]>
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); | |
9a664971 | 847 | path->search_for_extension = 1; |
85d4198e | 848 | path->keep_locks = 1; |
5d4f98a2 YZ |
849 | } else |
850 | extra_size = -1; | |
3173a18f JB |
851 | |
852 | /* | |
16d1c062 NB |
853 | * Owner is our level, so we can just add one to get the level for the |
854 | * block we are interested in. | |
3173a18f JB |
855 | */ |
856 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
857 | key.type = BTRFS_METADATA_ITEM_KEY; | |
858 | key.offset = owner; | |
859 | } | |
860 | ||
861 | again: | |
5d4f98a2 | 862 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 863 | if (ret < 0) { |
5d4f98a2 YZ |
864 | err = ret; |
865 | goto out; | |
866 | } | |
3173a18f JB |
867 | |
868 | /* | |
869 | * We may be a newly converted file system which still has the old fat | |
870 | * extent entries for metadata, so try and see if we have one of those. | |
871 | */ | |
872 | if (ret > 0 && skinny_metadata) { | |
873 | skinny_metadata = false; | |
874 | if (path->slots[0]) { | |
875 | path->slots[0]--; | |
876 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
877 | path->slots[0]); | |
878 | if (key.objectid == bytenr && | |
879 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
880 | key.offset == num_bytes) | |
881 | ret = 0; | |
882 | } | |
883 | if (ret) { | |
9ce49a0b | 884 | key.objectid = bytenr; |
3173a18f JB |
885 | key.type = BTRFS_EXTENT_ITEM_KEY; |
886 | key.offset = num_bytes; | |
887 | btrfs_release_path(path); | |
888 | goto again; | |
889 | } | |
890 | } | |
891 | ||
79787eaa JM |
892 | if (ret && !insert) { |
893 | err = -ENOENT; | |
894 | goto out; | |
fae7f21c | 895 | } else if (WARN_ON(ret)) { |
492104c8 | 896 | err = -EIO; |
492104c8 | 897 | goto out; |
79787eaa | 898 | } |
5d4f98a2 YZ |
899 | |
900 | leaf = path->nodes[0]; | |
901 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 902 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
903 | err = -EINVAL; |
904 | btrfs_print_v0_err(fs_info); | |
905 | btrfs_abort_transaction(trans, err); | |
906 | goto out; | |
907 | } | |
5d4f98a2 | 908 | |
5d4f98a2 YZ |
909 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
910 | flags = btrfs_extent_flags(leaf, ei); | |
911 | ||
912 | ptr = (unsigned long)(ei + 1); | |
913 | end = (unsigned long)ei + item_size; | |
914 | ||
3173a18f | 915 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
916 | ptr += sizeof(struct btrfs_tree_block_info); |
917 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
918 | } |
919 | ||
3de28d57 LB |
920 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
921 | needed = BTRFS_REF_TYPE_DATA; | |
922 | else | |
923 | needed = BTRFS_REF_TYPE_BLOCK; | |
924 | ||
5d4f98a2 YZ |
925 | err = -ENOENT; |
926 | while (1) { | |
927 | if (ptr >= end) { | |
928 | WARN_ON(ptr > end); | |
929 | break; | |
930 | } | |
931 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
932 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
933 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 934 | err = -EUCLEAN; |
3de28d57 LB |
935 | goto out; |
936 | } | |
937 | ||
5d4f98a2 YZ |
938 | if (want < type) |
939 | break; | |
940 | if (want > type) { | |
941 | ptr += btrfs_extent_inline_ref_size(type); | |
942 | continue; | |
943 | } | |
944 | ||
945 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
946 | struct btrfs_extent_data_ref *dref; | |
947 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
948 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
949 | owner, offset)) { | |
950 | err = 0; | |
951 | break; | |
952 | } | |
953 | if (hash_extent_data_ref_item(leaf, dref) < | |
954 | hash_extent_data_ref(root_objectid, owner, offset)) | |
955 | break; | |
956 | } else { | |
957 | u64 ref_offset; | |
958 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
959 | if (parent > 0) { | |
960 | if (parent == ref_offset) { | |
961 | err = 0; | |
962 | break; | |
963 | } | |
964 | if (ref_offset < parent) | |
965 | break; | |
966 | } else { | |
967 | if (root_objectid == ref_offset) { | |
968 | err = 0; | |
969 | break; | |
970 | } | |
971 | if (ref_offset < root_objectid) | |
972 | break; | |
973 | } | |
974 | } | |
975 | ptr += btrfs_extent_inline_ref_size(type); | |
976 | } | |
977 | if (err == -ENOENT && insert) { | |
978 | if (item_size + extra_size >= | |
979 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
980 | err = -EAGAIN; | |
981 | goto out; | |
982 | } | |
983 | /* | |
984 | * To add new inline back ref, we have to make sure | |
985 | * there is no corresponding back ref item. | |
986 | * For simplicity, we just do not add new inline back | |
987 | * ref if there is any kind of item for this block | |
988 | */ | |
2c47e605 YZ |
989 | if (find_next_key(path, 0, &key) == 0 && |
990 | key.objectid == bytenr && | |
85d4198e | 991 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
992 | err = -EAGAIN; |
993 | goto out; | |
994 | } | |
995 | } | |
996 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
997 | out: | |
85d4198e | 998 | if (insert) { |
5d4f98a2 | 999 | path->keep_locks = 0; |
9a664971 | 1000 | path->search_for_extension = 0; |
5d4f98a2 YZ |
1001 | btrfs_unlock_up_safe(path, 1); |
1002 | } | |
1003 | return err; | |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * helper to add new inline back ref | |
1008 | */ | |
1009 | static noinline_for_stack | |
87bde3cd | 1010 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1011 | struct btrfs_path *path, |
1012 | struct btrfs_extent_inline_ref *iref, | |
1013 | u64 parent, u64 root_objectid, | |
1014 | u64 owner, u64 offset, int refs_to_add, | |
1015 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1016 | { |
1017 | struct extent_buffer *leaf; | |
1018 | struct btrfs_extent_item *ei; | |
1019 | unsigned long ptr; | |
1020 | unsigned long end; | |
1021 | unsigned long item_offset; | |
1022 | u64 refs; | |
1023 | int size; | |
1024 | int type; | |
5d4f98a2 YZ |
1025 | |
1026 | leaf = path->nodes[0]; | |
1027 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1028 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1029 | ||
1030 | type = extent_ref_type(parent, owner); | |
1031 | size = btrfs_extent_inline_ref_size(type); | |
1032 | ||
c71dd880 | 1033 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1034 | |
1035 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1036 | refs = btrfs_extent_refs(leaf, ei); | |
1037 | refs += refs_to_add; | |
1038 | btrfs_set_extent_refs(leaf, ei, refs); | |
1039 | if (extent_op) | |
1040 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1041 | ||
1042 | ptr = (unsigned long)ei + item_offset; | |
1043 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1044 | if (ptr < end - size) | |
1045 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1046 | end - size - ptr); | |
1047 | ||
1048 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1049 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1050 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1051 | struct btrfs_extent_data_ref *dref; | |
1052 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1053 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1054 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1055 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1056 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1057 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1058 | struct btrfs_shared_data_ref *sref; | |
1059 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1060 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1061 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1062 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1063 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1064 | } else { | |
1065 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1066 | } | |
1067 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1068 | } |
1069 | ||
1070 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1071 | struct btrfs_path *path, |
1072 | struct btrfs_extent_inline_ref **ref_ret, | |
1073 | u64 bytenr, u64 num_bytes, u64 parent, | |
1074 | u64 root_objectid, u64 owner, u64 offset) | |
1075 | { | |
1076 | int ret; | |
1077 | ||
867cc1fb NB |
1078 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1079 | num_bytes, parent, root_objectid, | |
1080 | owner, offset, 0); | |
5d4f98a2 | 1081 | if (ret != -ENOENT) |
54aa1f4d | 1082 | return ret; |
5d4f98a2 | 1083 | |
b3b4aa74 | 1084 | btrfs_release_path(path); |
5d4f98a2 YZ |
1085 | *ref_ret = NULL; |
1086 | ||
1087 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1088 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1089 | root_objectid); | |
5d4f98a2 | 1090 | } else { |
bd1d53ef NB |
1091 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1092 | root_objectid, owner, offset); | |
b9473439 | 1093 | } |
5d4f98a2 YZ |
1094 | return ret; |
1095 | } | |
31840ae1 | 1096 | |
5d4f98a2 YZ |
1097 | /* |
1098 | * helper to update/remove inline back ref | |
1099 | */ | |
1100 | static noinline_for_stack | |
61a18f1c | 1101 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1102 | struct btrfs_extent_inline_ref *iref, |
1103 | int refs_to_mod, | |
fcebe456 JB |
1104 | struct btrfs_delayed_extent_op *extent_op, |
1105 | int *last_ref) | |
5d4f98a2 | 1106 | { |
61a18f1c | 1107 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1108 | struct btrfs_extent_item *ei; |
1109 | struct btrfs_extent_data_ref *dref = NULL; | |
1110 | struct btrfs_shared_data_ref *sref = NULL; | |
1111 | unsigned long ptr; | |
1112 | unsigned long end; | |
1113 | u32 item_size; | |
1114 | int size; | |
1115 | int type; | |
5d4f98a2 YZ |
1116 | u64 refs; |
1117 | ||
5d4f98a2 YZ |
1118 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1119 | refs = btrfs_extent_refs(leaf, ei); | |
1120 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1121 | refs += refs_to_mod; | |
1122 | btrfs_set_extent_refs(leaf, ei, refs); | |
1123 | if (extent_op) | |
1124 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1125 | ||
3de28d57 LB |
1126 | /* |
1127 | * If type is invalid, we should have bailed out after | |
1128 | * lookup_inline_extent_backref(). | |
1129 | */ | |
1130 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1131 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1132 | |
1133 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1134 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1135 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1136 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1137 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1138 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1139 | } else { | |
1140 | refs = 1; | |
1141 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1142 | } |
31840ae1 | 1143 | |
5d4f98a2 YZ |
1144 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1145 | refs += refs_to_mod; | |
1146 | ||
1147 | if (refs > 0) { | |
1148 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1149 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1150 | else | |
1151 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1152 | } else { | |
fcebe456 | 1153 | *last_ref = 1; |
5d4f98a2 YZ |
1154 | size = btrfs_extent_inline_ref_size(type); |
1155 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1156 | ptr = (unsigned long)iref; | |
1157 | end = (unsigned long)ei + item_size; | |
1158 | if (ptr + size < end) | |
1159 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1160 | end - ptr - size); | |
