref; | |
4292 | struct btrfs_path *path; | |
4293 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4294 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4295 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4296 | u64 num_bytes; |
21ebfbe7 | 4297 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4298 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4299 | |
4e6bd4e0 NB |
4300 | ref = btrfs_delayed_node_to_tree_ref(node); |
4301 | ||
4e6bd4e0 NB |
4302 | extent_key.objectid = node->bytenr; |
4303 | if (skinny_metadata) { | |
4304 | extent_key.offset = ref->level; | |
4305 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4306 | num_bytes = fs_info->nodesize; | |
4307 | } else { | |
4308 | extent_key.offset = node->num_bytes; | |
4309 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4310 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4311 | num_bytes = node->num_bytes; |
4312 | } | |
1c2308f8 | 4313 | |
5d4f98a2 | 4314 | path = btrfs_alloc_path(); |
80ee54bf | 4315 | if (!path) |
d8926bb3 | 4316 | return -ENOMEM; |
56bec294 | 4317 | |
5d4f98a2 YZ |
4318 | path->leave_spinning = 1; |
4319 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 4320 | &extent_key, size); |
79787eaa | 4321 | if (ret) { |
dd825259 | 4322 | btrfs_free_path(path); |
79787eaa JM |
4323 | return ret; |
4324 | } | |
5d4f98a2 YZ |
4325 | |
4326 | leaf = path->nodes[0]; | |
4327 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4328 | struct btrfs_extent_item); | |
4329 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4330 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4331 | btrfs_set_extent_flags(leaf, extent_item, | |
4332 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4333 | |
3173a18f JB |
4334 | if (skinny_metadata) { |
4335 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4336 | } else { | |
4337 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4338 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4339 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4340 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4341 | } | |
5d4f98a2 | 4342 | |
d4b20733 | 4343 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4344 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4345 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4346 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4347 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4348 | } else { |
4349 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4350 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4351 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4352 | } |
4353 | ||
4354 | btrfs_mark_buffer_dirty(leaf); | |
4355 | btrfs_free_path(path); | |
4356 | ||
4e6bd4e0 NB |
4357 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4358 | num_bytes); | |
1e144fb8 OS |
4359 | if (ret) |
4360 | return ret; | |
4361 | ||
ade4b516 JB |
4362 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4363 | fs_info->nodesize, 1); | |
79787eaa | 4364 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4365 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4366 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4367 | BUG(); |
4368 | } | |
0be5dc67 | 4369 | |
4e6bd4e0 | 4370 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4371 | fs_info->nodesize); |
5d4f98a2 YZ |
4372 | return ret; |
4373 | } | |
4374 | ||
4375 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4376 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4377 | u64 offset, u64 ram_bytes, |
4378 | struct btrfs_key *ins) | |
5d4f98a2 | 4379 | { |
76675593 | 4380 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
4381 | int ret; |
4382 | ||
84f7d8e6 | 4383 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4384 | |
76675593 QW |
4385 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4386 | ins->objectid, ins->offset, 0); | |
4387 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4388 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
4389 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
4390 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
4391 | return ret; |
4392 | } | |
e02119d5 CM |
4393 | |
4394 | /* | |
4395 | * this is used by the tree logging recovery code. It records that | |
4396 | * an extent has been allocated and makes sure to clear the free | |
4397 | * space cache bits as well | |
4398 | */ | |
5d4f98a2 | 4399 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4400 | u64 root_objectid, u64 owner, u64 offset, |
4401 | struct btrfs_key *ins) | |
e02119d5 | 4402 | { |
61da2abf | 4403 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4404 | int ret; |
32da5386 | 4405 | struct btrfs_block_group *block_group; |
ed7a6948 | 4406 | struct btrfs_space_info *space_info; |
11833d66 | 4407 | |
8c2a1a30 JB |
4408 | /* |
4409 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4410 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4411 | */ |
0b246afa | 4412 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4413 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4414 | ins->offset); | |
b50c6e25 | 4415 | if (ret) |
8c2a1a30 | 4416 | return ret; |
11833d66 YZ |
4417 | } |
4418 | ||
0b246afa | 4419 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4420 | if (!block_group) |
4421 | return -EINVAL; | |
4422 | ||
ed7a6948 WX |
4423 | space_info = block_group->space_info; |
4424 | spin_lock(&space_info->lock); | |
4425 | spin_lock(&block_group->lock); | |
4426 | space_info->bytes_reserved += ins->offset; | |
4427 | block_group->reserved += ins->offset; | |
4428 | spin_unlock(&block_group->lock); | |
4429 | spin_unlock(&space_info->lock); | |
4430 | ||
ef89b824 NB |
4431 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4432 | offset, ins, 1); | |
bd727173 JB |
4433 | if (ret) |
4434 | btrfs_pin_extent(fs_info, ins->objectid, ins->offset, 1); | |
b50c6e25 | 4435 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4436 | return ret; |
4437 | } | |
4438 | ||
48a3b636 ES |
4439 | static struct extent_buffer * |
4440 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 4441 | u64 bytenr, int level, u64 owner) |
65b51a00 | 4442 | { |
0b246afa | 4443 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4444 | struct extent_buffer *buf; |
4445 | ||
2ff7e61e | 4446 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4447 | if (IS_ERR(buf)) |
4448 | return buf; | |
4449 | ||
b72c3aba QW |
4450 | /* |
4451 | * Extra safety check in case the extent tree is corrupted and extent | |
4452 | * allocator chooses to use a tree block which is already used and | |
4453 | * locked. | |
4454 | */ | |
4455 | if (buf->lock_owner == current->pid) { | |
4456 | btrfs_err_rl(fs_info, | |
4457 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4458 | buf->start, btrfs_header_owner(buf), current->pid); | |
4459 | free_extent_buffer(buf); | |
4460 | return ERR_PTR(-EUCLEAN); | |
4461 | } | |
4462 | ||
85d4e461 | 4463 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 4464 | btrfs_tree_lock(buf); |
6a884d7d | 4465 | btrfs_clean_tree_block(buf); |
3083ee2e | 4466 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 4467 | |
