]>
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" |
c57dd1f2 | 36 | #include "rcu-string.h" |
fec577fb | 37 | |
709c0486 AJ |
38 | #undef SCRAMBLE_DELAYED_REFS |
39 | ||
9f9b8e8d | 40 | |
5d4f98a2 | 41 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
42 | struct btrfs_delayed_ref_node *node, u64 parent, |
43 | u64 root_objectid, u64 owner_objectid, | |
44 | u64 owner_offset, int refs_to_drop, | |
45 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
46 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
47 | struct extent_buffer *leaf, | |
48 | struct btrfs_extent_item *ei); | |
49 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
50 | u64 parent, u64 root_objectid, |
51 | u64 flags, u64 owner, u64 offset, | |
52 | struct btrfs_key *ins, int ref_mod); | |
53 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 54 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 55 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
56 | static int find_next_key(struct btrfs_path *path, int level, |
57 | struct btrfs_key *key); | |
6a63209f | 58 | |
32da5386 | 59 | static int block_group_bits(struct btrfs_block_group *cache, u64 bits) |
0f9dd46c JB |
60 | { |
61 | return (cache->flags & bits) == bits; | |
62 | } | |
63 | ||
6f410d1b JB |
64 | int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info, |
65 | u64 start, u64 num_bytes) | |
817d52f8 | 66 | { |
11833d66 | 67 | u64 end = start + num_bytes - 1; |
fe119a6e NB |
68 | set_extent_bits(&fs_info->excluded_extents, start, end, |
69 | EXTENT_UPTODATE); | |
11833d66 YZ |
70 | return 0; |
71 | } | |
817d52f8 | 72 | |
32da5386 | 73 | void btrfs_free_excluded_extents(struct btrfs_block_group *cache) |
11833d66 | 74 | { |
9e715da8 | 75 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 76 | u64 start, end; |
817d52f8 | 77 | |
b3470b5d DS |
78 | start = cache->start; |
79 | end = start + cache->length - 1; | |
11833d66 | 80 | |
fe119a6e NB |
81 | clear_extent_bits(&fs_info->excluded_extents, start, end, |
82 | EXTENT_UPTODATE); | |
817d52f8 JB |
83 | } |
84 | ||
78192442 | 85 | static u64 generic_ref_to_space_flags(struct btrfs_ref *ref) |
0d9f824d | 86 | { |
ddf30cf0 QW |
87 | if (ref->type == BTRFS_REF_METADATA) { |
88 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
78192442 | 89 | return BTRFS_BLOCK_GROUP_SYSTEM; |
0d9f824d | 90 | else |
78192442 | 91 | return BTRFS_BLOCK_GROUP_METADATA; |
0d9f824d | 92 | } |
78192442 QW |
93 | return BTRFS_BLOCK_GROUP_DATA; |
94 | } | |
95 | ||
96 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, | |
97 | struct btrfs_ref *ref) | |
98 | { | |
99 | struct btrfs_space_info *space_info; | |
100 | u64 flags = generic_ref_to_space_flags(ref); | |
101 | ||
280c2908 | 102 | space_info = btrfs_find_space_info(fs_info, flags); |
78192442 QW |
103 | ASSERT(space_info); |
104 | percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len, | |
105 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
106 | } | |
107 | ||
108 | static void sub_pinned_bytes(struct btrfs_fs_info *fs_info, | |
109 | struct btrfs_ref *ref) | |
110 | { | |
111 | struct btrfs_space_info *space_info; | |
112 | u64 flags = generic_ref_to_space_flags(ref); | |
0d9f824d | 113 | |
280c2908 | 114 | space_info = btrfs_find_space_info(fs_info, flags); |
55e8196a | 115 | ASSERT(space_info); |
78192442 | 116 | percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len, |
dec59fa3 | 117 | BTRFS_TOTAL_BYTES_PINNED_BATCH); |
0d9f824d OS |
118 | } |
119 | ||
1a4ed8fd | 120 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 121 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
122 | { |
123 | int ret; | |
124 | struct btrfs_key key; | |
31840ae1 | 125 | struct btrfs_path *path; |
e02119d5 | 126 | |
31840ae1 | 127 | path = btrfs_alloc_path(); |
d8926bb3 MF |
128 | if (!path) |
129 | return -ENOMEM; | |
130 | ||
e02119d5 CM |
131 | key.objectid = start; |
132 | key.offset = len; | |
3173a18f | 133 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 134 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 135 | btrfs_free_path(path); |
7bb86316 CM |
136 | return ret; |
137 | } | |
138 | ||
a22285a6 | 139 | /* |
3173a18f | 140 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
141 | * |
142 | * the head node for delayed ref is used to store the sum of all the | |
143 | * reference count modifications queued up in the rbtree. the head | |
144 | * node may also store the extent flags to set. This way you can check | |
145 | * to see what the reference count and extent flags would be if all of | |
146 | * the delayed refs are not processed. | |
147 | */ | |
148 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 149 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 150 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
151 | { |
152 | struct btrfs_delayed_ref_head *head; | |
153 | struct btrfs_delayed_ref_root *delayed_refs; | |
154 | struct btrfs_path *path; | |
155 | struct btrfs_extent_item *ei; | |
156 | struct extent_buffer *leaf; | |
157 | struct btrfs_key key; | |
158 | u32 item_size; | |
159 | u64 num_refs; | |
160 | u64 extent_flags; | |
161 | int ret; | |
162 | ||
3173a18f JB |
163 | /* |
164 | * If we don't have skinny metadata, don't bother doing anything | |
165 | * different | |
166 | */ | |
0b246afa JM |
167 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
168 | offset = fs_info->nodesize; | |
3173a18f JB |
169 | metadata = 0; |
170 | } | |
171 | ||
a22285a6 YZ |
172 | path = btrfs_alloc_path(); |
173 | if (!path) | |
174 | return -ENOMEM; | |
175 | ||
a22285a6 YZ |
176 | if (!trans) { |
177 | path->skip_locking = 1; | |
178 | path->search_commit_root = 1; | |
179 | } | |
639eefc8 FDBM |
180 | |
181 | search_again: | |
182 | key.objectid = bytenr; | |
183 | key.offset = offset; | |
184 | if (metadata) | |
185 | key.type = BTRFS_METADATA_ITEM_KEY; | |
186 | else | |
187 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
188 | ||
0b246afa | 189 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
190 | if (ret < 0) |
191 | goto out_free; | |
192 | ||
3173a18f | 193 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
194 | if (path->slots[0]) { |
195 | path->slots[0]--; | |
196 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
197 | path->slots[0]); | |
198 | if (key.objectid == bytenr && | |
199 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 200 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
201 | ret = 0; |
202 | } | |
3173a18f JB |
203 | } |
204 | ||
a22285a6 YZ |
205 | if (ret == 0) { |
206 | leaf = path->nodes[0]; | |
207 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
208 | if (item_size >= sizeof(*ei)) { | |
209 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
210 | struct btrfs_extent_item); | |
211 | num_refs = btrfs_extent_refs(leaf, ei); | |
212 | extent_flags = btrfs_extent_flags(leaf, ei); | |
213 | } else { | |
ba3c2b19 NB |
214 | ret = -EINVAL; |
215 | btrfs_print_v0_err(fs_info); | |
216 | if (trans) | |
217 | btrfs_abort_transaction(trans, ret); | |
218 | else | |
219 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
220 | ||
221 | goto out_free; | |
a22285a6 | 222 | } |
ba3c2b19 | 223 | |
a22285a6 YZ |
224 | BUG_ON(num_refs == 0); |
225 | } else { | |
226 | num_refs = 0; | |
227 | extent_flags = 0; | |
228 | ret = 0; | |
229 | } | |
230 | ||
231 | if (!trans) | |
232 | goto out; | |
233 | ||
234 | delayed_refs = &trans->transaction->delayed_refs; | |
235 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 236 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
237 | if (head) { |
238 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 239 | refcount_inc(&head->refs); |
a22285a6 YZ |
240 | spin_unlock(&delayed_refs->lock); |
241 | ||
b3b4aa74 | 242 | btrfs_release_path(path); |
a22285a6 | 243 | |
8cc33e5c DS |
244 | /* |
245 | * Mutex was contended, block until it's released and try | |
246 | * again | |
247 | */ | |
a22285a6 YZ |
248 | mutex_lock(&head->mutex); |
249 | mutex_unlock(&head->mutex); | |
d278850e | 250 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 251 | goto search_again; |
a22285a6 | 252 | } |
d7df2c79 | 253 | spin_lock(&head->lock); |
a22285a6 YZ |
254 | if (head->extent_op && head->extent_op->update_flags) |
255 | extent_flags |= head->extent_op->flags_to_set; | |
256 | else | |
257 | BUG_ON(num_refs == 0); | |
258 | ||
d278850e | 259 | num_refs += head->ref_mod; |
d7df2c79 | 260 | spin_unlock(&head->lock); |
a22285a6 YZ |
261 | mutex_unlock(&head->mutex); |
262 | } | |
263 | spin_unlock(&delayed_refs->lock); | |
264 | out: | |
265 | WARN_ON(num_refs == 0); | |
266 | if (refs) | |
267 | *refs = num_refs; | |
268 | if (flags) | |
269 | *flags = extent_flags; | |
270 | out_free: | |
271 | btrfs_free_path(path); | |
272 | return ret; | |
273 | } | |
274 | ||
d8d5f3e1 CM |
275 | /* |
276 | * Back reference rules. Back refs have three main goals: | |
277 | * | |
278 | * 1) differentiate between all holders of references to an extent so that | |
279 | * when a reference is dropped we can make sure it was a valid reference | |
280 | * before freeing the extent. | |
281 | * | |
282 | * 2) Provide enough information to quickly find the holders of an extent | |
283 | * if we notice a given block is corrupted or bad. | |
284 | * | |
285 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
286 | * maintenance. This is actually the same as #2, but with a slightly | |
287 | * different use case. | |
288 | * | |
5d4f98a2 YZ |
289 | * There are two kinds of back refs. The implicit back refs is optimized |
290 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
291 | * back refs of this kind provide information about the block's owner tree | |
292 | * and the pointer's key. These information allow us to find the block by | |
293 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
294 | * referenced by their owner trees. The location of tree block is recorded | |
295 | * in the back refs. Actually the full back refs is generic, and can be | |
296 | * used in all cases the implicit back refs is used. The major shortcoming | |
297 | * of the full back refs is its overhead. Every time a tree block gets | |
298 | * COWed, we have to update back refs entry for all pointers in it. | |
299 | * | |
300 | * For a newly allocated tree block, we use implicit back refs for | |
301 | * pointers in it. This means most tree related operations only involve | |
302 | * implicit back refs. For a tree block created in old transaction, the | |
303 | * only way to drop a reference to it is COW it. So we can detect the | |
304 | * event that tree block loses its owner tree's reference and do the | |
305 | * back refs conversion. | |
306 | * | |
01327610 | 307 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
308 | * |
309 | * The reference count of the block is one and the tree is the block's | |
310 | * owner tree. Nothing to do in this case. | |
311 | * | |
312 | * The reference count of the block is one and the tree is not the | |
313 | * block's owner tree. In this case, full back refs is used for pointers | |
314 | * in the block. Remove these full back refs, add implicit back refs for | |
315 | * every pointers in the new block. | |
316 | * | |
317 | * The reference count of the block is greater than one and the tree is | |
318 | * the block's owner tree. In this case, implicit back refs is used for | |
319 | * pointers in the block. Add full back refs for every pointers in the | |
320 | * block, increase lower level extents' reference counts. The original | |
321 | * implicit back refs are entailed to the new block. | |
322 | * | |
323 | * The reference count of the block is greater than one and the tree is | |
324 | * not the block's owner tree. Add implicit back refs for every pointer in | |
325 | * the new block, increase lower level extents' reference count. | |
326 | * | |
327 | * Back Reference Key composing: | |
328 | * | |
329 | * The key objectid corresponds to the first byte in the extent, | |
330 | * The key type is used to differentiate between types of back refs. | |
331 | * There are different meanings of the key offset for different types | |
332 | * of back refs. | |
333 | * | |
d8d5f3e1 CM |
334 | * File extents can be referenced by: |
335 | * | |
336 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 337 | * - different files inside a single subvolume |
d8d5f3e1 CM |
338 | * - different offsets inside a file (bookend extents in file.c) |
339 | * | |
5d4f98a2 | 340 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
341 | * |
342 | * - Objectid of the subvolume root | |
d8d5f3e1 | 343 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
344 | * - original offset in the file |
345 | * - how many bookend extents | |
d8d5f3e1 | 346 | * |
5d4f98a2 YZ |
347 | * The key offset for the implicit back refs is hash of the first |
348 | * three fields. | |
d8d5f3e1 | 349 | * |
5d4f98a2 | 350 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 351 | * |
5d4f98a2 | 352 | * - number of pointers in the tree leaf |
d8d5f3e1 | 353 | * |
5d4f98a2 YZ |
354 | * The key offset for the implicit back refs is the first byte of |
355 | * the tree leaf | |
d8d5f3e1 | 356 | * |
5d4f98a2 YZ |
357 | * When a file extent is allocated, The implicit back refs is used. |
358 | * the fields are filled in: | |
d8d5f3e1 | 359 | * |
5d4f98a2 | 360 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 361 | * |
5d4f98a2 YZ |
362 | * When a file extent is removed file truncation, we find the |
363 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 364 | * |
5d4f98a2 | 365 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 366 | * |
5d4f98a2 | 367 | * Btree extents can be referenced by: |
d8d5f3e1 | 368 | * |
5d4f98a2 | 369 | * - Different subvolumes |
d8d5f3e1 | 370 | * |
5d4f98a2 YZ |
371 | * Both the implicit back refs and the full back refs for tree blocks |
372 | * only consist of key. The key offset for the implicit back refs is | |
373 | * objectid of block's owner tree. The key offset for the full back refs | |
374 | * is the first byte of parent block. | |
d8d5f3e1 | 375 | * |
5d4f98a2 YZ |
376 | * When implicit back refs is used, information about the lowest key and |
377 | * level of the tree block are required. These information are stored in | |
378 | * tree block info structure. | |
d8d5f3e1 | 379 | */ |
31840ae1 | 380 | |
167ce953 LB |
381 | /* |
382 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 383 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
384 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
385 | */ | |
386 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
387 | struct btrfs_extent_inline_ref *iref, | |
388 | enum btrfs_inline_ref_type is_data) | |
389 | { | |
390 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 391 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
392 | |
393 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
394 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
395 | type == BTRFS_SHARED_DATA_REF_KEY || | |
396 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
397 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 398 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 399 | return type; |
64ecdb64 LB |
400 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
401 | ASSERT(eb->fs_info); | |
402 | /* | |
ea57788e QW |
403 | * Every shared one has parent tree block, |
404 | * which must be aligned to sector size. | |
64ecdb64 LB |
405 | */ |
406 | if (offset && | |
ea57788e | 407 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
408 | return type; |
409 | } | |
167ce953 | 410 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 411 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 412 | return type; |
64ecdb64 LB |
413 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
414 | ASSERT(eb->fs_info); | |
415 | /* | |
ea57788e QW |
416 | * Every shared one has parent tree block, |
417 | * which must be aligned to sector size. | |
64ecdb64 LB |
418 | */ |
419 | if (offset && | |
ea57788e | 420 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
421 | return type; |
422 | } | |
167ce953 LB |
423 | } else { |
424 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
425 | return type; | |
426 | } | |
427 | } | |
428 | ||
429 | btrfs_print_leaf((struct extent_buffer *)eb); | |
ea57788e QW |
430 | btrfs_err(eb->fs_info, |
431 | "eb %llu iref 0x%lx invalid extent inline ref type %d", | |
432 | eb->start, (unsigned long)iref, type); | |
167ce953 LB |
433 | WARN_ON(1); |
434 | ||
435 | return BTRFS_REF_TYPE_INVALID; | |
436 | } | |
437 | ||
0785a9aa | 438 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
439 | { |
440 | u32 high_crc = ~(u32)0; | |
441 | u32 low_crc = ~(u32)0; | |
442 | __le64 lenum; | |
443 | ||
444 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 445 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 446 | lenum = cpu_to_le64(owner); |
65019df8 | 447 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 448 | lenum = cpu_to_le64(offset); |
65019df8 | 449 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
450 | |
451 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
452 | } | |
453 | ||
454 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
455 | struct btrfs_extent_data_ref *ref) | |
456 | { | |
457 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
458 | btrfs_extent_data_ref_objectid(leaf, ref), | |
459 | btrfs_extent_data_ref_offset(leaf, ref)); | |
460 | } | |
461 | ||
462 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
463 | struct btrfs_extent_data_ref *ref, | |
464 | u64 root_objectid, u64 owner, u64 offset) | |
465 | { | |
466 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
467 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
468 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
469 | return 0; | |
470 | return 1; | |
471 | } | |
472 | ||
473 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
474 | struct btrfs_path *path, |
475 | u64 bytenr, u64 parent, | |
476 | u64 root_objectid, | |
477 | u64 owner, u64 offset) | |
478 | { | |
bd1d53ef | 479 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
480 | struct btrfs_key key; |
481 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 482 | struct extent_buffer *leaf; |
5d4f98a2 | 483 | u32 nritems; |
74493f7a | 484 | int ret; |
5d4f98a2 YZ |
485 | int recow; |
486 | int err = -ENOENT; | |
74493f7a | 487 | |
31840ae1 | 488 | key.objectid = bytenr; |
5d4f98a2 YZ |
489 | if (parent) { |
490 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
491 | key.offset = parent; | |
492 | } else { | |
493 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
494 | key.offset = hash_extent_data_ref(root_objectid, | |
495 | owner, offset); | |
496 | } | |
497 | again: | |
498 | recow = 0; | |
499 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
500 | if (ret < 0) { | |
501 | err = ret; | |
502 | goto fail; | |
503 | } | |
31840ae1 | 504 | |
5d4f98a2 YZ |
505 | if (parent) { |
506 | if (!ret) | |
507 | return 0; | |
5d4f98a2 | 508 | goto fail; |
31840ae1 ZY |
509 | } |
510 | ||
511 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
512 | nritems = btrfs_header_nritems(leaf); |
513 | while (1) { | |
514 | if (path->slots[0] >= nritems) { | |
515 | ret = btrfs_next_leaf(root, path); | |
516 | if (ret < 0) | |
517 | err = ret; | |
518 | if (ret) | |
519 | goto fail; | |
520 | ||
521 | leaf = path->nodes[0]; | |
522 | nritems = btrfs_header_nritems(leaf); | |
523 | recow = 1; | |
524 | } | |
525 | ||
526 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
527 | if (key.objectid != bytenr || | |
528 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
529 | goto fail; | |
530 | ||
531 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
532 | struct btrfs_extent_data_ref); | |
533 | ||
534 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
535 | owner, offset)) { | |
536 | if (recow) { | |
b3b4aa74 | 537 | btrfs_release_path(path); |
5d4f98a2 YZ |
538 | goto again; |
539 | } | |
540 | err = 0; | |
541 | break; | |
542 | } | |
543 | path->slots[0]++; | |
31840ae1 | 544 | } |
5d4f98a2 YZ |
545 | fail: |
546 | return err; | |
31840ae1 ZY |
547 | } |
548 | ||
5d4f98a2 | 549 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
550 | struct btrfs_path *path, |
551 | u64 bytenr, u64 parent, | |
552 | u64 root_objectid, u64 owner, | |
553 | u64 offset, int refs_to_add) | |
31840ae1 | 554 | { |
62b895af | 555 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
556 | struct btrfs_key key; |
557 | struct extent_buffer *leaf; | |
5d4f98a2 | 558 | u32 size; |
31840ae1 ZY |
559 | u32 num_refs; |
560 | int ret; | |
74493f7a | 561 | |
74493f7a | 562 | key.objectid = bytenr; |
5d4f98a2 YZ |
563 | if (parent) { |
564 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
565 | key.offset = parent; | |
566 | size = sizeof(struct btrfs_shared_data_ref); | |
567 | } else { | |
568 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
569 | key.offset = hash_extent_data_ref(root_objectid, | |
570 | owner, offset); | |
571 | size = sizeof(struct btrfs_extent_data_ref); | |
572 | } | |
74493f7a | 573 | |
5d4f98a2 YZ |
574 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
575 | if (ret && ret != -EEXIST) | |
576 | goto fail; | |
577 | ||
578 | leaf = path->nodes[0]; | |
579 | if (parent) { | |
580 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 581 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
582 | struct btrfs_shared_data_ref); |
583 | if (ret == 0) { | |
584 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
585 | } else { | |
586 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
587 | num_refs += refs_to_add; | |
588 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 589 | } |
5d4f98a2 YZ |
590 | } else { |
591 | struct btrfs_extent_data_ref *ref; | |
592 | while (ret == -EEXIST) { | |
593 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
594 | struct btrfs_extent_data_ref); | |
595 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
596 | owner, offset)) | |
597 | break; | |
b3b4aa74 | 598 | btrfs_release_path(path); |
5d4f98a2 YZ |
599 | key.offset++; |
600 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
601 | size); | |
602 | if (ret && ret != -EEXIST) | |
603 | goto fail; | |
31840ae1 | 604 | |
5d4f98a2 YZ |
605 | leaf = path->nodes[0]; |
606 | } | |
607 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
608 | struct btrfs_extent_data_ref); | |
609 | if (ret == 0) { | |
610 | btrfs_set_extent_data_ref_root(leaf, ref, | |
611 | root_objectid); | |
612 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
613 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
614 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
615 | } else { | |
616 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
617 | num_refs += refs_to_add; | |
618 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 619 | } |
31840ae1 | 620 | } |
5d4f98a2 YZ |
621 | btrfs_mark_buffer_dirty(leaf); |
622 | ret = 0; | |
623 | fail: | |
b3b4aa74 | 624 | btrfs_release_path(path); |
7bb86316 | 625 | return ret; |
74493f7a CM |
626 | } |
627 | ||
5d4f98a2 | 628 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 629 | struct btrfs_path *path, |
fcebe456 | 630 | int refs_to_drop, int *last_ref) |
31840ae1 | 631 | { |
5d4f98a2 YZ |
632 | struct btrfs_key key; |
633 | struct btrfs_extent_data_ref *ref1 = NULL; | |
634 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 635 | struct extent_buffer *leaf; |
5d4f98a2 | 636 | u32 num_refs = 0; |
31840ae1 ZY |
637 | int ret = 0; |
638 | ||
639 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
640 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
641 | ||
642 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
643 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
644 | struct btrfs_extent_data_ref); | |
645 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
646 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
647 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
648 | struct btrfs_shared_data_ref); | |
649 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 650 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
651 | btrfs_print_v0_err(trans->fs_info); |
652 | btrfs_abort_transaction(trans, -EINVAL); | |
653 | return -EINVAL; | |
5d4f98a2 YZ |
654 | } else { |
655 | BUG(); | |
656 | } | |
657 | ||
56bec294 CM |
658 | BUG_ON(num_refs < refs_to_drop); |
659 | num_refs -= refs_to_drop; | |
5d4f98a2 | 660 | |
31840ae1 | 661 | if (num_refs == 0) { |
e9f6290d | 662 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 663 | *last_ref = 1; |
31840ae1 | 664 | } else { |
5d4f98a2 YZ |
665 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
666 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
667 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
668 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
669 | btrfs_mark_buffer_dirty(leaf); |
670 | } | |
31840ae1 ZY |
671 | return ret; |
672 | } | |
673 | ||
9ed0dea0 | 674 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 675 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 676 | { |
5d4f98a2 YZ |
677 | struct btrfs_key key; |
678 | struct extent_buffer *leaf; | |
679 | struct btrfs_extent_data_ref *ref1; | |
680 | struct btrfs_shared_data_ref *ref2; | |
681 | u32 num_refs = 0; | |
3de28d57 | 682 | int type; |
5d4f98a2 YZ |
683 | |
684 | leaf = path->nodes[0]; | |
685 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
686 | |
687 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 688 | if (iref) { |
3de28d57 LB |
689 | /* |
690 | * If type is invalid, we should have bailed out earlier than | |
691 | * this call. | |
692 | */ | |
693 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
694 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
695 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
696 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
697 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
698 | } else { | |
699 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
700 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
701 | } | |
702 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
703 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
704 | struct btrfs_extent_data_ref); | |
705 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
706 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
707 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
708 | struct btrfs_shared_data_ref); | |
709 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
710 | } else { |
711 | WARN_ON(1); | |
712 | } | |
713 | return num_refs; | |
714 | } | |
15916de8 | 715 | |
5d4f98a2 | 716 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
717 | struct btrfs_path *path, |
718 | u64 bytenr, u64 parent, | |
719 | u64 root_objectid) | |
1f3c79a2 | 720 | { |
b8582eea | 721 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 722 | struct btrfs_key key; |
1f3c79a2 | 723 | int ret; |
1f3c79a2 | 724 | |
5d4f98a2 YZ |
725 | key.objectid = bytenr; |
726 | if (parent) { | |
727 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
728 | key.offset = parent; | |
729 | } else { | |
730 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
731 | key.offset = root_objectid; | |
1f3c79a2 LH |
732 | } |
733 | ||
5d4f98a2 YZ |
734 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
735 | if (ret > 0) | |
736 | ret = -ENOENT; | |
5d4f98a2 | 737 | return ret; |
1f3c79a2 LH |
738 | } |
739 | ||
5d4f98a2 | 740 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
741 | struct btrfs_path *path, |
742 | u64 bytenr, u64 parent, | |
743 | u64 root_objectid) | |
31840ae1 | 744 | { |
5d4f98a2 | 745 | struct btrfs_key key; |
31840ae1 | 746 | int ret; |
31840ae1 | 747 | |
5d4f98a2 YZ |
748 | key.objectid = bytenr; |
749 | if (parent) { | |
750 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
751 | key.offset = parent; | |
752 | } else { | |
753 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
754 | key.offset = root_objectid; | |
755 | } | |
756 | ||
10728404 | 757 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 758 | path, &key, 0); |