1161 | item_size -= size; | |
78ac4f9e | 1162 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1163 | } |
1164 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1165 | } |
1166 | ||
1167 | static noinline_for_stack | |
1168 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1169 | struct btrfs_path *path, |
1170 | u64 bytenr, u64 num_bytes, u64 parent, | |
1171 | u64 root_objectid, u64 owner, | |
1172 | u64 offset, int refs_to_add, | |
1173 | struct btrfs_delayed_extent_op *extent_op) | |
1174 | { | |
1175 | struct btrfs_extent_inline_ref *iref; | |
1176 | int ret; | |
1177 | ||
867cc1fb NB |
1178 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1179 | num_bytes, parent, root_objectid, | |
1180 | owner, offset, 1); | |
5d4f98a2 | 1181 | if (ret == 0) { |
07cce5cf QW |
1182 | /* |
1183 | * We're adding refs to a tree block we already own, this | |
1184 | * should not happen at all. | |
1185 | */ | |
1186 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1187 | btrfs_crit(trans->fs_info, | |
1188 | "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu", | |
1189 | bytenr, num_bytes, root_objectid); | |
1190 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
1191 | WARN_ON(1); | |
1192 | btrfs_crit(trans->fs_info, | |
1193 | "path->slots[0]=%d path->nodes[0]:", path->slots[0]); | |
1194 | btrfs_print_leaf(path->nodes[0]); | |
1195 | } | |
1196 | return -EUCLEAN; | |
1197 | } | |
61a18f1c NB |
1198 | update_inline_extent_backref(path, iref, refs_to_add, |
1199 | extent_op, NULL); | |
5d4f98a2 | 1200 | } else if (ret == -ENOENT) { |
a639cdeb | 1201 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1202 | root_objectid, owner, offset, |
1203 | refs_to_add, extent_op); | |
1204 | ret = 0; | |
771ed689 | 1205 | } |
5d4f98a2 YZ |
1206 | return ret; |
1207 | } | |
31840ae1 | 1208 | |
5d4f98a2 | 1209 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1210 | struct btrfs_path *path, |
1211 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1212 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1213 | { |
143bede5 | 1214 | int ret = 0; |
b9473439 | 1215 | |
5d4f98a2 YZ |
1216 | BUG_ON(!is_data && refs_to_drop != 1); |
1217 | if (iref) { | |
61a18f1c NB |
1218 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1219 | last_ref); | |
5d4f98a2 | 1220 | } else if (is_data) { |
e9f6290d | 1221 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1222 | last_ref); |
5d4f98a2 | 1223 | } else { |
fcebe456 | 1224 | *last_ref = 1; |
87cc7a8a | 1225 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1226 | } |
1227 | return ret; | |
1228 | } | |
1229 | ||
d04c6b88 JM |
1230 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1231 | u64 *discarded_bytes) | |
5d4f98a2 | 1232 | { |
86557861 JM |
1233 | int j, ret = 0; |
1234 | u64 bytes_left, end; | |
4d89d377 | 1235 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1236 | |
4d89d377 JM |
1237 | if (WARN_ON(start != aligned_start)) { |
1238 | len -= aligned_start - start; | |
1239 | len = round_down(len, 1 << 9); | |
1240 | start = aligned_start; | |
1241 | } | |
d04c6b88 | 1242 | |
4d89d377 | 1243 | *discarded_bytes = 0; |
86557861 JM |
1244 | |
1245 | if (!len) | |
1246 | return 0; | |
1247 | ||
1248 | end = start + len; | |
1249 | bytes_left = len; | |
1250 | ||
1251 | /* Skip any superblocks on this device. */ | |
1252 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1253 | u64 sb_start = btrfs_sb_offset(j); | |
1254 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1255 | u64 size = sb_start - start; | |
1256 | ||
1257 | if (!in_range(sb_start, start, bytes_left) && | |
1258 | !in_range(sb_end, start, bytes_left) && | |
1259 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1260 | continue; | |
1261 | ||
1262 | /* | |
1263 | * Superblock spans beginning of range. Adjust start and | |
1264 | * try again. | |
1265 | */ | |
1266 | if (sb_start <= start) { | |
1267 | start += sb_end - start; | |
1268 | if (start > end) { | |
1269 | bytes_left = 0; | |
1270 | break; | |
1271 | } | |
1272 | bytes_left = end - start; | |
1273 | continue; | |
1274 | } | |
1275 | ||
1276 | if (size) { | |
1277 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1278 | GFP_NOFS, 0); | |
1279 | if (!ret) | |
1280 | *discarded_bytes += size; | |
1281 | else if (ret != -EOPNOTSUPP) | |
1282 | return ret; | |
1283 | } | |
1284 | ||
1285 | start = sb_end; | |
1286 | if (start > end) { | |
1287 | bytes_left = 0; | |
1288 | break; | |
1289 | } | |
1290 | bytes_left = end - start; | |
1291 | } | |
1292 | ||
1293 | if (bytes_left) { | |
1294 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1295 | GFP_NOFS, 0); |
1296 | if (!ret) | |
86557861 | 1297 | *discarded_bytes += bytes_left; |
4d89d377 | 1298 | } |
d04c6b88 | 1299 | return ret; |
5d4f98a2 | 1300 | } |
5d4f98a2 | 1301 | |
2ff7e61e | 1302 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1303 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1304 | { |
6b7faadd | 1305 | int ret = 0; |
5378e607 | 1306 | u64 discarded_bytes = 0; |
6b7faadd QW |
1307 | u64 end = bytenr + num_bytes; |
1308 | u64 cur = bytenr; | |
a1d3c478 | 1309 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1310 | |
e244a0ae | 1311 | |
2999241d FM |
1312 | /* |
1313 | * Avoid races with device replace and make sure our bbio has devices | |
1314 | * associated to its stripes that don't go away while we are discarding. | |
1315 | */ | |
0b246afa | 1316 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd QW |
1317 | while (cur < end) { |
1318 | struct btrfs_bio_stripe *stripe; | |
5d4f98a2 YZ |
1319 | int i; |
1320 | ||
6b7faadd QW |
1321 | num_bytes = end - cur; |
1322 | /* Tell the block device(s) that the sectors can be discarded */ | |
1323 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
1324 | &num_bytes, &bbio, 0); | |
1325 | /* | |
1326 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1327 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1328 | * thus we can't continue anyway. | |
1329 | */ | |
1330 | if (ret < 0) | |
1331 | goto out; | |
5d4f98a2 | 1332 | |
6b7faadd | 1333 | stripe = bbio->stripes; |
a1d3c478 | 1334 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1335 | u64 bytes; |
38b5f68e AJ |
1336 | struct request_queue *req_q; |
1337 | ||
627e0873 FM |
1338 | if (!stripe->dev->bdev) { |
1339 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1340 | continue; | |
1341 | } | |
38b5f68e AJ |
1342 | req_q = bdev_get_queue(stripe->dev->bdev); |
1343 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1344 | continue; |
1345 | ||
5378e607 LD |
1346 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1347 | stripe->physical, | |
d04c6b88 JM |
1348 | stripe->length, |
1349 | &bytes); | |
6b7faadd | 1350 | if (!ret) { |
d04c6b88 | 1351 | discarded_bytes += bytes; |
6b7faadd QW |
1352 | } else if (ret != -EOPNOTSUPP) { |
1353 | /* | |
1354 | * Logic errors or -ENOMEM, or -EIO, but | |
1355 | * unlikely to happen. | |
1356 | * | |
1357 | * And since there are two loops, explicitly | |
1358 | * go to out to avoid confusion. | |
1359 | */ | |
1360 | btrfs_put_bbio(bbio); | |
1361 | goto out; | |
1362 | } | |
d5e2003c JB |
1363 | |
1364 | /* | |
1365 | * Just in case we get back EOPNOTSUPP for some reason, | |
1366 | * just ignore the return value so we don't screw up | |
1367 | * people calling discard_extent. | |
1368 | */ | |
1369 | ret = 0; | |
5d4f98a2 | 1370 | } |
6e9606d2 | 1371 | btrfs_put_bbio(bbio); |
6b7faadd | 1372 | cur += num_bytes; |
5d4f98a2 | 1373 | } |
6b7faadd | 1374 | out: |
0b246afa | 1375 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1376 | |
1377 | if (actual_bytes) | |
1378 | *actual_bytes = discarded_bytes; | |
1379 | ||
5d4f98a2 | 1380 | |
53b381b3 DW |
1381 | if (ret == -EOPNOTSUPP) |
1382 | ret = 0; | |
5d4f98a2 | 1383 | return ret; |
5d4f98a2 YZ |
1384 | } |
1385 | ||
79787eaa | 1386 | /* Can return -ENOMEM */ |
5d4f98a2 | 1387 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1388 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1389 | { |
82fa113f | 1390 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1391 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1392 | int ret; |
66d7e7f0 | 1393 | |
82fa113f QW |
1394 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1395 | generic_ref->action); | |
1396 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1397 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1398 | |
82fa113f QW |
1399 | if (generic_ref->type == BTRFS_REF_METADATA) |
1400 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1401 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1402 | else |
1403 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1404 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1405 | |
82fa113f | 1406 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1407 | |
ddf30cf0 | 1408 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1409 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 1410 | |
5d4f98a2 YZ |
1411 | return ret; |
1412 | } | |
1413 | ||
bd3c685e NB |
1414 | /* |
1415 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1416 | * | |
07cce5cf QW |
1417 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1418 | * how it works. | |
1419 | * | |
bd3c685e NB |
1420 | * @trans: Handle of transaction |
1421 | * | |
1422 | * @node: The delayed ref node used to get the bytenr/length for | |
1423 | * extent whose references are incremented. | |
1424 | * | |
1425 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1426 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1427 | * bytenr of the parent block. Since new extents are always | |
1428 | * created with indirect references, this will only be the case | |
1429 | * when relocating a shared extent. In that case, root_objectid | |
1430 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1431 | * be 0 | |
1432 | * | |
1433 | * @root_objectid: The id of the root where this modification has originated, | |
1434 | * this can be either one of the well-known metadata trees or | |
1435 | * the subvolume id which references this extent. | |
1436 | * | |
1437 | * @owner: For data extents it is the inode number of the owning file. | |
1438 | * For metadata extents this parameter holds the level in the | |
1439 | * tree of the extent. | |
1440 | * | |
1441 | * @offset: For metadata extents the offset is ignored and is currently | |
1442 | * always passed as 0. For data extents it is the fileoffset | |
1443 | * this extent belongs to. | |
1444 | * | |
1445 | * @refs_to_add Number of references to add | |
1446 | * | |
1447 | * @extent_op Pointer to a structure, holding information necessary when | |
1448 | * updating a tree block's flags | |
1449 | * | |
1450 | */ | |
5d4f98a2 | 1451 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1452 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1453 | u64 parent, u64 root_objectid, |
1454 | u64 owner, u64 offset, int refs_to_add, | |
1455 | struct btrfs_delayed_extent_op *extent_op) | |
1456 | { | |
1457 | struct btrfs_path *path; | |
1458 | struct extent_buffer *leaf; | |
1459 | struct btrfs_extent_item *item; | |
fcebe456 | 1460 | struct btrfs_key key; |
c682f9b3 QW |
1461 | u64 bytenr = node->bytenr; |
1462 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1463 | u64 refs; |
1464 | int ret; | |
5d4f98a2 YZ |
1465 | |
1466 | path = btrfs_alloc_path(); | |
1467 | if (!path) | |
1468 | return -ENOMEM; | |
1469 | ||
5d4f98a2 | 1470 | /* this will setup the path even if it fails to insert the back ref */ |