8bead258 | 4468 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 4469 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4470 | |
bc877d28 NB |
4471 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4472 | btrfs_set_header_level(buf, level); | |
4473 | btrfs_set_header_bytenr(buf, buf->start); | |
4474 | btrfs_set_header_generation(buf, trans->transid); | |
4475 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4476 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4477 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4478 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4479 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4480 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4481 | /* |
4482 | * we allow two log transactions at a time, use different | |
52042d8e | 4483 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4484 | */ |
656f30db | 4485 | if (buf->log_index == 0) |
8cef4e16 YZ |
4486 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4487 | buf->start + buf->len - 1, GFP_NOFS); | |
4488 | else | |
4489 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4490 | buf->start + buf->len - 1); |
d0c803c4 | 4491 | } else { |
656f30db | 4492 | buf->log_index = -1; |
d0c803c4 | 4493 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4494 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4495 | } |
64c12921 | 4496 | trans->dirty = true; |
b4ce94de | 4497 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4498 | return buf; |
4499 | } | |
4500 | ||
fec577fb | 4501 | /* |
f0486c68 | 4502 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4503 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4504 | */ |
4d75f8a9 | 4505 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4506 | struct btrfs_root *root, |
4507 | u64 parent, u64 root_objectid, | |
4508 | const struct btrfs_disk_key *key, | |
4509 | int level, u64 hint, | |
4510 | u64 empty_size) | |
fec577fb | 4511 | { |
0b246afa | 4512 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4513 | struct btrfs_key ins; |
f0486c68 | 4514 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4515 | struct extent_buffer *buf; |
67b7859e | 4516 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4517 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4518 | u64 flags = 0; |
4519 | int ret; | |
0b246afa JM |
4520 | u32 blocksize = fs_info->nodesize; |
4521 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4522 | |
05653ef3 | 4523 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4524 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4525 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 4526 | level, root_objectid); |
faa2dbf0 JB |
4527 | if (!IS_ERR(buf)) |
4528 | root->alloc_bytenr += blocksize; | |
4529 | return buf; | |
4530 | } | |
05653ef3 | 4531 | #endif |
fccb84c9 | 4532 | |
67f9c220 | 4533 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4534 | if (IS_ERR(block_rsv)) |
4535 | return ERR_CAST(block_rsv); | |
4536 | ||
18513091 | 4537 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4538 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4539 | if (ret) |
4540 | goto out_unuse; | |
55c69072 | 4541 | |
bc877d28 NB |
4542 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
4543 | root_objectid); | |
67b7859e OS |
4544 | if (IS_ERR(buf)) { |
4545 | ret = PTR_ERR(buf); | |
4546 | goto out_free_reserved; | |
4547 | } | |
f0486c68 YZ |
4548 | |
4549 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4550 | if (parent == 0) | |
4551 | parent = ins.objectid; | |
4552 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4553 | } else | |
4554 | BUG_ON(parent > 0); | |
4555 | ||
4556 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4557 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4558 | if (!extent_op) { |
4559 | ret = -ENOMEM; | |
4560 | goto out_free_buf; | |
4561 | } | |
f0486c68 YZ |
4562 | if (key) |
4563 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4564 | else | |
4565 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4566 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4567 | extent_op->update_key = skinny_metadata ? false : true; |
4568 | extent_op->update_flags = true; | |
4569 | extent_op->is_data = false; | |
b1c79e09 | 4570 | extent_op->level = level; |
f0486c68 | 4571 | |
ed4f255b QW |
4572 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4573 | ins.objectid, ins.offset, parent); | |
4574 | generic_ref.real_root = root->root_key.objectid; | |
4575 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4576 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 4577 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 4578 | extent_op, NULL, NULL); |
67b7859e OS |
4579 | if (ret) |
4580 | goto out_free_delayed; | |
f0486c68 | 4581 | } |
fec577fb | 4582 | return buf; |
67b7859e OS |
4583 | |
4584 | out_free_delayed: | |
4585 | btrfs_free_delayed_extent_op(extent_op); | |
4586 | out_free_buf: | |
4587 | free_extent_buffer(buf); | |
4588 | out_free_reserved: | |
2ff7e61e | 4589 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4590 | out_unuse: |
67f9c220 | 4591 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4592 | return ERR_PTR(ret); |
fec577fb | 4593 | } |
a28ec197 | 4594 | |
2c47e605 YZ |
4595 | struct walk_control { |
4596 | u64 refs[BTRFS_MAX_LEVEL]; | |
4597 | u64 flags[BTRFS_MAX_LEVEL]; | |
4598 | struct btrfs_key update_progress; | |
aea6f028 JB |
4599 | struct btrfs_key drop_progress; |
4600 | int drop_level; | |
2c47e605 YZ |
4601 | int stage; |
4602 | int level; | |
4603 | int shared_level; | |
4604 | int update_ref; | |
4605 | int keep_locks; | |
1c4850e2 YZ |
4606 | int reada_slot; |
4607 | int reada_count; | |
78c52d9e | 4608 | int restarted; |
2c47e605 YZ |
4609 | }; |
4610 | ||
4611 | #define DROP_REFERENCE 1 | |
4612 | #define UPDATE_BACKREF 2 | |
4613 | ||
1c4850e2 YZ |
4614 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4615 | struct btrfs_root *root, | |
4616 | struct walk_control *wc, | |
4617 | struct btrfs_path *path) | |
6407bf6d | 4618 | { |
0b246afa | 4619 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4620 | u64 bytenr; |
4621 | u64 generation; | |
4622 | u64 refs; | |
94fcca9f | 4623 | u64 flags; |
5d4f98a2 | 4624 | u32 nritems; |
1c4850e2 YZ |
4625 | struct btrfs_key key; |
4626 | struct extent_buffer *eb; | |
6407bf6d | 4627 | int ret; |
1c4850e2 YZ |
4628 | int slot; |
4629 | int nread = 0; | |
6407bf6d | 4630 | |
1c4850e2 YZ |
4631 | if (path->slots[wc->level] < wc->reada_slot) { |
4632 | wc->reada_count = wc->reada_count * 2 / 3; | |
4633 | wc->reada_count = max(wc->reada_count, 2); | |
4634 | } else { | |
4635 | wc->reada_count = wc->reada_count * 3 / 2; | |
4636 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4637 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4638 | } |
7bb86316 | 4639 | |
1c4850e2 YZ |
4640 | eb = path->nodes[wc->level]; |
4641 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4642 | |
1c4850e2 YZ |
4643 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4644 | if (nread >= wc->reada_count) | |
4645 | break; | |