b3b4aa74 | 759 | btrfs_release_path(path); |
31840ae1 ZY |
760 | return ret; |
761 | } | |
762 | ||
5d4f98a2 | 763 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 764 | { |
5d4f98a2 YZ |
765 | int type; |
766 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
767 | if (parent > 0) | |
768 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
769 | else | |
770 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
771 | } else { | |
772 | if (parent > 0) | |
773 | type = BTRFS_SHARED_DATA_REF_KEY; | |
774 | else | |
775 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
776 | } | |
777 | return type; | |
31840ae1 | 778 | } |
56bec294 | 779 | |
2c47e605 YZ |
780 | static int find_next_key(struct btrfs_path *path, int level, |
781 | struct btrfs_key *key) | |
56bec294 | 782 | |
02217ed2 | 783 | { |
2c47e605 | 784 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
785 | if (!path->nodes[level]) |
786 | break; | |
5d4f98a2 YZ |
787 | if (path->slots[level] + 1 >= |
788 | btrfs_header_nritems(path->nodes[level])) | |
789 | continue; | |
790 | if (level == 0) | |
791 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
792 | path->slots[level] + 1); | |
793 | else | |
794 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
795 | path->slots[level] + 1); | |
796 | return 0; | |
797 | } | |
798 | return 1; | |
799 | } | |
037e6390 | 800 | |
5d4f98a2 YZ |
801 | /* |
802 | * look for inline back ref. if back ref is found, *ref_ret is set | |
803 | * to the address of inline back ref, and 0 is returned. | |
804 | * | |
805 | * if back ref isn't found, *ref_ret is set to the address where it | |
806 | * should be inserted, and -ENOENT is returned. | |
807 | * | |
808 | * if insert is true and there are too many inline back refs, the path | |
809 | * points to the extent item, and -EAGAIN is returned. | |
810 | * | |
811 | * NOTE: inline back refs are ordered in the same way that back ref | |
812 | * items in the tree are ordered. | |
813 | */ | |
814 | static noinline_for_stack | |
815 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
816 | struct btrfs_path *path, |
817 | struct btrfs_extent_inline_ref **ref_ret, | |
818 | u64 bytenr, u64 num_bytes, | |
819 | u64 parent, u64 root_objectid, | |
820 | u64 owner, u64 offset, int insert) | |
821 | { | |
867cc1fb | 822 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 823 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
824 | struct btrfs_key key; |
825 | struct extent_buffer *leaf; | |
826 | struct btrfs_extent_item *ei; | |
827 | struct btrfs_extent_inline_ref *iref; | |
828 | u64 flags; | |
829 | u64 item_size; | |
830 | unsigned long ptr; | |
831 | unsigned long end; | |
832 | int extra_size; | |
833 | int type; | |
834 | int want; | |
835 | int ret; | |
836 | int err = 0; | |
0b246afa | 837 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 838 | int needed; |
26b8003f | 839 | |
db94535d | 840 | key.objectid = bytenr; |
31840ae1 | 841 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 842 | key.offset = num_bytes; |
31840ae1 | 843 | |
5d4f98a2 YZ |
844 | want = extent_ref_type(parent, owner); |
845 | if (insert) { | |
846 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 847 | path->keep_locks = 1; |
5d4f98a2 YZ |
848 | } else |
849 | extra_size = -1; | |
3173a18f JB |
850 | |
851 | /* | |
16d1c062 NB |
852 | * Owner is our level, so we can just add one to get the level for the |
853 | * block we are interested in. | |
3173a18f JB |
854 | */ |
855 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
856 | key.type = BTRFS_METADATA_ITEM_KEY; | |
857 | key.offset = owner; | |
858 | } | |
859 | ||
860 | again: | |
5d4f98a2 | 861 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 862 | if (ret < 0) { |
5d4f98a2 YZ |
863 | err = ret; |
864 | goto out; | |
865 | } | |
3173a18f JB |
866 | |
867 | /* | |
868 | * We may be a newly converted file system which still has the old fat | |
869 | * extent entries for metadata, so try and see if we have one of those. | |
870 | */ | |
871 | if (ret > 0 && skinny_metadata) { | |
872 | skinny_metadata = false; | |
873 | if (path->slots[0]) { | |
874 | path->slots[0]--; | |
875 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
876 | path->slots[0]); | |
877 | if (key.objectid == bytenr && | |
878 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
879 | key.offset == num_bytes) | |
880 | ret = 0; | |
881 | } | |
882 | if (ret) { | |
9ce49a0b | 883 | key.objectid = bytenr; |
3173a18f JB |
884 | key.type = BTRFS_EXTENT_ITEM_KEY; |
885 | key.offset = num_bytes; | |
886 | btrfs_release_path(path); | |
887 | goto again; | |
888 | } | |
889 | } | |
890 | ||
79787eaa JM |
891 | if (ret && !insert) { |
892 | err = -ENOENT; | |
893 | goto out; | |
fae7f21c | 894 | } else if (WARN_ON(ret)) { |
492104c8 | 895 | err = -EIO; |
492104c8 | 896 | goto out; |
79787eaa | 897 | } |
5d4f98a2 YZ |
898 | |
899 | leaf = path->nodes[0]; | |
900 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 901 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
902 | err = -EINVAL; |
903 | btrfs_print_v0_err(fs_info); | |
904 | btrfs_abort_transaction(trans, err); | |
905 | goto out; | |
906 | } | |
5d4f98a2 | 907 | |
5d4f98a2 YZ |
908 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
909 | flags = btrfs_extent_flags(leaf, ei); | |
910 | ||
911 | ptr = (unsigned long)(ei + 1); | |
912 | end = (unsigned long)ei + item_size; | |
913 | ||
3173a18f | 914 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
915 | ptr += sizeof(struct btrfs_tree_block_info); |
916 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
917 | } |
918 | ||
3de28d57 LB |
919 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
920 | needed = BTRFS_REF_TYPE_DATA; | |
921 | else | |
922 | needed = BTRFS_REF_TYPE_BLOCK; | |
923 | ||
5d4f98a2 YZ |
924 | err = -ENOENT; |
925 | while (1) { | |
926 | if (ptr >= end) { | |
927 | WARN_ON(ptr > end); | |
928 | break; | |
929 | } | |
930 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
931 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
932 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 933 | err = -EUCLEAN; |
3de28d57 LB |
934 | goto out; |
935 | } | |
936 | ||
5d4f98a2 YZ |
937 | if (want < type) |
938 | break; | |
939 | if (want > type) { | |
940 | ptr += btrfs_extent_inline_ref_size(type); | |
941 | continue; | |
942 | } | |
943 | ||
944 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
945 | struct btrfs_extent_data_ref *dref; | |
946 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
947 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
948 | owner, offset)) { | |
949 | err = 0; | |
950 | break; | |
951 | } | |
952 | if (hash_extent_data_ref_item(leaf, dref) < | |
953 | hash_extent_data_ref(root_objectid, owner, offset)) | |
954 | break; | |
955 | } else { | |
956 | u64 ref_offset; | |
957 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
958 | if (parent > 0) { | |
959 | if (parent == ref_offset) { | |
960 | err = 0; | |
961 | break; | |
962 | } | |
963 | if (ref_offset < parent) | |
964 | break; | |
965 | } else { | |
966 | if (root_objectid == ref_offset) { | |
967 | err = 0; | |
968 | break; | |
969 | } | |
970 | if (ref_offset < root_objectid) | |
971 | break; | |
972 | } | |
973 | } | |
974 | ptr += btrfs_extent_inline_ref_size(type); | |
975 | } | |
976 | if (err == -ENOENT && insert) { | |
977 | if (item_size + extra_size >= | |
978 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
979 | err = -EAGAIN; | |
980 | goto out; | |
981 | } | |
982 | /* | |
983 | * To add new inline back ref, we have to make sure | |
984 | * there is no corresponding back ref item. | |
985 | * For simplicity, we just do not add new inline back | |
986 | * ref if there is any kind of item for this block | |
987 | */ | |
2c47e605 YZ |
988 | if (find_next_key(path, 0, &key) == 0 && |
989 | key.objectid == bytenr && | |
85d4198e | 990 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
991 | err = -EAGAIN; |
992 | goto out; | |
993 | } | |
994 | } | |
995 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
996 | out: | |
85d4198e | 997 | if (insert) { |
5d4f98a2 YZ |
998 | path->keep_locks = 0; |
999 | btrfs_unlock_up_safe(path, 1); | |
1000 | } | |
1001 | return err; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * helper to add new inline back ref | |
1006 | */ | |
1007 | static noinline_for_stack | |
87bde3cd | 1008 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1009 | struct btrfs_path *path, |
1010 | struct btrfs_extent_inline_ref *iref, | |
1011 | u64 parent, u64 root_objectid, | |
1012 | u64 owner, u64 offset, int refs_to_add, | |
1013 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1014 | { |
1015 | struct extent_buffer *leaf; | |
1016 | struct btrfs_extent_item *ei; | |
1017 | unsigned long ptr; | |
1018 | unsigned long end; | |
1019 | unsigned long item_offset; | |
1020 | u64 refs; | |
1021 | int size; | |
1022 | int type; | |
5d4f98a2 YZ |
1023 | |
1024 | leaf = path->nodes[0]; | |
1025 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1026 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1027 | ||
1028 | type = extent_ref_type(parent, owner); | |
1029 | size = btrfs_extent_inline_ref_size(type); | |
1030 | ||
c71dd880 | 1031 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1032 | |
1033 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1034 | refs = btrfs_extent_refs(leaf, ei); | |
1035 | refs += refs_to_add; | |
1036 | btrfs_set_extent_refs(leaf, ei, refs); | |
1037 | if (extent_op) | |
1038 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1039 | ||
1040 | ptr = (unsigned long)ei + item_offset; | |
1041 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1042 | if (ptr < end - size) | |
1043 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1044 | end - size - ptr); | |
1045 | ||
1046 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1047 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1048 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1049 | struct btrfs_extent_data_ref *dref; | |
1050 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1051 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1052 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1053 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1054 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1055 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1056 | struct btrfs_shared_data_ref *sref; | |
1057 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1058 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1059 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1060 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1061 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1062 | } else { | |
1063 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1064 | } | |
1065 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1066 | } |
1067 | ||
1068 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1069 | struct btrfs_path *path, |
1070 | struct btrfs_extent_inline_ref **ref_ret, | |
1071 | u64 bytenr, u64 num_bytes, u64 parent, | |
1072 | u64 root_objectid, u64 owner, u64 offset) | |
1073 | { | |
1074 | int ret; | |
1075 | ||
867cc1fb NB |
1076 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1077 | num_bytes, parent, root_objectid, | |
1078 | owner, offset, 0); | |
5d4f98a2 | 1079 | if (ret != -ENOENT) |
54aa1f4d | 1080 | return ret; |
5d4f98a2 | 1081 | |
b3b4aa74 | 1082 | btrfs_release_path(path); |
5d4f98a2 YZ |
1083 | *ref_ret = NULL; |
1084 | ||
1085 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1086 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1087 | root_objectid); | |
5d4f98a2 | 1088 | } else { |
bd1d53ef NB |
1089 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1090 | root_objectid, owner, offset); | |
b9473439 | 1091 | } |
5d4f98a2 YZ |
1092 | return ret; |
1093 | } | |
31840ae1 | 1094 | |
5d4f98a2 YZ |
1095 | /* |
1096 | * helper to update/remove inline back ref | |
1097 | */ | |
1098 | static noinline_for_stack | |
61a18f1c | 1099 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1100 | struct btrfs_extent_inline_ref *iref, |
1101 | int refs_to_mod, | |
fcebe456 JB |
1102 | struct btrfs_delayed_extent_op *extent_op, |
1103 | int *last_ref) | |
5d4f98a2 | 1104 | { |
61a18f1c | 1105 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1106 | struct btrfs_extent_item *ei; |
1107 | struct btrfs_extent_data_ref *dref = NULL; | |
1108 | struct btrfs_shared_data_ref *sref = NULL; | |
1109 | unsigned long ptr; | |
1110 | unsigned long end; | |
1111 | u32 item_size; | |
1112 | int size; | |
1113 | int type; | |
5d4f98a2 YZ |
1114 | u64 refs; |
1115 | ||
5d4f98a2 YZ |
1116 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1117 | refs = btrfs_extent_refs(leaf, ei); | |
1118 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1119 | refs += refs_to_mod; | |
1120 | btrfs_set_extent_refs(leaf, ei, refs); | |
1121 | if (extent_op) | |
1122 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1123 | ||
3de28d57 LB |
1124 | /* |
1125 | * If type is invalid, we should have bailed out after | |
1126 | * lookup_inline_extent_backref(). | |
1127 | */ | |
1128 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1129 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1130 | |
1131 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1132 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1133 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1134 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1135 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1136 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1137 | } else { | |
1138 | refs = 1; | |
1139 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1140 | } |
31840ae1 | 1141 | |
5d4f98a2 YZ |
1142 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1143 | refs += refs_to_mod; | |
1144 | ||
1145 | if (refs > 0) { | |
1146 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1147 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1148 | else | |
1149 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1150 | } else { | |
fcebe456 | 1151 | *last_ref = 1; |
5d4f98a2 YZ |
1152 | size = btrfs_extent_inline_ref_size(type); |
1153 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1154 | ptr = (unsigned long)iref; | |
1155 | end = (unsigned long)ei + item_size; | |
1156 | if (ptr + size < end) | |
1157 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1158 | end - ptr - size); | |
1159 | item_size -= size; | |
78ac4f9e | 1160 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1161 | } |
1162 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1163 | } |
1164 | ||
1165 | static noinline_for_stack | |
1166 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1167 | struct btrfs_path *path, |
1168 | u64 bytenr, u64 num_bytes, u64 parent, | |
1169 | u64 root_objectid, u64 owner, | |
1170 | u64 offset, int refs_to_add, | |
1171 | struct btrfs_delayed_extent_op *extent_op) | |
1172 | { | |
1173 | struct btrfs_extent_inline_ref *iref; | |
1174 | int ret; | |
1175 | ||
867cc1fb NB |
1176 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1177 | num_bytes, parent, root_objectid, | |
1178 | owner, offset, 1); | |
5d4f98a2 | 1179 | if (ret == 0) { |
07cce5cf QW |
1180 | /* |
1181 | * We're adding refs to a tree block we already own, this | |
1182 | * should not happen at all. | |
1183 | */ | |
1184 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1185 | btrfs_crit(trans->fs_info, | |
1186 | "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu", | |
1187 | bytenr, num_bytes, root_objectid); | |
1188 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
1189 | WARN_ON(1); | |
1190 | btrfs_crit(trans->fs_info, | |
1191 | "path->slots[0]=%d path->nodes[0]:", path->slots[0]); | |
1192 | btrfs_print_leaf(path->nodes[0]); | |
1193 | } | |
1194 | return -EUCLEAN; | |
1195 | } | |
61a18f1c NB |
1196 | update_inline_extent_backref(path, iref, refs_to_add, |
1197 | extent_op, NULL); | |
5d4f98a2 | 1198 | } else if (ret == -ENOENT) { |
a639cdeb | 1199 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1200 | root_objectid, owner, offset, |
1201 | refs_to_add, extent_op); | |
1202 | ret = 0; | |
771ed689 | 1203 | } |
5d4f98a2 YZ |
1204 | return ret; |
1205 | } | |
31840ae1 | 1206 | |
5d4f98a2 | 1207 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1208 | struct btrfs_path *path, |
1209 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1210 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1211 | { |
143bede5 | 1212 | int ret = 0; |
b9473439 | 1213 | |
5d4f98a2 YZ |
1214 | BUG_ON(!is_data && refs_to_drop != 1); |
1215 | if (iref) { | |
61a18f1c NB |
1216 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1217 | last_ref); | |
5d4f98a2 | 1218 | } else if (is_data) { |
e9f6290d | 1219 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1220 | last_ref); |
5d4f98a2 | 1221 | } else { |
fcebe456 | 1222 | *last_ref = 1; |
87cc7a8a | 1223 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1224 | } |
1225 | return ret; | |
1226 | } | |
1227 | ||
d04c6b88 JM |
1228 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1229 | u64 *discarded_bytes) | |
5d4f98a2 | 1230 | { |
86557861 JM |
1231 | int j, ret = 0; |
1232 | u64 bytes_left, end; | |
4d89d377 | 1233 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1234 | |
4d89d377 JM |
1235 | if (WARN_ON(start != aligned_start)) { |
1236 | len -= aligned_start - start; | |
1237 | len = round_down(len, 1 << 9); | |
1238 | start = aligned_start; | |
1239 | } | |
d04c6b88 | 1240 | |
4d89d377 | 1241 | *discarded_bytes = 0; |
86557861 JM |
1242 | |
1243 | if (!len) | |
1244 | return 0; | |
1245 | ||
1246 | end = start + len; | |
1247 | bytes_left = len; | |
1248 | ||
1249 | /* Skip any superblocks on this device. */ | |
1250 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1251 | u64 sb_start = btrfs_sb_offset(j); | |
1252 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1253 | u64 size = sb_start - start; | |
1254 | ||
1255 | if (!in_range(sb_start, start, bytes_left) && | |
1256 | !in_range(sb_end, start, bytes_left) && | |
1257 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1258 | continue; | |
1259 | ||
1260 | /* | |
1261 | * Superblock spans beginning of range. Adjust start and | |
1262 | * try again. | |
1263 | */ | |
1264 | if (sb_start <= start) { | |
1265 | start += sb_end - start; | |
1266 | if (start > end) { | |
1267 | bytes_left = 0; | |
1268 | break; | |
1269 | } | |
1270 | bytes_left = end - start; | |
1271 | continue; | |
1272 | } | |
1273 | ||
1274 | if (size) { | |
1275 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1276 | GFP_NOFS, 0); | |
1277 | if (!ret) | |
1278 | *discarded_bytes += size; | |
1279 | else if (ret != -EOPNOTSUPP) | |
1280 | return ret; | |
1281 | } | |
1282 | ||
1283 | start = sb_end; | |
1284 | if (start > end) { | |
1285 | bytes_left = 0; | |
1286 | break; | |
1287 | } | |
1288 | bytes_left = end - start; | |
1289 | } | |
1290 | ||
1291 | if (bytes_left) { | |
1292 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1293 | GFP_NOFS, 0); |
1294 | if (!ret) | |
86557861 | 1295 | *discarded_bytes += bytes_left; |
4d89d377 | 1296 | } |
d04c6b88 | 1297 | return ret; |
5d4f98a2 | 1298 | } |
5d4f98a2 | 1299 | |
2ff7e61e | 1300 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1301 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1302 | { |
6b7faadd | 1303 | int ret = 0; |
5378e607 | 1304 | u64 discarded_bytes = 0; |
6b7faadd QW |
1305 | u64 end = bytenr + num_bytes; |
1306 | u64 cur = bytenr; | |
a1d3c478 | 1307 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1308 | |
e244a0ae | 1309 | |
2999241d FM |
1310 | /* |
1311 | * Avoid races with device replace and make sure our bbio has devices | |
1312 | * associated to its stripes that don't go away while we are discarding. | |
1313 | */ | |
0b246afa | 1314 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd QW |
1315 | while (cur < end) { |
1316 | struct btrfs_bio_stripe *stripe; | |
5d4f98a2 YZ |
1317 | int i; |
1318 | ||
6b7faadd QW |
1319 | num_bytes = end - cur; |
1320 | /* Tell the block device(s) that the sectors can be discarded */ | |
1321 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
1322 | &num_bytes, &bbio, 0); | |
1323 | /* | |
1324 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1325 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1326 | * thus we can't continue anyway. | |
1327 | */ | |
1328 | if (ret < 0) | |
1329 | goto out; | |
5d4f98a2 | 1330 | |
6b7faadd | 1331 | stripe = bbio->stripes; |
a1d3c478 | 1332 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1333 | u64 bytes; |
38b5f68e AJ |
1334 | struct request_queue *req_q; |
1335 | ||
627e0873 FM |
1336 | if (!stripe->dev->bdev) { |
1337 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1338 | continue; | |
1339 | } | |
38b5f68e AJ |
1340 | req_q = bdev_get_queue(stripe->dev->bdev); |
1341 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1342 | continue; |
1343 | ||
5378e607 LD |
1344 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1345 | stripe->physical, | |
d04c6b88 JM |
1346 | stripe->length, |
1347 | &bytes); | |
6b7faadd | 1348 | if (!ret) { |
d04c6b88 | 1349 | discarded_bytes += bytes; |
6b7faadd QW |
1350 | } else if (ret != -EOPNOTSUPP) { |
1351 | /* | |
1352 | * Logic errors or -ENOMEM, or -EIO, but | |
1353 | * unlikely to happen. | |
1354 | * | |
1355 | * And since there are two loops, explicitly | |
1356 | * go to out to avoid confusion. | |
1357 | */ | |
1358 | btrfs_put_bbio(bbio); | |
1359 | goto out; | |
1360 | } | |
d5e2003c JB |
1361 | |
1362 | /* | |
1363 | * Just in case we get back EOPNOTSUPP for some reason, | |
1364 | * just ignore the return value so we don't screw up | |
1365 | * people calling discard_extent. | |
1366 | */ | |
1367 | ret = 0; | |
5d4f98a2 | 1368 | } |
6e9606d2 | 1369 | btrfs_put_bbio(bbio); |
6b7faadd | 1370 | cur += num_bytes; |
5d4f98a2 | 1371 | } |
6b7faadd | 1372 | out: |
0b246afa | 1373 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1374 | |
1375 | if (actual_bytes) | |
1376 | *actual_bytes = discarded_bytes; | |
1377 | ||
5d4f98a2 | 1378 | |
53b381b3 DW |
1379 | if (ret == -EOPNOTSUPP) |
1380 | ret = 0; | |
5d4f98a2 | 1381 | return ret; |
5d4f98a2 YZ |
1382 | } |
1383 | ||
79787eaa | 1384 | /* Can return -ENOMEM */ |
5d4f98a2 | 1385 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1386 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1387 | { |
82fa113f | 1388 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1389 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1390 | int ret; |
66d7e7f0 | 1391 | |
82fa113f QW |
1392 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1393 | generic_ref->action); | |
1394 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1395 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1396 | |
82fa113f QW |
1397 | if (generic_ref->type == BTRFS_REF_METADATA) |
1398 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1399 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1400 | else |
1401 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1402 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1403 | |
82fa113f | 1404 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1405 | |
ddf30cf0 | 1406 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1407 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 1408 | |
5d4f98a2 YZ |
1409 | return ret; |
1410 | } | |
1411 | ||
bd3c685e NB |
1412 | /* |
1413 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1414 | * | |
07cce5cf QW |
1415 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1416 | * how it works. | |
1417 | * | |
bd3c685e NB |
1418 | * @trans: Handle of transaction |
1419 | * | |
1420 | * @node: The delayed ref node used to get the bytenr/length for | |
1421 | * extent whose references are incremented. | |
1422 | * | |
1423 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1424 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1425 | * bytenr of the parent block. Since new extents are always | |
1426 | * created with indirect references, this will only be the case | |
1427 | * when relocating a shared extent. In that case, root_objectid | |
1428 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1429 | * be 0 | |
1430 | * | |
1431 | * @root_objectid: The id of the root where this modification has originated, | |
1432 | * this can be either one of the well-known metadata trees or | |
1433 | * the subvolume id which references this extent. | |
1434 | * | |
1435 | * @owner: For data extents it is the inode number of the owning file. | |
1436 | * For metadata extents this parameter holds the level in the | |
1437 | * tree of the extent. | |
1438 | * | |
1439 | * @offset: For metadata extents the offset is ignored and is currently | |
1440 | * always passed as 0. For data extents it is the fileoffset | |
1441 | * this extent belongs to. | |
1442 | * | |
1443 | * @refs_to_add Number of references to add | |
1444 | * | |
1445 | * @extent_op Pointer to a structure, holding information necessary when | |
1446 | * updating a tree block's flags | |
1447 | * | |
1448 | */ | |
5d4f98a2 | 1449 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1450 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1451 | u64 parent, u64 root_objectid, |
1452 | u64 owner, u64 offset, int refs_to_add, | |
1453 | struct btrfs_delayed_extent_op *extent_op) | |
1454 | { | |
1455 | struct btrfs_path *path; | |
1456 | struct extent_buffer *leaf; | |
1457 | struct btrfs_extent_item *item; | |
fcebe456 | 1458 | struct btrfs_key key; |
c682f9b3 QW |
1459 | u64 bytenr = node->bytenr; |
1460 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1461 | u64 refs; |
1462 | int ret; | |
5d4f98a2 YZ |
1463 | |
1464 | path = btrfs_alloc_path(); | |
1465 | if (!path) | |
1466 | return -ENOMEM; | |
1467 | ||
5d4f98a2 | 1468 | /* this will setup the path even if it fails to insert the back ref */ |
a639cdeb NB |
1469 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1470 | parent, root_objectid, owner, | |
1471 | offset, refs_to_add, extent_op); | |
0ed4792a | 1472 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1473 | goto out; |
fcebe456 JB |
1474 | |
1475 | /* | |
1476 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1477 | * inline extent ref, so just update the reference count and add a | |
1478 | * normal backref. | |
1479 | */ | |
5d4f98a2 | 1480 | leaf = path->nodes[0]; |
fcebe456 | 1481 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1482 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1483 | refs = btrfs_extent_refs(leaf, item); | |
1484 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1485 | if (extent_op) | |
1486 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1487 | |
5d4f98a2 | 1488 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1489 | btrfs_release_path(path); |
56bec294 | 1490 | |
56bec294 | 1491 | /* now insert the actual backref */ |
65cd6d9e NB |
1492 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
1493 | BUG_ON(refs_to_add != 1); | |
1494 | ret = insert_tree_block_ref(trans, path, bytenr, parent, | |
1495 | root_objectid); | |
1496 | } else { | |
1497 | ret = insert_extent_data_ref(trans, path, bytenr, parent, | |
1498 | root_objectid, owner, offset, | |
1499 | refs_to_add); | |
1500 | } | |
79787eaa | 1501 | if (ret) |
66642832 | 1502 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1503 | out: |
56bec294 | 1504 | btrfs_free_path(path); |
30d133fc | 1505 | return ret; |
56bec294 CM |
1506 | } |
1507 | ||
5d4f98a2 | 1508 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1509 | struct btrfs_delayed_ref_node *node, |
1510 | struct btrfs_delayed_extent_op *extent_op, | |
1511 | int insert_reserved) | |
56bec294 | 1512 | { |
5d4f98a2 YZ |
1513 | int ret = 0; |
1514 | struct btrfs_delayed_data_ref *ref; | |
1515 | struct btrfs_key ins; | |
1516 | u64 parent = 0; | |
1517 | u64 ref_root = 0; | |
1518 | u64 flags = 0; | |
1519 | ||
1520 | ins.objectid = node->bytenr; | |