a639cdeb NB |
1471 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1472 | parent, root_objectid, owner, | |
1473 | offset, refs_to_add, extent_op); | |
0ed4792a | 1474 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1475 | goto out; |
fcebe456 JB |
1476 | |
1477 | /* | |
1478 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1479 | * inline extent ref, so just update the reference count and add a | |
1480 | * normal backref. | |
1481 | */ | |
5d4f98a2 | 1482 | leaf = path->nodes[0]; |
fcebe456 | 1483 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1484 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1485 | refs = btrfs_extent_refs(leaf, item); | |
1486 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1487 | if (extent_op) | |
1488 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1489 | |
5d4f98a2 | 1490 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1491 | btrfs_release_path(path); |
56bec294 | 1492 | |
56bec294 | 1493 | /* now insert the actual backref */ |
65cd6d9e NB |
1494 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
1495 | BUG_ON(refs_to_add != 1); | |
1496 | ret = insert_tree_block_ref(trans, path, bytenr, parent, | |
1497 | root_objectid); | |
1498 | } else { | |
1499 | ret = insert_extent_data_ref(trans, path, bytenr, parent, | |
1500 | root_objectid, owner, offset, | |
1501 | refs_to_add); | |
1502 | } | |
79787eaa | 1503 | if (ret) |
66642832 | 1504 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1505 | out: |
56bec294 | 1506 | btrfs_free_path(path); |
30d133fc | 1507 | return ret; |
56bec294 CM |
1508 | } |
1509 | ||
5d4f98a2 | 1510 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1511 | struct btrfs_delayed_ref_node *node, |
1512 | struct btrfs_delayed_extent_op *extent_op, | |
1513 | int insert_reserved) | |
56bec294 | 1514 | { |
5d4f98a2 YZ |
1515 | int ret = 0; |
1516 | struct btrfs_delayed_data_ref *ref; | |
1517 | struct btrfs_key ins; | |
1518 | u64 parent = 0; | |
1519 | u64 ref_root = 0; | |
1520 | u64 flags = 0; | |
1521 | ||
1522 | ins.objectid = node->bytenr; | |
1523 | ins.offset = node->num_bytes; | |
1524 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1525 | ||
1526 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1527 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1528 | |
5d4f98a2 YZ |
1529 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1530 | parent = ref->parent; | |
fcebe456 | 1531 | ref_root = ref->root; |
5d4f98a2 YZ |
1532 | |
1533 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1534 | if (extent_op) |
5d4f98a2 | 1535 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1536 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1537 | flags, ref->objectid, | |
1538 | ref->offset, &ins, | |
1539 | node->ref_mod); | |
5d4f98a2 | 1540 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1541 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1542 | ref->objectid, ref->offset, | |
1543 | node->ref_mod, extent_op); | |
5d4f98a2 | 1544 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1545 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1546 | ref_root, ref->objectid, |
1547 | ref->offset, node->ref_mod, | |
c682f9b3 | 1548 | extent_op); |
5d4f98a2 YZ |
1549 | } else { |
1550 | BUG(); | |
1551 | } | |
1552 | return ret; | |
1553 | } | |
1554 | ||
1555 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1556 | struct extent_buffer *leaf, | |
1557 | struct btrfs_extent_item *ei) | |
1558 | { | |
1559 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1560 | if (extent_op->update_flags) { | |
1561 | flags |= extent_op->flags_to_set; | |
1562 | btrfs_set_extent_flags(leaf, ei, flags); | |
1563 | } | |
1564 | ||
1565 | if (extent_op->update_key) { | |
1566 | struct btrfs_tree_block_info *bi; | |
1567 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1568 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1569 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1570 | } | |
1571 | } | |
1572 | ||
1573 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1574 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1575 | struct btrfs_delayed_extent_op *extent_op) |
1576 | { | |
20b9a2d6 | 1577 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1578 | struct btrfs_key key; |
1579 | struct btrfs_path *path; | |
1580 | struct btrfs_extent_item *ei; | |
1581 | struct extent_buffer *leaf; | |
1582 | u32 item_size; | |
56bec294 | 1583 | int ret; |
5d4f98a2 | 1584 | int err = 0; |
b1c79e09 | 1585 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1586 | |
bf31f87f | 1587 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1588 | return 0; |
1589 | ||
0b246afa | 1590 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1591 | metadata = 0; |
1592 | ||
5d4f98a2 YZ |
1593 | path = btrfs_alloc_path(); |
1594 | if (!path) | |
1595 | return -ENOMEM; | |
1596 | ||
d278850e | 1597 | key.objectid = head->bytenr; |
5d4f98a2 | 1598 | |
3173a18f | 1599 | if (metadata) { |
3173a18f | 1600 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1601 | key.offset = extent_op->level; |
3173a18f JB |
1602 | } else { |
1603 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1604 | key.offset = head->num_bytes; |
3173a18f JB |
1605 | } |
1606 | ||
1607 | again: | |
0b246afa | 1608 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1609 | if (ret < 0) { |
1610 | err = ret; | |
1611 | goto out; | |
1612 | } | |
1613 | if (ret > 0) { | |
3173a18f | 1614 | if (metadata) { |
55994887 FDBM |
1615 | if (path->slots[0] > 0) { |
1616 | path->slots[0]--; | |
1617 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1618 | path->slots[0]); | |
d278850e | 1619 | if (key.objectid == head->bytenr && |
55994887 | 1620 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1621 | key.offset == head->num_bytes) |
55994887 FDBM |
1622 | ret = 0; |
1623 | } | |
1624 | if (ret > 0) { | |
1625 | btrfs_release_path(path); | |
1626 | metadata = 0; | |
3173a18f | 1627 | |
d278850e JB |
1628 | key.objectid = head->bytenr; |
1629 | key.offset = head->num_bytes; | |
55994887 FDBM |
1630 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1631 | goto again; | |
1632 | } | |
1633 | } else { | |
1634 | err = -EIO; | |
1635 | goto out; | |
3173a18f | 1636 | } |
5d4f98a2 YZ |
1637 | } |
1638 | ||
1639 | leaf = path->nodes[0]; | |
1640 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1641 | |
6d8ff4e4 | 1642 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1643 | err = -EINVAL; |
1644 | btrfs_print_v0_err(fs_info); | |
1645 | btrfs_abort_transaction(trans, err); | |
1646 | goto out; | |
1647 | } | |
1648 | ||
5d4f98a2 YZ |
1649 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1650 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1651 | |
5d4f98a2 YZ |
1652 | btrfs_mark_buffer_dirty(leaf); |
1653 | out: | |
1654 | btrfs_free_path(path); | |
1655 | return err; | |
56bec294 CM |
1656 | } |
1657 | ||
5d4f98a2 | 1658 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1659 | struct btrfs_delayed_ref_node *node, |
1660 | struct btrfs_delayed_extent_op *extent_op, | |
1661 | int insert_reserved) | |
56bec294 CM |
1662 | { |
1663 | int ret = 0; | |
5d4f98a2 | 1664 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1665 | u64 parent = 0; |
1666 | u64 ref_root = 0; | |
56bec294 | 1667 | |
5d4f98a2 | 1668 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1669 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1670 | |
5d4f98a2 YZ |
1671 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1672 | parent = ref->parent; | |
fcebe456 | 1673 | ref_root = ref->root; |
5d4f98a2 | 1674 | |
02794222 | 1675 | if (node->ref_mod != 1) { |
f97806f2 | 1676 | btrfs_err(trans->fs_info, |
02794222 LB |
1677 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1678 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1679 | parent); | |
1680 | return -EIO; | |
1681 | } | |
5d4f98a2 | 1682 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1683 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1684 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1685 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1686 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1687 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1688 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1689 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1690 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1691 | } else { |
1692 | BUG(); | |
1693 | } | |
56bec294 CM |
1694 | return ret; |
1695 | } | |
1696 | ||
1697 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1698 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1699 | struct btrfs_delayed_ref_node *node, |
1700 | struct btrfs_delayed_extent_op *extent_op, | |
1701 | int insert_reserved) | |
56bec294 | 1702 | { |
79787eaa JM |
1703 | int ret = 0; |
1704 | ||
bf31f87f | 1705 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1706 | if (insert_reserved) |
b25c36f8 | 1707 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1708 | return 0; |
857cc2fc | 1709 | } |
79787eaa | 1710 | |
5d4f98a2 YZ |
1711 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1712 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1713 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1714 | insert_reserved); |
1715 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1716 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1717 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1718 | insert_reserved); |
1719 | else | |
1720 | BUG(); | |
80ee54bf | 1721 | if (ret && insert_reserved) |
b25c36f8 | 1722 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
5d4f98a2 | 1723 | return ret; |
56bec294 CM |
1724 | } |
1725 | ||
c6fc2454 | 1726 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1727 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1728 | { | |
cffc3374 FM |
1729 | struct btrfs_delayed_ref_node *ref; |
1730 | ||
e3d03965 | 1731 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1732 | return NULL; |
d7df2c79 | 1733 | |
cffc3374 FM |
1734 | /* |
1735 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1736 | * This is to prevent a ref count from going down to zero, which deletes | |
1737 | * the extent item from the extent tree, when there still are references | |
1738 | * to add, which would fail because they would not find the extent item. | |
1739 | */ | |
1d57ee94 WX |
1740 | if (!list_empty(&head->ref_add_list)) |
1741 | return list_first_entry(&head->ref_add_list, | |
1742 | struct btrfs_delayed_ref_node, add_list); | |
1743 | ||
e3d03965 | 1744 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1745 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1746 | ASSERT(list_empty(&ref->add_list)); |
1747 | return ref; | |
56bec294 CM |
1748 | } |
1749 | ||
2eadaa22 JB |
1750 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1751 | struct btrfs_delayed_ref_head *head) | |
1752 | { | |
1753 | spin_lock(&delayed_refs->lock); | |
1754 | head->processing = 0; | |
1755 | delayed_refs->num_heads_ready++; | |
1756 | spin_unlock(&delayed_refs->lock); | |
1757 | btrfs_delayed_ref_unlock(head); | |
1758 | } | |
1759 | ||
bedc6617 JB |
1760 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1761 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1762 | { |
1763 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1764 | |
1765 | if (!extent_op) | |
bedc6617 JB |
1766 | return NULL; |
1767 | ||
b00e6250 | 1768 | if (head->must_insert_reserved) { |
bedc6617 | 1769 | head->extent_op = NULL; |