bd56b302 | 4646 | |
2dd3e67b | 4647 | cond_resched(); |
1c4850e2 YZ |
4648 | bytenr = btrfs_node_blockptr(eb, slot); |
4649 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4650 | |
1c4850e2 YZ |
4651 | if (slot == path->slots[wc->level]) |
4652 | goto reada; | |
5d4f98a2 | 4653 | |
1c4850e2 YZ |
4654 | if (wc->stage == UPDATE_BACKREF && |
4655 | generation <= root->root_key.offset) | |
bd56b302 CM |
4656 | continue; |
4657 | ||
94fcca9f | 4658 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4659 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4660 | wc->level - 1, 1, &refs, |
4661 | &flags); | |
79787eaa JM |
4662 | /* We don't care about errors in readahead. */ |
4663 | if (ret < 0) | |
4664 | continue; | |
94fcca9f YZ |
4665 | BUG_ON(refs == 0); |
4666 | ||
1c4850e2 | 4667 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4668 | if (refs == 1) |
4669 | goto reada; | |
bd56b302 | 4670 | |
94fcca9f YZ |
4671 | if (wc->level == 1 && |
4672 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4673 | continue; | |
1c4850e2 YZ |
4674 | if (!wc->update_ref || |
4675 | generation <= root->root_key.offset) | |
4676 | continue; | |
4677 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4678 | ret = btrfs_comp_cpu_keys(&key, | |
4679 | &wc->update_progress); | |
4680 | if (ret < 0) | |
4681 | continue; | |
94fcca9f YZ |
4682 | } else { |
4683 | if (wc->level == 1 && | |
4684 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4685 | continue; | |
6407bf6d | 4686 | } |
1c4850e2 | 4687 | reada: |
2ff7e61e | 4688 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 4689 | nread++; |
20524f02 | 4690 | } |
1c4850e2 | 4691 | wc->reada_slot = slot; |
20524f02 | 4692 | } |
2c47e605 | 4693 | |
f82d02d9 | 4694 | /* |
2c016dc2 | 4695 | * helper to process tree block while walking down the tree. |
2c47e605 | 4696 | * |
2c47e605 YZ |
4697 | * when wc->stage == UPDATE_BACKREF, this function updates |
4698 | * back refs for pointers in the block. | |
4699 | * | |
4700 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4701 | */ |
2c47e605 | 4702 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4703 | struct btrfs_root *root, |
2c47e605 | 4704 | struct btrfs_path *path, |
94fcca9f | 4705 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4706 | { |
2ff7e61e | 4707 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4708 | int level = wc->level; |
4709 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4710 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4711 | int ret; |
4712 | ||
2c47e605 YZ |
4713 | if (wc->stage == UPDATE_BACKREF && |
4714 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4715 | return 1; | |
f82d02d9 | 4716 | |
2c47e605 YZ |
4717 | /* |
4718 | * when reference count of tree block is 1, it won't increase | |
4719 | * again. once full backref flag is set, we never clear it. | |
4720 | */ | |
94fcca9f YZ |
4721 | if (lookup_info && |
4722 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4723 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4724 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4725 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4726 | eb->start, level, 1, |
2c47e605 YZ |
4727 | &wc->refs[level], |
4728 | &wc->flags[level]); | |
79787eaa JM |
4729 | BUG_ON(ret == -ENOMEM); |
4730 | if (ret) | |
4731 | return ret; | |
2c47e605 YZ |
4732 | BUG_ON(wc->refs[level] == 0); |
4733 | } | |
5d4f98a2 | 4734 | |
2c47e605 YZ |
4735 | if (wc->stage == DROP_REFERENCE) { |
4736 | if (wc->refs[level] > 1) | |
4737 | return 1; | |
f82d02d9 | 4738 | |
2c47e605 | 4739 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4740 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4741 | path->locks[level] = 0; |
4742 | } | |
4743 | return 0; | |
4744 | } | |
f82d02d9 | 4745 | |
2c47e605 YZ |
4746 | /* wc->stage == UPDATE_BACKREF */ |
4747 | if (!(wc->flags[level] & flag)) { | |
4748 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4749 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4750 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4751 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4752 | BUG_ON(ret); /* -ENOMEM */ |
f5c8daa5 | 4753 | ret = btrfs_set_disk_extent_flags(trans, eb->start, |
b1c79e09 JB |
4754 | eb->len, flag, |
4755 | btrfs_header_level(eb), 0); | |
79787eaa | 4756 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4757 | wc->flags[level] |= flag; |
4758 | } | |
4759 | ||
4760 | /* | |
4761 | * the block is shared by multiple trees, so it's not good to | |
4762 | * keep the tree lock | |
4763 | */ | |
4764 | if (path->locks[level] && level > 0) { | |
bd681513 | 4765 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4766 | path->locks[level] = 0; |
4767 | } | |
4768 | return 0; | |
4769 | } | |
4770 | ||
78c52d9e JB |
4771 | /* |
4772 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4773 | * would have a drop_progress key that hadn't been updated properly. | |
4774 | */ | |
4775 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4776 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4777 | int level) | |
4778 | { | |
4779 | struct btrfs_path *path; | |
4780 | struct btrfs_extent_inline_ref *iref; | |
4781 | int ret; | |
4782 | ||
4783 | path = btrfs_alloc_path(); | |
4784 | if (!path) | |
4785 | return -ENOMEM; | |
4786 | ||
4787 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
4788 | root->fs_info->nodesize, parent, | |
4789 | root->root_key.objectid, level, 0); | |
4790 | btrfs_free_path(path); | |
4791 | if (ret == -ENOENT) | |
4792 | return 0; | |
4793 | if (ret < 0) | |
4794 | return ret; | |
4795 | return 1; | |
4796 | } | |
4797 | ||
1c4850e2 | 4798 | /* |
2c016dc2 | 4799 | * helper to process tree block pointer. |
1c4850e2 YZ |
4800 | * |
4801 | * when wc->stage == DROP_REFERENCE, this function checks | |
4802 | * reference count of the block pointed to. if the block | |
4803 | * is shared and we need update back refs for the subtree | |
4804 | * rooted at the block, this function changes wc->stage to | |
4805 | * UPDATE_BACKREF. if the block is shared and there is no | |
4806 | * need to update back, this function drops the reference | |
4807 | * to the block. | |
4808 | * | |
4809 | * NOTE: return value 1 means we should stop walking down. | |
4810 | */ | |
4811 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
4812 | struct btrfs_root *root, | |
4813 | struct btrfs_path *path, | |
94fcca9f | 4814 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 4815 | { |
0b246afa | 4816 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4817 | u64 bytenr; |
4818 | u64 generation; | |
4819 | u64 parent; | |
1c4850e2 | 4820 | struct btrfs_key key; |
581c1760 | 4821 | struct btrfs_key first_key; |
ffd4bb2a | 4822 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
4823 | struct extent_buffer *next; |
4824 | int level = wc->level; | |
4825 | int reada = 0; | |
4826 | int ret = 0; | |
1152651a | 4827 | bool need_account = false; |
1c4850e2 YZ |
4828 | |
4829 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
4830 | path->slots[level]); | |
4831 | /* | |