1521 | ins.offset = node->num_bytes; | |
1522 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1523 | ||
1524 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1525 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1526 | |
5d4f98a2 YZ |
1527 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1528 | parent = ref->parent; | |
fcebe456 | 1529 | ref_root = ref->root; |
5d4f98a2 YZ |
1530 | |
1531 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1532 | if (extent_op) |
5d4f98a2 | 1533 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1534 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1535 | flags, ref->objectid, | |
1536 | ref->offset, &ins, | |
1537 | node->ref_mod); | |
5d4f98a2 | 1538 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1539 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1540 | ref->objectid, ref->offset, | |
1541 | node->ref_mod, extent_op); | |
5d4f98a2 | 1542 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1543 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1544 | ref_root, ref->objectid, |
1545 | ref->offset, node->ref_mod, | |
c682f9b3 | 1546 | extent_op); |
5d4f98a2 YZ |
1547 | } else { |
1548 | BUG(); | |
1549 | } | |
1550 | return ret; | |
1551 | } | |
1552 | ||
1553 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1554 | struct extent_buffer *leaf, | |
1555 | struct btrfs_extent_item *ei) | |
1556 | { | |
1557 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1558 | if (extent_op->update_flags) { | |
1559 | flags |= extent_op->flags_to_set; | |
1560 | btrfs_set_extent_flags(leaf, ei, flags); | |
1561 | } | |
1562 | ||
1563 | if (extent_op->update_key) { | |
1564 | struct btrfs_tree_block_info *bi; | |
1565 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1566 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1567 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1568 | } | |
1569 | } | |
1570 | ||
1571 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1572 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1573 | struct btrfs_delayed_extent_op *extent_op) |
1574 | { | |
20b9a2d6 | 1575 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1576 | struct btrfs_key key; |
1577 | struct btrfs_path *path; | |
1578 | struct btrfs_extent_item *ei; | |
1579 | struct extent_buffer *leaf; | |
1580 | u32 item_size; | |
56bec294 | 1581 | int ret; |
5d4f98a2 | 1582 | int err = 0; |
b1c79e09 | 1583 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1584 | |
bf31f87f | 1585 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1586 | return 0; |
1587 | ||
0b246afa | 1588 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1589 | metadata = 0; |
1590 | ||
5d4f98a2 YZ |
1591 | path = btrfs_alloc_path(); |
1592 | if (!path) | |
1593 | return -ENOMEM; | |
1594 | ||
d278850e | 1595 | key.objectid = head->bytenr; |
5d4f98a2 | 1596 | |
3173a18f | 1597 | if (metadata) { |
3173a18f | 1598 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1599 | key.offset = extent_op->level; |
3173a18f JB |
1600 | } else { |
1601 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1602 | key.offset = head->num_bytes; |
3173a18f JB |
1603 | } |
1604 | ||
1605 | again: | |
0b246afa | 1606 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1607 | if (ret < 0) { |
1608 | err = ret; | |
1609 | goto out; | |
1610 | } | |
1611 | if (ret > 0) { | |
3173a18f | 1612 | if (metadata) { |
55994887 FDBM |
1613 | if (path->slots[0] > 0) { |
1614 | path->slots[0]--; | |
1615 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1616 | path->slots[0]); | |
d278850e | 1617 | if (key.objectid == head->bytenr && |
55994887 | 1618 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1619 | key.offset == head->num_bytes) |
55994887 FDBM |
1620 | ret = 0; |
1621 | } | |
1622 | if (ret > 0) { | |
1623 | btrfs_release_path(path); | |
1624 | metadata = 0; | |
3173a18f | 1625 | |
d278850e JB |
1626 | key.objectid = head->bytenr; |
1627 | key.offset = head->num_bytes; | |
55994887 FDBM |
1628 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1629 | goto again; | |
1630 | } | |
1631 | } else { | |
1632 | err = -EIO; | |
1633 | goto out; | |
3173a18f | 1634 | } |
5d4f98a2 YZ |
1635 | } |
1636 | ||
1637 | leaf = path->nodes[0]; | |
1638 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1639 | |
6d8ff4e4 | 1640 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1641 | err = -EINVAL; |
1642 | btrfs_print_v0_err(fs_info); | |
1643 | btrfs_abort_transaction(trans, err); | |
1644 | goto out; | |
1645 | } | |
1646 | ||
5d4f98a2 YZ |
1647 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1648 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1649 | |
5d4f98a2 YZ |
1650 | btrfs_mark_buffer_dirty(leaf); |
1651 | out: | |
1652 | btrfs_free_path(path); | |
1653 | return err; | |
56bec294 CM |
1654 | } |
1655 | ||
5d4f98a2 | 1656 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1657 | struct btrfs_delayed_ref_node *node, |
1658 | struct btrfs_delayed_extent_op *extent_op, | |
1659 | int insert_reserved) | |
56bec294 CM |
1660 | { |
1661 | int ret = 0; | |
5d4f98a2 | 1662 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1663 | u64 parent = 0; |
1664 | u64 ref_root = 0; | |
56bec294 | 1665 | |
5d4f98a2 | 1666 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1667 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1668 | |
5d4f98a2 YZ |
1669 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1670 | parent = ref->parent; | |
fcebe456 | 1671 | ref_root = ref->root; |
5d4f98a2 | 1672 | |
02794222 | 1673 | if (node->ref_mod != 1) { |
f97806f2 | 1674 | btrfs_err(trans->fs_info, |
02794222 LB |
1675 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1676 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1677 | parent); | |
1678 | return -EIO; | |
1679 | } | |
5d4f98a2 | 1680 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1681 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1682 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1683 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1684 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1685 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1686 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1687 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1688 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1689 | } else { |
1690 | BUG(); | |
1691 | } | |
56bec294 CM |
1692 | return ret; |
1693 | } | |
1694 | ||
1695 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1696 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1697 | struct btrfs_delayed_ref_node *node, |
1698 | struct btrfs_delayed_extent_op *extent_op, | |
1699 | int insert_reserved) | |
56bec294 | 1700 | { |
79787eaa JM |
1701 | int ret = 0; |
1702 | ||
bf31f87f | 1703 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1704 | if (insert_reserved) |
b25c36f8 | 1705 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1706 | return 0; |
857cc2fc | 1707 | } |
79787eaa | 1708 | |
5d4f98a2 YZ |
1709 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1710 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1711 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1712 | insert_reserved); |
1713 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1714 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1715 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1716 | insert_reserved); |
1717 | else | |
1718 | BUG(); | |
80ee54bf | 1719 | if (ret && insert_reserved) |
b25c36f8 | 1720 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
5d4f98a2 | 1721 | return ret; |
56bec294 CM |
1722 | } |
1723 | ||
c6fc2454 | 1724 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1725 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1726 | { | |
cffc3374 FM |
1727 | struct btrfs_delayed_ref_node *ref; |
1728 | ||
e3d03965 | 1729 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1730 | return NULL; |
d7df2c79 | 1731 | |
cffc3374 FM |
1732 | /* |
1733 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1734 | * This is to prevent a ref count from going down to zero, which deletes | |
1735 | * the extent item from the extent tree, when there still are references | |
1736 | * to add, which would fail because they would not find the extent item. | |
1737 | */ | |
1d57ee94 WX |
1738 | if (!list_empty(&head->ref_add_list)) |
1739 | return list_first_entry(&head->ref_add_list, | |
1740 | struct btrfs_delayed_ref_node, add_list); | |
1741 | ||
e3d03965 | 1742 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1743 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1744 | ASSERT(list_empty(&ref->add_list)); |
1745 | return ref; | |
56bec294 CM |
1746 | } |
1747 | ||
2eadaa22 JB |
1748 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1749 | struct btrfs_delayed_ref_head *head) | |
1750 | { | |
1751 | spin_lock(&delayed_refs->lock); | |
1752 | head->processing = 0; | |
1753 | delayed_refs->num_heads_ready++; | |
1754 | spin_unlock(&delayed_refs->lock); | |
1755 | btrfs_delayed_ref_unlock(head); | |
1756 | } | |
1757 | ||
bedc6617 JB |
1758 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1759 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1760 | { |
1761 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1762 | |
1763 | if (!extent_op) | |
bedc6617 JB |
1764 | return NULL; |
1765 | ||
b00e6250 | 1766 | if (head->must_insert_reserved) { |
bedc6617 | 1767 | head->extent_op = NULL; |
b00e6250 | 1768 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1769 | return NULL; |
b00e6250 | 1770 | } |
bedc6617 JB |
1771 | return extent_op; |
1772 | } | |
1773 | ||
1774 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1775 | struct btrfs_delayed_ref_head *head) | |
1776 | { | |
1777 | struct btrfs_delayed_extent_op *extent_op; | |
1778 | int ret; | |
1779 | ||
1780 | extent_op = cleanup_extent_op(head); | |
1781 | if (!extent_op) | |
1782 | return 0; | |
1783 | head->extent_op = NULL; | |
b00e6250 | 1784 | spin_unlock(&head->lock); |
20b9a2d6 | 1785 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1786 | btrfs_free_delayed_extent_op(extent_op); |
1787 | return ret ? ret : 1; | |
1788 | } | |
1789 | ||
31890da0 JB |
1790 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1791 | struct btrfs_delayed_ref_root *delayed_refs, | |
1792 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1793 | { |
ba2c4d4e | 1794 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
1795 | |
1796 | if (head->total_ref_mod < 0) { | |
1797 | struct btrfs_space_info *space_info; | |
1798 | u64 flags; | |
1799 | ||
1800 | if (head->is_data) | |
1801 | flags = BTRFS_BLOCK_GROUP_DATA; | |
1802 | else if (head->is_system) | |
1803 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
1804 | else | |
1805 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 1806 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
1807 | ASSERT(space_info); |
1808 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1809 | -head->num_bytes, | |
1810 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1811 | ||
ba2c4d4e JB |
1812 | /* |
1813 | * We had csum deletions accounted for in our delayed refs rsv, | |
1814 | * we need to drop the csum leaves for this update from our | |
1815 | * delayed_refs_rsv. | |
1816 | */ | |
07c47775 JB |
1817 | if (head->is_data) { |
1818 | spin_lock(&delayed_refs->lock); | |
1819 | delayed_refs->pending_csums -= head->num_bytes; | |
1820 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
1821 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
1822 | head->num_bytes); | |
07c47775 JB |
1823 | } |
1824 | } | |
1825 | ||
ba2c4d4e | 1826 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1827 | } |
1828 | ||
194ab0bc | 1829 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1830 | struct btrfs_delayed_ref_head *head) |
1831 | { | |
f9871edd NB |
1832 | |
1833 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1834 | struct btrfs_delayed_ref_root *delayed_refs; |
1835 | int ret; | |
1836 | ||
1837 | delayed_refs = &trans->transaction->delayed_refs; | |
1838 | ||
bedc6617 | 1839 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1840 | if (ret < 0) { |
1841 | unselect_delayed_ref_head(delayed_refs, head); | |
1842 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1843 | return ret; | |
1844 | } else if (ret) { | |
1845 | return ret; | |
1846 | } | |
1847 | ||
1848 | /* | |
1849 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1850 | * and then re-check to make sure nobody got added. | |
1851 | */ | |
1852 | spin_unlock(&head->lock); | |
1853 | spin_lock(&delayed_refs->lock); | |
1854 | spin_lock(&head->lock); | |
e3d03965 | 1855 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1856 | spin_unlock(&head->lock); |
1857 | spin_unlock(&delayed_refs->lock); | |
1858 | return 1; | |
1859 | } | |
d7baffda | 1860 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1861 | spin_unlock(&head->lock); |
1e7a1421 | 1862 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1863 | |
c1103f7a | 1864 | if (head->must_insert_reserved) { |
b25c36f8 | 1865 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1866 | if (head->is_data) { |
40e046ac FM |
1867 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1868 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1869 | } |
1870 | } | |
1871 | ||
31890da0 | 1872 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1873 | |
1874 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1875 | btrfs_delayed_ref_unlock(head); |
d278850e | 1876 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1877 | return 0; |
1878 | } | |
1879 | ||
b1cdbcb5 NB |
1880 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1881 | struct btrfs_trans_handle *trans) | |
1882 | { | |
1883 | struct btrfs_delayed_ref_root *delayed_refs = | |
1884 | &trans->transaction->delayed_refs; | |
1885 | struct btrfs_delayed_ref_head *head = NULL; | |
1886 | int ret; | |
1887 | ||
1888 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1889 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1890 | if (!head) { |
1891 | spin_unlock(&delayed_refs->lock); | |
1892 | return head; | |
1893 | } | |
1894 | ||
1895 | /* | |
1896 | * Grab the lock that says we are going to process all the refs for | |
1897 | * this head | |
1898 | */ | |
9e920a6f | 1899 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1900 | spin_unlock(&delayed_refs->lock); |
1901 | ||
1902 | /* | |
1903 | * We may have dropped the spin lock to get the head mutex lock, and | |
1904 | * that might have given someone else time to free the head. If that's | |
1905 | * true, it has been removed from our list and we can move on. | |
1906 | */ | |
1907 | if (ret == -EAGAIN) | |
1908 | head = ERR_PTR(-EAGAIN); | |
1909 | ||
1910 | return head; | |
1911 | } | |
1912 | ||
e7261386 NB |
1913 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1914 | struct btrfs_delayed_ref_head *locked_ref, | |
1915 | unsigned long *run_refs) | |
1916 | { | |
1917 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1918 | struct btrfs_delayed_ref_root *delayed_refs; | |
1919 | struct btrfs_delayed_extent_op *extent_op; | |
1920 | struct btrfs_delayed_ref_node *ref; | |
1921 | int must_insert_reserved = 0; | |
1922 | int ret; | |
1923 | ||
1924 | delayed_refs = &trans->transaction->delayed_refs; | |
1925 | ||
0110a4c4 NB |
1926 | lockdep_assert_held(&locked_ref->mutex); |
1927 | lockdep_assert_held(&locked_ref->lock); | |
1928 | ||
e7261386 NB |
1929 | while ((ref = select_delayed_ref(locked_ref))) { |
1930 | if (ref->seq && | |
1931 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1932 | spin_unlock(&locked_ref->lock); | |
1933 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1934 | return -EAGAIN; | |
1935 | } | |
1936 | ||
1937 | (*run_refs)++; | |
1938 | ref->in_tree = 0; | |
1939 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1940 | RB_CLEAR_NODE(&ref->ref_node); | |
1941 | if (!list_empty(&ref->add_list)) | |
1942 | list_del(&ref->add_list); | |
1943 | /* | |
1944 | * When we play the delayed ref, also correct the ref_mod on | |
1945 | * head | |
1946 | */ | |
1947 | switch (ref->action) { | |
1948 | case BTRFS_ADD_DELAYED_REF: | |
1949 | case BTRFS_ADD_DELAYED_EXTENT: | |
1950 | locked_ref->ref_mod -= ref->ref_mod; | |
1951 | break; | |
1952 | case BTRFS_DROP_DELAYED_REF: | |
1953 | locked_ref->ref_mod += ref->ref_mod; | |
1954 | break; | |
1955 | default: | |
1956 | WARN_ON(1); | |
1957 | } | |
1958 | atomic_dec(&delayed_refs->num_entries); | |
1959 | ||
1960 | /* | |
1961 | * Record the must_insert_reserved flag before we drop the | |
1962 | * spin lock. | |
1963 | */ | |
1964 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1965 | locked_ref->must_insert_reserved = 0; | |
1966 | ||
1967 | extent_op = locked_ref->extent_op; | |
1968 | locked_ref->extent_op = NULL; | |
1969 | spin_unlock(&locked_ref->lock); | |
1970 | ||
1971 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1972 | must_insert_reserved); | |
1973 | ||
1974 | btrfs_free_delayed_extent_op(extent_op); | |
1975 | if (ret) { | |
1976 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1977 | btrfs_put_delayed_ref(ref); | |
1978 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1979 | ret); | |
1980 | return ret; | |
1981 | } | |
1982 | ||
1983 | btrfs_put_delayed_ref(ref); | |
1984 | cond_resched(); | |
1985 | ||
1986 | spin_lock(&locked_ref->lock); | |
1987 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1988 | } | |
1989 | ||
1990 | return 0; | |
1991 | } | |
1992 | ||
79787eaa JM |
1993 | /* |
1994 | * Returns 0 on success or if called with an already aborted transaction. | |
1995 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1996 | */ | |
d7df2c79 | 1997 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 1998 | unsigned long nr) |
56bec294 | 1999 | { |
0a1e458a | 2000 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2001 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2002 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2003 | ktime_t start = ktime_get(); |
56bec294 | 2004 | int ret; |
d7df2c79 | 2005 | unsigned long count = 0; |
0a2b2a84 | 2006 | unsigned long actual_count = 0; |
56bec294 CM |
2007 | |
2008 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2009 | do { |
56bec294 | 2010 | if (!locked_ref) { |
b1cdbcb5 | 2011 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2012 | if (IS_ERR_OR_NULL(locked_ref)) { |
2013 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2014 | continue; | |
2015 | } else { | |
2016 | break; | |
2017 | } | |
56bec294 | 2018 | } |
0110a4c4 | 2019 | count++; |
56bec294 | 2020 | } |
2c3cf7d5 FM |
2021 | /* |
2022 | * We need to try and merge add/drops of the same ref since we | |
2023 | * can run into issues with relocate dropping the implicit ref | |
2024 | * and then it being added back again before the drop can | |
2025 | * finish. If we merged anything we need to re-loop so we can | |
2026 | * get a good ref. | |
2027 | * Or we can get node references of the same type that weren't | |
2028 | * merged when created due to bumps in the tree mod seq, and | |
2029 | * we need to merge them to prevent adding an inline extent | |
2030 | * backref before dropping it (triggering a BUG_ON at | |
2031 | * insert_inline_extent_backref()). | |
2032 | */ | |
d7df2c79 | 2033 | spin_lock(&locked_ref->lock); |
be97f133 | 2034 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2035 | |
0110a4c4 NB |
2036 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2037 | &actual_count); | |
2038 | if (ret < 0 && ret != -EAGAIN) { | |
2039 | /* | |
2040 | * Error, btrfs_run_delayed_refs_for_head already | |
2041 | * unlocked everything so just bail out | |
2042 | */ | |
2043 | return ret; | |
2044 | } else if (!ret) { | |
2045 | /* | |
2046 | * Success, perform the usual cleanup of a processed | |
2047 | * head | |
2048 | */ | |
f9871edd | 2049 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2050 | if (ret > 0 ) { |
b00e6250 JB |
2051 | /* We dropped our lock, we need to loop. */ |
2052 | ret = 0; | |
d7df2c79 | 2053 | continue; |
194ab0bc JB |
2054 | } else if (ret) { |
2055 | return ret; | |
5d4f98a2 | 2056 | } |
22cd2e7d | 2057 | } |
1ce7a5ec | 2058 | |
b00e6250 | 2059 | /* |
0110a4c4 NB |
2060 | * Either success case or btrfs_run_delayed_refs_for_head |
2061 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2062 | */ |
b00e6250 | 2063 | |
0110a4c4 | 2064 | locked_ref = NULL; |
c3e69d58 | 2065 | cond_resched(); |
0110a4c4 | 2066 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2067 | |
2068 | /* | |
2069 | * We don't want to include ref heads since we can have empty ref heads | |
2070 | * and those will drastically skew our runtime down since we just do | |
2071 | * accounting, no actual extent tree updates. | |
2072 | */ | |
2073 | if (actual_count > 0) { | |
2074 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2075 | u64 avg; | |
2076 | ||
2077 | /* | |
2078 | * We weigh the current average higher than our current runtime | |
2079 | * to avoid large swings in the average. | |
2080 | */ | |
2081 | spin_lock(&delayed_refs->lock); | |
2082 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2083 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2084 | spin_unlock(&delayed_refs->lock); |
2085 | } | |
d7df2c79 | 2086 | return 0; |
c3e69d58 CM |
2087 | } |
2088 | ||
709c0486 AJ |
2089 | #ifdef SCRAMBLE_DELAYED_REFS |
2090 | /* | |
2091 | * Normally delayed refs get processed in ascending bytenr order. This | |
2092 | * correlates in most cases to the order added. To expose dependencies on this | |
2093 | * order, we start to process the tree in the middle instead of the beginning | |
2094 | */ | |
2095 | static u64 find_middle(struct rb_root *root) | |
2096 | { | |
2097 | struct rb_node *n = root->rb_node; | |
2098 | struct btrfs_delayed_ref_node *entry; | |
2099 | int alt = 1; | |
2100 | u64 middle; | |
2101 | u64 first = 0, last = 0; | |
2102 | ||
2103 | n = rb_first(root); | |
2104 | if (n) { | |
2105 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2106 | first = entry->bytenr; | |
2107 | } | |
2108 | n = rb_last(root); | |
2109 | if (n) { | |
2110 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2111 | last = entry->bytenr; | |
2112 | } | |
2113 | n = root->rb_node; | |
2114 | ||
2115 | while (n) { | |
2116 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2117 | WARN_ON(!entry->in_tree); | |
2118 | ||
2119 | middle = entry->bytenr; | |
2120 | ||
2121 | if (alt) | |
2122 | n = n->rb_left; | |
2123 | else | |
2124 | n = n->rb_right; | |
2125 | ||
2126 | alt = 1 - alt; | |
2127 | } | |
2128 | return middle; | |
2129 | } | |
2130 | #endif | |
2131 | ||
c3e69d58 CM |
2132 | /* |
2133 | * this starts processing the delayed reference count updates and | |
2134 | * extent insertions we have queued up so far. count can be | |
2135 | * 0, which means to process everything in the tree at the start | |
2136 | * of the run (but not newly added entries), or it can be some target | |
2137 | * number you'd like to process. | |
79787eaa JM |
2138 | * |
2139 | * Returns 0 on success or if called with an aborted transaction | |
2140 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2141 | */ |
2142 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2143 | unsigned long count) |
c3e69d58 | 2144 | { |
c79a70b1 | 2145 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2146 | struct rb_node *node; |
2147 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2148 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2149 | int ret; |
2150 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2151 | |
79787eaa | 2152 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2153 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2154 | return 0; |
2155 | ||
0b246afa | 2156 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2157 | return 0; |
2158 | ||
c3e69d58 | 2159 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2160 | if (count == 0) |
d7df2c79 | 2161 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2162 | |
c3e69d58 | 2163 | again: |
709c0486 AJ |
2164 | #ifdef SCRAMBLE_DELAYED_REFS |
2165 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2166 | #endif | |
0a1e458a | 2167 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2168 | if (ret < 0) { |
66642832 | 2169 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2170 | return ret; |
eb099670 | 2171 | } |
c3e69d58 | 2172 | |
56bec294 | 2173 | if (run_all) { |
119e80df | 2174 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2175 | |
d7df2c79 | 2176 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2177 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2178 | if (!node) { |
2179 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2180 | goto out; |
d7df2c79 | 2181 | } |
d278850e JB |
2182 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2183 | href_node); | |
2184 | refcount_inc(&head->refs); | |
2185 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2186 | |
d278850e JB |
2187 | /* Mutex was contended, block until it's released and retry. */ |
2188 | mutex_lock(&head->mutex); | |
2189 | mutex_unlock(&head->mutex); | |
56bec294 | 2190 | |
d278850e | 2191 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2192 | cond_resched(); |
56bec294 | 2193 | goto again; |
5f39d397 | 2194 | } |
54aa1f4d | 2195 | out: |
a28ec197 CM |
2196 | return 0; |
2197 | } | |
2198 | ||
5d4f98a2 | 2199 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
42c9d0b5 | 2200 | struct extent_buffer *eb, u64 flags, |
b1c79e09 | 2201 | int level, int is_data) |
5d4f98a2 YZ |
2202 | { |
2203 | struct btrfs_delayed_extent_op *extent_op; | |
2204 | int ret; | |
2205 | ||
78a6184a | 2206 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2207 | if (!extent_op) |
2208 | return -ENOMEM; | |
2209 | ||
2210 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2211 | extent_op->update_flags = true; |
2212 | extent_op->update_key = false; | |
2213 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2214 | extent_op->level = level; |
5d4f98a2 | 2215 | |
42c9d0b5 | 2216 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2217 | if (ret) |
78a6184a | 2218 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2219 | return ret; |
2220 | } | |
2221 | ||
e4c3b2dc | 2222 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2223 | struct btrfs_path *path, |
2224 | u64 objectid, u64 offset, u64 bytenr) | |
2225 | { | |
2226 | struct btrfs_delayed_ref_head *head; | |
2227 | struct btrfs_delayed_ref_node *ref; | |
2228 | struct btrfs_delayed_data_ref *data_ref; | |
2229 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2230 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2231 | struct rb_node *node; |
5d4f98a2 YZ |
2232 | int ret = 0; |
2233 | ||
998ac6d2 | 2234 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2235 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2236 | if (cur_trans) |
2237 | refcount_inc(&cur_trans->use_count); | |
2238 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2239 | if (!cur_trans) |
2240 | return 0; | |
2241 | ||
2242 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2243 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2244 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2245 | if (!head) { |
2246 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2247 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2248 | return 0; |
2249 | } | |
5d4f98a2 YZ |
2250 | |
2251 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2252 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2253 | spin_unlock(&delayed_refs->lock); |
2254 | ||
b3b4aa74 | 2255 | btrfs_release_path(path); |
5d4f98a2 | 2256 | |
8cc33e5c DS |
2257 | /* |
2258 | * Mutex was contended, block until it's released and let | |
2259 | * caller try again | |
2260 | */ | |
5d4f98a2 YZ |
2261 | mutex_lock(&head->mutex); |
2262 | mutex_unlock(&head->mutex); | |
d278850e | 2263 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2264 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2265 | return -EAGAIN; |
2266 | } | |
d7df2c79 | 2267 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2268 | |