b00e6250 | 1770 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1771 | return NULL; |
b00e6250 | 1772 | } |
bedc6617 JB |
1773 | return extent_op; |
1774 | } | |
1775 | ||
1776 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1777 | struct btrfs_delayed_ref_head *head) | |
1778 | { | |
1779 | struct btrfs_delayed_extent_op *extent_op; | |
1780 | int ret; | |
1781 | ||
1782 | extent_op = cleanup_extent_op(head); | |
1783 | if (!extent_op) | |
1784 | return 0; | |
1785 | head->extent_op = NULL; | |
b00e6250 | 1786 | spin_unlock(&head->lock); |
20b9a2d6 | 1787 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1788 | btrfs_free_delayed_extent_op(extent_op); |
1789 | return ret ? ret : 1; | |
1790 | } | |
1791 | ||
31890da0 JB |
1792 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1793 | struct btrfs_delayed_ref_root *delayed_refs, | |
1794 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1795 | { |
ba2c4d4e | 1796 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
1797 | |
1798 | if (head->total_ref_mod < 0) { | |
1799 | struct btrfs_space_info *space_info; | |
1800 | u64 flags; | |
1801 | ||
1802 | if (head->is_data) | |
1803 | flags = BTRFS_BLOCK_GROUP_DATA; | |
1804 | else if (head->is_system) | |
1805 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
1806 | else | |
1807 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 1808 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
1809 | ASSERT(space_info); |
1810 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1811 | -head->num_bytes, | |
1812 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1813 | ||
ba2c4d4e JB |
1814 | /* |
1815 | * We had csum deletions accounted for in our delayed refs rsv, | |
1816 | * we need to drop the csum leaves for this update from our | |
1817 | * delayed_refs_rsv. | |
1818 | */ | |
07c47775 JB |
1819 | if (head->is_data) { |
1820 | spin_lock(&delayed_refs->lock); | |
1821 | delayed_refs->pending_csums -= head->num_bytes; | |
1822 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
1823 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
1824 | head->num_bytes); | |
07c47775 JB |
1825 | } |
1826 | } | |
1827 | ||
ba2c4d4e | 1828 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1829 | } |
1830 | ||
194ab0bc | 1831 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1832 | struct btrfs_delayed_ref_head *head) |
1833 | { | |
f9871edd NB |
1834 | |
1835 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1836 | struct btrfs_delayed_ref_root *delayed_refs; |
1837 | int ret; | |
1838 | ||
1839 | delayed_refs = &trans->transaction->delayed_refs; | |
1840 | ||
bedc6617 | 1841 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1842 | if (ret < 0) { |
1843 | unselect_delayed_ref_head(delayed_refs, head); | |
1844 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1845 | return ret; | |
1846 | } else if (ret) { | |
1847 | return ret; | |
1848 | } | |
1849 | ||
1850 | /* | |
1851 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1852 | * and then re-check to make sure nobody got added. | |
1853 | */ | |
1854 | spin_unlock(&head->lock); | |
1855 | spin_lock(&delayed_refs->lock); | |
1856 | spin_lock(&head->lock); | |
e3d03965 | 1857 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1858 | spin_unlock(&head->lock); |
1859 | spin_unlock(&delayed_refs->lock); | |
1860 | return 1; | |
1861 | } | |
d7baffda | 1862 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1863 | spin_unlock(&head->lock); |
1e7a1421 | 1864 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1865 | |
c1103f7a | 1866 | if (head->must_insert_reserved) { |
b25c36f8 | 1867 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1868 | if (head->is_data) { |
40e046ac FM |
1869 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1870 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1871 | } |
1872 | } | |
1873 | ||
31890da0 | 1874 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1875 | |
1876 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1877 | btrfs_delayed_ref_unlock(head); |
d278850e | 1878 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1879 | return 0; |
1880 | } | |
1881 | ||
b1cdbcb5 NB |
1882 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1883 | struct btrfs_trans_handle *trans) | |
1884 | { | |
1885 | struct btrfs_delayed_ref_root *delayed_refs = | |
1886 | &trans->transaction->delayed_refs; | |
1887 | struct btrfs_delayed_ref_head *head = NULL; | |
1888 | int ret; | |
1889 | ||
1890 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1891 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1892 | if (!head) { |
1893 | spin_unlock(&delayed_refs->lock); | |
1894 | return head; | |
1895 | } | |
1896 | ||
1897 | /* | |
1898 | * Grab the lock that says we are going to process all the refs for | |
1899 | * this head | |
1900 | */ | |
9e920a6f | 1901 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1902 | spin_unlock(&delayed_refs->lock); |
1903 | ||
1904 | /* | |
1905 | * We may have dropped the spin lock to get the head mutex lock, and | |
1906 | * that might have given someone else time to free the head. If that's | |
1907 | * true, it has been removed from our list and we can move on. | |
1908 | */ | |
1909 | if (ret == -EAGAIN) | |
1910 | head = ERR_PTR(-EAGAIN); | |
1911 | ||
1912 | return head; | |
1913 | } | |
1914 | ||
e7261386 NB |
1915 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1916 | struct btrfs_delayed_ref_head *locked_ref, | |
1917 | unsigned long *run_refs) | |
1918 | { | |
1919 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1920 | struct btrfs_delayed_ref_root *delayed_refs; | |
1921 | struct btrfs_delayed_extent_op *extent_op; | |
1922 | struct btrfs_delayed_ref_node *ref; | |
1923 | int must_insert_reserved = 0; | |
1924 | int ret; | |
1925 | ||
1926 | delayed_refs = &trans->transaction->delayed_refs; | |
1927 | ||
0110a4c4 NB |
1928 | lockdep_assert_held(&locked_ref->mutex); |
1929 | lockdep_assert_held(&locked_ref->lock); | |
1930 | ||
e7261386 NB |
1931 | while ((ref = select_delayed_ref(locked_ref))) { |
1932 | if (ref->seq && | |
1933 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1934 | spin_unlock(&locked_ref->lock); | |
1935 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1936 | return -EAGAIN; | |
1937 | } | |
1938 | ||
1939 | (*run_refs)++; | |
1940 | ref->in_tree = 0; | |
1941 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1942 | RB_CLEAR_NODE(&ref->ref_node); | |
1943 | if (!list_empty(&ref->add_list)) | |
1944 | list_del(&ref->add_list); | |
1945 | /* | |
1946 | * When we play the delayed ref, also correct the ref_mod on | |
1947 | * head | |
1948 | */ | |
1949 | switch (ref->action) { | |
1950 | case BTRFS_ADD_DELAYED_REF: | |
1951 | case BTRFS_ADD_DELAYED_EXTENT: | |
1952 | locked_ref->ref_mod -= ref->ref_mod; | |
1953 | break; | |
1954 | case BTRFS_DROP_DELAYED_REF: | |
1955 | locked_ref->ref_mod += ref->ref_mod; | |
1956 | break; | |
1957 | default: | |
1958 | WARN_ON(1); | |
1959 | } | |
1960 | atomic_dec(&delayed_refs->num_entries); | |
1961 | ||
1962 | /* | |
1963 | * Record the must_insert_reserved flag before we drop the | |
1964 | * spin lock. | |
1965 | */ | |
1966 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1967 | locked_ref->must_insert_reserved = 0; | |
1968 | ||
1969 | extent_op = locked_ref->extent_op; | |
1970 | locked_ref->extent_op = NULL; | |
1971 | spin_unlock(&locked_ref->lock); | |
1972 | ||
1973 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1974 | must_insert_reserved); | |
1975 | ||
1976 | btrfs_free_delayed_extent_op(extent_op); | |
1977 | if (ret) { | |
1978 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1979 | btrfs_put_delayed_ref(ref); | |
1980 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1981 | ret); | |
1982 | return ret; | |
1983 | } | |
1984 | ||
1985 | btrfs_put_delayed_ref(ref); | |
1986 | cond_resched(); | |
1987 | ||
1988 | spin_lock(&locked_ref->lock); | |
1989 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1990 | } | |
1991 | ||
1992 | return 0; | |
1993 | } | |
1994 | ||
79787eaa JM |
1995 | /* |
1996 | * Returns 0 on success or if called with an already aborted transaction. | |
1997 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1998 | */ | |
d7df2c79 | 1999 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2000 | unsigned long nr) |
56bec294 | 2001 | { |
0a1e458a | 2002 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2003 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2004 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2005 | ktime_t start = ktime_get(); |
56bec294 | 2006 | int ret; |
d7df2c79 | 2007 | unsigned long count = 0; |
0a2b2a84 | 2008 | unsigned long actual_count = 0; |
56bec294 CM |
2009 | |
2010 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2011 | do { |
56bec294 | 2012 | if (!locked_ref) { |
b1cdbcb5 | 2013 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2014 | if (IS_ERR_OR_NULL(locked_ref)) { |
2015 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2016 | continue; | |
2017 | } else { | |
2018 | break; | |
2019 | } | |
56bec294 | 2020 | } |
0110a4c4 | 2021 | count++; |
56bec294 | 2022 | } |
2c3cf7d5 FM |
2023 | /* |
2024 | * We need to try and merge add/drops of the same ref since we | |
2025 | * can run into issues with relocate dropping the implicit ref | |
2026 | * and then it being added back again before the drop can | |
2027 | * finish. If we merged anything we need to re-loop so we can | |
2028 | * get a good ref. | |
2029 | * Or we can get node references of the same type that weren't | |
2030 | * merged when created due to bumps in the tree mod seq, and | |
2031 | * we need to merge them to prevent adding an inline extent | |
2032 | * backref before dropping it (triggering a BUG_ON at | |
2033 | * insert_inline_extent_backref()). | |
2034 | */ | |
d7df2c79 | 2035 | spin_lock(&locked_ref->lock); |
be97f133 | 2036 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2037 | |
0110a4c4 NB |
2038 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2039 | &actual_count); | |
2040 | if (ret < 0 && ret != -EAGAIN) { | |
2041 | /* | |
2042 | * Error, btrfs_run_delayed_refs_for_head already | |
2043 | * unlocked everything so just bail out | |
2044 | */ | |
2045 | return ret; | |
2046 | } else if (!ret) { | |
2047 | /* | |
2048 | * Success, perform the usual cleanup of a processed | |
2049 | * head | |
2050 | */ | |
f9871edd | 2051 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2052 | if (ret > 0 ) { |
b00e6250 JB |
2053 | /* We dropped our lock, we need to loop. */ |
2054 | ret = 0; | |
d7df2c79 | 2055 | continue; |
194ab0bc JB |
2056 | } else if (ret) { |
2057 | return ret; | |
5d4f98a2 | 2058 | } |
22cd2e7d | 2059 | } |
1ce7a5ec | 2060 | |
b00e6250 | 2061 | /* |
0110a4c4 NB |
2062 | * Either success case or btrfs_run_delayed_refs_for_head |
2063 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2064 | */ |
b00e6250 | 2065 | |
0110a4c4 | 2066 | locked_ref = NULL; |
c3e69d58 | 2067 | cond_resched(); |
0110a4c4 | 2068 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2069 | |
2070 | /* | |
2071 | * We don't want to include ref heads since we can have empty ref heads | |
2072 | * and those will drastically skew our runtime down since we just do | |
2073 | * accounting, no actual extent tree updates. | |
2074 | */ | |
2075 | if (actual_count > 0) { | |
2076 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2077 | u64 avg; | |
2078 | ||
2079 | /* | |