4832 | * if the lower level block was created before the snapshot | |
4833 | * was created, we know there is no need to update back refs | |
4834 | * for the subtree | |
4835 | */ | |
4836 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
4837 | generation <= root->root_key.offset) { |
4838 | *lookup_info = 1; | |
1c4850e2 | 4839 | return 1; |
94fcca9f | 4840 | } |
1c4850e2 YZ |
4841 | |
4842 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
4843 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
4844 | path->slots[level]); | |
1c4850e2 | 4845 | |
0b246afa | 4846 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 4847 | if (!next) { |
2ff7e61e | 4848 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4849 | if (IS_ERR(next)) |
4850 | return PTR_ERR(next); | |
4851 | ||
b2aaaa3b JB |
4852 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
4853 | level - 1); | |
1c4850e2 YZ |
4854 | reada = 1; |
4855 | } | |
4856 | btrfs_tree_lock(next); | |
8bead258 | 4857 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 4858 | |
2ff7e61e | 4859 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
4860 | &wc->refs[level - 1], |
4861 | &wc->flags[level - 1]); | |
4867268c JB |
4862 | if (ret < 0) |
4863 | goto out_unlock; | |
79787eaa | 4864 | |
c2cf52eb | 4865 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 4866 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
4867 | ret = -EIO; |
4868 | goto out_unlock; | |
c2cf52eb | 4869 | } |
94fcca9f | 4870 | *lookup_info = 0; |
1c4850e2 | 4871 | |
94fcca9f | 4872 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 4873 | if (wc->refs[level - 1] > 1) { |
1152651a | 4874 | need_account = true; |
94fcca9f YZ |
4875 | if (level == 1 && |
4876 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4877 | goto skip; | |
4878 | ||
1c4850e2 YZ |
4879 | if (!wc->update_ref || |
4880 | generation <= root->root_key.offset) | |
4881 | goto skip; | |
4882 | ||
4883 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
4884 | path->slots[level]); | |
4885 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
4886 | if (ret < 0) | |
4887 | goto skip; | |
4888 | ||
4889 | wc->stage = UPDATE_BACKREF; | |
4890 | wc->shared_level = level - 1; | |
4891 | } | |
94fcca9f YZ |
4892 | } else { |
4893 | if (level == 1 && | |
4894 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4895 | goto skip; | |
1c4850e2 YZ |
4896 | } |
4897 | ||
b9fab919 | 4898 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
4899 | btrfs_tree_unlock(next); |
4900 | free_extent_buffer(next); | |
4901 | next = NULL; | |
94fcca9f | 4902 | *lookup_info = 1; |
1c4850e2 YZ |
4903 | } |
4904 | ||
4905 | if (!next) { | |
4906 | if (reada && level == 1) | |
4907 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
4908 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
4909 | &first_key); | |
64c043de LB |
4910 | if (IS_ERR(next)) { |
4911 | return PTR_ERR(next); | |
4912 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 4913 | free_extent_buffer(next); |
97d9a8a4 | 4914 | return -EIO; |
416bc658 | 4915 | } |
1c4850e2 | 4916 | btrfs_tree_lock(next); |
8bead258 | 4917 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
4918 | } |
4919 | ||
4920 | level--; | |
4867268c JB |
4921 | ASSERT(level == btrfs_header_level(next)); |
4922 | if (level != btrfs_header_level(next)) { | |
4923 | btrfs_err(root->fs_info, "mismatched level"); | |
4924 | ret = -EIO; | |
4925 | goto out_unlock; | |
4926 | } | |
1c4850e2 YZ |
4927 | path->nodes[level] = next; |
4928 | path->slots[level] = 0; | |
bd681513 | 4929 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
4930 | wc->level = level; |
4931 | if (wc->level == 1) | |
4932 | wc->reada_slot = 0; | |
4933 | return 0; | |
4934 | skip: | |
4935 | wc->refs[level - 1] = 0; | |
4936 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
4937 | if (wc->stage == DROP_REFERENCE) { |
4938 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
4939 | parent = path->nodes[level]->start; | |
4940 | } else { | |
4867268c | 4941 | ASSERT(root->root_key.objectid == |
94fcca9f | 4942 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
4943 | if (root->root_key.objectid != |
4944 | btrfs_header_owner(path->nodes[level])) { | |
4945 | btrfs_err(root->fs_info, | |
4946 | "mismatched block owner"); | |
4947 | ret = -EIO; | |
4948 | goto out_unlock; | |
4949 | } | |
94fcca9f YZ |
4950 | parent = 0; |
4951 | } | |
1c4850e2 | 4952 | |
78c52d9e JB |
4953 | /* |
4954 | * If we had a drop_progress we need to verify the refs are set | |
4955 | * as expected. If we find our ref then we know that from here | |
4956 | * on out everything should be correct, and we can clear the | |
4957 | * ->restarted flag. | |
4958 | */ | |
4959 | if (wc->restarted) { | |
4960 | ret = check_ref_exists(trans, root, bytenr, parent, | |
4961 | level - 1); | |
4962 | if (ret < 0) | |
4963 | goto out_unlock; | |
4964 | if (ret == 0) | |
4965 | goto no_delete; | |
4966 | ret = 0; | |
4967 | wc->restarted = 0; | |
4968 | } | |
4969 | ||
2cd86d30 QW |
4970 | /* |
4971 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
4972 | * already accounted them at merge time (replace_path), | |
4973 | * thus we could skip expensive subtree trace here. | |
4974 | */ | |
4975 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
4976 | need_account) { | |
deb40627 | 4977 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 4978 | generation, level - 1); |
1152651a | 4979 | if (ret) { |
0b246afa | 4980 | btrfs_err_rl(fs_info, |
5d163e0e JM |
4981 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
4982 | ret); | |
1152651a MF |
4983 | } |
4984 | } | |
aea6f028 JB |
4985 | |
4986 | /* | |
4987 | * We need to update the next key in our walk control so we can | |
4988 | * update the drop_progress key accordingly. We don't care if | |
4989 | * find_next_key doesn't find a key because that means we're at | |
4990 | * the end and are going to clean up now. | |
4991 | */ | |
4992 | wc->drop_level = level; | |
4993 | find_next_key(path, level, &wc->drop_progress); | |
4994 | ||
ffd4bb2a QW |
4995 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
4996 | fs_info->nodesize, parent); | |
4997 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
4998 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
4999 | if (ret) |
5000 | goto out_unlock; | |
1c4850e2 | 5001 | } |
78c52d9e | 5002 | no_delete: |
4867268c JB |
5003 | *lookup_info = 1; |
5004 | ret = 1; | |
5005 | ||
5006 | out_unlock: | |
1c4850e2 YZ |
5007 | btrfs_tree_unlock(next); |
5008 | free_extent_buffer(next); | |
4867268c JB |
5009 | |
5010 | return ret; | |
1c4850e2 YZ |
5011 | } |
5012 | ||
2c47e605 | 5013 | /* |
2c016dc2 | 5014 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5015 | * |
5016 | * when wc->stage == DROP_REFERENCE, this function drops | |
5017 | * reference count on the block. | |
5018 | * | |
5019 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5020 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5021 | * to UPDATE_BACKREF previously while processing the block. | |
5022 | * | |
5023 | * NOTE: return value 1 means we should stop walking up. | |