d7df2c79 | 2269 | spin_lock(&head->lock); |
0e0adbcf JB |
2270 | /* |
2271 | * XXX: We should replace this with a proper search function in the | |
2272 | * future. | |
2273 | */ | |
e3d03965 LB |
2274 | for (node = rb_first_cached(&head->ref_tree); node; |
2275 | node = rb_next(node)) { | |
0e0adbcf | 2276 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2277 | /* If it's a shared ref we know a cross reference exists */ |
2278 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2279 | ret = 1; | |
2280 | break; | |
2281 | } | |
5d4f98a2 | 2282 | |
d7df2c79 | 2283 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2284 | |
d7df2c79 JB |
2285 | /* |
2286 | * If our ref doesn't match the one we're currently looking at | |
2287 | * then we have a cross reference. | |
2288 | */ | |
2289 | if (data_ref->root != root->root_key.objectid || | |
2290 | data_ref->objectid != objectid || | |
2291 | data_ref->offset != offset) { | |
2292 | ret = 1; | |
2293 | break; | |
2294 | } | |
5d4f98a2 | 2295 | } |
d7df2c79 | 2296 | spin_unlock(&head->lock); |
5d4f98a2 | 2297 | mutex_unlock(&head->mutex); |
998ac6d2 | 2298 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2299 | return ret; |
2300 | } | |
2301 | ||
e4c3b2dc | 2302 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2303 | struct btrfs_path *path, |
a84d5d42 BB |
2304 | u64 objectid, u64 offset, u64 bytenr, |
2305 | bool strict) | |
be20aa9d | 2306 | { |
0b246afa JM |
2307 | struct btrfs_fs_info *fs_info = root->fs_info; |
2308 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2309 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2310 | struct btrfs_extent_data_ref *ref; |
2311 | struct btrfs_extent_inline_ref *iref; | |
2312 | struct btrfs_extent_item *ei; | |
f321e491 | 2313 | struct btrfs_key key; |
5d4f98a2 | 2314 | u32 item_size; |
3de28d57 | 2315 | int type; |
be20aa9d | 2316 | int ret; |
925baedd | 2317 | |
be20aa9d | 2318 | key.objectid = bytenr; |
31840ae1 | 2319 | key.offset = (u64)-1; |
f321e491 | 2320 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2321 | |
be20aa9d CM |
2322 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2323 | if (ret < 0) | |
2324 | goto out; | |
79787eaa | 2325 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2326 | |
2327 | ret = -ENOENT; | |
2328 | if (path->slots[0] == 0) | |
31840ae1 | 2329 | goto out; |
be20aa9d | 2330 | |
31840ae1 | 2331 | path->slots[0]--; |
f321e491 | 2332 | leaf = path->nodes[0]; |
5d4f98a2 | 2333 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2334 | |
5d4f98a2 | 2335 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2336 | goto out; |
f321e491 | 2337 | |
5d4f98a2 YZ |
2338 | ret = 1; |
2339 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2340 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2341 | |
a6bd9cd1 | 2342 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2343 | if (item_size != sizeof(*ei) + |
2344 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2345 | goto out; | |
be20aa9d | 2346 | |
a84d5d42 BB |
2347 | /* |
2348 | * If extent created before last snapshot => it's shared unless the | |
2349 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2350 | */ | |
2351 | if (!strict && | |
2352 | (btrfs_extent_generation(leaf, ei) <= | |
2353 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2354 | goto out; |
2355 | ||
2356 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2357 | |
a6bd9cd1 | 2358 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2359 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2360 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2361 | goto out; |
2362 | ||
2363 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2364 | if (btrfs_extent_refs(leaf, ei) != | |
2365 | btrfs_extent_data_ref_count(leaf, ref) || | |
2366 | btrfs_extent_data_ref_root(leaf, ref) != | |
2367 | root->root_key.objectid || | |
2368 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2369 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2370 | goto out; | |
2371 | ||
2372 | ret = 0; | |
2373 | out: | |
2374 | return ret; | |
2375 | } | |
2376 | ||
e4c3b2dc | 2377 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
a84d5d42 | 2378 | u64 bytenr, bool strict) |
5d4f98a2 YZ |
2379 | { |
2380 | struct btrfs_path *path; | |
2381 | int ret; | |
5d4f98a2 YZ |
2382 | |
2383 | path = btrfs_alloc_path(); | |
2384 | if (!path) | |
9132c4ff | 2385 | return -ENOMEM; |
5d4f98a2 YZ |
2386 | |
2387 | do { | |
e4c3b2dc | 2388 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2389 | offset, bytenr, strict); |
5d4f98a2 | 2390 | if (ret && ret != -ENOENT) |
f321e491 | 2391 | goto out; |
80ff3856 | 2392 | |
380fd066 MT |
2393 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2394 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2395 | |
be20aa9d | 2396 | out: |
80ff3856 | 2397 | btrfs_free_path(path); |
f0486c68 YZ |
2398 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2399 | WARN_ON(ret > 0); | |
f321e491 | 2400 | return ret; |
be20aa9d | 2401 | } |
c5739bba | 2402 | |
5d4f98a2 | 2403 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2404 | struct btrfs_root *root, |
5d4f98a2 | 2405 | struct extent_buffer *buf, |
e339a6b0 | 2406 | int full_backref, int inc) |
31840ae1 | 2407 | { |
0b246afa | 2408 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2409 | u64 bytenr; |
5d4f98a2 YZ |
2410 | u64 num_bytes; |
2411 | u64 parent; | |
31840ae1 | 2412 | u64 ref_root; |
31840ae1 | 2413 | u32 nritems; |
31840ae1 ZY |
2414 | struct btrfs_key key; |
2415 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2416 | struct btrfs_ref generic_ref = { 0 }; |
2417 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2418 | int i; |
82fa113f | 2419 | int action; |
31840ae1 ZY |
2420 | int level; |
2421 | int ret = 0; | |
fccb84c9 | 2422 | |
0b246afa | 2423 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2424 | return 0; |
fccb84c9 | 2425 | |
31840ae1 | 2426 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2427 | nritems = btrfs_header_nritems(buf); |
2428 | level = btrfs_header_level(buf); | |
2429 | ||
92a7cc42 | 2430 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2431 | return 0; |
31840ae1 | 2432 | |
5d4f98a2 YZ |
2433 | if (full_backref) |
2434 | parent = buf->start; | |
2435 | else | |
2436 | parent = 0; | |
82fa113f QW |
2437 | if (inc) |
2438 | action = BTRFS_ADD_DELAYED_REF; | |
2439 | else | |
2440 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2441 | |
2442 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2443 | if (level == 0) { |
5d4f98a2 | 2444 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2445 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2446 | continue; |
5d4f98a2 | 2447 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2448 | struct btrfs_file_extent_item); |
2449 | if (btrfs_file_extent_type(buf, fi) == | |
2450 | BTRFS_FILE_EXTENT_INLINE) | |
2451 | continue; | |
2452 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2453 | if (bytenr == 0) | |
2454 | continue; | |
5d4f98a2 YZ |
2455 | |
2456 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2457 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2458 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2459 | num_bytes, parent); | |
2460 | generic_ref.real_root = root->root_key.objectid; | |
2461 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2462 | key.offset); | |
2463 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2464 | if (inc) |
82fa113f | 2465 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2466 | else |
ffd4bb2a | 2467 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2468 | if (ret) |
2469 | goto fail; | |
2470 | } else { | |
5d4f98a2 | 2471 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2472 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2473 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2474 | num_bytes, parent); | |
2475 | generic_ref.real_root = root->root_key.objectid; | |
2476 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2477 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2478 | if (inc) |
82fa113f | 2479 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2480 | else |
ffd4bb2a | 2481 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2482 | if (ret) |
2483 | goto fail; | |
2484 | } | |
2485 | } | |
2486 | return 0; | |
2487 | fail: | |
5d4f98a2 YZ |
2488 | return ret; |
2489 | } | |
2490 | ||
2491 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2492 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2493 | { |
e339a6b0 | 2494 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2495 | } |
2496 | ||
2497 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2498 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2499 | { |
e339a6b0 | 2500 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2501 | } |
2502 | ||
2ff7e61e | 2503 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 | 2504 | { |
32da5386 | 2505 | struct btrfs_block_group *block_group; |
d2fb3437 YZ |
2506 | int readonly = 0; |
2507 | ||
0b246afa | 2508 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2509 | if (!block_group || block_group->ro) |
2510 | readonly = 1; | |
2511 | if (block_group) | |
fa9c0d79 | 2512 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2513 | return readonly; |
2514 | } | |
2515 | ||
1b86826d | 2516 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2517 | { |
0b246afa | 2518 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2519 | u64 flags; |
53b381b3 | 2520 | u64 ret; |
9ed74f2d | 2521 | |
b742bb82 YZ |
2522 | if (data) |
2523 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2524 | else if (root == fs_info->chunk_root) |
b742bb82 | 2525 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2526 | else |
b742bb82 | 2527 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2528 | |
878d7b67 | 2529 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2530 | return ret; |
6a63209f | 2531 | } |
9ed74f2d | 2532 | |
2ff7e61e | 2533 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2534 | { |
32da5386 | 2535 | struct btrfs_block_group *cache; |
d2fb3437 | 2536 | u64 bytenr; |
0f9dd46c | 2537 | |
0b246afa JM |
2538 | spin_lock(&fs_info->block_group_cache_lock); |
2539 | bytenr = fs_info->first_logical_byte; | |
2540 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2541 | |
2542 | if (bytenr < (u64)-1) | |
2543 | return bytenr; | |
2544 | ||
0b246afa | 2545 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2546 | if (!cache) |
a061fc8d | 2547 | return 0; |
0f9dd46c | 2548 | |
b3470b5d | 2549 | bytenr = cache->start; |
fa9c0d79 | 2550 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2551 | |
2552 | return bytenr; | |
a061fc8d CM |
2553 | } |
2554 | ||
6690d071 NB |
2555 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2556 | struct btrfs_block_group *cache, | |
f0486c68 | 2557 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2558 | { |
fdf08605 DS |
2559 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2560 | ||
11833d66 YZ |
2561 | spin_lock(&cache->space_info->lock); |
2562 | spin_lock(&cache->lock); | |
2563 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2564 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2565 | num_bytes); | |
11833d66 YZ |
2566 | if (reserved) { |
2567 | cache->reserved -= num_bytes; | |
2568 | cache->space_info->bytes_reserved -= num_bytes; | |
2569 | } | |
2570 | spin_unlock(&cache->lock); | |
2571 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2572 | |
dec59fa3 EL |
2573 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
2574 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
fe119a6e | 2575 | set_extent_dirty(&trans->transaction->pinned_extents, bytenr, |
f0486c68 YZ |
2576 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2577 | return 0; | |
2578 | } | |
68b38550 | 2579 | |
b25c36f8 | 2580 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2581 | u64 bytenr, u64 num_bytes, int reserved) |
2582 | { | |
32da5386 | 2583 | struct btrfs_block_group *cache; |
68b38550 | 2584 | |
b25c36f8 | 2585 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2586 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2587 | |
6690d071 | 2588 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2589 | |
2590 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2591 | return 0; |
2592 | } | |
2593 | ||
f0486c68 | 2594 | /* |
e688b725 CM |
2595 | * this function must be called within transaction |
2596 | */ | |
9fce5704 | 2597 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
e688b725 CM |
2598 | u64 bytenr, u64 num_bytes) |
2599 | { | |
32da5386 | 2600 | struct btrfs_block_group *cache; |
b50c6e25 | 2601 | int ret; |
e688b725 | 2602 | |
f2fb7298 NB |
2603 | btrfs_add_excluded_extent(trans->fs_info, bytenr, num_bytes); |
2604 | ||
9fce5704 | 2605 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
b50c6e25 JB |
2606 | if (!cache) |
2607 | return -EINVAL; | |
e688b725 CM |
2608 | |
2609 | /* | |
2610 | * pull in the free space cache (if any) so that our pin | |
2611 | * removes the free space from the cache. We have load_only set | |
2612 | * to one because the slow code to read in the free extents does check | |
2613 | * the pinned extents. | |
2614 | */ | |
676f1f75 | 2615 | btrfs_cache_block_group(cache, 1); |
e688b725 | 2616 | |
6690d071 | 2617 | pin_down_extent(trans, cache, bytenr, num_bytes, 0); |
e688b725 CM |
2618 | |
2619 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2620 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 2621 | btrfs_put_block_group(cache); |
b50c6e25 | 2622 | return ret; |
e688b725 CM |
2623 | } |
2624 | ||
2ff7e61e JM |
2625 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2626 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2627 | { |
2628 | int ret; | |
32da5386 | 2629 | struct btrfs_block_group *block_group; |
8c2a1a30 JB |
2630 | struct btrfs_caching_control *caching_ctl; |
2631 | ||
0b246afa | 2632 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2633 | if (!block_group) |
2634 | return -EINVAL; | |
2635 | ||
676f1f75 | 2636 | btrfs_cache_block_group(block_group, 0); |
e3cb339f | 2637 | caching_ctl = btrfs_get_caching_control(block_group); |
8c2a1a30 JB |
2638 | |
2639 | if (!caching_ctl) { | |
2640 | /* Logic error */ | |
32da5386 | 2641 | BUG_ON(!btrfs_block_group_done(block_group)); |
8c2a1a30 JB |
2642 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2643 | } else { | |
2644 | mutex_lock(&caching_ctl->mutex); | |
2645 | ||
2646 | if (start >= caching_ctl->progress) { | |
6f410d1b JB |
2647 | ret = btrfs_add_excluded_extent(fs_info, start, |
2648 | num_bytes); | |
8c2a1a30 JB |
2649 | } else if (start + num_bytes <= caching_ctl->progress) { |
2650 | ret = btrfs_remove_free_space(block_group, | |
2651 | start, num_bytes); | |
2652 | } else { | |
2653 | num_bytes = caching_ctl->progress - start; | |
2654 | ret = btrfs_remove_free_space(block_group, | |
2655 | start, num_bytes); | |
2656 | if (ret) | |
2657 | goto out_lock; | |
2658 | ||
2659 | num_bytes = (start + num_bytes) - | |
2660 | caching_ctl->progress; | |
2661 | start = caching_ctl->progress; | |
6f410d1b JB |
2662 | ret = btrfs_add_excluded_extent(fs_info, start, |
2663 | num_bytes); | |
8c2a1a30 JB |
2664 | } |
2665 | out_lock: | |
2666 | mutex_unlock(&caching_ctl->mutex); | |
e3cb339f | 2667 | btrfs_put_caching_control(caching_ctl); |
8c2a1a30 JB |
2668 | } |
2669 | btrfs_put_block_group(block_group); | |
2670 | return ret; | |
2671 | } | |
2672 | ||
bcdc428c | 2673 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2674 | { |
bcdc428c | 2675 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2676 | struct btrfs_file_extent_item *item; |
2677 | struct btrfs_key key; | |
2678 | int found_type; | |
2679 | int i; | |
b89311ef | 2680 | int ret = 0; |
8c2a1a30 | 2681 | |
2ff7e61e | 2682 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2683 | return 0; |
2684 | ||
2685 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2686 | btrfs_item_key_to_cpu(eb, &key, i); | |
2687 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2688 | continue; | |
2689 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2690 | found_type = btrfs_file_extent_type(eb, item); | |
2691 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2692 | continue; | |
2693 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2694 | continue; | |
2695 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2696 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2697 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2698 | if (ret) | |
2699 | break; | |
8c2a1a30 JB |
2700 | } |
2701 | ||
b89311ef | 2702 | return ret; |
8c2a1a30 JB |
2703 | } |
2704 | ||
9cfa3e34 | 2705 | static void |
32da5386 | 2706 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2707 | { |
2708 | atomic_inc(&bg->reservations); | |
2709 | } | |
2710 | ||
c759c4e1 JB |
2711 | /* |
2712 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2713 | * what it should be based on the mount options. | |
2714 | */ | |
2715 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2716 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2717 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2718 | { |
2719 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2720 | |
2721 | *empty_cluster = 0; | |
2722 | if (btrfs_mixed_space_info(space_info)) | |
2723 | return ret; | |
2724 | ||
c759c4e1 | 2725 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2726 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2727 | if (btrfs_test_opt(fs_info, SSD)) |
2728 | *empty_cluster = SZ_2M; | |
2729 | else | |
ee22184b | 2730 | *empty_cluster = SZ_64K; |
583b7231 HK |
2731 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2732 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2733 | *empty_cluster = SZ_2M; | |
0b246afa | 2734 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2735 | } |
2736 | ||
2737 | return ret; | |
2738 | } | |
2739 | ||
2ff7e61e JM |
2740 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2741 | u64 start, u64 end, | |
678886bd | 2742 | const bool return_free_space) |
ccd467d6 | 2743 | { |
32da5386 | 2744 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2745 | struct btrfs_space_info *space_info; |
2746 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2747 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2748 | u64 len; |
c759c4e1 JB |
2749 | u64 total_unpinned = 0; |
2750 | u64 empty_cluster = 0; | |
7b398f8e | 2751 | bool readonly; |
ccd467d6 | 2752 | |
11833d66 | 2753 | while (start <= end) { |
7b398f8e | 2754 | readonly = false; |
11833d66 | 2755 | if (!cache || |
b3470b5d | 2756 | start >= cache->start + cache->length) { |
11833d66 YZ |
2757 | if (cache) |
2758 | btrfs_put_block_group(cache); | |
c759c4e1 | 2759 | total_unpinned = 0; |
11833d66 | 2760 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2761 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2762 | |
2ff7e61e | 2763 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2764 | cache->space_info, |
2765 | &empty_cluster); | |
2766 | empty_cluster <<= 1; | |
11833d66 YZ |
2767 | } |
2768 | ||
b3470b5d | 2769 | len = cache->start + cache->length - start; |
11833d66 YZ |
2770 | len = min(len, end + 1 - start); |
2771 | ||
9076dbd5 JB |
2772 | if (start < cache->last_byte_to_unpin && return_free_space) { |
2773 | u64 add_len = min(len, cache->last_byte_to_unpin - start); | |
2774 | ||
2775 | btrfs_add_free_space(cache, start, add_len); | |
11833d66 YZ |
2776 | } |
2777 | ||
f0486c68 | 2778 | start += len; |
c759c4e1 | 2779 | total_unpinned += len; |
7b398f8e | 2780 | space_info = cache->space_info; |
f0486c68 | 2781 | |
c759c4e1 JB |
2782 | /* |
2783 | * If this space cluster has been marked as fragmented and we've | |
2784 | * unpinned enough in this block group to potentially allow a | |
2785 | * cluster to be created inside of it go ahead and clear the | |
2786 | * fragmented check. | |
2787 | */ | |
2788 | if (cluster && cluster->fragmented && | |
2789 | total_unpinned > empty_cluster) { | |
2790 | spin_lock(&cluster->lock); | |
2791 | cluster->fragmented = 0; | |
2792 | spin_unlock(&cluster->lock); | |
2793 | } | |
2794 | ||
7b398f8e | 2795 | spin_lock(&space_info->lock); |
11833d66 YZ |
2796 | spin_lock(&cache->lock); |
2797 | cache->pinned -= len; | |
bb96c4e5 | 2798 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2799 | space_info->max_extent_size = 0; |
dec59fa3 EL |
2800 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2801 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
2802 | if (cache->ro) { |
2803 | space_info->bytes_readonly += len; | |
2804 | readonly = true; | |
2805 | } | |
11833d66 | 2806 | spin_unlock(&cache->lock); |
957780eb JB |
2807 | if (!readonly && return_free_space && |
2808 | global_rsv->space_info == space_info) { | |
2809 | u64 to_add = len; | |
92ac58ec | 2810 | |
7b398f8e JB |
2811 | spin_lock(&global_rsv->lock); |
2812 | if (!global_rsv->full) { | |
957780eb JB |
2813 | to_add = min(len, global_rsv->size - |
2814 | global_rsv->reserved); | |
2815 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2816 | btrfs_space_info_update_bytes_may_use(fs_info, |
2817 | space_info, to_add); | |
7b398f8e JB |
2818 | if (global_rsv->reserved >= global_rsv->size) |
2819 | global_rsv->full = 1; | |
957780eb | 2820 | len -= to_add; |
7b398f8e JB |
2821 | } |
2822 | spin_unlock(&global_rsv->lock); | |
2823 | } | |
2732798c JB |
2824 | /* Add to any tickets we may have */ |
2825 | if (!readonly && return_free_space && len) | |
2826 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2827 | spin_unlock(&space_info->lock); |
ccd467d6 | 2828 | } |
11833d66 YZ |
2829 | |
2830 | if (cache) | |
2831 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2832 | return 0; |
2833 | } | |
2834 | ||
5ead2dd0 | 2835 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2836 | { |
5ead2dd0 | 2837 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2838 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2839 | struct list_head *deleted_bgs; |
11833d66 | 2840 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2841 | u64 start; |
2842 | u64 end; | |
a28ec197 | 2843 | int ret; |
a28ec197 | 2844 | |
fe119a6e | 2845 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2846 | |
bf31f87f | 2847 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2848 | struct extent_state *cached_state = NULL; |
2849 | ||
d4b450cd | 2850 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2851 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2852 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2853 | if (ret) { |
2854 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2855 | break; |
d4b450cd | 2856 | } |
fe119a6e NB |
2857 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
2858 | clear_extent_bits(&fs_info->excluded_extents, start, | |
f2fb7298 | 2859 | end, EXTENT_UPTODATE); |
1f3c79a2 | 2860 | |
46b27f50 | 2861 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2862 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2863 | end + 1 - start, NULL); |
1f3c79a2 | 2864 | |
0e6ec385 | 2865 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2866 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2867 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2868 | free_extent_state(cached_state); |
b9473439 | 2869 | cond_resched(); |
a28ec197 | 2870 | } |
817d52f8 | 2871 | |
a2309300 DZ |
2872 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2873 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2874 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2875 | } |
b0643e59 | 2876 | |
e33e17ee JM |
2877 | /* |
2878 | * Transaction is finished. We don't need the lock anymore. We | |
2879 | * do need to clean up the block groups in case of a transaction | |
2880 | * abort. | |
2881 | */ | |
2882 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2883 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2884 | u64 trimmed = 0; | |
2885 | ||
2886 | ret = -EROFS; | |
bf31f87f | 2887 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2888 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2889 | block_group->start, |
2890 | block_group->length, | |
e33e17ee JM |
2891 | &trimmed); |
2892 | ||
2893 | list_del_init(&block_group->bg_list); | |
6b7304af | 2894 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2895 | btrfs_put_block_group(block_group); |
2896 | ||
2897 | if (ret) { | |
2898 | const char *errstr = btrfs_decode_error(ret); | |
2899 | btrfs_warn(fs_info, | |
913e1535 | 2900 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2901 | ret, errstr); |
2902 | } | |
2903 | } | |
2904 | ||
e20d96d6 CM |
2905 | return 0; |
2906 | } | |
2907 | ||
1c2a07f5 QW |
2908 | /* |
2909 | * Drop one or more refs of @node. | |
2910 | * | |
2911 | * 1. Locate the extent refs. | |
2912 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
2913 | * Locate it, then reduce the refs number or remove the ref line completely. | |
2914 | * | |
2915 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
2916 | * | |
2917 | * Inline backref case: | |
2918 | * | |
2919 | * in extent tree we have: | |
2920 | * | |
2921 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2922 | * refs 2 gen 6 flags DATA | |
2923 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2924 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
2925 | * | |
2926 | * This function gets called with: | |
2927 | * | |
2928 | * node->bytenr = 13631488 | |
2929 | * node->num_bytes = 1048576 | |
2930 | * root_objectid = FS_TREE | |
2931 | * owner_objectid = 257 | |
2932 | * owner_offset = 0 | |
2933 | * refs_to_drop = 1 | |
2934 | * | |
2935 | * Then we should get some like: | |
2936 | * | |
2937 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2938 | * refs 1 gen 6 flags DATA | |
2939 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2940 | * | |
2941 | * Keyed backref case: | |
2942 | * | |
2943 | * in extent tree we have: | |
2944 | * | |
2945 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2946 | * refs 754 gen 6 flags DATA | |
2947 | * [...] | |
2948 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
2949 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
2950 | * | |
2951 | * This function get called with: | |
2952 | * | |
2953 | * node->bytenr = 13631488 | |
2954 | * node->num_bytes = 1048576 | |
2955 | * root_objectid = FS_TREE | |
2956 | * owner_objectid = 866 | |
2957 | * owner_offset = 0 | |
2958 | * refs_to_drop = 1 | |
2959 | * | |
2960 | * Then we should get some like: | |
2961 | * | |
2962 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2963 | * refs 753 gen 6 flags DATA | |
2964 | * | |
2965 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
2966 | */ | |
5d4f98a2 | 2967 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2968 | struct btrfs_delayed_ref_node *node, u64 parent, |
2969 | u64 root_objectid, u64 owner_objectid, | |
2970 | u64 owner_offset, int refs_to_drop, | |
2971 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2972 | { |
e72cb923 | 2973 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2974 | struct btrfs_key key; |
5d4f98a2 | 2975 | struct btrfs_path *path; |
1261ec42 | 2976 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2977 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2978 | struct btrfs_extent_item *ei; |
2979 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2980 | int ret; |
5d4f98a2 | 2981 | int is_data; |
952fccac CM |
2982 | int extent_slot = 0; |
2983 | int found_extent = 0; | |
2984 | int num_to_del = 1; | |
5d4f98a2 YZ |
2985 | u32 item_size; |
2986 | u64 refs; | |
c682f9b3 QW |
2987 | u64 bytenr = node->bytenr; |
2988 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2989 | int last_ref = 0; |
0b246afa | 2990 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 2991 | |
5caf2a00 | 2992 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2993 | if (!path) |
2994 | return -ENOMEM; | |
5f26f772 | 2995 | |
5d4f98a2 | 2996 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
1c2a07f5 QW |
2997 | |
2998 | if (!is_data && refs_to_drop != 1) { | |