2080 | * We weigh the current average higher than our current runtime | |
2081 | * to avoid large swings in the average. | |
2082 | */ | |
2083 | spin_lock(&delayed_refs->lock); | |
2084 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2085 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2086 | spin_unlock(&delayed_refs->lock); |
2087 | } | |
d7df2c79 | 2088 | return 0; |
c3e69d58 CM |
2089 | } |
2090 | ||
709c0486 AJ |
2091 | #ifdef SCRAMBLE_DELAYED_REFS |
2092 | /* | |
2093 | * Normally delayed refs get processed in ascending bytenr order. This | |
2094 | * correlates in most cases to the order added. To expose dependencies on this | |
2095 | * order, we start to process the tree in the middle instead of the beginning | |
2096 | */ | |
2097 | static u64 find_middle(struct rb_root *root) | |
2098 | { | |
2099 | struct rb_node *n = root->rb_node; | |
2100 | struct btrfs_delayed_ref_node *entry; | |
2101 | int alt = 1; | |
2102 | u64 middle; | |
2103 | u64 first = 0, last = 0; | |
2104 | ||
2105 | n = rb_first(root); | |
2106 | if (n) { | |
2107 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2108 | first = entry->bytenr; | |
2109 | } | |
2110 | n = rb_last(root); | |
2111 | if (n) { | |
2112 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2113 | last = entry->bytenr; | |
2114 | } | |
2115 | n = root->rb_node; | |
2116 | ||
2117 | while (n) { | |
2118 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2119 | WARN_ON(!entry->in_tree); | |
2120 | ||
2121 | middle = entry->bytenr; | |
2122 | ||
2123 | if (alt) | |
2124 | n = n->rb_left; | |
2125 | else | |
2126 | n = n->rb_right; | |
2127 | ||
2128 | alt = 1 - alt; | |
2129 | } | |
2130 | return middle; | |
2131 | } | |
2132 | #endif | |
2133 | ||
c3e69d58 CM |
2134 | /* |
2135 | * this starts processing the delayed reference count updates and | |
2136 | * extent insertions we have queued up so far. count can be | |
2137 | * 0, which means to process everything in the tree at the start | |
2138 | * of the run (but not newly added entries), or it can be some target | |
2139 | * number you'd like to process. | |
79787eaa JM |
2140 | * |
2141 | * Returns 0 on success or if called with an aborted transaction | |
2142 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2143 | */ |
2144 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2145 | unsigned long count) |
c3e69d58 | 2146 | { |
c79a70b1 | 2147 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2148 | struct rb_node *node; |
2149 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2150 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2151 | int ret; |
2152 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2153 | |
79787eaa | 2154 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2155 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2156 | return 0; |
2157 | ||
0b246afa | 2158 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2159 | return 0; |
2160 | ||
c3e69d58 | 2161 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2162 | if (count == 0) |
d7df2c79 | 2163 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2164 | |
c3e69d58 | 2165 | again: |
709c0486 AJ |
2166 | #ifdef SCRAMBLE_DELAYED_REFS |
2167 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2168 | #endif | |
0a1e458a | 2169 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2170 | if (ret < 0) { |
66642832 | 2171 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2172 | return ret; |
eb099670 | 2173 | } |
c3e69d58 | 2174 | |
56bec294 | 2175 | if (run_all) { |
119e80df | 2176 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2177 | |
d7df2c79 | 2178 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2179 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2180 | if (!node) { |
2181 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2182 | goto out; |
d7df2c79 | 2183 | } |
d278850e JB |
2184 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2185 | href_node); | |
2186 | refcount_inc(&head->refs); | |
2187 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2188 | |
d278850e JB |
2189 | /* Mutex was contended, block until it's released and retry. */ |
2190 | mutex_lock(&head->mutex); | |
2191 | mutex_unlock(&head->mutex); | |
56bec294 | 2192 | |
d278850e | 2193 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2194 | cond_resched(); |
56bec294 | 2195 | goto again; |
5f39d397 | 2196 | } |
54aa1f4d | 2197 | out: |
a28ec197 CM |
2198 | return 0; |
2199 | } | |
2200 | ||
5d4f98a2 | 2201 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
42c9d0b5 | 2202 | struct extent_buffer *eb, u64 flags, |
b1c79e09 | 2203 | int level, int is_data) |
5d4f98a2 YZ |
2204 | { |
2205 | struct btrfs_delayed_extent_op *extent_op; | |
2206 | int ret; | |
2207 | ||
78a6184a | 2208 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2209 | if (!extent_op) |
2210 | return -ENOMEM; | |
2211 | ||
2212 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2213 | extent_op->update_flags = true; |
2214 | extent_op->update_key = false; | |
2215 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2216 | extent_op->level = level; |
5d4f98a2 | 2217 | |
42c9d0b5 | 2218 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2219 | if (ret) |
78a6184a | 2220 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2221 | return ret; |
2222 | } | |
2223 | ||
e4c3b2dc | 2224 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2225 | struct btrfs_path *path, |
2226 | u64 objectid, u64 offset, u64 bytenr) | |
2227 | { | |
2228 | struct btrfs_delayed_ref_head *head; | |
2229 | struct btrfs_delayed_ref_node *ref; | |
2230 | struct btrfs_delayed_data_ref *data_ref; | |
2231 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2232 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2233 | struct rb_node *node; |
5d4f98a2 YZ |
2234 | int ret = 0; |
2235 | ||
998ac6d2 | 2236 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2237 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2238 | if (cur_trans) |
2239 | refcount_inc(&cur_trans->use_count); | |
2240 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2241 | if (!cur_trans) |
2242 | return 0; | |
2243 | ||
2244 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2245 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2246 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2247 | if (!head) { |
2248 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2249 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2250 | return 0; |
2251 | } | |
5d4f98a2 YZ |
2252 | |
2253 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2254 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2255 | spin_unlock(&delayed_refs->lock); |
2256 | ||
b3b4aa74 | 2257 | btrfs_release_path(path); |
5d4f98a2 | 2258 | |
8cc33e5c DS |
2259 | /* |
2260 | * Mutex was contended, block until it's released and let | |
2261 | * caller try again | |
2262 | */ | |
5d4f98a2 YZ |
2263 | mutex_lock(&head->mutex); |
2264 | mutex_unlock(&head->mutex); | |
d278850e | 2265 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2266 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2267 | return -EAGAIN; |
2268 | } | |
d7df2c79 | 2269 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2270 | |
d7df2c79 | 2271 | spin_lock(&head->lock); |
0e0adbcf JB |
2272 | /* |
2273 | * XXX: We should replace this with a proper search function in the | |
2274 | * future. | |
2275 | */ | |
e3d03965 LB |
2276 | for (node = rb_first_cached(&head->ref_tree); node; |
2277 | node = rb_next(node)) { | |
0e0adbcf | 2278 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2279 | /* If it's a shared ref we know a cross reference exists */ |
2280 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2281 | ret = 1; | |
2282 | break; | |
2283 | } | |
5d4f98a2 | 2284 | |
d7df2c79 | 2285 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2286 | |
d7df2c79 JB |
2287 | /* |
2288 | * If our ref doesn't match the one we're currently looking at | |
2289 | * then we have a cross reference. | |
2290 | */ | |
2291 | if (data_ref->root != root->root_key.objectid || | |
2292 | data_ref->objectid != objectid || | |
2293 | data_ref->offset != offset) { | |
2294 | ret = 1; | |
2295 | break; | |
2296 | } | |
5d4f98a2 | 2297 | } |
d7df2c79 | 2298 | spin_unlock(&head->lock); |
5d4f98a2 | 2299 | mutex_unlock(&head->mutex); |
998ac6d2 | 2300 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2301 | return ret; |
2302 | } | |
2303 | ||
e4c3b2dc | 2304 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2305 | struct btrfs_path *path, |
a84d5d42 BB |
2306 | u64 objectid, u64 offset, u64 bytenr, |
2307 | bool strict) | |
be20aa9d | 2308 | { |
0b246afa JM |
2309 | struct btrfs_fs_info *fs_info = root->fs_info; |
2310 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2311 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2312 | struct btrfs_extent_data_ref *ref; |
2313 | struct btrfs_extent_inline_ref *iref; | |
2314 | struct btrfs_extent_item *ei; | |
f321e491 | 2315 | struct btrfs_key key; |
5d4f98a2 | 2316 | u32 item_size; |
3de28d57 | 2317 | int type; |
be20aa9d | 2318 | int ret; |
925baedd | 2319 | |
be20aa9d | 2320 | key.objectid = bytenr; |
31840ae1 | 2321 | key.offset = (u64)-1; |
f321e491 | 2322 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2323 | |
be20aa9d CM |
2324 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2325 | if (ret < 0) | |
2326 | goto out; | |
79787eaa | 2327 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2328 | |
2329 | ret = -ENOENT; | |
2330 | if (path->slots[0] == 0) | |
31840ae1 | 2331 | goto out; |
be20aa9d | 2332 | |
31840ae1 | 2333 | path->slots[0]--; |
f321e491 | 2334 | leaf = path->nodes[0]; |
5d4f98a2 | 2335 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2336 | |
5d4f98a2 | 2337 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2338 | goto out; |
f321e491 | 2339 | |
5d4f98a2 YZ |
2340 | ret = 1; |
2341 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2342 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2343 | |
a6bd9cd1 | 2344 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2345 | if (item_size != sizeof(*ei) + |
2346 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2347 | goto out; | |
be20aa9d | 2348 | |
a84d5d42 BB |
2349 | /* |
2350 | * If extent created before last snapshot => it's shared unless the | |
2351 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2352 | */ | |
2353 | if (!strict && | |
2354 | (btrfs_extent_generation(leaf, ei) <= | |
2355 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2356 | goto out; |
2357 | ||
2358 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2359 | |
a6bd9cd1 | 2360 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2361 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2362 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2363 | goto out; |
2364 | ||
2365 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2366 | if (btrfs_extent_refs(leaf, ei) != | |
2367 | btrfs_extent_data_ref_count(leaf, ref) || | |
2368 | btrfs_extent_data_ref_root(leaf, ref) != | |
2369 | root->root_key.objectid || | |
2370 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2371 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2372 | goto out; | |
2373 | ||
2374 | ret = 0; | |
2375 | out: | |
2376 | return ret; | |
2377 | } | |
2378 | ||
e4c3b2dc | 2379 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
a84d5d42 | 2380 | u64 bytenr, bool strict) |
5d4f98a2 YZ |
2381 | { |
2382 | struct btrfs_path *path; | |