5024 | */ | |
5025 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5026 | struct btrfs_root *root, | |
5027 | struct btrfs_path *path, | |
5028 | struct walk_control *wc) | |
5029 | { | |
0b246afa | 5030 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5031 | int ret; |
2c47e605 YZ |
5032 | int level = wc->level; |
5033 | struct extent_buffer *eb = path->nodes[level]; | |
5034 | u64 parent = 0; | |
5035 | ||
5036 | if (wc->stage == UPDATE_BACKREF) { | |
5037 | BUG_ON(wc->shared_level < level); | |
5038 | if (level < wc->shared_level) | |
5039 | goto out; | |
5040 | ||
2c47e605 YZ |
5041 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5042 | if (ret > 0) | |
5043 | wc->update_ref = 0; | |
5044 | ||
5045 | wc->stage = DROP_REFERENCE; | |
5046 | wc->shared_level = -1; | |
5047 | path->slots[level] = 0; | |
5048 | ||
5049 | /* | |
5050 | * check reference count again if the block isn't locked. | |
5051 | * we should start walking down the tree again if reference | |
5052 | * count is one. | |
5053 | */ | |
5054 | if (!path->locks[level]) { | |
5055 | BUG_ON(level == 0); | |
5056 | btrfs_tree_lock(eb); | |
8bead258 | 5057 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5058 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5059 | |
2ff7e61e | 5060 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5061 | eb->start, level, 1, |
2c47e605 YZ |
5062 | &wc->refs[level], |
5063 | &wc->flags[level]); | |
79787eaa JM |
5064 | if (ret < 0) { |
5065 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5066 | path->locks[level] = 0; |
79787eaa JM |
5067 | return ret; |
5068 | } | |
2c47e605 YZ |
5069 | BUG_ON(wc->refs[level] == 0); |
5070 | if (wc->refs[level] == 1) { | |
bd681513 | 5071 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5072 | path->locks[level] = 0; |
2c47e605 YZ |
5073 | return 1; |
5074 | } | |
f82d02d9 | 5075 | } |
2c47e605 | 5076 | } |
f82d02d9 | 5077 | |
2c47e605 YZ |
5078 | /* wc->stage == DROP_REFERENCE */ |
5079 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5080 | |
2c47e605 YZ |
5081 | if (wc->refs[level] == 1) { |
5082 | if (level == 0) { | |
5083 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5084 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5085 | else |
e339a6b0 | 5086 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5087 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5088 | if (is_fstree(root->root_key.objectid)) { |
5089 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5090 | if (ret) { | |
5091 | btrfs_err_rl(fs_info, | |
5092 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5093 | ret); |
c4140cbf | 5094 | } |
1152651a | 5095 | } |
2c47e605 | 5096 | } |
6a884d7d | 5097 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5098 | if (!path->locks[level] && |
5099 | btrfs_header_generation(eb) == trans->transid) { | |
5100 | btrfs_tree_lock(eb); | |
8bead258 | 5101 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5102 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5103 | } |
6a884d7d | 5104 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5105 | } |
5106 | ||
5107 | if (eb == root->node) { | |
5108 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5109 | parent = eb->start; | |
65c6e82b QW |
5110 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5111 | goto owner_mismatch; | |
2c47e605 YZ |
5112 | } else { |
5113 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5114 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5115 | else if (root->root_key.objectid != |
5116 | btrfs_header_owner(path->nodes[level + 1])) | |
5117 | goto owner_mismatch; | |
f82d02d9 | 5118 | } |
f82d02d9 | 5119 | |
5581a51a | 5120 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5121 | out: |
5122 | wc->refs[level] = 0; | |
5123 | wc->flags[level] = 0; | |
f0486c68 | 5124 | return 0; |
65c6e82b QW |
5125 | |
5126 | owner_mismatch: | |
5127 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5128 | btrfs_header_owner(eb), root->root_key.objectid); | |
5129 | return -EUCLEAN; | |
2c47e605 YZ |
5130 | } |
5131 | ||
5132 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5133 | struct btrfs_root *root, | |
5134 | struct btrfs_path *path, | |
5135 | struct walk_control *wc) | |
5136 | { | |
2c47e605 | 5137 | int level = wc->level; |
94fcca9f | 5138 | int lookup_info = 1; |
2c47e605 YZ |
5139 | int ret; |
5140 | ||
5141 | while (level >= 0) { | |
94fcca9f | 5142 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5143 | if (ret > 0) |
5144 | break; | |
5145 | ||
5146 | if (level == 0) | |
5147 | break; | |
5148 | ||
7a7965f8 YZ |
5149 | if (path->slots[level] >= |
5150 | btrfs_header_nritems(path->nodes[level])) | |
5151 | break; | |
5152 | ||
94fcca9f | 5153 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5154 | if (ret > 0) { |
5155 | path->slots[level]++; | |
5156 | continue; | |
90d2c51d MX |
5157 | } else if (ret < 0) |
5158 | return ret; | |
1c4850e2 | 5159 | level = wc->level; |
f82d02d9 | 5160 | } |
f82d02d9 YZ |
5161 | return 0; |
5162 | } | |
5163 | ||
d397712b | 5164 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5165 | struct btrfs_root *root, |
f82d02d9 | 5166 | struct btrfs_path *path, |
2c47e605 | 5167 | struct walk_control *wc, int max_level) |
20524f02 | 5168 | { |
2c47e605 | 5169 | int level = wc->level; |
20524f02 | 5170 | int ret; |
9f3a7427 | 5171 | |
2c47e605 YZ |
5172 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5173 | while (level < max_level && path->nodes[level]) { | |
5174 | wc->level = level; | |
5175 | if (path->slots[level] + 1 < | |
5176 | btrfs_header_nritems(path->nodes[level])) { | |
5177 | path->slots[level]++; | |
20524f02 CM |
5178 | return 0; |
5179 | } else { | |
2c47e605 YZ |
5180 | ret = walk_up_proc(trans, root, path, wc); |
5181 | if (ret > 0) | |
5182 | return 0; | |
65c6e82b QW |
5183 | if (ret < 0) |
5184 | return ret; | |
bd56b302 | 5185 | |
2c47e605 | 5186 | if (path->locks[level]) { |
bd681513 CM |
5187 | btrfs_tree_unlock_rw(path->nodes[level], |
5188 | path->locks[level]); | |
2c47e605 | 5189 | path->locks[level] = 0; |
f82d02d9 | 5190 | } |
2c47e605 YZ |
5191 | free_extent_buffer(path->nodes[level]); |
5192 | path->nodes[level] = NULL; | |
5193 | level++; | |
20524f02 CM |
5194 | } |
5195 | } | |
5196 | return 1; | |
5197 | } | |
5198 | ||
9aca1d51 | 5199 | /* |
2c47e605 YZ |
5200 | * drop a subvolume tree. |
5201 | * | |
5202 | * this function traverses the tree freeing any blocks that only | |
5203 | * referenced by the tree. | |
5204 | * | |
5205 | * when a shared tree block is found. this function decreases its | |
5206 | * reference count by one. if update_ref is true, this function | |
5207 | * also make sure backrefs for the shared block and all lower level | |
5208 | * blocks are properly updated. | |
9d1a2a3a DS |
5209 | * |
5210 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5211 | */ |
2c536799 | 5212 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
5213 | struct btrfs_block_rsv *block_rsv, int update_ref, |
5214 | int for_reloc) | |
20524f02 | 5215 | { |