2999 | btrfs_crit(info, | |
3000 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
3001 | node->bytenr, refs_to_drop); | |
3002 | ret = -EINVAL; | |
3003 | btrfs_abort_transaction(trans, ret); | |
3004 | goto out; | |
3005 | } | |
5d4f98a2 | 3006 | |
3173a18f | 3007 | if (is_data) |
897ca819 | 3008 | skinny_metadata = false; |
3173a18f | 3009 | |
fbe4801b NB |
3010 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
3011 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 3012 | owner_offset); |
7bb86316 | 3013 | if (ret == 0) { |
1c2a07f5 QW |
3014 | /* |
3015 | * Either the inline backref or the SHARED_DATA_REF/ | |
3016 | * SHARED_BLOCK_REF is found | |
3017 | * | |
3018 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
3019 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
3020 | */ | |
952fccac | 3021 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3022 | while (extent_slot >= 0) { |
3023 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3024 | extent_slot); |
5d4f98a2 | 3025 | if (key.objectid != bytenr) |
952fccac | 3026 | break; |
5d4f98a2 YZ |
3027 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3028 | key.offset == num_bytes) { | |
952fccac CM |
3029 | found_extent = 1; |
3030 | break; | |
3031 | } | |
3173a18f JB |
3032 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3033 | key.offset == owner_objectid) { | |
3034 | found_extent = 1; | |
3035 | break; | |
3036 | } | |
1c2a07f5 QW |
3037 | |
3038 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
3039 | if (path->slots[0] - extent_slot > 5) |
3040 | break; | |
5d4f98a2 | 3041 | extent_slot--; |
952fccac | 3042 | } |
a79865c6 | 3043 | |
31840ae1 | 3044 | if (!found_extent) { |
1c2a07f5 QW |
3045 | if (iref) { |
3046 | btrfs_crit(info, | |
3047 | "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref"); | |
3048 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3049 | goto err_dump; | |
3050 | } | |
3051 | /* Must be SHARED_* item, remove the backref first */ | |
87cc7a8a | 3052 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3053 | refs_to_drop, |
fcebe456 | 3054 | is_data, &last_ref); |
005d6427 | 3055 | if (ret) { |
66642832 | 3056 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3057 | goto out; |
3058 | } | |
b3b4aa74 | 3059 | btrfs_release_path(path); |
5d4f98a2 | 3060 | |
1c2a07f5 | 3061 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3062 | key.objectid = bytenr; |
3063 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3064 | key.offset = num_bytes; | |
3065 | ||
3173a18f JB |
3066 | if (!is_data && skinny_metadata) { |
3067 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3068 | key.offset = owner_objectid; | |
3069 | } | |
3070 | ||
31840ae1 ZY |
3071 | ret = btrfs_search_slot(trans, extent_root, |
3072 | &key, path, -1, 1); | |
3173a18f JB |
3073 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3074 | /* | |
3075 | * Couldn't find our skinny metadata item, | |
3076 | * see if we have ye olde extent item. | |
3077 | */ | |
3078 | path->slots[0]--; | |
3079 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3080 | path->slots[0]); | |
3081 | if (key.objectid == bytenr && | |
3082 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3083 | key.offset == num_bytes) | |
3084 | ret = 0; | |
3085 | } | |
3086 | ||
3087 | if (ret > 0 && skinny_metadata) { | |
3088 | skinny_metadata = false; | |
9ce49a0b | 3089 | key.objectid = bytenr; |
3173a18f JB |
3090 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3091 | key.offset = num_bytes; | |
3092 | btrfs_release_path(path); | |
3093 | ret = btrfs_search_slot(trans, extent_root, | |
3094 | &key, path, -1, 1); | |
3095 | } | |
3096 | ||
f3465ca4 | 3097 | if (ret) { |
5d163e0e JM |
3098 | btrfs_err(info, |
3099 | "umm, got %d back from search, was looking for %llu", | |
3100 | ret, bytenr); | |
b783e62d | 3101 | if (ret > 0) |
a4f78750 | 3102 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3103 | } |
005d6427 | 3104 | if (ret < 0) { |
66642832 | 3105 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3106 | goto out; |
3107 | } | |
31840ae1 ZY |
3108 | extent_slot = path->slots[0]; |
3109 | } | |
fae7f21c | 3110 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3111 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3112 | btrfs_err(info, |
3113 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3114 | bytenr, parent, root_objectid, owner_objectid, |
3115 | owner_offset); | |
66642832 | 3116 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3117 | goto out; |
79787eaa | 3118 | } else { |
66642832 | 3119 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3120 | goto out; |
7bb86316 | 3121 | } |
5f39d397 CM |
3122 | |
3123 | leaf = path->nodes[0]; | |
5d4f98a2 | 3124 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3125 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3126 | ret = -EINVAL; |
3127 | btrfs_print_v0_err(info); | |
3128 | btrfs_abort_transaction(trans, ret); | |
3129 | goto out; | |
3130 | } | |
952fccac | 3131 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3132 | struct btrfs_extent_item); |
3173a18f JB |
3133 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3134 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3135 | struct btrfs_tree_block_info *bi; |
1c2a07f5 QW |
3136 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
3137 | btrfs_crit(info, | |
cad69d13 | 3138 | "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu", |
1c2a07f5 QW |
3139 | key.objectid, key.type, key.offset, |
3140 | owner_objectid, item_size, | |
3141 | sizeof(*ei) + sizeof(*bi)); | |
3142 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3143 | goto err_dump; | |
3144 | } | |
5d4f98a2 YZ |
3145 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3146 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3147 | } | |
56bec294 | 3148 | |
5d4f98a2 | 3149 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3150 | if (refs < refs_to_drop) { |
1c2a07f5 QW |
3151 | btrfs_crit(info, |
3152 | "trying to drop %d refs but we only have %llu for bytenr %llu", | |
5d163e0e | 3153 | refs_to_drop, refs, bytenr); |
1c2a07f5 QW |
3154 | btrfs_abort_transaction(trans, -EUCLEAN); |
3155 | goto err_dump; | |
32b02538 | 3156 | } |
56bec294 | 3157 | refs -= refs_to_drop; |
5f39d397 | 3158 | |
5d4f98a2 YZ |
3159 | if (refs > 0) { |
3160 | if (extent_op) | |
3161 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3162 | /* | |
3163 | * In the case of inline back ref, reference count will | |
3164 | * be updated by remove_extent_backref | |
952fccac | 3165 | */ |
5d4f98a2 | 3166 | if (iref) { |
1c2a07f5 QW |
3167 | if (!found_extent) { |
3168 | btrfs_crit(info, | |
3169 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found"); | |
3170 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3171 | goto err_dump; | |
3172 | } | |
5d4f98a2 YZ |
3173 | } else { |
3174 | btrfs_set_extent_refs(leaf, ei, refs); | |
3175 | btrfs_mark_buffer_dirty(leaf); | |
3176 | } | |
3177 | if (found_extent) { | |
87cc7a8a NB |
3178 | ret = remove_extent_backref(trans, path, iref, |
3179 | refs_to_drop, is_data, | |
3180 | &last_ref); | |
005d6427 | 3181 | if (ret) { |
66642832 | 3182 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3183 | goto out; |
3184 | } | |
952fccac | 3185 | } |
5d4f98a2 | 3186 | } else { |
1c2a07f5 | 3187 | /* In this branch refs == 1 */ |
5d4f98a2 | 3188 | if (found_extent) { |
1c2a07f5 QW |
3189 | if (is_data && refs_to_drop != |
3190 | extent_data_ref_count(path, iref)) { | |
3191 | btrfs_crit(info, | |
3192 | "invalid refs_to_drop, current refs %u refs_to_drop %u", | |
3193 | extent_data_ref_count(path, iref), | |
3194 | refs_to_drop); | |
3195 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3196 | goto err_dump; | |
3197 | } | |
5d4f98a2 | 3198 | if (iref) { |
1c2a07f5 QW |
3199 | if (path->slots[0] != extent_slot) { |
3200 | btrfs_crit(info, | |
3201 | "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref", | |
3202 | key.objectid, key.type, | |
3203 | key.offset); | |
3204 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3205 | goto err_dump; | |
3206 | } | |
5d4f98a2 | 3207 | } else { |
1c2a07f5 QW |
3208 | /* |
3209 | * No inline ref, we must be at SHARED_* item, | |
3210 | * And it's single ref, it must be: | |
3211 | * | extent_slot ||extent_slot + 1| | |
3212 | * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ] | |
3213 | */ | |
3214 | if (path->slots[0] != extent_slot + 1) { | |
3215 | btrfs_crit(info, | |
3216 | "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM"); | |
3217 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3218 | goto err_dump; | |
3219 | } | |
5d4f98a2 YZ |
3220 | path->slots[0] = extent_slot; |
3221 | num_to_del = 2; | |
3222 | } | |
78fae27e | 3223 | } |
b9473439 | 3224 | |
fcebe456 | 3225 | last_ref = 1; |
952fccac CM |
3226 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3227 | num_to_del); | |
005d6427 | 3228 | if (ret) { |
66642832 | 3229 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3230 | goto out; |
3231 | } | |
b3b4aa74 | 3232 | btrfs_release_path(path); |
21af804c | 3233 | |
5d4f98a2 | 3234 | if (is_data) { |
40e046ac FM |
3235 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3236 | num_bytes); | |
005d6427 | 3237 | if (ret) { |
66642832 | 3238 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3239 | goto out; |
3240 | } | |
459931ec CM |
3241 | } |
3242 | ||
e7355e50 | 3243 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3244 | if (ret) { |
66642832 | 3245 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3246 | goto out; |
3247 | } | |
3248 | ||
ade4b516 | 3249 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3250 | if (ret) { |
66642832 | 3251 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3252 | goto out; |
3253 | } | |
a28ec197 | 3254 | } |
fcebe456 JB |
3255 | btrfs_release_path(path); |
3256 | ||
79787eaa | 3257 | out: |
5caf2a00 | 3258 | btrfs_free_path(path); |
a28ec197 | 3259 | return ret; |
1c2a07f5 QW |
3260 | err_dump: |
3261 | /* | |
3262 | * Leaf dump can take up a lot of log buffer, so we only do full leaf | |
3263 | * dump for debug build. | |
3264 | */ | |
3265 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
3266 | btrfs_crit(info, "path->slots[0]=%d extent_slot=%d", | |
3267 | path->slots[0], extent_slot); | |
3268 | btrfs_print_leaf(path->nodes[0]); | |
3269 | } | |
3270 | ||
3271 | btrfs_free_path(path); | |
3272 | return -EUCLEAN; | |
a28ec197 CM |
3273 | } |
3274 | ||
1887be66 | 3275 | /* |
f0486c68 | 3276 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3277 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3278 | * a given extent, and if there are no other delayed refs to be processed, it | |
3279 | * removes it from the tree. | |
3280 | */ | |
3281 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3282 | u64 bytenr) |
1887be66 CM |
3283 | { |
3284 | struct btrfs_delayed_ref_head *head; | |
3285 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3286 | int ret = 0; |
1887be66 CM |
3287 | |
3288 | delayed_refs = &trans->transaction->delayed_refs; | |
3289 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3290 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3291 | if (!head) |
cf93da7b | 3292 | goto out_delayed_unlock; |
1887be66 | 3293 | |
d7df2c79 | 3294 | spin_lock(&head->lock); |
e3d03965 | 3295 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3296 | goto out; |
3297 | ||
bedc6617 JB |
3298 | if (cleanup_extent_op(head) != NULL) |
3299 | goto out; | |
5d4f98a2 | 3300 | |
1887be66 CM |
3301 | /* |
3302 | * waiting for the lock here would deadlock. If someone else has it | |
3303 | * locked they are already in the process of dropping it anyway | |
3304 | */ | |
3305 | if (!mutex_trylock(&head->mutex)) | |
3306 | goto out; | |
3307 | ||
d7baffda | 3308 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3309 | head->processing = 0; |
d7baffda | 3310 | |
d7df2c79 | 3311 | spin_unlock(&head->lock); |
1887be66 CM |
3312 | spin_unlock(&delayed_refs->lock); |
3313 | ||
f0486c68 YZ |
3314 | BUG_ON(head->extent_op); |
3315 | if (head->must_insert_reserved) | |
3316 | ret = 1; | |
3317 | ||
31890da0 | 3318 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3319 | mutex_unlock(&head->mutex); |
d278850e | 3320 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3321 | return ret; |
1887be66 | 3322 | out: |
d7df2c79 | 3323 | spin_unlock(&head->lock); |
cf93da7b CM |
3324 | |
3325 | out_delayed_unlock: | |
1887be66 CM |
3326 | spin_unlock(&delayed_refs->lock); |
3327 | return 0; | |
3328 | } | |
3329 | ||
f0486c68 YZ |
3330 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3331 | struct btrfs_root *root, | |
3332 | struct extent_buffer *buf, | |
5581a51a | 3333 | u64 parent, int last_ref) |
f0486c68 | 3334 | { |
0b246afa | 3335 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3336 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 3337 | int pin = 1; |
f0486c68 YZ |
3338 | int ret; |
3339 | ||
ed4f255b QW |
3340 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3341 | buf->start, buf->len, parent); | |
3342 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3343 | root->root_key.objectid); | |
3344 | ||
f0486c68 | 3345 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
3346 | int old_ref_mod, new_ref_mod; |
3347 | ||
8a5040f7 | 3348 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 3349 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 3350 | &old_ref_mod, &new_ref_mod); |
79787eaa | 3351 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 3352 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
3353 | } |
3354 | ||
0a16c7d7 | 3355 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3356 | struct btrfs_block_group *cache; |
6219872d | 3357 | |
f0486c68 | 3358 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3359 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 3360 | if (!ret) |
37be25bc | 3361 | goto out; |
f0486c68 YZ |
3362 | } |
3363 | ||
4da8b76d | 3364 | pin = 0; |
0b246afa | 3365 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3366 | |
f0486c68 | 3367 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3368 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3369 | btrfs_put_block_group(cache); |
37be25bc | 3370 | goto out; |
f0486c68 YZ |
3371 | } |
3372 | ||
3373 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3374 | ||
3375 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3376 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3377 | btrfs_put_block_group(cache); |
71ff6437 | 3378 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3379 | } |
3380 | out: | |
b150a4f1 | 3381 | if (pin) |
78192442 | 3382 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 3383 | |
0a16c7d7 OS |
3384 | if (last_ref) { |
3385 | /* | |
3386 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3387 | * matter anymore. | |
3388 | */ | |
3389 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3390 | } | |
f0486c68 YZ |
3391 | } |
3392 | ||
79787eaa | 3393 | /* Can return -ENOMEM */ |
ffd4bb2a | 3394 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3395 | { |
ffd4bb2a | 3396 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 3397 | int old_ref_mod, new_ref_mod; |
925baedd CM |
3398 | int ret; |
3399 | ||
f5ee5c9a | 3400 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3401 | return 0; |
fccb84c9 | 3402 | |
56bec294 CM |
3403 | /* |
3404 | * tree log blocks never actually go into the extent allocation | |
3405 | * tree, just update pinning info and exit early. | |
56bec294 | 3406 | */ |
ffd4bb2a QW |
3407 | if ((ref->type == BTRFS_REF_METADATA && |
3408 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3409 | (ref->type == BTRFS_REF_DATA && | |
3410 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3411 | /* unlocks the pinned mutex */ |
b25c36f8 | 3412 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
d7eae340 | 3413 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 3414 | ret = 0; |
ffd4bb2a QW |
3415 | } else if (ref->type == BTRFS_REF_METADATA) { |
3416 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 3417 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 3418 | } else { |
ffd4bb2a | 3419 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 3420 | &old_ref_mod, &new_ref_mod); |
56bec294 | 3421 | } |
d7eae340 | 3422 | |
ffd4bb2a QW |
3423 | if (!((ref->type == BTRFS_REF_METADATA && |
3424 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3425 | (ref->type == BTRFS_REF_DATA && | |
3426 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3427 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3428 | |
ddf30cf0 | 3429 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 3430 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 3431 | |
925baedd CM |
3432 | return ret; |
3433 | } | |
3434 | ||
817d52f8 | 3435 | enum btrfs_loop_type { |
f262fa8d DS |
3436 | LOOP_CACHING_NOWAIT, |
3437 | LOOP_CACHING_WAIT, | |
3438 | LOOP_ALLOC_CHUNK, | |
3439 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3440 | }; |
3441 | ||
e570fd27 | 3442 | static inline void |
32da5386 | 3443 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3444 | int delalloc) |
3445 | { | |
3446 | if (delalloc) | |
3447 | down_read(&cache->data_rwsem); | |
3448 | } | |
3449 | ||
32da5386 | 3450 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3451 | int delalloc) |
3452 | { | |
3453 | btrfs_get_block_group(cache); | |
3454 | if (delalloc) | |
3455 | down_read(&cache->data_rwsem); | |
3456 | } | |
3457 | ||
32da5386 DS |
3458 | static struct btrfs_block_group *btrfs_lock_cluster( |
3459 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3460 | struct btrfs_free_cluster *cluster, |
3461 | int delalloc) | |
c142c6a4 | 3462 | __acquires(&cluster->refill_lock) |
e570fd27 | 3463 | { |
32da5386 | 3464 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3465 | |
e570fd27 | 3466 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3467 | while (1) { |
3468 | used_bg = cluster->block_group; | |
3469 | if (!used_bg) | |
3470 | return NULL; | |
3471 | ||
3472 | if (used_bg == block_group) | |
e570fd27 MX |
3473 | return used_bg; |
3474 | ||
6719afdc | 3475 | btrfs_get_block_group(used_bg); |
e570fd27 | 3476 | |
6719afdc GU |
3477 | if (!delalloc) |
3478 | return used_bg; | |
e570fd27 | 3479 | |
6719afdc GU |
3480 | if (down_read_trylock(&used_bg->data_rwsem)) |
3481 | return used_bg; | |
e570fd27 | 3482 | |
6719afdc | 3483 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3484 | |
e321f8a8 LB |
3485 | /* We should only have one-level nested. */ |
3486 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3487 | |
6719afdc GU |
3488 | spin_lock(&cluster->refill_lock); |
3489 | if (used_bg == cluster->block_group) | |
3490 | return used_bg; | |
e570fd27 | 3491 | |
6719afdc GU |
3492 | up_read(&used_bg->data_rwsem); |
3493 | btrfs_put_block_group(used_bg); | |
3494 | } | |
e570fd27 MX |
3495 | } |
3496 | ||
3497 | static inline void | |
32da5386 | 3498 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3499 | int delalloc) |
3500 | { | |
3501 | if (delalloc) | |
3502 | up_read(&cache->data_rwsem); | |
3503 | btrfs_put_block_group(cache); | |
3504 | } | |
3505 | ||
cb2f96f8 NA |
3506 | enum btrfs_extent_allocation_policy { |
3507 | BTRFS_EXTENT_ALLOC_CLUSTERED, | |
3508 | }; | |
3509 | ||
b4bd745d QW |
3510 | /* |
3511 | * Structure used internally for find_free_extent() function. Wraps needed | |
3512 | * parameters. | |
3513 | */ | |
3514 | struct find_free_extent_ctl { | |
3515 | /* Basic allocation info */ | |
b4bd745d QW |
3516 | u64 num_bytes; |
3517 | u64 empty_size; | |
3518 | u64 flags; | |
3519 | int delalloc; | |
3520 | ||
3521 | /* Where to start the search inside the bg */ | |
3522 | u64 search_start; | |
3523 | ||
3524 | /* For clustered allocation */ | |
3525 | u64 empty_cluster; | |
c10859be NA |
3526 | struct btrfs_free_cluster *last_ptr; |
3527 | bool use_cluster; | |
b4bd745d QW |
3528 | |
3529 | bool have_caching_bg; | |
3530 | bool orig_have_caching_bg; | |
3531 | ||
3532 | /* RAID index, converted from flags */ | |
3533 | int index; | |
3534 | ||
e72d79d6 QW |
3535 | /* |
3536 | * Current loop number, check find_free_extent_update_loop() for details | |
3537 | */ | |
b4bd745d QW |
3538 | int loop; |
3539 | ||
3540 | /* | |
3541 | * Whether we're refilling a cluster, if true we need to re-search | |
3542 | * current block group but don't try to refill the cluster again. | |
3543 | */ | |
3544 | bool retry_clustered; | |
3545 | ||
3546 | /* | |
3547 | * Whether we're updating free space cache, if true we need to re-search | |
3548 | * current block group but don't try updating free space cache again. | |
3549 | */ | |
3550 | bool retry_unclustered; | |
3551 | ||
3552 | /* If current block group is cached */ | |
3553 | int cached; | |
3554 | ||
3555 | /* Max contiguous hole found */ | |
3556 | u64 max_extent_size; | |
3557 | ||
3558 | /* Total free space from free space cache, not always contiguous */ | |
3559 | u64 total_free_space; | |
3560 | ||
3561 | /* Found result */ | |
3562 | u64 found_offset; | |
cb2f96f8 | 3563 | |
ea544149 NA |
3564 | /* Hint where to start looking for an empty space */ |
3565 | u64 hint_byte; | |
3566 | ||
cb2f96f8 NA |
3567 | /* Allocation policy */ |
3568 | enum btrfs_extent_allocation_policy policy; | |
b4bd745d QW |
3569 | }; |
3570 | ||
d06e3bb6 QW |
3571 | |
3572 | /* | |
3573 | * Helper function for find_free_extent(). | |
3574 | * | |
3575 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3576 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3577 | * Return >0 to inform caller that we find nothing | |
3578 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3579 | */ | |
32da5386 | 3580 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3581 | struct find_free_extent_ctl *ffe_ctl, |
3582 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3583 | { |
32da5386 | 3584 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3585 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3586 | u64 aligned_cluster; |
3587 | u64 offset; | |
3588 | int ret; | |
3589 | ||
3590 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3591 | if (!cluster_bg) | |
3592 | goto refill_cluster; | |
3593 | if (cluster_bg != bg && (cluster_bg->ro || | |
3594 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3595 | goto release_cluster; | |
3596 | ||
3597 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3598 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3599 | &ffe_ctl->max_extent_size); |
3600 | if (offset) { | |
3601 | /* We have a block, we're done */ | |
3602 | spin_unlock(&last_ptr->refill_lock); | |
3603 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3604 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3605 | *cluster_bg_ret = cluster_bg; | |
3606 | ffe_ctl->found_offset = offset; | |
3607 | return 0; | |
3608 | } | |
3609 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3610 | ||
3611 | release_cluster: | |
3612 | /* | |
3613 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3614 | * lets just skip it and let the allocator find whatever block it can | |
3615 | * find. If we reach this point, we will have tried the cluster | |
3616 | * allocator plenty of times and not have found anything, so we are | |
3617 | * likely way too fragmented for the clustering stuff to find anything. | |
3618 | * | |
3619 | * However, if the cluster is taken from the current block group, | |
3620 | * release the cluster first, so that we stand a better chance of | |
3621 | * succeeding in the unclustered allocation. | |
3622 | */ | |
3623 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3624 | spin_unlock(&last_ptr->refill_lock); | |
3625 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3626 | return -ENOENT; | |
3627 | } | |
3628 | ||
3629 | /* This cluster didn't work out, free it and start over */ | |
3630 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3631 | ||
3632 | if (cluster_bg != bg) | |
3633 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3634 | ||
3635 | refill_cluster: | |
3636 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3637 | spin_unlock(&last_ptr->refill_lock); | |
3638 | return -ENOENT; | |
3639 | } | |
3640 | ||
3641 | aligned_cluster = max_t(u64, | |
3642 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3643 | bg->full_stripe_len); | |
2ceeae2e DS |
3644 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3645 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3646 | if (ret == 0) { |
3647 | /* Now pull our allocation out of this cluster */ | |
3648 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3649 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3650 | &ffe_ctl->max_extent_size); | |
3651 | if (offset) { | |
3652 | /* We found one, proceed */ | |
3653 | spin_unlock(&last_ptr->refill_lock); | |
3654 | trace_btrfs_reserve_extent_cluster(bg, | |
3655 | ffe_ctl->search_start, | |
3656 | ffe_ctl->num_bytes); | |
3657 | ffe_ctl->found_offset = offset; | |
3658 | return 0; | |
3659 | } | |
3660 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3661 | !ffe_ctl->retry_clustered) { | |
3662 | spin_unlock(&last_ptr->refill_lock); | |
3663 | ||
3664 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3665 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3666 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3667 | return -EAGAIN; | |
3668 | } | |
3669 | /* | |
3670 | * At this point we either didn't find a cluster or we weren't able to | |
3671 | * allocate a block from our cluster. Free the cluster we've been | |
3672 | * trying to use, and go to the next block group. | |
3673 | */ | |
3674 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3675 | spin_unlock(&last_ptr->refill_lock); | |
3676 | return 1; | |
3677 | } | |
3678 | ||
e1a41848 QW |
3679 | /* |
3680 | * Return >0 to inform caller that we find nothing | |
3681 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3682 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3683 | */ | |
32da5386 | 3684 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3685 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3686 | { |
897cae79 | 3687 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3688 | u64 offset; |
3689 | ||
3690 | /* | |
3691 | * We are doing an unclustered allocation, set the fragmented flag so | |
3692 | * we don't bother trying to setup a cluster again until we get more | |
3693 | * space. | |
3694 | */ | |
3695 | if (unlikely(last_ptr)) { | |
3696 | spin_lock(&last_ptr->lock); | |
3697 | last_ptr->fragmented = 1; | |
3698 | spin_unlock(&last_ptr->lock); | |
3699 | } | |
3700 | if (ffe_ctl->cached) { | |
3701 | struct btrfs_free_space_ctl *free_space_ctl; | |
3702 | ||
3703 | free_space_ctl = bg->free_space_ctl; | |
3704 | spin_lock(&free_space_ctl->tree_lock); | |
3705 | if (free_space_ctl->free_space < | |
3706 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3707 | ffe_ctl->empty_size) { | |
3708 | ffe_ctl->total_free_space = max_t(u64, | |
3709 | ffe_ctl->total_free_space, | |
3710 | free_space_ctl->free_space); | |
3711 | spin_unlock(&free_space_ctl->tree_lock); | |
3712 | return 1; | |
3713 | } | |
3714 | spin_unlock(&free_space_ctl->tree_lock); | |
3715 | } | |
3716 | ||
3717 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3718 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3719 | &ffe_ctl->max_extent_size); | |
3720 | ||
3721 | /* | |
3722 | * If we didn't find a chunk, and we haven't failed on this block group | |
3723 | * before, and this block group is in the middle of caching and we are | |
3724 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3725 | * set @retry_unclustered to true. | |
3726 | * | |
3727 | * If @retry_unclustered is true then we've already waited on this | |
3728 | * block group once and should move on to the next block group. | |
3729 | */ | |
3730 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3731 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3732 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3733 | ffe_ctl->empty_size); | |