2383 | int ret; | |
5d4f98a2 YZ |
2384 | |
2385 | path = btrfs_alloc_path(); | |
2386 | if (!path) | |
9132c4ff | 2387 | return -ENOMEM; |
5d4f98a2 YZ |
2388 | |
2389 | do { | |
e4c3b2dc | 2390 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2391 | offset, bytenr, strict); |
5d4f98a2 | 2392 | if (ret && ret != -ENOENT) |
f321e491 | 2393 | goto out; |
80ff3856 | 2394 | |
380fd066 MT |
2395 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2396 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2397 | |
be20aa9d | 2398 | out: |
80ff3856 | 2399 | btrfs_free_path(path); |
f0486c68 YZ |
2400 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2401 | WARN_ON(ret > 0); | |
f321e491 | 2402 | return ret; |
be20aa9d | 2403 | } |
c5739bba | 2404 | |
5d4f98a2 | 2405 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2406 | struct btrfs_root *root, |
5d4f98a2 | 2407 | struct extent_buffer *buf, |
e339a6b0 | 2408 | int full_backref, int inc) |
31840ae1 | 2409 | { |
0b246afa | 2410 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2411 | u64 bytenr; |
5d4f98a2 YZ |
2412 | u64 num_bytes; |
2413 | u64 parent; | |
31840ae1 | 2414 | u64 ref_root; |
31840ae1 | 2415 | u32 nritems; |
31840ae1 ZY |
2416 | struct btrfs_key key; |
2417 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2418 | struct btrfs_ref generic_ref = { 0 }; |
2419 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2420 | int i; |
82fa113f | 2421 | int action; |
31840ae1 ZY |
2422 | int level; |
2423 | int ret = 0; | |
fccb84c9 | 2424 | |
0b246afa | 2425 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2426 | return 0; |
fccb84c9 | 2427 | |
31840ae1 | 2428 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2429 | nritems = btrfs_header_nritems(buf); |
2430 | level = btrfs_header_level(buf); | |
2431 | ||
92a7cc42 | 2432 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2433 | return 0; |
31840ae1 | 2434 | |
5d4f98a2 YZ |
2435 | if (full_backref) |
2436 | parent = buf->start; | |
2437 | else | |
2438 | parent = 0; | |
82fa113f QW |
2439 | if (inc) |
2440 | action = BTRFS_ADD_DELAYED_REF; | |
2441 | else | |
2442 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2443 | |
2444 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2445 | if (level == 0) { |
5d4f98a2 | 2446 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2447 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2448 | continue; |
5d4f98a2 | 2449 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2450 | struct btrfs_file_extent_item); |
2451 | if (btrfs_file_extent_type(buf, fi) == | |
2452 | BTRFS_FILE_EXTENT_INLINE) | |
2453 | continue; | |
2454 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2455 | if (bytenr == 0) | |
2456 | continue; | |
5d4f98a2 YZ |
2457 | |
2458 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2459 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2460 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2461 | num_bytes, parent); | |
2462 | generic_ref.real_root = root->root_key.objectid; | |
2463 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2464 | key.offset); | |
2465 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2466 | if (inc) |
82fa113f | 2467 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2468 | else |
ffd4bb2a | 2469 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2470 | if (ret) |
2471 | goto fail; | |
2472 | } else { | |
5d4f98a2 | 2473 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2474 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2475 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2476 | num_bytes, parent); | |
2477 | generic_ref.real_root = root->root_key.objectid; | |
2478 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2479 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2480 | if (inc) |
82fa113f | 2481 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2482 | else |
ffd4bb2a | 2483 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2484 | if (ret) |
2485 | goto fail; | |
2486 | } | |
2487 | } | |
2488 | return 0; | |
2489 | fail: | |
5d4f98a2 YZ |
2490 | return ret; |
2491 | } | |
2492 | ||
2493 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2494 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2495 | { |
e339a6b0 | 2496 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2497 | } |
2498 | ||
2499 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2500 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2501 | { |
e339a6b0 | 2502 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2503 | } |
2504 | ||
2ff7e61e | 2505 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 | 2506 | { |
32da5386 | 2507 | struct btrfs_block_group *block_group; |
d2fb3437 YZ |
2508 | int readonly = 0; |
2509 | ||
0b246afa | 2510 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2511 | if (!block_group || block_group->ro) |
2512 | readonly = 1; | |
2513 | if (block_group) | |
fa9c0d79 | 2514 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2515 | return readonly; |
2516 | } | |
2517 | ||
1b86826d | 2518 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2519 | { |
0b246afa | 2520 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2521 | u64 flags; |
53b381b3 | 2522 | u64 ret; |
9ed74f2d | 2523 | |
b742bb82 YZ |
2524 | if (data) |
2525 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2526 | else if (root == fs_info->chunk_root) |
b742bb82 | 2527 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2528 | else |
b742bb82 | 2529 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2530 | |
878d7b67 | 2531 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2532 | return ret; |
6a63209f | 2533 | } |
9ed74f2d | 2534 | |
2ff7e61e | 2535 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2536 | { |
32da5386 | 2537 | struct btrfs_block_group *cache; |
d2fb3437 | 2538 | u64 bytenr; |
0f9dd46c | 2539 | |
0b246afa JM |
2540 | spin_lock(&fs_info->block_group_cache_lock); |
2541 | bytenr = fs_info->first_logical_byte; | |
2542 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2543 | |
2544 | if (bytenr < (u64)-1) | |
2545 | return bytenr; | |
2546 | ||
0b246afa | 2547 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2548 | if (!cache) |
a061fc8d | 2549 | return 0; |
0f9dd46c | 2550 | |
b3470b5d | 2551 | bytenr = cache->start; |
fa9c0d79 | 2552 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2553 | |
2554 | return bytenr; | |
a061fc8d CM |
2555 | } |
2556 | ||
6690d071 NB |
2557 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2558 | struct btrfs_block_group *cache, | |
f0486c68 | 2559 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2560 | { |
fdf08605 DS |
2561 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2562 | ||
11833d66 YZ |
2563 | spin_lock(&cache->space_info->lock); |
2564 | spin_lock(&cache->lock); | |
2565 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2566 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2567 | num_bytes); | |
11833d66 YZ |
2568 | if (reserved) { |
2569 | cache->reserved -= num_bytes; | |
2570 | cache->space_info->bytes_reserved -= num_bytes; | |
2571 | } | |
2572 | spin_unlock(&cache->lock); | |
2573 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2574 | |
dec59fa3 EL |
2575 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
2576 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
fe119a6e | 2577 | set_extent_dirty(&trans->transaction->pinned_extents, bytenr, |
f0486c68 YZ |
2578 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2579 | return 0; | |
2580 | } | |
68b38550 | 2581 | |
b25c36f8 | 2582 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2583 | u64 bytenr, u64 num_bytes, int reserved) |
2584 | { | |
32da5386 | 2585 | struct btrfs_block_group *cache; |
68b38550 | 2586 | |
b25c36f8 | 2587 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2588 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2589 | |
6690d071 | 2590 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2591 | |
2592 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2593 | return 0; |
2594 | } | |
2595 | ||
f0486c68 | 2596 | /* |
e688b725 CM |
2597 | * this function must be called within transaction |
2598 | */ | |
9fce5704 | 2599 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
e688b725 CM |
2600 | u64 bytenr, u64 num_bytes) |
2601 | { | |
32da5386 | 2602 | struct btrfs_block_group *cache; |
b50c6e25 | 2603 | int ret; |
e688b725 | 2604 | |
f2fb7298 NB |
2605 | btrfs_add_excluded_extent(trans->fs_info, bytenr, num_bytes); |
2606 | ||
9fce5704 | 2607 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
b50c6e25 JB |
2608 | if (!cache) |
2609 | return -EINVAL; | |
e688b725 CM |
2610 | |
2611 | /* | |
2612 | * pull in the free space cache (if any) so that our pin | |
2613 | * removes the free space from the cache. We have load_only set | |
2614 | * to one because the slow code to read in the free extents does check | |
2615 | * the pinned extents. | |
2616 | */ | |
676f1f75 | 2617 | btrfs_cache_block_group(cache, 1); |
e688b725 | 2618 | |
6690d071 | 2619 | pin_down_extent(trans, cache, bytenr, num_bytes, 0); |
e688b725 CM |
2620 | |
2621 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2622 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 2623 | btrfs_put_block_group(cache); |
b50c6e25 | 2624 | return ret; |
e688b725 CM |
2625 | } |
2626 | ||
2ff7e61e JM |
2627 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2628 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2629 | { |
2630 | int ret; | |
32da5386 | 2631 | struct btrfs_block_group *block_group; |
8c2a1a30 JB |
2632 | struct btrfs_caching_control *caching_ctl; |
2633 | ||
0b246afa | 2634 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2635 | if (!block_group) |
2636 | return -EINVAL; | |
2637 | ||
676f1f75 | 2638 | btrfs_cache_block_group(block_group, 0); |
e3cb339f | 2639 | caching_ctl = btrfs_get_caching_control(block_group); |
8c2a1a30 JB |
2640 | |
2641 | if (!caching_ctl) { | |
2642 | /* Logic error */ | |
32da5386 | 2643 | BUG_ON(!btrfs_block_group_done(block_group)); |
8c2a1a30 JB |
2644 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2645 | } else { | |
e747853c JB |
2646 | /* |
2647 | * We must wait for v1 caching to finish, otherwise we may not | |
2648 | * remove our space. | |
2649 | */ | |
2650 | btrfs_wait_space_cache_v1_finished(block_group, caching_ctl); | |
8c2a1a30 JB |
2651 | mutex_lock(&caching_ctl->mutex); |
2652 | ||
2653 | if (start >= caching_ctl->progress) { | |
6f410d1b JB |
2654 | ret = btrfs_add_excluded_extent(fs_info, start, |
2655 | num_bytes); | |
8c2a1a30 JB |
2656 | } else if (start + num_bytes <= caching_ctl->progress) { |
2657 | ret = btrfs_remove_free_space(block_group, | |
2658 | start, num_bytes); | |
2659 | } else { | |
2660 | num_bytes = caching_ctl->progress - start; | |
2661 | ret = btrfs_remove_free_space(block_group, | |
2662 | start, num_bytes); | |
2663 | if (ret) | |
2664 | goto out_lock; | |
2665 | ||
2666 | num_bytes = (start + num_bytes) - | |
2667 | caching_ctl->progress; | |
2668 | start = caching_ctl->progress; | |
6f410d1b JB |
2669 | ret = btrfs_add_excluded_extent(fs_info, start, |
2670 | num_bytes); | |
8c2a1a30 JB |
2671 | } |
2672 | out_lock: | |
2673 | mutex_unlock(&caching_ctl->mutex); | |
e3cb339f | 2674 | btrfs_put_caching_control(caching_ctl); |
8c2a1a30 JB |
2675 | } |
2676 | btrfs_put_block_group(block_group); | |
2677 | return ret; | |
2678 | } | |
2679 | ||
bcdc428c | 2680 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2681 | { |