ab8d0fc4 | 5216 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5217 | struct btrfs_path *path; |
2c47e605 | 5218 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5219 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5220 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5221 | struct walk_control *wc; |
5222 | struct btrfs_key key; | |
5223 | int err = 0; | |
5224 | int ret; | |
5225 | int level; | |
d29a9f62 | 5226 | bool root_dropped = false; |
20524f02 | 5227 | |
4fd786e6 | 5228 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5229 | |
5caf2a00 | 5230 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5231 | if (!path) { |
5232 | err = -ENOMEM; | |
5233 | goto out; | |
5234 | } | |
20524f02 | 5235 | |
2c47e605 | 5236 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5237 | if (!wc) { |
5238 | btrfs_free_path(path); | |
cb1b69f4 TI |
5239 | err = -ENOMEM; |
5240 | goto out; | |
38a1a919 | 5241 | } |
2c47e605 | 5242 | |
a22285a6 | 5243 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
5244 | if (IS_ERR(trans)) { |
5245 | err = PTR_ERR(trans); | |
5246 | goto out_free; | |
5247 | } | |
98d5dc13 | 5248 | |
0568e82d JB |
5249 | err = btrfs_run_delayed_items(trans); |
5250 | if (err) | |
5251 | goto out_end_trans; | |
5252 | ||
3fd0a558 YZ |
5253 | if (block_rsv) |
5254 | trans->block_rsv = block_rsv; | |
2c47e605 | 5255 | |
83354f07 JB |
5256 | /* |
5257 | * This will help us catch people modifying the fs tree while we're | |
5258 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5259 | * dropped as we unlock the root node and parent nodes as we walk down | |
5260 | * the tree, assuming nothing will change. If something does change | |
5261 | * then we'll have stale information and drop references to blocks we've | |
5262 | * already dropped. | |
5263 | */ | |
5264 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5265 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5266 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5267 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 5268 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 5269 | path->slots[level] = 0; |
bd681513 | 5270 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5271 | memset(&wc->update_progress, 0, |
5272 | sizeof(wc->update_progress)); | |
9f3a7427 | 5273 | } else { |
9f3a7427 | 5274 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5275 | memcpy(&wc->update_progress, &key, |
5276 | sizeof(wc->update_progress)); | |
5277 | ||
6702ed49 | 5278 | level = root_item->drop_level; |
2c47e605 | 5279 | BUG_ON(level == 0); |
6702ed49 | 5280 | path->lowest_level = level; |
2c47e605 YZ |
5281 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5282 | path->lowest_level = 0; | |
5283 | if (ret < 0) { | |
5284 | err = ret; | |
79787eaa | 5285 | goto out_end_trans; |
9f3a7427 | 5286 | } |
1c4850e2 | 5287 | WARN_ON(ret > 0); |
2c47e605 | 5288 | |
7d9eb12c CM |
5289 | /* |
5290 | * unlock our path, this is safe because only this | |
5291 | * function is allowed to delete this snapshot | |
5292 | */ | |
5d4f98a2 | 5293 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5294 | |
5295 | level = btrfs_header_level(root->node); | |
5296 | while (1) { | |
5297 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 5298 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 5299 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5300 | |
2ff7e61e | 5301 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5302 | path->nodes[level]->start, |
3173a18f | 5303 | level, 1, &wc->refs[level], |
2c47e605 | 5304 | &wc->flags[level]); |
79787eaa JM |
5305 | if (ret < 0) { |
5306 | err = ret; | |
5307 | goto out_end_trans; | |
5308 | } | |
2c47e605 YZ |
5309 | BUG_ON(wc->refs[level] == 0); |
5310 | ||
5311 | if (level == root_item->drop_level) | |
5312 | break; | |
5313 | ||
5314 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5315 | path->locks[level] = 0; |
2c47e605 YZ |
5316 | WARN_ON(wc->refs[level] != 1); |
5317 | level--; | |
5318 | } | |
9f3a7427 | 5319 | } |
2c47e605 | 5320 | |
78c52d9e | 5321 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5322 | wc->level = level; |
5323 | wc->shared_level = -1; | |
5324 | wc->stage = DROP_REFERENCE; | |
5325 | wc->update_ref = update_ref; | |
5326 | wc->keep_locks = 0; | |
0b246afa | 5327 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5328 | |
d397712b | 5329 | while (1) { |
9d1a2a3a | 5330 | |
2c47e605 YZ |
5331 | ret = walk_down_tree(trans, root, path, wc); |
5332 | if (ret < 0) { | |
5333 | err = ret; | |
20524f02 | 5334 | break; |
2c47e605 | 5335 | } |
9aca1d51 | 5336 | |
2c47e605 YZ |
5337 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5338 | if (ret < 0) { | |
5339 | err = ret; | |
20524f02 | 5340 | break; |
2c47e605 YZ |
5341 | } |
5342 | ||
5343 | if (ret > 0) { | |
5344 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5345 | break; |
5346 | } | |
2c47e605 YZ |
5347 | |
5348 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5349 | wc->drop_level = wc->level; |
5350 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5351 | &wc->drop_progress, | |
5352 | path->slots[wc->drop_level]); | |
5353 | } | |
5354 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5355 | &wc->drop_progress); | |
5356 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
5357 | |
5358 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5359 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5360 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5361 | ret = btrfs_update_root(trans, tree_root, |
5362 | &root->root_key, | |
5363 | root_item); | |
79787eaa | 5364 | if (ret) { |
66642832 | 5365 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5366 | err = ret; |
5367 | goto out_end_trans; | |
5368 | } | |
2c47e605 | 5369 | |
3a45bb20 | 5370 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5371 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5372 | btrfs_debug(fs_info, |
5373 | "drop snapshot early exit"); | |
3c8f2422 JB |
5374 | err = -EAGAIN; |
5375 | goto out_free; | |
5376 | } | |
5377 | ||
a22285a6 | 5378 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
5379 | if (IS_ERR(trans)) { |
5380 | err = PTR_ERR(trans); | |
5381 | goto out_free; | |
5382 | } | |
3fd0a558 YZ |
5383 | if (block_rsv) |
5384 | trans->block_rsv = block_rsv; | |
c3e69d58 | 5385 | } |
20524f02 | 5386 | } |
b3b4aa74 | 5387 | btrfs_release_path(path); |
79787eaa JM |
5388 | if (err) |
5389 | goto out_end_trans; | |
2c47e605 | 5390 | |
ab9ce7d4 | 5391 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5392 | if (ret) { |
66642832 | 5393 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5394 | err = ret; |
79787eaa JM |
5395 | goto out_end_trans; |
5396 | } | |
2c47e605 | 5397 | |
76dda93c | 5398 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5399 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5400 | NULL, NULL); | |
79787eaa | 5401 | if (ret < 0) { |
66642832 | 5402 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5403 | err = ret; |
5404 | goto out_end_trans; | |
5405 | } else if (ret > 0) { | |
84cd948c JB |
5406 | /* if we fail to delete the orphan item this time |