e1a41848 QW |
3734 | ffe_ctl->retry_unclustered = true; |
3735 | return -EAGAIN; | |
3736 | } else if (!offset) { | |
3737 | return 1; | |
3738 | } | |
3739 | ffe_ctl->found_offset = offset; | |
3740 | return 0; | |
3741 | } | |
3742 | ||
c668690d NA |
3743 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3744 | struct find_free_extent_ctl *ffe_ctl, | |
3745 | struct btrfs_block_group **bg_ret) | |
3746 | { | |
3747 | int ret; | |
3748 | ||
3749 | /* We want to try and use the cluster allocator, so lets look there */ | |
3750 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3751 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
c668690d NA |
3752 | if (ret >= 0 || ret == -EAGAIN) |
3753 | return ret; | |
3754 | /* ret == -ENOENT case falls through */ | |
3755 | } | |
3756 | ||
897cae79 | 3757 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3758 | } |
3759 | ||
3760 | static int do_allocation(struct btrfs_block_group *block_group, | |
3761 | struct find_free_extent_ctl *ffe_ctl, | |
3762 | struct btrfs_block_group **bg_ret) | |
3763 | { | |
3764 | switch (ffe_ctl->policy) { | |
3765 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3766 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
3767 | default: | |
3768 | BUG(); | |
3769 | } | |
3770 | } | |
3771 | ||
baba5062 NA |
3772 | static void release_block_group(struct btrfs_block_group *block_group, |
3773 | struct find_free_extent_ctl *ffe_ctl, | |
3774 | int delalloc) | |
3775 | { | |
3776 | switch (ffe_ctl->policy) { | |
3777 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3778 | ffe_ctl->retry_clustered = false; | |
3779 | ffe_ctl->retry_unclustered = false; | |
3780 | break; | |
3781 | default: | |
3782 | BUG(); | |
3783 | } | |
3784 | ||
3785 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
3786 | ffe_ctl->index); | |
3787 | btrfs_release_block_group(block_group, delalloc); | |
3788 | } | |
3789 | ||
0ab9724b NA |
3790 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
3791 | struct btrfs_key *ins) | |
3792 | { | |
3793 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3794 | ||
3795 | if (!ffe_ctl->use_cluster && last_ptr) { | |
3796 | spin_lock(&last_ptr->lock); | |
3797 | last_ptr->window_start = ins->objectid; | |
3798 | spin_unlock(&last_ptr->lock); | |
3799 | } | |
3800 | } | |
3801 | ||
3802 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
3803 | struct btrfs_key *ins) | |
3804 | { | |
3805 | switch (ffe_ctl->policy) { | |
3806 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3807 | found_extent_clustered(ffe_ctl, ins); | |
3808 | break; | |
3809 | default: | |
3810 | BUG(); | |
3811 | } | |
3812 | } | |
3813 | ||
c70e2139 NA |
3814 | static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl) |
3815 | { | |
3816 | switch (ffe_ctl->policy) { | |
3817 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3818 | /* | |
3819 | * If we can't allocate a new chunk we've already looped through | |
3820 | * at least once, move on to the NO_EMPTY_SIZE case. | |
3821 | */ | |
3822 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3823 | return 0; | |
3824 | default: | |
3825 | BUG(); | |
3826 | } | |
3827 | } | |
3828 | ||
e72d79d6 QW |
3829 | /* |
3830 | * Return >0 means caller needs to re-search for free extent | |
3831 | * Return 0 means we have the needed free extent. | |
3832 | * Return <0 means we failed to locate any free extent. | |
3833 | */ | |
3834 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
3835 | struct btrfs_key *ins, |
3836 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 3837 | bool full_search) |
e72d79d6 QW |
3838 | { |
3839 | struct btrfs_root *root = fs_info->extent_root; | |
3840 | int ret; | |
3841 | ||
3842 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3843 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3844 | ffe_ctl->orig_have_caching_bg = true; | |
3845 | ||
3846 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3847 | ffe_ctl->have_caching_bg) | |
3848 | return 1; | |
3849 | ||
3850 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3851 | return 1; | |
3852 | ||
3853 | if (ins->objectid) { | |
0ab9724b | 3854 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
3855 | return 0; |
3856 | } | |
3857 | ||
3858 | /* | |
3859 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3860 | * caching kthreads as we move along | |
3861 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3862 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3863 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3864 | * again | |
3865 | */ | |
3866 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3867 | ffe_ctl->index = 0; | |
3868 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3869 | /* | |
3870 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3871 | * don't have any uncached bgs and we've already done a | |
3872 | * full search through. | |
3873 | */ | |
3874 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3875 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3876 | else | |
3877 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3878 | } else { | |
3879 | ffe_ctl->loop++; | |
3880 | } | |
3881 | ||
3882 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3883 | struct btrfs_trans_handle *trans; | |
3884 | int exist = 0; | |
3885 | ||
3886 | trans = current->journal_info; | |
3887 | if (trans) | |
3888 | exist = 1; | |
3889 | else | |
3890 | trans = btrfs_join_transaction(root); | |
3891 | ||
3892 | if (IS_ERR(trans)) { | |
3893 | ret = PTR_ERR(trans); | |
3894 | return ret; | |
3895 | } | |
3896 | ||
fc471cb0 JB |
3897 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3898 | CHUNK_ALLOC_FORCE); | |
e72d79d6 | 3899 | |
e72d79d6 | 3900 | /* Do not bail out on ENOSPC since we can do more. */ |
c70e2139 NA |
3901 | if (ret == -ENOSPC) |
3902 | ret = chunk_allocation_failed(ffe_ctl); | |
3903 | else if (ret < 0) | |
e72d79d6 QW |
3904 | btrfs_abort_transaction(trans, ret); |
3905 | else | |
3906 | ret = 0; | |
3907 | if (!exist) | |
3908 | btrfs_end_transaction(trans); | |
3909 | if (ret) | |
3910 | return ret; | |
3911 | } | |
3912 | ||
3913 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
3914 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
3915 | return -ENOSPC; | |
3916 | ||
e72d79d6 QW |
3917 | /* |
3918 | * Don't loop again if we already have no empty_size and | |
3919 | * no empty_cluster. | |
3920 | */ | |
3921 | if (ffe_ctl->empty_size == 0 && | |
3922 | ffe_ctl->empty_cluster == 0) | |
3923 | return -ENOSPC; | |
3924 | ffe_ctl->empty_size = 0; | |
3925 | ffe_ctl->empty_cluster = 0; | |
3926 | } | |
3927 | return 1; | |
3928 | } | |
3929 | return -ENOSPC; | |
3930 | } | |
3931 | ||
7e895409 NA |
3932 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
3933 | struct find_free_extent_ctl *ffe_ctl, | |
3934 | struct btrfs_space_info *space_info, | |
3935 | struct btrfs_key *ins) | |
3936 | { | |
3937 | /* | |
3938 | * If our free space is heavily fragmented we may not be able to make | |
3939 | * big contiguous allocations, so instead of doing the expensive search | |
3940 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3941 | * can go ahead and search for a more manageable chunk. | |
3942 | * | |
3943 | * If our max_extent_size is large enough for our allocation simply | |
3944 | * disable clustering since we will likely not be able to find enough | |
3945 | * space to create a cluster and induce latency trying. | |
3946 | */ | |
3947 | if (space_info->max_extent_size) { | |
3948 | spin_lock(&space_info->lock); | |
3949 | if (space_info->max_extent_size && | |
3950 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
3951 | ins->offset = space_info->max_extent_size; | |
3952 | spin_unlock(&space_info->lock); | |
3953 | return -ENOSPC; | |
3954 | } else if (space_info->max_extent_size) { | |
3955 | ffe_ctl->use_cluster = false; | |
3956 | } | |
3957 | spin_unlock(&space_info->lock); | |
3958 | } | |
3959 | ||
3960 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
3961 | &ffe_ctl->empty_cluster); | |
3962 | if (ffe_ctl->last_ptr) { | |
3963 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3964 | ||
3965 | spin_lock(&last_ptr->lock); | |
3966 | if (last_ptr->block_group) | |
3967 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3968 | if (last_ptr->fragmented) { | |
3969 | /* | |
3970 | * We still set window_start so we can keep track of the | |
3971 | * last place we found an allocation to try and save | |
3972 | * some time. | |
3973 | */ | |
3974 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3975 | ffe_ctl->use_cluster = false; | |
3976 | } | |
3977 | spin_unlock(&last_ptr->lock); | |
3978 | } | |
3979 | ||
3980 | return 0; | |
3981 | } | |
3982 | ||
3983 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
3984 | struct find_free_extent_ctl *ffe_ctl, | |
3985 | struct btrfs_space_info *space_info, | |
3986 | struct btrfs_key *ins) | |
3987 | { | |
3988 | switch (ffe_ctl->policy) { | |
3989 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3990 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
3991 | space_info, ins); | |
3992 | default: | |
3993 | BUG(); | |
3994 | } | |
3995 | } | |
3996 | ||
fec577fb CM |
3997 | /* |
3998 | * walks the btree of allocated extents and find a hole of a given size. | |
3999 | * The key ins is changed to record the hole: | |
a4820398 | 4000 | * ins->objectid == start position |
62e2749e | 4001 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4002 | * ins->offset == the size of the hole. |
fec577fb | 4003 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4004 | * |
4005 | * If there is no suitable free space, we will record the max size of | |
4006 | * the free space extent currently. | |
e72d79d6 QW |
4007 | * |
4008 | * The overall logic and call chain: | |
4009 | * | |
4010 | * find_free_extent() | |
4011 | * |- Iterate through all block groups | |
4012 | * | |- Get a valid block group | |
4013 | * | |- Try to do clustered allocation in that block group | |
4014 | * | |- Try to do unclustered allocation in that block group | |
4015 | * | |- Check if the result is valid | |
4016 | * | | |- If valid, then exit | |
4017 | * | |- Jump to next block group | |
4018 | * | | |
4019 | * |- Push harder to find free extents | |
4020 | * |- If not found, re-iterate all block groups | |
fec577fb | 4021 | */ |
437490fe | 4022 | static noinline int find_free_extent(struct btrfs_root *root, |
18513091 | 4023 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
ea544149 | 4024 | u64 hint_byte_orig, struct btrfs_key *ins, |
18513091 | 4025 | u64 flags, int delalloc) |
fec577fb | 4026 | { |
437490fe | 4027 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4028 | int ret = 0; |
db8fe64f | 4029 | int cache_block_group_error = 0; |
32da5386 | 4030 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 4031 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 4032 | struct btrfs_space_info *space_info; |
a5e681d9 | 4033 | bool full_search = false; |
fec577fb | 4034 | |
0b246afa | 4035 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 4036 | |
b4bd745d QW |
4037 | ffe_ctl.num_bytes = num_bytes; |
4038 | ffe_ctl.empty_size = empty_size; | |
4039 | ffe_ctl.flags = flags; | |
4040 | ffe_ctl.search_start = 0; | |
b4bd745d QW |
4041 | ffe_ctl.delalloc = delalloc; |
4042 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
4043 | ffe_ctl.have_caching_bg = false; | |
4044 | ffe_ctl.orig_have_caching_bg = false; | |
4045 | ffe_ctl.found_offset = 0; | |
ea544149 | 4046 | ffe_ctl.hint_byte = hint_byte_orig; |
cb2f96f8 | 4047 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED; |
b4bd745d | 4048 | |
c10859be NA |
4049 | /* For clustered allocation */ |
4050 | ffe_ctl.retry_clustered = false; | |
4051 | ffe_ctl.retry_unclustered = false; | |
4052 | ffe_ctl.last_ptr = NULL; | |
4053 | ffe_ctl.use_cluster = true; | |
4054 | ||
962a298f | 4055 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4056 | ins->objectid = 0; |
4057 | ins->offset = 0; | |
b1a4d965 | 4058 | |
437490fe | 4059 | trace_find_free_extent(root, num_bytes, empty_size, flags); |
3f7de037 | 4060 | |
280c2908 | 4061 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 4062 | if (!space_info) { |
0b246afa | 4063 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
4064 | return -ENOSPC; |
4065 | } | |
2552d17e | 4066 | |
7e895409 NA |
4067 | ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins); |
4068 | if (ret < 0) | |
4069 | return ret; | |
fa9c0d79 | 4070 | |
b4bd745d QW |
4071 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
4072 | first_logical_byte(fs_info, 0)); | |
ea544149 NA |
4073 | ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte); |
4074 | if (ffe_ctl.search_start == ffe_ctl.hint_byte) { | |
b4bd745d QW |
4075 | block_group = btrfs_lookup_block_group(fs_info, |
4076 | ffe_ctl.search_start); | |
817d52f8 JB |
4077 | /* |
4078 | * we don't want to use the block group if it doesn't match our | |
4079 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4080 | * |
4081 | * However if we are re-searching with an ideal block group | |
4082 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4083 | */ |
b6919a58 | 4084 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 4085 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4086 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4087 | if (list_empty(&block_group->list) || |
4088 | block_group->ro) { | |
4089 | /* | |
4090 | * someone is removing this block group, | |
4091 | * we can't jump into the have_block_group | |
4092 | * target because our list pointers are not | |
4093 | * valid | |
4094 | */ | |
4095 | btrfs_put_block_group(block_group); | |
4096 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4097 | } else { |
b4bd745d | 4098 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 4099 | block_group->flags); |
e570fd27 | 4100 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 4101 | goto have_block_group; |
ccf0e725 | 4102 | } |
2552d17e | 4103 | } else if (block_group) { |
fa9c0d79 | 4104 | btrfs_put_block_group(block_group); |
2552d17e | 4105 | } |
42e70e7a | 4106 | } |
2552d17e | 4107 | search: |
b4bd745d QW |
4108 | ffe_ctl.have_caching_bg = false; |
4109 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
4110 | ffe_ctl.index == 0) | |
a5e681d9 | 4111 | full_search = true; |
80eb234a | 4112 | down_read(&space_info->groups_sem); |
b4bd745d QW |
4113 | list_for_each_entry(block_group, |
4114 | &space_info->block_groups[ffe_ctl.index], list) { | |
c668690d NA |
4115 | struct btrfs_block_group *bg_ret; |
4116 | ||
14443937 JM |
4117 | /* If the block group is read-only, we can skip it entirely. */ |
4118 | if (unlikely(block_group->ro)) | |
4119 | continue; | |
4120 | ||
e570fd27 | 4121 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 4122 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 4123 | |
83a50de9 CM |
4124 | /* |
4125 | * this can happen if we end up cycling through all the | |
4126 | * raid types, but we want to make sure we only allocate | |
4127 | * for the proper type. | |
4128 | */ | |
b6919a58 | 4129 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 4130 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4131 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4132 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4133 | BTRFS_BLOCK_GROUP_RAID10; |
4134 | ||
4135 | /* | |
4136 | * if they asked for extra copies and this block group | |
4137 | * doesn't provide them, bail. This does allow us to | |
4138 | * fill raid0 from raid1. | |
4139 | */ | |
b6919a58 | 4140 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4141 | goto loop; |
2a28468e QW |
4142 | |
4143 | /* | |
4144 | * This block group has different flags than we want. | |
4145 | * It's possible that we have MIXED_GROUP flag but no | |
4146 | * block group is mixed. Just skip such block group. | |
4147 | */ | |
4148 | btrfs_release_block_group(block_group, delalloc); | |
4149 | continue; | |
83a50de9 CM |
4150 | } |
4151 | ||
2552d17e | 4152 | have_block_group: |
32da5386 | 4153 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
4154 | if (unlikely(!ffe_ctl.cached)) { |
4155 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 4156 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
4157 | |
4158 | /* | |
4159 | * If we get ENOMEM here or something else we want to | |
4160 | * try other block groups, because it may not be fatal. | |
4161 | * However if we can't find anything else we need to | |
4162 | * save our return here so that we return the actual | |
4163 | * error that caused problems, not ENOSPC. | |
4164 | */ | |
4165 | if (ret < 0) { | |
4166 | if (!cache_block_group_error) | |
4167 | cache_block_group_error = ret; | |
4168 | ret = 0; | |
4169 | goto loop; | |
4170 | } | |
1d4284bd | 4171 | ret = 0; |
817d52f8 JB |
4172 | } |
4173 | ||
36cce922 JB |
4174 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
4175 | goto loop; | |
0f9dd46c | 4176 | |
c668690d NA |
4177 | bg_ret = NULL; |
4178 | ret = do_allocation(block_group, &ffe_ctl, &bg_ret); | |
4179 | if (ret == 0) { | |
4180 | if (bg_ret && bg_ret != block_group) { | |
4181 | btrfs_release_block_group(block_group, delalloc); | |
4182 | block_group = bg_ret; | |
fa9c0d79 | 4183 | } |
c668690d | 4184 | } else if (ret == -EAGAIN) { |
817d52f8 | 4185 | goto have_block_group; |
c668690d | 4186 | } else if (ret > 0) { |
1cdda9b8 | 4187 | goto loop; |
c668690d NA |
4188 | } |
4189 | ||
4190 | /* Checks */ | |
b4bd745d QW |
4191 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4192 | fs_info->stripesize); | |
25179201 | 4193 | |
2552d17e | 4194 | /* move on to the next group */ |
b4bd745d | 4195 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4196 | block_group->start + block_group->length) { |
b4bd745d QW |
4197 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4198 | num_bytes); | |
2552d17e | 4199 | goto loop; |
6226cb0a | 4200 | } |
f5a31e16 | 4201 | |
b4bd745d QW |
4202 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
4203 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
4204 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4205 | |
18513091 WX |
4206 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4207 | num_bytes, delalloc); | |
f0486c68 | 4208 | if (ret == -EAGAIN) { |
b4bd745d QW |
4209 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4210 | num_bytes); | |
2552d17e | 4211 | goto loop; |
0f9dd46c | 4212 | } |
9cfa3e34 | 4213 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4214 | |
f0486c68 | 4215 | /* we are all good, lets return */ |
b4bd745d | 4216 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4217 | ins->offset = num_bytes; |
d2fb3437 | 4218 | |
b4bd745d QW |
4219 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4220 | num_bytes); | |
e570fd27 | 4221 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4222 | break; |
4223 | loop: | |
baba5062 | 4224 | release_block_group(block_group, &ffe_ctl, delalloc); |
14443937 | 4225 | cond_resched(); |
2552d17e JB |
4226 | } |
4227 | up_read(&space_info->groups_sem); | |
4228 | ||
15b7ee65 | 4229 | ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search); |
e72d79d6 | 4230 | if (ret > 0) |
b742bb82 YZ |
4231 | goto search; |
4232 | ||
db8fe64f | 4233 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4234 | /* |
4235 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4236 | * any contiguous hole. | |
4237 | */ | |
4238 | if (!ffe_ctl.max_extent_size) | |
4239 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4240 | spin_lock(&space_info->lock); |
b4bd745d | 4241 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4242 | spin_unlock(&space_info->lock); |
b4bd745d | 4243 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4244 | } else if (ret == -ENOSPC) { |
4245 | ret = cache_block_group_error; | |
4f4db217 | 4246 | } |
0f70abe2 | 4247 | return ret; |
fec577fb | 4248 | } |
ec44a35c | 4249 | |
6f47c706 NB |
4250 | /* |
4251 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4252 | * hole that is at least as big as @num_bytes. | |
4253 | * | |
4254 | * @root - The root that will contain this extent | |
4255 | * | |
4256 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4257 | * is used for accounting purposes. This value differs | |
4258 | * from @num_bytes only in the case of compressed extents. | |
4259 | * | |
4260 | * @num_bytes - Number of bytes to allocate on-disk. | |
4261 | * | |
4262 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4263 | * allocator should try to satisfy. In some cases | |
4264 | * @num_bytes may be larger than what is required and if | |
4265 | * the filesystem is fragmented then allocation fails. | |
4266 | * However, the presence of @min_alloc_size gives a | |
4267 | * chance to try and satisfy the smaller allocation. | |
4268 | * | |
4269 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4270 | * size in bytes the allocator should try to find free | |
4271 | * next to the block it returns. This is just a hint and | |
4272 | * may be ignored by the allocator. | |
4273 | * | |
4274 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4275 | * address passed. It might be ignored. | |
4276 | * | |
4277 | * @ins - This key is modified to record the found hole. It will | |
4278 | * have the following values: | |
4279 | * ins->objectid == start position | |
4280 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4281 | * ins->offset == the size of the hole. | |
4282 | * | |
4283 | * @is_data - Boolean flag indicating whether an extent is | |
4284 | * allocated for data (true) or metadata (false) | |
4285 | * | |
4286 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4287 | * delalloc or not. If 'true' data_rwsem of block groups | |
4288 | * is going to be acquired. | |
4289 | * | |
4290 | * | |
4291 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4292 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4293 | * largest available hole the allocator managed to find. | |
4294 | */ | |
18513091 | 4295 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4296 | u64 num_bytes, u64 min_alloc_size, |
4297 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4298 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4299 | { |
ab8d0fc4 | 4300 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4301 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4302 | u64 flags; |
fec577fb | 4303 | int ret; |
925baedd | 4304 | |
1b86826d | 4305 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4306 | again: |
0b246afa | 4307 | WARN_ON(num_bytes < fs_info->sectorsize); |
437490fe | 4308 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
18513091 | 4309 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4310 | if (!ret && !is_data) { |
ab8d0fc4 | 4311 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4312 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4313 | if (!final_tried && ins->offset) { |
4314 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4315 | num_bytes = round_down(num_bytes, |
0b246afa | 4316 | fs_info->sectorsize); |
9e622d6b | 4317 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4318 | ram_bytes = num_bytes; |
9e622d6b MX |
4319 | if (num_bytes == min_alloc_size) |
4320 | final_tried = true; | |
4321 | goto again; | |
ab8d0fc4 | 4322 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4323 | struct btrfs_space_info *sinfo; |
4324 | ||
280c2908 | 4325 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4326 | btrfs_err(fs_info, |
5d163e0e JM |
4327 | "allocation failed flags %llu, wanted %llu", |
4328 | flags, num_bytes); | |
53804280 | 4329 | if (sinfo) |
5da6afeb JB |
4330 | btrfs_dump_space_info(fs_info, sinfo, |
4331 | num_bytes, 1); | |
9e622d6b | 4332 | } |
925baedd | 4333 | } |
0f9dd46c JB |
4334 | |
4335 | return ret; | |
e6dcd2dc CM |
4336 | } |
4337 | ||
a0fbf736 NB |
4338 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4339 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4340 | { |
32da5386 | 4341 | struct btrfs_block_group *cache; |
0f9dd46c | 4342 | |
0b246afa | 4343 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4344 | if (!cache) { |
0b246afa JM |
4345 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4346 | start); | |
0f9dd46c JB |
4347 | return -ENOSPC; |
4348 | } | |
1f3c79a2 | 4349 | |
a0fbf736 NB |
4350 | btrfs_add_free_space(cache, start, len); |
4351 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4352 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4353 | ||
fa9c0d79 | 4354 | btrfs_put_block_group(cache); |
a0fbf736 | 4355 | return 0; |
e6dcd2dc CM |
4356 | } |
4357 | ||
7bfc1007 NB |
4358 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, |
4359 | u64 len) | |
e688b725 | 4360 | { |
7ef54d54 | 4361 | struct btrfs_block_group *cache; |
a0fbf736 | 4362 | int ret = 0; |
7ef54d54 | 4363 | |
7bfc1007 | 4364 | cache = btrfs_lookup_block_group(trans->fs_info, start); |
7ef54d54 | 4365 | if (!cache) { |
7bfc1007 NB |
4366 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
4367 | start); | |
7ef54d54 NB |
4368 | return -ENOSPC; |
4369 | } | |
4370 | ||
6690d071 | 4371 | ret = pin_down_extent(trans, cache, start, len, 1); |
7ef54d54 | 4372 | btrfs_put_block_group(cache); |
a0fbf736 | 4373 | return ret; |
e688b725 CM |
4374 | } |
4375 | ||
5d4f98a2 | 4376 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4377 | u64 parent, u64 root_objectid, |
4378 | u64 flags, u64 owner, u64 offset, | |
4379 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4380 | { |
ef89b824 | 4381 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4382 | int ret; |
e6dcd2dc | 4383 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4384 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4385 | struct btrfs_path *path; |
5d4f98a2 YZ |
4386 | struct extent_buffer *leaf; |
4387 | int type; | |
4388 | u32 size; | |
26b8003f | 4389 | |
5d4f98a2 YZ |
4390 | if (parent > 0) |
4391 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4392 | else | |
4393 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4394 | |
5d4f98a2 | 4395 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4396 | |
4397 | path = btrfs_alloc_path(); | |
db5b493a TI |
4398 | if (!path) |
4399 | return -ENOMEM; | |
47e4bb98 | 4400 | |
5d4f98a2 YZ |
4401 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4402 | ins, size); | |
79787eaa JM |
4403 | if (ret) { |
4404 | btrfs_free_path(path); | |
4405 | return ret; | |
4406 | } | |
0f9dd46c | 4407 | |
5d4f98a2 YZ |
4408 | leaf = path->nodes[0]; |
4409 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4410 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4411 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4412 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4413 | btrfs_set_extent_flags(leaf, extent_item, | |
4414 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4415 | ||
4416 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4417 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4418 | if (parent > 0) { | |
4419 | struct btrfs_shared_data_ref *ref; | |
4420 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4421 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4422 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4423 | } else { | |
4424 | struct btrfs_extent_data_ref *ref; | |
4425 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4426 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4427 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4428 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4429 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4430 | } | |
47e4bb98 CM |
4431 | |
4432 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4433 | btrfs_free_path(path); |
f510cfec | 4434 | |
25a356d3 | 4435 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4436 | if (ret) |
4437 | return ret; | |
4438 | ||
ade4b516 | 4439 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4440 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4441 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4442 | ins->objectid, ins->offset); |
f5947066 CM |
4443 | BUG(); |
4444 | } | |