bcdc428c | 2682 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2683 | struct btrfs_file_extent_item *item; |
2684 | struct btrfs_key key; | |
2685 | int found_type; | |
2686 | int i; | |
b89311ef | 2687 | int ret = 0; |
8c2a1a30 | 2688 | |
2ff7e61e | 2689 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2690 | return 0; |
2691 | ||
2692 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2693 | btrfs_item_key_to_cpu(eb, &key, i); | |
2694 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2695 | continue; | |
2696 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2697 | found_type = btrfs_file_extent_type(eb, item); | |
2698 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2699 | continue; | |
2700 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2701 | continue; | |
2702 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2703 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2704 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2705 | if (ret) | |
2706 | break; | |
8c2a1a30 JB |
2707 | } |
2708 | ||
b89311ef | 2709 | return ret; |
8c2a1a30 JB |
2710 | } |
2711 | ||
9cfa3e34 | 2712 | static void |
32da5386 | 2713 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2714 | { |
2715 | atomic_inc(&bg->reservations); | |
2716 | } | |
2717 | ||
c759c4e1 JB |
2718 | /* |
2719 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2720 | * what it should be based on the mount options. | |
2721 | */ | |
2722 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2723 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2724 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2725 | { |
2726 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2727 | |
2728 | *empty_cluster = 0; | |
2729 | if (btrfs_mixed_space_info(space_info)) | |
2730 | return ret; | |
2731 | ||
c759c4e1 | 2732 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2733 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2734 | if (btrfs_test_opt(fs_info, SSD)) |
2735 | *empty_cluster = SZ_2M; | |
2736 | else | |
ee22184b | 2737 | *empty_cluster = SZ_64K; |
583b7231 HK |
2738 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2739 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2740 | *empty_cluster = SZ_2M; | |
0b246afa | 2741 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2742 | } |
2743 | ||
2744 | return ret; | |
2745 | } | |
2746 | ||
2ff7e61e JM |
2747 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2748 | u64 start, u64 end, | |
678886bd | 2749 | const bool return_free_space) |
ccd467d6 | 2750 | { |
32da5386 | 2751 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2752 | struct btrfs_space_info *space_info; |
2753 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2754 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2755 | u64 len; |
c759c4e1 JB |
2756 | u64 total_unpinned = 0; |
2757 | u64 empty_cluster = 0; | |
7b398f8e | 2758 | bool readonly; |
ccd467d6 | 2759 | |
11833d66 | 2760 | while (start <= end) { |
7b398f8e | 2761 | readonly = false; |
11833d66 | 2762 | if (!cache || |
b3470b5d | 2763 | start >= cache->start + cache->length) { |
11833d66 YZ |
2764 | if (cache) |
2765 | btrfs_put_block_group(cache); | |
c759c4e1 | 2766 | total_unpinned = 0; |
11833d66 | 2767 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2768 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2769 | |
2ff7e61e | 2770 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2771 | cache->space_info, |
2772 | &empty_cluster); | |
2773 | empty_cluster <<= 1; | |
11833d66 YZ |
2774 | } |
2775 | ||
b3470b5d | 2776 | len = cache->start + cache->length - start; |
11833d66 YZ |
2777 | len = min(len, end + 1 - start); |
2778 | ||
2ca08c56 | 2779 | down_read(&fs_info->commit_root_sem); |
9076dbd5 JB |
2780 | if (start < cache->last_byte_to_unpin && return_free_space) { |
2781 | u64 add_len = min(len, cache->last_byte_to_unpin - start); | |
2782 | ||
2783 | btrfs_add_free_space(cache, start, add_len); | |
11833d66 | 2784 | } |
2ca08c56 | 2785 | up_read(&fs_info->commit_root_sem); |
11833d66 | 2786 | |
f0486c68 | 2787 | start += len; |
c759c4e1 | 2788 | total_unpinned += len; |
7b398f8e | 2789 | space_info = cache->space_info; |
f0486c68 | 2790 | |
c759c4e1 JB |
2791 | /* |
2792 | * If this space cluster has been marked as fragmented and we've | |
2793 | * unpinned enough in this block group to potentially allow a | |
2794 | * cluster to be created inside of it go ahead and clear the | |
2795 | * fragmented check. | |
2796 | */ | |
2797 | if (cluster && cluster->fragmented && | |
2798 | total_unpinned > empty_cluster) { | |
2799 | spin_lock(&cluster->lock); | |
2800 | cluster->fragmented = 0; | |
2801 | spin_unlock(&cluster->lock); | |
2802 | } | |
2803 | ||
7b398f8e | 2804 | spin_lock(&space_info->lock); |
11833d66 YZ |
2805 | spin_lock(&cache->lock); |
2806 | cache->pinned -= len; | |
bb96c4e5 | 2807 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2808 | space_info->max_extent_size = 0; |
dec59fa3 EL |
2809 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2810 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
2811 | if (cache->ro) { |
2812 | space_info->bytes_readonly += len; | |
2813 | readonly = true; | |
2814 | } | |
11833d66 | 2815 | spin_unlock(&cache->lock); |
957780eb JB |
2816 | if (!readonly && return_free_space && |
2817 | global_rsv->space_info == space_info) { | |
2818 | u64 to_add = len; | |
92ac58ec | 2819 | |
7b398f8e JB |
2820 | spin_lock(&global_rsv->lock); |
2821 | if (!global_rsv->full) { | |
957780eb JB |
2822 | to_add = min(len, global_rsv->size - |
2823 | global_rsv->reserved); | |
2824 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2825 | btrfs_space_info_update_bytes_may_use(fs_info, |
2826 | space_info, to_add); | |
7b398f8e JB |
2827 | if (global_rsv->reserved >= global_rsv->size) |
2828 | global_rsv->full = 1; | |
957780eb | 2829 | len -= to_add; |
7b398f8e JB |
2830 | } |
2831 | spin_unlock(&global_rsv->lock); | |
2832 | } | |
2732798c JB |
2833 | /* Add to any tickets we may have */ |
2834 | if (!readonly && return_free_space && len) | |
2835 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2836 | spin_unlock(&space_info->lock); |
ccd467d6 | 2837 | } |
11833d66 YZ |
2838 | |
2839 | if (cache) | |
2840 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2841 | return 0; |
2842 | } | |
2843 | ||
5ead2dd0 | 2844 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2845 | { |
5ead2dd0 | 2846 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2847 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2848 | struct list_head *deleted_bgs; |
11833d66 | 2849 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2850 | u64 start; |
2851 | u64 end; | |
a28ec197 | 2852 | int ret; |
a28ec197 | 2853 | |
fe119a6e | 2854 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2855 | |
bf31f87f | 2856 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2857 | struct extent_state *cached_state = NULL; |
2858 | ||
d4b450cd | 2859 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2860 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2861 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2862 | if (ret) { |
2863 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2864 | break; |
d4b450cd | 2865 | } |
fe119a6e NB |
2866 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
2867 | clear_extent_bits(&fs_info->excluded_extents, start, | |
f2fb7298 | 2868 | end, EXTENT_UPTODATE); |
1f3c79a2 | 2869 | |
46b27f50 | 2870 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2871 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2872 | end + 1 - start, NULL); |
1f3c79a2 | 2873 | |
0e6ec385 | 2874 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2875 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2876 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2877 | free_extent_state(cached_state); |
b9473439 | 2878 | cond_resched(); |
a28ec197 | 2879 | } |
817d52f8 | 2880 | |
a2309300 DZ |
2881 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2882 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2883 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2884 | } |
b0643e59 | 2885 | |
e33e17ee JM |
2886 | /* |
2887 | * Transaction is finished. We don't need the lock anymore. We | |
2888 | * do need to clean up the block groups in case of a transaction | |
2889 | * abort. | |
2890 | */ | |
2891 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2892 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2893 | u64 trimmed = 0; | |
2894 | ||
2895 | ret = -EROFS; | |
bf31f87f | 2896 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2897 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2898 | block_group->start, |
2899 | block_group->length, | |
e33e17ee JM |
2900 | &trimmed); |
2901 | ||
2902 | list_del_init(&block_group->bg_list); | |
6b7304af | 2903 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2904 | btrfs_put_block_group(block_group); |
2905 | ||
2906 | if (ret) { | |
2907 | const char *errstr = btrfs_decode_error(ret); | |
2908 | btrfs_warn(fs_info, | |
913e1535 | 2909 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2910 | ret, errstr); |
2911 | } | |
2912 | } | |
2913 | ||
e20d96d6 CM |
2914 | return 0; |
2915 | } | |
2916 | ||
1c2a07f5 QW |
2917 | /* |
2918 | * Drop one or more refs of @node. | |
2919 | * | |
2920 | * 1. Locate the extent refs. | |
2921 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
2922 | * Locate it, then reduce the refs number or remove the ref line completely. | |
2923 | * | |
2924 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
2925 | * | |
2926 | * Inline backref case: | |
2927 | * | |
2928 | * in extent tree we have: | |
2929 | * | |
2930 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2931 | * refs 2 gen 6 flags DATA | |
2932 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2933 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
2934 | * | |
2935 | * This function gets called with: | |
2936 | * | |
2937 | * node->bytenr = 13631488 | |
2938 | * node->num_bytes = 1048576 | |
2939 | * root_objectid = FS_TREE | |
2940 | * owner_objectid = 257 | |
2941 | * owner_offset = 0 | |
2942 | * refs_to_drop = 1 | |
2943 | * | |
2944 | * Then we should get some like: | |
2945 | * | |
2946 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2947 | * refs 1 gen 6 flags DATA | |
2948 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2949 | * | |
2950 | * Keyed backref case: | |
2951 | * | |
2952 | * in extent tree we have: | |
2953 | * | |
2954 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2955 | * refs 754 gen 6 flags DATA | |
2956 | * [...] | |
2957 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
2958 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
2959 | * | |
2960 | * This function get called with: | |
2961 | * | |
2962 | * node->bytenr = 13631488 | |
2963 | * node->num_bytes = 1048576 | |
2964 | * root_objectid = FS_TREE | |
2965 | * owner_objectid = 866 | |
2966 | * owner_offset = 0 | |
2967 | * refs_to_drop = 1 | |
2968 | * | |
2969 | * Then we should get some like: | |
2970 | * | |
2971 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2972 | * refs 753 gen 6 flags DATA | |
2973 | * | |
2974 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
2975 | */ | |
5d4f98a2 | 2976 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2977 | struct btrfs_delayed_ref_node *node, u64 parent, |
2978 | u64 root_objectid, u64 owner_objectid, | |
2979 | u64 owner_offset, int refs_to_drop, | |