5407 | * around, it'll get picked up the next time. | |
5408 | * | |
5409 | * The most common failure here is just -ENOENT. | |
5410 | */ | |
5411 | btrfs_del_orphan_item(trans, tree_root, | |
5412 | root->root_key.objectid); | |
76dda93c YZ |
5413 | } |
5414 | } | |
5415 | ||
27cdeb70 | 5416 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 5417 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
5418 | } else { |
5419 | free_extent_buffer(root->node); | |
5420 | free_extent_buffer(root->commit_root); | |
00246528 | 5421 | btrfs_put_root(root); |
76dda93c | 5422 | } |
d29a9f62 | 5423 | root_dropped = true; |
79787eaa | 5424 | out_end_trans: |
3a45bb20 | 5425 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5426 | out_free: |
2c47e605 | 5427 | kfree(wc); |
5caf2a00 | 5428 | btrfs_free_path(path); |
cb1b69f4 | 5429 | out: |
d29a9f62 JB |
5430 | /* |
5431 | * So if we need to stop dropping the snapshot for whatever reason we | |
5432 | * need to make sure to add it back to the dead root list so that we | |
5433 | * keep trying to do the work later. This also cleans up roots if we | |
5434 | * don't have it in the radix (like when we recover after a power fail | |
5435 | * or unmount) so we don't leak memory. | |
5436 | */ | |
897ca819 | 5437 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5438 | btrfs_add_dead_root(root); |
90515e7f | 5439 | if (err && err != -EAGAIN) |
ab8d0fc4 | 5440 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 5441 | return err; |
20524f02 | 5442 | } |
9078a3e1 | 5443 | |
2c47e605 YZ |
5444 | /* |
5445 | * drop subtree rooted at tree block 'node'. | |
5446 | * | |
5447 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5448 | * only used by relocation code |
2c47e605 | 5449 | */ |
f82d02d9 YZ |
5450 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5451 | struct btrfs_root *root, | |
5452 | struct extent_buffer *node, | |
5453 | struct extent_buffer *parent) | |
5454 | { | |
0b246afa | 5455 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5456 | struct btrfs_path *path; |
2c47e605 | 5457 | struct walk_control *wc; |
f82d02d9 YZ |
5458 | int level; |
5459 | int parent_level; | |
5460 | int ret = 0; | |
5461 | int wret; | |
5462 | ||
2c47e605 YZ |
5463 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5464 | ||
f82d02d9 | 5465 | path = btrfs_alloc_path(); |
db5b493a TI |
5466 | if (!path) |
5467 | return -ENOMEM; | |
f82d02d9 | 5468 | |
2c47e605 | 5469 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5470 | if (!wc) { |
5471 | btrfs_free_path(path); | |
5472 | return -ENOMEM; | |
5473 | } | |
2c47e605 | 5474 | |
b9447ef8 | 5475 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5476 | parent_level = btrfs_header_level(parent); |
67439dad | 5477 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5478 | path->nodes[parent_level] = parent; |
5479 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5480 | ||
b9447ef8 | 5481 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5482 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5483 | path->nodes[level] = node; |
5484 | path->slots[level] = 0; | |
bd681513 | 5485 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5486 | |
5487 | wc->refs[parent_level] = 1; | |
5488 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5489 | wc->level = level; | |
5490 | wc->shared_level = -1; | |
5491 | wc->stage = DROP_REFERENCE; | |
5492 | wc->update_ref = 0; | |
5493 | wc->keep_locks = 1; | |
0b246afa | 5494 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5495 | |
5496 | while (1) { | |
2c47e605 YZ |
5497 | wret = walk_down_tree(trans, root, path, wc); |
5498 | if (wret < 0) { | |
f82d02d9 | 5499 | ret = wret; |
f82d02d9 | 5500 | break; |
2c47e605 | 5501 | } |
f82d02d9 | 5502 | |
2c47e605 | 5503 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5504 | if (wret < 0) |
5505 | ret = wret; | |
5506 | if (wret != 0) | |
5507 | break; | |
5508 | } | |
5509 | ||
2c47e605 | 5510 | kfree(wc); |
f82d02d9 YZ |
5511 | btrfs_free_path(path); |
5512 | return ret; | |
5513 | } | |
5514 | ||
6d07bcec MX |
5515 | /* |
5516 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5517 | * space_info. takes mirrors into account. |
6d07bcec | 5518 | */ |
633c0aad | 5519 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5520 | { |
32da5386 | 5521 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5522 | u64 free_bytes = 0; |
5523 | int factor; | |
5524 | ||
01327610 | 5525 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5526 | if (list_empty(&sinfo->ro_bgs)) |
5527 | return 0; | |
5528 | ||
5529 | spin_lock(&sinfo->lock); | |
5530 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5531 | spin_lock(&block_group->lock); |
5532 | ||
5533 | if (!block_group->ro) { | |
5534 | spin_unlock(&block_group->lock); | |
5535 | continue; | |
5536 | } | |
5537 | ||
46df06b8 | 5538 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5539 | free_bytes += (block_group->length - |
bf38be65 | 5540 | block_group->used) * factor; |
6d07bcec MX |
5541 | |
5542 | spin_unlock(&block_group->lock); | |
5543 | } | |
6d07bcec MX |
5544 | spin_unlock(&sinfo->lock); |
5545 | ||
5546 | return free_bytes; | |
5547 | } | |
5548 | ||
2ff7e61e JM |
5549 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5550 | u64 start, u64 end) | |
acce952b | 5551 | { |
2ff7e61e | 5552 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5553 | } |
5554 | ||
499f377f JM |
5555 | /* |
5556 | * It used to be that old block groups would be left around forever. | |
5557 | * Iterating over them would be enough to trim unused space. Since we | |
5558 | * now automatically remove them, we also need to iterate over unallocated | |
5559 | * space. | |
5560 | * | |
5561 | * We don't want a transaction for this since the discard may take a | |
5562 | * substantial amount of time. We don't require that a transaction be | |
5563 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5564 | * to ensure that we're not discarding chunks that were released or |
5565 | * allocated in the current transaction. | |
499f377f JM |
5566 | * |
5567 | * Holding the chunks lock will prevent other threads from allocating | |
5568 | * or releasing chunks, but it won't prevent a running transaction | |
5569 | * from committing and releasing the memory that the pending chunks | |
5570 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5571 | * transaction and hold the commit root sem. We only need to hold |
5572 | * it while performing the free space search since we have already | |
5573 | * held back allocations. | |
499f377f | 5574 | */ |
8103d10b | 5575 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5576 | { |
8103d10b | 5577 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5578 | int ret; |
5579 | ||
5580 | *trimmed = 0; | |
5581 | ||
0be88e36 JM |
5582 | /* Discard not supported = nothing to do. */ |
5583 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5584 | return 0; | |
5585 | ||
52042d8e | 5586 | /* Not writable = nothing to do. */ |
ebbede42 | 5587 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5588 | return 0; |
5589 | ||