71ff6437 | 4445 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4446 | return ret; |
4447 | } | |
4448 | ||
5d4f98a2 | 4449 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4450 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4451 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4452 | { |
9dcdbe01 | 4453 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4454 | int ret; |
5d4f98a2 | 4455 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4456 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4457 | struct btrfs_tree_block_info *block_info; |
4458 | struct btrfs_extent_inline_ref *iref; | |
4459 | struct btrfs_path *path; | |
4460 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4461 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4462 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4463 | u64 num_bytes; |
21ebfbe7 | 4464 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4465 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4466 | |
4e6bd4e0 NB |
4467 | ref = btrfs_delayed_node_to_tree_ref(node); |
4468 | ||
4e6bd4e0 NB |
4469 | extent_key.objectid = node->bytenr; |
4470 | if (skinny_metadata) { | |
4471 | extent_key.offset = ref->level; | |
4472 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4473 | num_bytes = fs_info->nodesize; | |
4474 | } else { | |
4475 | extent_key.offset = node->num_bytes; | |
4476 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4477 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4478 | num_bytes = node->num_bytes; |
4479 | } | |
1c2308f8 | 4480 | |
5d4f98a2 | 4481 | path = btrfs_alloc_path(); |
80ee54bf | 4482 | if (!path) |
d8926bb3 | 4483 | return -ENOMEM; |
56bec294 | 4484 | |
5d4f98a2 | 4485 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4e6bd4e0 | 4486 | &extent_key, size); |
79787eaa | 4487 | if (ret) { |
dd825259 | 4488 | btrfs_free_path(path); |
79787eaa JM |
4489 | return ret; |
4490 | } | |
5d4f98a2 YZ |
4491 | |
4492 | leaf = path->nodes[0]; | |
4493 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4494 | struct btrfs_extent_item); | |
4495 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4496 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4497 | btrfs_set_extent_flags(leaf, extent_item, | |
4498 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4499 | |
3173a18f JB |
4500 | if (skinny_metadata) { |
4501 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4502 | } else { | |
4503 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4504 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4505 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4506 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4507 | } | |
5d4f98a2 | 4508 | |
d4b20733 | 4509 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4510 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4511 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4512 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4513 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4514 | } else { |
4515 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4516 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4517 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4518 | } |
4519 | ||
4520 | btrfs_mark_buffer_dirty(leaf); | |
4521 | btrfs_free_path(path); | |
4522 | ||
4e6bd4e0 NB |
4523 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4524 | num_bytes); | |
1e144fb8 OS |
4525 | if (ret) |
4526 | return ret; | |
4527 | ||
ade4b516 JB |
4528 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4529 | fs_info->nodesize, 1); | |
79787eaa | 4530 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4531 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4532 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4533 | BUG(); |
4534 | } | |
0be5dc67 | 4535 | |
4e6bd4e0 | 4536 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4537 | fs_info->nodesize); |
5d4f98a2 YZ |
4538 | return ret; |
4539 | } | |
4540 | ||
4541 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4542 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4543 | u64 offset, u64 ram_bytes, |
4544 | struct btrfs_key *ins) | |
5d4f98a2 | 4545 | { |
76675593 | 4546 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
4547 | int ret; |
4548 | ||
84f7d8e6 | 4549 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4550 | |
76675593 QW |
4551 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4552 | ins->objectid, ins->offset, 0); | |
4553 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4554 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
4555 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
4556 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
4557 | return ret; |
4558 | } | |
e02119d5 CM |
4559 | |
4560 | /* | |
4561 | * this is used by the tree logging recovery code. It records that | |
4562 | * an extent has been allocated and makes sure to clear the free | |
4563 | * space cache bits as well | |
4564 | */ | |
5d4f98a2 | 4565 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4566 | u64 root_objectid, u64 owner, u64 offset, |
4567 | struct btrfs_key *ins) | |
e02119d5 | 4568 | { |
61da2abf | 4569 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4570 | int ret; |
32da5386 | 4571 | struct btrfs_block_group *block_group; |
ed7a6948 | 4572 | struct btrfs_space_info *space_info; |
11833d66 | 4573 | |
8c2a1a30 JB |
4574 | /* |
4575 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4576 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4577 | */ |
0b246afa | 4578 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4579 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4580 | ins->offset); | |
b50c6e25 | 4581 | if (ret) |
8c2a1a30 | 4582 | return ret; |
11833d66 YZ |
4583 | } |
4584 | ||
0b246afa | 4585 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4586 | if (!block_group) |
4587 | return -EINVAL; | |
4588 | ||
ed7a6948 WX |
4589 | space_info = block_group->space_info; |
4590 | spin_lock(&space_info->lock); | |
4591 | spin_lock(&block_group->lock); | |
4592 | space_info->bytes_reserved += ins->offset; | |
4593 | block_group->reserved += ins->offset; | |
4594 | spin_unlock(&block_group->lock); | |
4595 | spin_unlock(&space_info->lock); | |
4596 | ||
ef89b824 NB |
4597 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4598 | offset, ins, 1); | |
bd727173 | 4599 | if (ret) |
ab9b2c7b | 4600 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
b50c6e25 | 4601 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4602 | return ret; |
4603 | } | |
4604 | ||
48a3b636 ES |
4605 | static struct extent_buffer * |
4606 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
9631e4cc JB |
4607 | u64 bytenr, int level, u64 owner, |
4608 | enum btrfs_lock_nesting nest) | |
65b51a00 | 4609 | { |
0b246afa | 4610 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4611 | struct extent_buffer *buf; |
4612 | ||
2ff7e61e | 4613 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4614 | if (IS_ERR(buf)) |
4615 | return buf; | |
4616 | ||
b72c3aba QW |
4617 | /* |
4618 | * Extra safety check in case the extent tree is corrupted and extent | |
4619 | * allocator chooses to use a tree block which is already used and | |
4620 | * locked. | |
4621 | */ | |
4622 | if (buf->lock_owner == current->pid) { | |
4623 | btrfs_err_rl(fs_info, | |
4624 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4625 | buf->start, btrfs_header_owner(buf), current->pid); | |
4626 | free_extent_buffer(buf); | |
4627 | return ERR_PTR(-EUCLEAN); | |
4628 | } | |
4629 | ||
ad244665 | 4630 | btrfs_set_buffer_lockdep_class(owner, buf, level); |
9631e4cc | 4631 | __btrfs_tree_lock(buf, nest); |
6a884d7d | 4632 | btrfs_clean_tree_block(buf); |
3083ee2e | 4633 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 4634 | |
4db8c528 | 4635 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4636 | |
bc877d28 NB |
4637 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4638 | btrfs_set_header_level(buf, level); | |
4639 | btrfs_set_header_bytenr(buf, buf->start); | |
4640 | btrfs_set_header_generation(buf, trans->transid); | |
4641 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4642 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4643 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4644 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4645 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4646 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4647 | /* |
4648 | * we allow two log transactions at a time, use different | |
52042d8e | 4649 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4650 | */ |
656f30db | 4651 | if (buf->log_index == 0) |
8cef4e16 YZ |
4652 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4653 | buf->start + buf->len - 1, GFP_NOFS); | |
4654 | else | |
4655 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4656 | buf->start + buf->len - 1); |
d0c803c4 | 4657 | } else { |
656f30db | 4658 | buf->log_index = -1; |
d0c803c4 | 4659 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4660 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4661 | } |
64c12921 | 4662 | trans->dirty = true; |
b4ce94de | 4663 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4664 | return buf; |
4665 | } | |
4666 | ||
fec577fb | 4667 | /* |
f0486c68 | 4668 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4669 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4670 | */ |
4d75f8a9 | 4671 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4672 | struct btrfs_root *root, |
4673 | u64 parent, u64 root_objectid, | |
4674 | const struct btrfs_disk_key *key, | |
4675 | int level, u64 hint, | |
9631e4cc JB |
4676 | u64 empty_size, |
4677 | enum btrfs_lock_nesting nest) | |
fec577fb | 4678 | { |
0b246afa | 4679 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4680 | struct btrfs_key ins; |
f0486c68 | 4681 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4682 | struct extent_buffer *buf; |
67b7859e | 4683 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4684 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4685 | u64 flags = 0; |
4686 | int ret; | |
0b246afa JM |
4687 | u32 blocksize = fs_info->nodesize; |
4688 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4689 | |
05653ef3 | 4690 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4691 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4692 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
9631e4cc | 4693 | level, root_objectid, nest); |
faa2dbf0 JB |
4694 | if (!IS_ERR(buf)) |
4695 | root->alloc_bytenr += blocksize; | |
4696 | return buf; | |
4697 | } | |
05653ef3 | 4698 | #endif |
fccb84c9 | 4699 | |
67f9c220 | 4700 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4701 | if (IS_ERR(block_rsv)) |
4702 | return ERR_CAST(block_rsv); | |
4703 | ||
18513091 | 4704 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4705 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4706 | if (ret) |
4707 | goto out_unuse; | |
55c69072 | 4708 | |
bc877d28 | 4709 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
9631e4cc | 4710 | root_objectid, nest); |
67b7859e OS |
4711 | if (IS_ERR(buf)) { |
4712 | ret = PTR_ERR(buf); | |
4713 | goto out_free_reserved; | |
4714 | } | |
f0486c68 YZ |
4715 | |
4716 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4717 | if (parent == 0) | |
4718 | parent = ins.objectid; | |
4719 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4720 | } else | |
4721 | BUG_ON(parent > 0); | |
4722 | ||
4723 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4724 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4725 | if (!extent_op) { |
4726 | ret = -ENOMEM; | |
4727 | goto out_free_buf; | |
4728 | } | |
f0486c68 YZ |
4729 | if (key) |
4730 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4731 | else | |
4732 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4733 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4734 | extent_op->update_key = skinny_metadata ? false : true; |
4735 | extent_op->update_flags = true; | |
4736 | extent_op->is_data = false; | |
b1c79e09 | 4737 | extent_op->level = level; |
f0486c68 | 4738 | |
ed4f255b QW |
4739 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4740 | ins.objectid, ins.offset, parent); | |
4741 | generic_ref.real_root = root->root_key.objectid; | |
4742 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4743 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 4744 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 4745 | extent_op, NULL, NULL); |
67b7859e OS |
4746 | if (ret) |
4747 | goto out_free_delayed; | |
f0486c68 | 4748 | } |
fec577fb | 4749 | return buf; |
67b7859e OS |
4750 | |
4751 | out_free_delayed: | |
4752 | btrfs_free_delayed_extent_op(extent_op); | |
4753 | out_free_buf: | |
4754 | free_extent_buffer(buf); | |
4755 | out_free_reserved: | |
2ff7e61e | 4756 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4757 | out_unuse: |
67f9c220 | 4758 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4759 | return ERR_PTR(ret); |
fec577fb | 4760 | } |
a28ec197 | 4761 | |
2c47e605 YZ |
4762 | struct walk_control { |
4763 | u64 refs[BTRFS_MAX_LEVEL]; | |
4764 | u64 flags[BTRFS_MAX_LEVEL]; | |
4765 | struct btrfs_key update_progress; | |
aea6f028 JB |
4766 | struct btrfs_key drop_progress; |
4767 | int drop_level; | |
2c47e605 YZ |
4768 | int stage; |
4769 | int level; | |
4770 | int shared_level; | |
4771 | int update_ref; | |
4772 | int keep_locks; | |
1c4850e2 YZ |
4773 | int reada_slot; |
4774 | int reada_count; | |
78c52d9e | 4775 | int restarted; |
2c47e605 YZ |
4776 | }; |
4777 | ||
4778 | #define DROP_REFERENCE 1 | |
4779 | #define UPDATE_BACKREF 2 | |
4780 | ||
1c4850e2 YZ |
4781 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4782 | struct btrfs_root *root, | |
4783 | struct walk_control *wc, | |
4784 | struct btrfs_path *path) | |
6407bf6d | 4785 | { |
0b246afa | 4786 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4787 | u64 bytenr; |
4788 | u64 generation; | |
4789 | u64 refs; | |
94fcca9f | 4790 | u64 flags; |
5d4f98a2 | 4791 | u32 nritems; |
1c4850e2 YZ |
4792 | struct btrfs_key key; |
4793 | struct extent_buffer *eb; | |
6407bf6d | 4794 | int ret; |
1c4850e2 YZ |
4795 | int slot; |
4796 | int nread = 0; | |
6407bf6d | 4797 | |
1c4850e2 YZ |
4798 | if (path->slots[wc->level] < wc->reada_slot) { |
4799 | wc->reada_count = wc->reada_count * 2 / 3; | |
4800 | wc->reada_count = max(wc->reada_count, 2); | |
4801 | } else { | |
4802 | wc->reada_count = wc->reada_count * 3 / 2; | |
4803 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4804 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4805 | } |
7bb86316 | 4806 | |
1c4850e2 YZ |
4807 | eb = path->nodes[wc->level]; |
4808 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4809 | |
1c4850e2 YZ |
4810 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4811 | if (nread >= wc->reada_count) | |
4812 | break; | |
bd56b302 | 4813 | |
2dd3e67b | 4814 | cond_resched(); |
1c4850e2 YZ |
4815 | bytenr = btrfs_node_blockptr(eb, slot); |
4816 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4817 | |
1c4850e2 YZ |
4818 | if (slot == path->slots[wc->level]) |
4819 | goto reada; | |
5d4f98a2 | 4820 | |
1c4850e2 YZ |
4821 | if (wc->stage == UPDATE_BACKREF && |
4822 | generation <= root->root_key.offset) | |
bd56b302 CM |
4823 | continue; |
4824 | ||
94fcca9f | 4825 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4826 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4827 | wc->level - 1, 1, &refs, |
4828 | &flags); | |
79787eaa JM |
4829 | /* We don't care about errors in readahead. */ |
4830 | if (ret < 0) | |
4831 | continue; | |
94fcca9f YZ |
4832 | BUG_ON(refs == 0); |
4833 | ||
1c4850e2 | 4834 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4835 | if (refs == 1) |
4836 | goto reada; | |
bd56b302 | 4837 | |
94fcca9f YZ |
4838 | if (wc->level == 1 && |
4839 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4840 | continue; | |
1c4850e2 YZ |
4841 | if (!wc->update_ref || |
4842 | generation <= root->root_key.offset) | |
4843 | continue; | |
4844 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4845 | ret = btrfs_comp_cpu_keys(&key, | |
4846 | &wc->update_progress); | |
4847 | if (ret < 0) | |
4848 | continue; | |
94fcca9f YZ |
4849 | } else { |
4850 | if (wc->level == 1 && | |
4851 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4852 | continue; | |
6407bf6d | 4853 | } |
1c4850e2 | 4854 | reada: |
2ff7e61e | 4855 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 4856 | nread++; |
20524f02 | 4857 | } |
1c4850e2 | 4858 | wc->reada_slot = slot; |
20524f02 | 4859 | } |
2c47e605 | 4860 | |
f82d02d9 | 4861 | /* |
2c016dc2 | 4862 | * helper to process tree block while walking down the tree. |
2c47e605 | 4863 | * |
2c47e605 YZ |
4864 | * when wc->stage == UPDATE_BACKREF, this function updates |
4865 | * back refs for pointers in the block. | |
4866 | * | |
4867 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4868 | */ |
2c47e605 | 4869 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4870 | struct btrfs_root *root, |
2c47e605 | 4871 | struct btrfs_path *path, |
94fcca9f | 4872 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4873 | { |
2ff7e61e | 4874 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4875 | int level = wc->level; |
4876 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4877 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4878 | int ret; |
4879 | ||
2c47e605 YZ |
4880 | if (wc->stage == UPDATE_BACKREF && |
4881 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4882 | return 1; | |
f82d02d9 | 4883 | |
2c47e605 YZ |
4884 | /* |
4885 | * when reference count of tree block is 1, it won't increase | |
4886 | * again. once full backref flag is set, we never clear it. | |
4887 | */ | |
94fcca9f YZ |
4888 | if (lookup_info && |
4889 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4890 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4891 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4892 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4893 | eb->start, level, 1, |
2c47e605 YZ |
4894 | &wc->refs[level], |
4895 | &wc->flags[level]); | |
79787eaa JM |
4896 | BUG_ON(ret == -ENOMEM); |
4897 | if (ret) | |
4898 | return ret; | |
2c47e605 YZ |
4899 | BUG_ON(wc->refs[level] == 0); |
4900 | } | |
5d4f98a2 | 4901 | |
2c47e605 YZ |
4902 | if (wc->stage == DROP_REFERENCE) { |
4903 | if (wc->refs[level] > 1) | |
4904 | return 1; | |
f82d02d9 | 4905 | |
2c47e605 | 4906 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4907 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4908 | path->locks[level] = 0; |
4909 | } | |
4910 | return 0; | |
4911 | } | |
f82d02d9 | 4912 | |
2c47e605 YZ |
4913 | /* wc->stage == UPDATE_BACKREF */ |
4914 | if (!(wc->flags[level] & flag)) { | |
4915 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4916 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4917 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4918 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4919 | BUG_ON(ret); /* -ENOMEM */ |
42c9d0b5 | 4920 | ret = btrfs_set_disk_extent_flags(trans, eb, flag, |
b1c79e09 | 4921 | btrfs_header_level(eb), 0); |
79787eaa | 4922 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4923 | wc->flags[level] |= flag; |
4924 | } | |
4925 | ||
4926 | /* | |
4927 | * the block is shared by multiple trees, so it's not good to | |
4928 | * keep the tree lock | |
4929 | */ | |
4930 | if (path->locks[level] && level > 0) { | |
bd681513 | 4931 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4932 | path->locks[level] = 0; |
4933 | } | |
4934 | return 0; | |
4935 | } | |
4936 | ||
78c52d9e JB |
4937 | /* |
4938 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4939 | * would have a drop_progress key that hadn't been updated properly. | |
4940 | */ | |
4941 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4942 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4943 | int level) | |
4944 | { | |
4945 | struct btrfs_path *path; | |
4946 | struct btrfs_extent_inline_ref *iref; | |
4947 | int ret; | |
4948 | ||
4949 | path = btrfs_alloc_path(); | |
4950 | if (!path) | |
4951 | return -ENOMEM; | |
4952 | ||
4953 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
4954 | root->fs_info->nodesize, parent, | |
4955 | root->root_key.objectid, level, 0); | |
4956 | btrfs_free_path(path); | |
4957 | if (ret == -ENOENT) | |
4958 | return 0; | |
4959 | if (ret < 0) | |
4960 | return ret; | |
4961 | return 1; | |
4962 | } | |
4963 | ||
1c4850e2 | 4964 | /* |
2c016dc2 | 4965 | * helper to process tree block pointer. |
1c4850e2 YZ |
4966 | * |
4967 | * when wc->stage == DROP_REFERENCE, this function checks | |
4968 | * reference count of the block pointed to. if the block | |
4969 | * is shared and we need update back refs for the subtree | |
4970 | * rooted at the block, this function changes wc->stage to | |
4971 | * UPDATE_BACKREF. if the block is shared and there is no | |
4972 | * need to update back, this function drops the reference | |
4973 | * to the block. | |
4974 | * | |
4975 | * NOTE: return value 1 means we should stop walking down. | |
4976 | */ | |
4977 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
4978 | struct btrfs_root *root, | |
4979 | struct btrfs_path *path, | |
94fcca9f | 4980 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 4981 | { |
0b246afa | 4982 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4983 | u64 bytenr; |
4984 | u64 generation; | |
4985 | u64 parent; | |
1c4850e2 | 4986 | struct btrfs_key key; |
581c1760 | 4987 | struct btrfs_key first_key; |
ffd4bb2a | 4988 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
4989 | struct extent_buffer *next; |
4990 | int level = wc->level; | |
4991 | int reada = 0; | |
4992 | int ret = 0; | |
1152651a | 4993 | bool need_account = false; |
1c4850e2 YZ |
4994 | |
4995 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
4996 | path->slots[level]); | |
4997 | /* | |
4998 | * if the lower level block was created before the snapshot | |
4999 | * was created, we know there is no need to update back refs | |
5000 | * for the subtree | |
5001 | */ | |
5002 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5003 | generation <= root->root_key.offset) { |
5004 | *lookup_info = 1; | |
1c4850e2 | 5005 | return 1; |
94fcca9f | 5006 | } |
1c4850e2 YZ |
5007 | |
5008 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
5009 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
5010 | path->slots[level]); | |
1c4850e2 | 5011 | |
0b246afa | 5012 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5013 | if (!next) { |
2ff7e61e | 5014 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
5015 | if (IS_ERR(next)) |
5016 | return PTR_ERR(next); | |
5017 | ||
b2aaaa3b JB |
5018 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
5019 | level - 1); | |
1c4850e2 YZ |
5020 | reada = 1; |
5021 | } | |
5022 | btrfs_tree_lock(next); | |
1c4850e2 | 5023 | |
2ff7e61e | 5024 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
5025 | &wc->refs[level - 1], |
5026 | &wc->flags[level - 1]); | |
4867268c JB |
5027 | if (ret < 0) |
5028 | goto out_unlock; | |
79787eaa | 5029 | |
c2cf52eb | 5030 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5031 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5032 | ret = -EIO; |
5033 | goto out_unlock; | |
c2cf52eb | 5034 | } |
94fcca9f | 5035 | *lookup_info = 0; |
1c4850e2 | 5036 | |
94fcca9f | 5037 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5038 | if (wc->refs[level - 1] > 1) { |
1152651a | 5039 | need_account = true; |
94fcca9f YZ |
5040 | if (level == 1 && |
5041 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5042 | goto skip; | |
5043 | ||
1c4850e2 YZ |
5044 | if (!wc->update_ref || |
5045 | generation <= root->root_key.offset) | |
5046 | goto skip; | |
5047 | ||
5048 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5049 | path->slots[level]); | |
5050 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5051 | if (ret < 0) | |
5052 | goto skip; | |
5053 | ||
5054 | wc->stage = UPDATE_BACKREF; | |
5055 | wc->shared_level = level - 1; | |
5056 | } | |
94fcca9f YZ |
5057 | } else { |
5058 | if (level == 1 && | |
5059 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5060 | goto skip; | |
1c4850e2 YZ |
5061 | } |
5062 | ||
b9fab919 | 5063 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5064 | btrfs_tree_unlock(next); |
5065 | free_extent_buffer(next); | |
5066 | next = NULL; | |
94fcca9f | 5067 | *lookup_info = 1; |
1c4850e2 YZ |
5068 | } |
5069 | ||
5070 | if (!next) { | |
5071 | if (reada && level == 1) | |
5072 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
5073 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
5074 | &first_key); | |
64c043de LB |
5075 | if (IS_ERR(next)) { |
5076 | return PTR_ERR(next); | |
5077 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5078 | free_extent_buffer(next); |
97d9a8a4 | 5079 | return -EIO; |
416bc658 | 5080 | } |
1c4850e2 | 5081 | btrfs_tree_lock(next); |
1c4850e2 YZ |
5082 | } |
5083 | ||
5084 | level--; | |
4867268c JB |
5085 | ASSERT(level == btrfs_header_level(next)); |
5086 | if (level != btrfs_header_level(next)) { | |
5087 | btrfs_err(root->fs_info, "mismatched level"); | |
5088 | ret = -EIO; | |
5089 | goto out_unlock; | |
5090 | } | |
1c4850e2 YZ |
5091 | path->nodes[level] = next; |
5092 | path->slots[level] = 0; | |
ac5887c8 | 5093 | path->locks[level] = BTRFS_WRITE_LOCK; |
1c4850e2 YZ |
5094 | wc->level = level; |
5095 | if (wc->level == 1) | |
5096 | wc->reada_slot = 0; | |
5097 | return 0; | |
5098 | skip: | |
5099 | wc->refs[level - 1] = 0; | |
5100 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5101 | if (wc->stage == DROP_REFERENCE) { |
5102 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5103 | parent = path->nodes[level]->start; | |
5104 | } else { | |
4867268c | 5105 | ASSERT(root->root_key.objectid == |
94fcca9f | 5106 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5107 | if (root->root_key.objectid != |
5108 | btrfs_header_owner(path->nodes[level])) { | |
5109 | btrfs_err(root->fs_info, | |
5110 | "mismatched block owner"); | |
5111 | ret = -EIO; | |
5112 | goto out_unlock; | |
5113 | } | |
94fcca9f YZ |
5114 | parent = 0; |
5115 | } | |
1c4850e2 | 5116 | |
78c52d9e JB |
5117 | /* |
5118 | * If we had a drop_progress we need to verify the refs are set | |
5119 | * as expected. If we find our ref then we know that from here | |
5120 | * on out everything should be correct, and we can clear the | |
5121 | * ->restarted flag. | |
5122 | */ | |
5123 | if (wc->restarted) { | |
5124 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5125 | level - 1); | |
5126 | if (ret < 0) | |
5127 | goto out_unlock; | |
5128 | if (ret == 0) | |
5129 | goto no_delete; | |
5130 | ret = 0; | |
5131 | wc->restarted = 0; | |
5132 | } | |
5133 | ||
2cd86d30 QW |
5134 | /* |
5135 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5136 | * already accounted them at merge time (replace_path), | |
5137 | * thus we could skip expensive subtree trace here. | |
5138 | */ | |
5139 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5140 | need_account) { | |
deb40627 | 5141 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5142 | generation, level - 1); |
1152651a | 5143 | if (ret) { |
0b246afa | 5144 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5145 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5146 | ret); | |
1152651a MF |
5147 | } |
5148 | } | |
aea6f028 JB |
5149 | |
5150 | /* | |
5151 | * We need to update the next key in our walk control so we can | |
5152 | * update the drop_progress key accordingly. We don't care if | |
5153 | * find_next_key doesn't find a key because that means we're at | |
5154 | * the end and are going to clean up now. | |
5155 | */ | |
5156 | wc->drop_level = level; | |
5157 | find_next_key(path, level, &wc->drop_progress); | |
5158 | ||
ffd4bb2a QW |
5159 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
5160 | fs_info->nodesize, parent); | |
5161 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
5162 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
5163 | if (ret) |
5164 | goto out_unlock; | |
1c4850e2 | 5165 | } |
78c52d9e | 5166 | no_delete: |
4867268c JB |
5167 | *lookup_info = 1; |
5168 | ret = 1; | |
5169 | ||
5170 | out_unlock: | |
1c4850e2 YZ |
5171 | btrfs_tree_unlock(next); |
5172 | free_extent_buffer(next); | |
4867268c JB |
5173 | |
5174 | return ret; | |
1c4850e2 YZ |
5175 | } |
5176 | ||
2c47e605 | 5177 | /* |
2c016dc2 | 5178 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5179 | * |
5180 | * when wc->stage == DROP_REFERENCE, this function drops | |
5181 | * reference count on the block. | |
5182 | * | |
5183 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5184 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5185 | * to UPDATE_BACKREF previously while processing the block. | |
5186 | * | |
5187 | * NOTE: return value 1 means we should stop walking up. | |
5188 | */ | |
5189 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5190 | struct btrfs_root *root, | |
5191 | struct btrfs_path *path, | |
5192 | struct walk_control *wc) | |
5193 | { | |
0b246afa | 5194 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5195 | int ret; |
2c47e605 YZ |
5196 | int level = wc->level; |
5197 | struct extent_buffer *eb = path->nodes[level]; | |
5198 | u64 parent = 0; | |
5199 | ||
5200 | if (wc->stage == UPDATE_BACKREF) { | |
5201 | BUG_ON(wc->shared_level < level); | |
5202 | if (level < wc->shared_level) | |
5203 | goto out; | |
5204 | ||
2c47e605 YZ |
5205 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5206 | if (ret > 0) | |
5207 | wc->update_ref = 0; | |
5208 | ||
5209 | wc->stage = DROP_REFERENCE; | |
5210 | wc->shared_level = -1; | |
5211 | path->slots[level] = 0; | |
5212 | ||
5213 | /* | |
5214 | * check reference count again if the block isn't locked. | |
5215 | * we should start walking down the tree again if reference | |
5216 | * count is one. | |
5217 | */ | |
5218 | if (!path->locks[level]) { | |
5219 | BUG_ON(level == 0); | |
5220 | btrfs_tree_lock(eb); | |
ac5887c8 | 5221 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5222 | |
2ff7e61e | 5223 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5224 | eb->start, level, 1, |
2c47e605 YZ |
5225 | &wc->refs[level], |
5226 | &wc->flags[level]); | |
79787eaa JM |
5227 | if (ret < 0) { |
5228 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5229 | path->locks[level] = 0; |
79787eaa JM |
5230 | return ret; |
5231 | } | |
2c47e605 YZ |
5232 | BUG_ON(wc->refs[level] == 0); |
5233 | if (wc->refs[level] == 1) { | |
bd681513 | 5234 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5235 | path->locks[level] = 0; |
2c47e605 YZ |
5236 | return 1; |
5237 | } | |
f82d02d9 | 5238 | } |
2c47e605 | 5239 | } |
f82d02d9 | 5240 | |
2c47e605 YZ |
5241 | /* wc->stage == DROP_REFERENCE */ |
5242 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5243 | |
2c47e605 YZ |
5244 | if (wc->refs[level] == 1) { |
5245 | if (level == 0) { | |
5246 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5247 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5248 | else |
e339a6b0 | 5249 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5250 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5251 | if (is_fstree(root->root_key.objectid)) { |
5252 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5253 | if (ret) { | |
5254 | btrfs_err_rl(fs_info, | |
5255 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5256 | ret); |
c4140cbf | 5257 | } |
1152651a | 5258 | } |
2c47e605 | 5259 | } |
6a884d7d | 5260 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5261 | if (!path->locks[level] && |
5262 | btrfs_header_generation(eb) == trans->transid) { | |
5263 | btrfs_tree_lock(eb); | |
ac5887c8 | 5264 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5265 | } |
6a884d7d | 5266 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5267 | } |
5268 | ||
5269 | if (eb == root->node) { | |
5270 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5271 | parent = eb->start; | |
65c6e82b QW |
5272 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5273 | goto owner_mismatch; | |
2c47e605 YZ |
5274 | } else { |
5275 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5276 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5277 | else if (root->root_key.objectid != |
5278 | btrfs_header_owner(path->nodes[level + 1])) | |
5279 | goto owner_mismatch; | |
f82d02d9 | 5280 | } |
f82d02d9 | 5281 | |
5581a51a | 5282 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5283 | out: |
5284 | wc->refs[level] = 0; | |
5285 | wc->flags[level] = 0; | |
f0486c68 | 5286 | return 0; |
65c6e82b QW |
5287 | |
5288 | owner_mismatch: | |
5289 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5290 | btrfs_header_owner(eb), root->root_key.objectid); | |
5291 | return -EUCLEAN; | |
2c47e605 YZ |
5292 | } |
5293 | ||
5294 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5295 | struct btrfs_root *root, | |
5296 | struct btrfs_path *path, | |
5297 | struct walk_control *wc) | |
5298 | { | |
2c47e605 | 5299 | int level = wc->level; |
94fcca9f | 5300 | int lookup_info = 1; |
2c47e605 YZ |
5301 | int ret; |
5302 | ||
5303 | while (level >= 0) { | |
94fcca9f | 5304 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5305 | if (ret > 0) |
5306 | break; | |
5307 | ||
5308 | if (level == 0) | |
5309 | break; | |
5310 | ||
7a7965f8 YZ |
5311 | if (path->slots[level] >= |
5312 | btrfs_header_nritems(path->nodes[level])) | |
5313 | break; | |
5314 | ||
94fcca9f | 5315 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5316 | if (ret > 0) { |
5317 | path->slots[level]++; | |
5318 | continue; | |
90d2c51d MX |
5319 | } else if (ret < 0) |
5320 | return ret; | |
1c4850e2 | 5321 | level = wc->level; |
f82d02d9 | 5322 | } |
f82d02d9 YZ |
5323 | return 0; |
5324 | } | |
5325 | ||
d397712b | 5326 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5327 | struct btrfs_root *root, |
f82d02d9 | 5328 | struct btrfs_path *path, |
2c47e605 | 5329 | struct walk_control *wc, int max_level) |
20524f02 | 5330 | { |
2c47e605 | 5331 | int level = wc->level; |
20524f02 | 5332 | int ret; |
9f3a7427 | 5333 | |
2c47e605 YZ |
5334 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5335 | while (level < max_level && path->nodes[level]) { | |
5336 | wc->level = level; | |
5337 | if (path->slots[level] + 1 < | |
5338 | btrfs_header_nritems(path->nodes[level])) { | |
5339 | path->slots[level]++; | |
20524f02 CM |
5340 | return 0; |
5341 | } else { | |
2c47e605 YZ |
5342 | ret = walk_up_proc(trans, root, path, wc); |
5343 | if (ret > 0) | |
5344 | return 0; | |
65c6e82b QW |
5345 | if (ret < 0) |
5346 | return ret; | |
bd56b302 | 5347 | |
2c47e605 | 5348 | if (path->locks[level]) { |
bd681513 CM |
5349 | btrfs_tree_unlock_rw(path->nodes[level], |
5350 | path->locks[level]); | |
2c47e605 | 5351 | path->locks[level] = 0; |
f82d02d9 | 5352 | } |
2c47e605 YZ |
5353 | free_extent_buffer(path->nodes[level]); |
5354 | path->nodes[level] = NULL; | |
5355 | level++; | |
20524f02 CM |
5356 | } |
5357 | } | |
5358 | return 1; | |
5359 | } | |
5360 | ||
9aca1d51 | 5361 | /* |
2c47e605 YZ |
5362 | * drop a subvolume tree. |
5363 | * | |
5364 | * this function traverses the tree freeing any blocks that only | |
5365 | * referenced by the tree. | |
5366 | * | |
5367 | * when a shared tree block is found. this function decreases its | |
5368 | * reference count by one. if update_ref is true, this function | |
5369 | * also make sure backrefs for the shared block and all lower level | |
5370 | * blocks are properly updated. | |
9d1a2a3a DS |
5371 | * |
5372 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5373 | */ |
0078a9f9 | 5374 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5375 | { |
ab8d0fc4 | 5376 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5377 | struct btrfs_path *path; |
2c47e605 | 5378 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5379 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5380 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5381 | struct walk_control *wc; |
5382 | struct btrfs_key key; | |
5383 | int err = 0; | |
5384 | int ret; | |
5385 | int level; | |
d29a9f62 | 5386 | bool root_dropped = false; |
20524f02 | 5387 | |
4fd786e6 | 5388 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5389 | |
5caf2a00 | 5390 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5391 | if (!path) { |
5392 | err = -ENOMEM; | |
5393 | goto out; | |
5394 | } | |
20524f02 | 5395 | |
2c47e605 | 5396 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5397 | if (!wc) { |
5398 | btrfs_free_path(path); | |
cb1b69f4 TI |
5399 | err = -ENOMEM; |
5400 | goto out; | |
38a1a919 | 5401 | } |
2c47e605 | 5402 | |
f3e3d9cc QW |
5403 | /* |
5404 | * Use join to avoid potential EINTR from transaction start. See | |
5405 | * wait_reserve_ticket and the whole reservation callchain. | |
5406 | */ | |
5407 | if (for_reloc) | |
5408 | trans = btrfs_join_transaction(tree_root); | |
5409 | else | |
5410 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5411 | if (IS_ERR(trans)) { |
5412 | err = PTR_ERR(trans); | |
5413 | goto out_free; | |
5414 | } | |
98d5dc13 | 5415 | |
0568e82d JB |
5416 | err = btrfs_run_delayed_items(trans); |
5417 | if (err) | |
5418 | goto out_end_trans; | |
5419 | ||
83354f07 JB |
5420 | /* |
5421 | * This will help us catch people modifying the fs tree while we're | |
5422 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5423 | * dropped as we unlock the root node and parent nodes as we walk down | |
5424 | * the tree, assuming nothing will change. If something does change | |
5425 | * then we'll have stale information and drop references to blocks we've | |
5426 | * already dropped. | |
5427 | */ | |
5428 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5429 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5430 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5431 | path->nodes[level] = btrfs_lock_root_node(root); |
9f3a7427 | 5432 | path->slots[level] = 0; |
ac5887c8 | 5433 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5434 | memset(&wc->update_progress, 0, |
5435 | sizeof(wc->update_progress)); | |
9f3a7427 | 5436 | } else { |
9f3a7427 | 5437 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5438 | memcpy(&wc->update_progress, &key, |
5439 | sizeof(wc->update_progress)); | |
5440 | ||
c8422684 | 5441 | level = btrfs_root_drop_level(root_item); |
2c47e605 | 5442 | BUG_ON(level == 0); |
6702ed49 | 5443 | path->lowest_level = level; |
2c47e605 YZ |
5444 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5445 | path->lowest_level = 0; | |
5446 | if (ret < 0) { | |
5447 | err = ret; | |
79787eaa | 5448 | goto out_end_trans; |
9f3a7427 | 5449 | } |
1c4850e2 | 5450 | WARN_ON(ret > 0); |
2c47e605 | 5451 | |
7d9eb12c CM |
5452 | /* |
5453 | * unlock our path, this is safe because only this | |
5454 | * function is allowed to delete this snapshot | |
5455 | */ | |
5d4f98a2 | 5456 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5457 | |
5458 | level = btrfs_header_level(root->node); | |
5459 | while (1) { | |
5460 | btrfs_tree_lock(path->nodes[level]); | |
ac5887c8 | 5461 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5462 | |
2ff7e61e | 5463 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5464 | path->nodes[level]->start, |
3173a18f | 5465 | level, 1, &wc->refs[level], |
2c47e605 | 5466 | &wc->flags[level]); |
79787eaa JM |
5467 | if (ret < 0) { |
5468 | err = ret; | |
5469 | goto out_end_trans; | |
5470 | } | |
2c47e605 YZ |
5471 | BUG_ON(wc->refs[level] == 0); |
5472 | ||
c8422684 | 5473 | if (level == btrfs_root_drop_level(root_item)) |
2c47e605 YZ |
5474 | break; |
5475 | ||
5476 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5477 | path->locks[level] = 0; |
2c47e605 YZ |
5478 | WARN_ON(wc->refs[level] != 1); |
5479 | level--; | |
5480 | } | |
9f3a7427 | 5481 | } |
2c47e605 | 5482 | |
78c52d9e | 5483 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5484 | wc->level = level; |
5485 | wc->shared_level = -1; | |
5486 | wc->stage = DROP_REFERENCE; | |
5487 | wc->update_ref = update_ref; | |
5488 | wc->keep_locks = 0; | |
0b246afa | 5489 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5490 | |
d397712b | 5491 | while (1) { |
9d1a2a3a | 5492 | |
2c47e605 YZ |
5493 | ret = walk_down_tree(trans, root, path, wc); |
5494 | if (ret < 0) { | |
5495 | err = ret; | |
20524f02 | 5496 | break; |
2c47e605 | 5497 | } |
9aca1d51 | 5498 | |
2c47e605 YZ |
5499 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5500 | if (ret < 0) { | |
5501 | err = ret; | |
20524f02 | 5502 | break; |
2c47e605 YZ |
5503 | } |
5504 | ||
5505 | if (ret > 0) { | |
5506 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5507 | break; |
5508 | } | |
2c47e605 YZ |
5509 | |
5510 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5511 | wc->drop_level = wc->level; |
5512 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5513 | &wc->drop_progress, | |
5514 | path->slots[wc->drop_level]); | |
5515 | } | |
5516 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5517 | &wc->drop_progress); | |
c8422684 | 5518 | btrfs_set_root_drop_level(root_item, wc->drop_level); |
2c47e605 YZ |
5519 | |
5520 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5521 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5522 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5523 | ret = btrfs_update_root(trans, tree_root, |
5524 | &root->root_key, | |
5525 | root_item); | |
79787eaa | 5526 | if (ret) { |
66642832 | 5527 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5528 | err = ret; |
5529 | goto out_end_trans; | |
5530 | } | |
2c47e605 | 5531 | |
3a45bb20 | 5532 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5533 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5534 | btrfs_debug(fs_info, |
5535 | "drop snapshot early exit"); | |
3c8f2422 JB |
5536 | err = -EAGAIN; |
5537 | goto out_free; | |
5538 | } | |
5539 | ||
a22285a6 | 5540 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
5541 | if (IS_ERR(trans)) { |
5542 | err = PTR_ERR(trans); | |
5543 | goto out_free; | |
5544 | } | |
c3e69d58 | 5545 | } |
20524f02 | 5546 | } |
b3b4aa74 | 5547 | btrfs_release_path(path); |
79787eaa JM |
5548 | if (err) |
5549 | goto out_end_trans; | |
2c47e605 | 5550 | |
ab9ce7d4 | 5551 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5552 | if (ret) { |
66642832 | 5553 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5554 | err = ret; |
79787eaa JM |
5555 | goto out_end_trans; |
5556 | } | |
2c47e605 | 5557 | |
76dda93c | 5558 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5559 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5560 | NULL, NULL); | |
79787eaa | 5561 | if (ret < 0) { |
66642832 | 5562 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5563 | err = ret; |
5564 | goto out_end_trans; | |
5565 | } else if (ret > 0) { | |
84cd948c JB |
5566 | /* if we fail to delete the orphan item this time |
5567 | * around, it'll get picked up the next time. | |
5568 | * | |
5569 | * The most common failure here is just -ENOENT. | |
5570 | */ | |
5571 | btrfs_del_orphan_item(trans, tree_root, | |
5572 | root->root_key.objectid); | |
76dda93c YZ |
5573 | } |
5574 | } | |
5575 | ||
a3cf0e43 QW |
5576 | /* |
5577 | * This subvolume is going to be completely dropped, and won't be | |
5578 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
5579 | * commit transaction time. So free it here manually. | |
5580 | */ | |
5581 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
5582 | btrfs_qgroup_free_meta_all_pertrans(root); | |
5583 | ||
8c38938c | 5584 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 5585 | btrfs_add_dropped_root(trans, root); |
8c38938c | 5586 | else |
00246528 | 5587 | btrfs_put_root(root); |
d29a9f62 | 5588 | root_dropped = true; |
79787eaa | 5589 | out_end_trans: |
3a45bb20 | 5590 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5591 | out_free: |
2c47e605 | 5592 | kfree(wc); |
5caf2a00 | 5593 | btrfs_free_path(path); |
cb1b69f4 | 5594 | out: |
d29a9f62 JB |
5595 | /* |
5596 | * So if we need to stop dropping the snapshot for whatever reason we | |
5597 | * need to make sure to add it back to the dead root list so that we | |
5598 | * keep trying to do the work later. This also cleans up roots if we | |
5599 | * don't have it in the radix (like when we recover after a power fail | |
5600 | * or unmount) so we don't leak memory. | |
5601 | */ | |
897ca819 | 5602 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5603 | btrfs_add_dead_root(root); |
2c536799 | 5604 | return err; |
20524f02 | 5605 | } |
9078a3e1 | 5606 | |
2c47e605 YZ |
5607 | /* |
5608 | * drop subtree rooted at tree block 'node'. | |
5609 | * | |
5610 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5611 | * only used by relocation code |
2c47e605 | 5612 | */ |
f82d02d9 YZ |
5613 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5614 | struct btrfs_root *root, | |
5615 | struct extent_buffer *node, | |
5616 | struct extent_buffer *parent) | |
5617 | { | |
0b246afa | 5618 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5619 | struct btrfs_path *path; |
2c47e605 | 5620 | struct walk_control *wc; |
f82d02d9 YZ |
5621 | int level; |
5622 | int parent_level; | |
5623 | int ret = 0; | |
5624 | int wret; | |
5625 | ||
2c47e605 YZ |
5626 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5627 | ||
f82d02d9 | 5628 | path = btrfs_alloc_path(); |
db5b493a TI |
5629 | if (!path) |
5630 | return -ENOMEM; | |
f82d02d9 | 5631 | |
2c47e605 | 5632 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5633 | if (!wc) { |
5634 | btrfs_free_path(path); | |
5635 | return -ENOMEM; | |
5636 | } | |
2c47e605 | 5637 | |
b9447ef8 | 5638 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5639 | parent_level = btrfs_header_level(parent); |
67439dad | 5640 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5641 | path->nodes[parent_level] = parent; |
5642 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5643 | ||
b9447ef8 | 5644 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5645 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5646 | path->nodes[level] = node; |
5647 | path->slots[level] = 0; | |
ac5887c8 | 5648 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5649 | |
5650 | wc->refs[parent_level] = 1; | |
5651 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5652 | wc->level = level; | |
5653 | wc->shared_level = -1; | |
5654 | wc->stage = DROP_REFERENCE; | |
5655 | wc->update_ref = 0; | |
5656 | wc->keep_locks = 1; | |
0b246afa | 5657 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5658 | |
5659 | while (1) { | |
2c47e605 YZ |
5660 | wret = walk_down_tree(trans, root, path, wc); |
5661 | if (wret < 0) { | |
f82d02d9 | 5662 | ret = wret; |
f82d02d9 | 5663 | break; |
2c47e605 | 5664 | } |
f82d02d9 | 5665 | |
2c47e605 | 5666 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5667 | if (wret < 0) |
5668 | ret = wret; | |
5669 | if (wret != 0) | |
5670 | break; | |
5671 | } | |
5672 | ||
2c47e605 | 5673 | kfree(wc); |
f82d02d9 YZ |
5674 | btrfs_free_path(path); |
5675 | return ret; | |
5676 | } | |
5677 | ||
6d07bcec MX |
5678 | /* |
5679 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5680 | * space_info. takes mirrors into account. |
6d07bcec | 5681 | */ |
633c0aad | 5682 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5683 | { |
32da5386 | 5684 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5685 | u64 free_bytes = 0; |
5686 | int factor; | |
5687 | ||
01327610 | 5688 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5689 | if (list_empty(&sinfo->ro_bgs)) |
5690 | return 0; | |
5691 | ||
5692 | spin_lock(&sinfo->lock); | |
5693 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5694 | spin_lock(&block_group->lock); |
5695 | ||
5696 | if (!block_group->ro) { | |
5697 | spin_unlock(&block_group->lock); | |
5698 | continue; | |
5699 | } | |
5700 | ||
46df06b8 | 5701 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5702 | free_bytes += (block_group->length - |
bf38be65 | 5703 | block_group->used) * factor; |
6d07bcec MX |
5704 | |
5705 | spin_unlock(&block_group->lock); | |
5706 | } | |
6d07bcec MX |
5707 | spin_unlock(&sinfo->lock); |
5708 | ||
5709 | return free_bytes; | |
5710 | } | |
5711 | ||
2ff7e61e JM |
5712 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5713 | u64 start, u64 end) | |
acce952b | 5714 | { |
2ff7e61e | 5715 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5716 | } |
5717 | ||
499f377f JM |
5718 | /* |
5719 | * It used to be that old block groups would be left around forever. | |
5720 | * Iterating over them would be enough to trim unused space. Since we | |
5721 | * now automatically remove them, we also need to iterate over unallocated | |
5722 | * space. | |
5723 | * | |
5724 | * We don't want a transaction for this since the discard may take a | |
5725 | * substantial amount of time. We don't require that a transaction be | |
5726 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5727 | * to ensure that we're not discarding chunks that were released or |
5728 | * allocated in the current transaction. | |
499f377f JM |
5729 | * |
5730 | * Holding the chunks lock will prevent other threads from allocating | |
5731 | * or releasing chunks, but it won't prevent a running transaction | |
5732 | * from committing and releasing the memory that the pending chunks | |
5733 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5734 | * transaction and hold the commit root sem. We only need to hold |
5735 | * it while performing the free space search since we have already | |
5736 | * held back allocations. | |
499f377f | 5737 | */ |
8103d10b | 5738 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5739 | { |
8103d10b | 5740 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5741 | int ret; |
5742 | ||
5743 | *trimmed = 0; | |
5744 | ||
0be88e36 JM |
5745 | /* Discard not supported = nothing to do. */ |
5746 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5747 | return 0; | |
5748 | ||
52042d8e | 5749 | /* Not writable = nothing to do. */ |
ebbede42 | 5750 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5751 | return 0; |
5752 | ||
5753 | /* No free space = nothing to do. */ | |
5754 | if (device->total_bytes <= device->bytes_used) | |
5755 | return 0; | |
5756 | ||
5757 | ret = 0; | |
5758 | ||
5759 | while (1) { | |
fb456252 | 5760 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5761 | u64 bytes; |
5762 | ||
5763 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5764 | if (ret) | |
fee7acc3 | 5765 | break; |
499f377f | 5766 | |
929be17a NB |
5767 | find_first_clear_extent_bit(&device->alloc_state, start, |
5768 | &start, &end, | |
5769 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 5770 | |
c57dd1f2 QW |
5771 | /* Check if there are any CHUNK_* bits left */ |
5772 | if (start > device->total_bytes) { | |
5773 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5774 | btrfs_warn_in_rcu(fs_info, | |
5775 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
5776 | start, end - start + 1, | |
5777 | rcu_str_deref(device->name), | |
5778 | device->total_bytes); | |
5779 | mutex_unlock(&fs_info->chunk_mutex); | |
5780 | ret = 0; | |
5781 | break; | |
5782 | } | |
5783 | ||
53460a45 NB |
5784 | /* Ensure we skip the reserved area in the first 1M */ |
5785 | start = max_t(u64, start, SZ_1M); | |
5786 | ||
929be17a NB |
5787 | /* |
5788 | * If find_first_clear_extent_bit find a range that spans the | |
5789 | * end of the device it will set end to -1, in this case it's up | |
5790 | * to the caller to trim the value to the size of the device. | |
5791 | */ | |
5792 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5793 | |
929be17a | 5794 | len = end - start + 1; |
499f377f | 5795 | |
929be17a NB |
5796 | /* We didn't find any extents */ |
5797 | if (!len) { | |
499f377f | 5798 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5799 | ret = 0; |
499f377f JM |
5800 | break; |
5801 | } | |
5802 | ||
929be17a NB |
5803 | ret = btrfs_issue_discard(device->bdev, start, len, |
5804 | &bytes); | |
5805 | if (!ret) | |
5806 | set_extent_bits(&device->alloc_state, start, | |
5807 | start + bytes - 1, | |
5808 | CHUNK_TRIMMED); | |
499f377f JM |
5809 | mutex_unlock(&fs_info->chunk_mutex); |
5810 | ||
5811 | if (ret) | |
5812 | break; | |
5813 | ||
5814 | start += len; | |
5815 | *trimmed += bytes; | |
5816 | ||
5817 | if (fatal_signal_pending(current)) { | |
5818 | ret = -ERESTARTSYS; | |
5819 | break; | |
5820 | } | |
5821 | ||
5822 | cond_resched(); | |
5823 | } | |
5824 | ||
5825 | return ret; | |
5826 | } | |
5827 | ||
93bba24d QW |
5828 | /* |
5829 | * Trim the whole filesystem by: | |
5830 | * 1) trimming the free space in each block group | |
5831 | * 2) trimming the unallocated space on each device | |
5832 | * | |
5833 | * This will also continue trimming even if a block group or device encounters | |
5834 | * an error. The return value will be the last error, or 0 if nothing bad | |
5835 | * happens. | |
5836 | */ | |
2ff7e61e | 5837 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5838 | { |
32da5386 | 5839 | struct btrfs_block_group *cache = NULL; |
499f377f JM |
5840 | struct btrfs_device *device; |
5841 | struct list_head *devices; | |
f7039b1d | 5842 | u64 group_trimmed; |
07301df7 | 5843 | u64 range_end = U64_MAX; |
f7039b1d LD |
5844 | u64 start; |
5845 | u64 end; | |
5846 | u64 trimmed = 0; | |
93bba24d QW |
5847 | u64 bg_failed = 0; |
5848 | u64 dev_failed = 0; | |
5849 | int bg_ret = 0; | |
5850 | int dev_ret = 0; | |
f7039b1d LD |
5851 | int ret = 0; |
5852 | ||
07301df7 QW |
5853 | /* |
5854 | * Check range overflow if range->len is set. | |
5855 | * The default range->len is U64_MAX. | |
5856 | */ | |
5857 | if (range->len != U64_MAX && | |
5858 | check_add_overflow(range->start, range->len, &range_end)) | |
5859 | return -EINVAL; | |
5860 | ||
6ba9fc8e | 5861 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5862 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 5863 | if (cache->start >= range_end) { |
f7039b1d LD |
5864 | btrfs_put_block_group(cache); |
5865 | break; | |
5866 | } | |
5867 | ||
b3470b5d DS |
5868 | start = max(range->start, cache->start); |
5869 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
5870 | |
5871 | if (end - start >= range->minlen) { | |
32da5386 | 5872 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 5873 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 5874 | if (ret) { |
93bba24d QW |
5875 | bg_failed++; |
5876 | bg_ret = ret; | |
5877 | continue; | |
1be41b78 | 5878 | } |
676f1f75 | 5879 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5880 | if (ret) { |
93bba24d QW |
5881 | bg_failed++; |
5882 | bg_ret = ret; | |
5883 | continue; | |
1be41b78 | 5884 | } |
f7039b1d LD |
5885 | } |
5886 | ret = btrfs_trim_block_group(cache, | |
5887 | &group_trimmed, | |
5888 | start, | |
5889 | end, | |
5890 | range->minlen); | |
5891 | ||
5892 | trimmed += group_trimmed; | |
5893 | if (ret) { | |
93bba24d QW |
5894 | bg_failed++; |
5895 | bg_ret = ret; | |
5896 | continue; | |
f7039b1d LD |
5897 | } |
5898 | } | |
f7039b1d LD |
5899 | } |
5900 | ||
93bba24d QW |
5901 | if (bg_failed) |
5902 | btrfs_warn(fs_info, | |
5903 | "failed to trim %llu block group(s), last error %d", | |
5904 | bg_failed, bg_ret); | |
0b246afa | 5905 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5906 | devices = &fs_info->fs_devices->devices; |
5907 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5908 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5909 | if (ret) { |
5910 | dev_failed++; | |
5911 | dev_ret = ret; | |
499f377f | 5912 | break; |
93bba24d | 5913 | } |
499f377f JM |
5914 | |
5915 | trimmed += group_trimmed; | |
5916 | } | |
0b246afa | 5917 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5918 | |
93bba24d QW |
5919 | if (dev_failed) |
5920 | btrfs_warn(fs_info, | |
5921 | "failed to trim %llu device(s), last error %d", | |
5922 | dev_failed, dev_ret); | |
f7039b1d | 5923 | range->len = trimmed; |
93bba24d QW |
5924 | if (bg_ret) |
5925 | return bg_ret; | |
5926 | return dev_ret; | |
f7039b1d | 5927 | } |