5590 | /* No free space = nothing to do. */ | |
5591 | if (device->total_bytes <= device->bytes_used) | |
5592 | return 0; | |
5593 | ||
5594 | ret = 0; | |
5595 | ||
5596 | while (1) { | |
fb456252 | 5597 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5598 | u64 bytes; |
5599 | ||
5600 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5601 | if (ret) | |
fee7acc3 | 5602 | break; |
499f377f | 5603 | |
929be17a NB |
5604 | find_first_clear_extent_bit(&device->alloc_state, start, |
5605 | &start, &end, | |
5606 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 NB |
5607 | |
5608 | /* Ensure we skip the reserved area in the first 1M */ | |
5609 | start = max_t(u64, start, SZ_1M); | |
5610 | ||
929be17a NB |
5611 | /* |
5612 | * If find_first_clear_extent_bit find a range that spans the | |
5613 | * end of the device it will set end to -1, in this case it's up | |
5614 | * to the caller to trim the value to the size of the device. | |
5615 | */ | |
5616 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5617 | |
929be17a | 5618 | len = end - start + 1; |
499f377f | 5619 | |
929be17a NB |
5620 | /* We didn't find any extents */ |
5621 | if (!len) { | |
499f377f | 5622 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5623 | ret = 0; |
499f377f JM |
5624 | break; |
5625 | } | |
5626 | ||
929be17a NB |
5627 | ret = btrfs_issue_discard(device->bdev, start, len, |
5628 | &bytes); | |
5629 | if (!ret) | |
5630 | set_extent_bits(&device->alloc_state, start, | |
5631 | start + bytes - 1, | |
5632 | CHUNK_TRIMMED); | |
499f377f JM |
5633 | mutex_unlock(&fs_info->chunk_mutex); |
5634 | ||
5635 | if (ret) | |
5636 | break; | |
5637 | ||
5638 | start += len; | |
5639 | *trimmed += bytes; | |
5640 | ||
5641 | if (fatal_signal_pending(current)) { | |
5642 | ret = -ERESTARTSYS; | |
5643 | break; | |
5644 | } | |
5645 | ||
5646 | cond_resched(); | |
5647 | } | |
5648 | ||
5649 | return ret; | |
5650 | } | |
5651 | ||
93bba24d QW |
5652 | /* |
5653 | * Trim the whole filesystem by: | |
5654 | * 1) trimming the free space in each block group | |
5655 | * 2) trimming the unallocated space on each device | |
5656 | * | |
5657 | * This will also continue trimming even if a block group or device encounters | |
5658 | * an error. The return value will be the last error, or 0 if nothing bad | |
5659 | * happens. | |
5660 | */ | |
2ff7e61e | 5661 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5662 | { |
32da5386 | 5663 | struct btrfs_block_group *cache = NULL; |
499f377f JM |
5664 | struct btrfs_device *device; |
5665 | struct list_head *devices; | |
f7039b1d | 5666 | u64 group_trimmed; |
07301df7 | 5667 | u64 range_end = U64_MAX; |
f7039b1d LD |
5668 | u64 start; |
5669 | u64 end; | |
5670 | u64 trimmed = 0; | |
93bba24d QW |
5671 | u64 bg_failed = 0; |
5672 | u64 dev_failed = 0; | |
5673 | int bg_ret = 0; | |
5674 | int dev_ret = 0; | |
f7039b1d LD |
5675 | int ret = 0; |
5676 | ||
07301df7 QW |
5677 | /* |
5678 | * Check range overflow if range->len is set. | |
5679 | * The default range->len is U64_MAX. | |
5680 | */ | |
5681 | if (range->len != U64_MAX && | |
5682 | check_add_overflow(range->start, range->len, &range_end)) | |
5683 | return -EINVAL; | |
5684 | ||
6ba9fc8e | 5685 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5686 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 5687 | if (cache->start >= range_end) { |
f7039b1d LD |
5688 | btrfs_put_block_group(cache); |
5689 | break; | |
5690 | } | |
5691 | ||
b3470b5d DS |
5692 | start = max(range->start, cache->start); |
5693 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
5694 | |
5695 | if (end - start >= range->minlen) { | |
32da5386 | 5696 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 5697 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 5698 | if (ret) { |
93bba24d QW |
5699 | bg_failed++; |
5700 | bg_ret = ret; | |
5701 | continue; | |
1be41b78 | 5702 | } |
676f1f75 | 5703 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5704 | if (ret) { |
93bba24d QW |
5705 | bg_failed++; |
5706 | bg_ret = ret; | |
5707 | continue; | |
1be41b78 | 5708 | } |
f7039b1d LD |
5709 | } |
5710 | ret = btrfs_trim_block_group(cache, | |
5711 | &group_trimmed, | |
5712 | start, | |
5713 | end, | |
5714 | range->minlen); | |
5715 | ||
5716 | trimmed += group_trimmed; | |
5717 | if (ret) { | |
93bba24d QW |
5718 | bg_failed++; |
5719 | bg_ret = ret; | |
5720 | continue; | |
f7039b1d LD |
5721 | } |
5722 | } | |
f7039b1d LD |
5723 | } |
5724 | ||
93bba24d QW |
5725 | if (bg_failed) |
5726 | btrfs_warn(fs_info, | |
5727 | "failed to trim %llu block group(s), last error %d", | |
5728 | bg_failed, bg_ret); | |
0b246afa | 5729 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5730 | devices = &fs_info->fs_devices->devices; |
5731 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5732 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5733 | if (ret) { |
5734 | dev_failed++; | |
5735 | dev_ret = ret; | |
499f377f | 5736 | break; |
93bba24d | 5737 | } |
499f377f JM |
5738 | |
5739 | trimmed += group_trimmed; | |
5740 | } | |
0b246afa | 5741 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5742 | |
93bba24d QW |
5743 | if (dev_failed) |
5744 | btrfs_warn(fs_info, | |
5745 | "failed to trim %llu device(s), last error %d", | |
5746 | dev_failed, dev_ret); | |
f7039b1d | 5747 | range->len = trimmed; |
93bba24d QW |
5748 | if (bg_ret) |
5749 | return bg_ret; | |
5750 | return dev_ret; | |
f7039b1d | 5751 | } |
8257b2dc MX |
5752 | |
5753 | /* | |
ea14b57f | 5754 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
5755 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
5756 | * data into the page cache through nocow before the subvolume is snapshoted, | |
5757 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 5758 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 5759 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 5760 | */ |
ea14b57f | 5761 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
5762 | { |
5763 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 5764 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
5765 | } |
5766 | ||
ea14b57f | 5767 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 5768 | { |
ea14b57f | 5769 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
5770 | return 0; |
5771 | ||
5772 | percpu_counter_inc(&root->subv_writers->counter); | |
5773 | /* | |
5774 | * Make sure counter is updated before we check for snapshot creation. | |
5775 | */ | |
5776 | smp_mb(); | |
ea14b57f DS |
5777 | if (atomic_read(&root->will_be_snapshotted)) { |
5778 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
5779 | return 0; |
5780 | } | |
5781 | return 1; | |
5782 | } | |
0bc19f90 | 5783 | |
0bc19f90 ZL |
5784 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
5785 | { | |
5786 | while (true) { | |
5787 | int ret; | |
5788 | ||
ea14b57f | 5789 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
5790 | if (ret) |
5791 | break; | |
4625956a PZ |
5792 | wait_var_event(&root->will_be_snapshotted, |
5793 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
5794 | } |
5795 | } |