2980 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2981 | { |
e72cb923 | 2982 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2983 | struct btrfs_key key; |
5d4f98a2 | 2984 | struct btrfs_path *path; |
1261ec42 | 2985 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2986 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2987 | struct btrfs_extent_item *ei; |
2988 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2989 | int ret; |
5d4f98a2 | 2990 | int is_data; |
952fccac CM |
2991 | int extent_slot = 0; |
2992 | int found_extent = 0; | |
2993 | int num_to_del = 1; | |
5d4f98a2 YZ |
2994 | u32 item_size; |
2995 | u64 refs; | |
c682f9b3 QW |
2996 | u64 bytenr = node->bytenr; |
2997 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2998 | int last_ref = 0; |
0b246afa | 2999 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 3000 | |
5caf2a00 | 3001 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3002 | if (!path) |
3003 | return -ENOMEM; | |
5f26f772 | 3004 | |
5d4f98a2 | 3005 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
1c2a07f5 QW |
3006 | |
3007 | if (!is_data && refs_to_drop != 1) { | |
3008 | btrfs_crit(info, | |
3009 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
3010 | node->bytenr, refs_to_drop); | |
3011 | ret = -EINVAL; | |
3012 | btrfs_abort_transaction(trans, ret); | |
3013 | goto out; | |
3014 | } | |
5d4f98a2 | 3015 | |
3173a18f | 3016 | if (is_data) |
897ca819 | 3017 | skinny_metadata = false; |
3173a18f | 3018 | |
fbe4801b NB |
3019 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
3020 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 3021 | owner_offset); |
7bb86316 | 3022 | if (ret == 0) { |
1c2a07f5 QW |
3023 | /* |
3024 | * Either the inline backref or the SHARED_DATA_REF/ | |
3025 | * SHARED_BLOCK_REF is found | |
3026 | * | |
3027 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
3028 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
3029 | */ | |
952fccac | 3030 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3031 | while (extent_slot >= 0) { |
3032 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3033 | extent_slot); |
5d4f98a2 | 3034 | if (key.objectid != bytenr) |
952fccac | 3035 | break; |
5d4f98a2 YZ |
3036 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3037 | key.offset == num_bytes) { | |
952fccac CM |
3038 | found_extent = 1; |
3039 | break; | |
3040 | } | |
3173a18f JB |
3041 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3042 | key.offset == owner_objectid) { | |
3043 | found_extent = 1; | |
3044 | break; | |
3045 | } | |
1c2a07f5 QW |
3046 | |
3047 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
3048 | if (path->slots[0] - extent_slot > 5) |
3049 | break; | |
5d4f98a2 | 3050 | extent_slot--; |
952fccac | 3051 | } |
a79865c6 | 3052 | |
31840ae1 | 3053 | if (!found_extent) { |
1c2a07f5 QW |
3054 | if (iref) { |
3055 | btrfs_crit(info, | |
3056 | "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref"); | |
3057 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3058 | goto err_dump; | |
3059 | } | |
3060 | /* Must be SHARED_* item, remove the backref first */ | |
87cc7a8a | 3061 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3062 | refs_to_drop, |
fcebe456 | 3063 | is_data, &last_ref); |
005d6427 | 3064 | if (ret) { |
66642832 | 3065 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3066 | goto out; |
3067 | } | |
b3b4aa74 | 3068 | btrfs_release_path(path); |
5d4f98a2 | 3069 | |
1c2a07f5 | 3070 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3071 | key.objectid = bytenr; |
3072 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3073 | key.offset = num_bytes; | |
3074 | ||
3173a18f JB |
3075 | if (!is_data && skinny_metadata) { |
3076 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3077 | key.offset = owner_objectid; | |
3078 | } | |
3079 | ||
31840ae1 ZY |
3080 | ret = btrfs_search_slot(trans, extent_root, |
3081 | &key, path, -1, 1); | |
3173a18f JB |
3082 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3083 | /* | |
3084 | * Couldn't find our skinny metadata item, | |
3085 | * see if we have ye olde extent item. | |
3086 | */ | |
3087 | path->slots[0]--; | |
3088 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3089 | path->slots[0]); | |
3090 | if (key.objectid == bytenr && | |
3091 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3092 | key.offset == num_bytes) | |
3093 | ret = 0; | |
3094 | } | |
3095 | ||
3096 | if (ret > 0 && skinny_metadata) { | |
3097 | skinny_metadata = false; | |
9ce49a0b | 3098 | key.objectid = bytenr; |
3173a18f JB |
3099 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3100 | key.offset = num_bytes; | |
3101 | btrfs_release_path(path); | |
3102 | ret = btrfs_search_slot(trans, extent_root, | |
3103 | &key, path, -1, 1); | |
3104 | } | |
3105 | ||
f3465ca4 | 3106 | if (ret) { |
5d163e0e JM |
3107 | btrfs_err(info, |
3108 | "umm, got %d back from search, was looking for %llu", | |
3109 | ret, bytenr); | |
b783e62d | 3110 | if (ret > 0) |
a4f78750 | 3111 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3112 | } |
005d6427 | 3113 | if (ret < 0) { |
66642832 | 3114 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3115 | goto out; |
3116 | } | |
31840ae1 ZY |
3117 | extent_slot = path->slots[0]; |
3118 | } | |
fae7f21c | 3119 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3120 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3121 | btrfs_err(info, |
3122 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3123 | bytenr, parent, root_objectid, owner_objectid, |
3124 | owner_offset); | |
66642832 | 3125 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3126 | goto out; |
79787eaa | 3127 | } else { |
66642832 | 3128 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3129 | goto out; |
7bb86316 | 3130 | } |
5f39d397 CM |
3131 | |
3132 | leaf = path->nodes[0]; | |
5d4f98a2 | 3133 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3134 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3135 | ret = -EINVAL; |
3136 | btrfs_print_v0_err(info); | |
3137 | btrfs_abort_transaction(trans, ret); | |
3138 | goto out; | |
3139 | } | |
952fccac | 3140 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3141 | struct btrfs_extent_item); |
3173a18f JB |
3142 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3143 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3144 | struct btrfs_tree_block_info *bi; |
1c2a07f5 QW |
3145 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
3146 | btrfs_crit(info, | |
cad69d13 | 3147 | "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu", |
1c2a07f5 QW |
3148 | key.objectid, key.type, key.offset, |
3149 | owner_objectid, item_size, | |
3150 | sizeof(*ei) + sizeof(*bi)); | |
3151 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3152 | goto err_dump; | |
3153 | } | |
5d4f98a2 YZ |
3154 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3155 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3156 | } | |
56bec294 | 3157 | |
5d4f98a2 | 3158 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3159 | if (refs < refs_to_drop) { |
1c2a07f5 QW |
3160 | btrfs_crit(info, |
3161 | "trying to drop %d refs but we only have %llu for bytenr %llu", | |
5d163e0e | 3162 | refs_to_drop, refs, bytenr); |
1c2a07f5 QW |
3163 | btrfs_abort_transaction(trans, -EUCLEAN); |
3164 | goto err_dump; | |
32b02538 | 3165 | } |
56bec294 | 3166 | refs -= refs_to_drop; |
5f39d397 | 3167 | |
5d4f98a2 YZ |
3168 | if (refs > 0) { |
3169 | if (extent_op) | |
3170 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3171 | /* | |
3172 | * In the case of inline back ref, reference count will | |
3173 | * be updated by remove_extent_backref | |
952fccac | 3174 | */ |
5d4f98a2 | 3175 | if (iref) { |
1c2a07f5 QW |
3176 | if (!found_extent) { |
3177 | btrfs_crit(info, | |
3178 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found"); | |
3179 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3180 | goto err_dump; | |
3181 | } | |
5d4f98a2 YZ |
3182 | } else { |
3183 | btrfs_set_extent_refs(leaf, ei, refs); | |
3184 | btrfs_mark_buffer_dirty(leaf); | |
3185 | } | |
3186 | if (found_extent) { | |
87cc7a8a NB |
3187 | ret = remove_extent_backref(trans, path, iref, |
3188 | refs_to_drop, is_data, | |
3189 | &last_ref); | |
005d6427 | 3190 | if (ret) { |
66642832 | 3191 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3192 | goto out; |
3193 | } | |
952fccac | 3194 | } |
5d4f98a2 | 3195 | } else { |
1c2a07f5 | 3196 | /* In this branch refs == 1 */ |
5d4f98a2 | 3197 | if (found_extent) { |
1c2a07f5 QW |
3198 | if (is_data && refs_to_drop != |
3199 | extent_data_ref_count(path, iref)) { | |
3200 | btrfs_crit(info, | |
3201 | "invalid refs_to_drop, current refs %u refs_to_drop %u", | |
3202 | extent_data_ref_count(path, iref), | |
3203 | refs_to_drop); | |
3204 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3205 | goto err_dump; | |
3206 | } | |
5d4f98a2 | 3207 | if (iref) { |
1c2a07f5 QW |
3208 | if (path->slots[0] != extent_slot) { |
3209 | btrfs_crit(info, | |
3210 | "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref", | |
3211 | key.objectid, key.type, | |
3212 | key.offset); | |
3213 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3214 | goto err_dump; | |
3215 | } | |
5d4f98a2 | 3216 | } else { |
1c2a07f5 QW |
3217 | /* |
3218 | * No inline ref, we must be at SHARED_* item, | |
3219 | * And it's single ref, it must be: | |
3220 | * | extent_slot ||extent_slot + 1| | |
3221 | * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ] | |
3222 | */ | |
3223 | if (path->slots[0] != extent_slot + 1) { | |
3224 | btrfs_crit(info, | |
3225 | "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM"); | |
3226 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3227 | goto err_dump; | |
3228 | } | |
5d4f98a2 YZ |
3229 | path->slots[0] = extent_slot; |
3230 | num_to_del = 2; | |
3231 | } | |
78fae27e | 3232 | } |
b9473439 | 3233 | |
fcebe456 | 3234 | last_ref = 1; |
952fccac CM |
3235 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3236 | num_to_del); | |
005d6427 | 3237 | if (ret) { |
66642832 | 3238 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3239 | goto out; |
3240 | } | |
b3b4aa74 | 3241 | btrfs_release_path(path); |
21af804c | 3242 | |
5d4f98a2 | 3243 | if (is_data) { |
40e046ac FM |
3244 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3245 | num_bytes); | |
005d6427 | 3246 | if (ret) { |
66642832 | 3247 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3248 | goto out; |
3249 | } | |
459931ec CM |
3250 | } |
3251 | ||
e7355e50 | 3252 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3253 | if (ret) { |
66642832 | 3254 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3255 | goto out; |
3256 | } | |
3257 | ||
ade4b516 | 3258 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3259 | if (ret) { |
66642832 | 3260 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3261 | goto out; |
3262 | } | |
a28ec197 | 3263 | } |
fcebe456 JB |
3264 | btrfs_release_path(path); |
3265 | ||
79787eaa | 3266 | out: |
5caf2a00 | 3267 | btrfs_free_path(path); |
a28ec197 | 3268 | return ret; |
1c2a07f5 QW |
3269 | err_dump: |
3270 | /* | |
3271 | * Leaf dump can take up a lot of log buffer, so we only do full leaf | |
3272 | * dump for debug build. | |
3273 | */ | |
3274 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
3275 | btrfs_crit(info, "path->slots[0]=%d extent_slot=%d", | |
3276 | path->slots[0], extent_slot); | |
3277 | btrfs_print_leaf(path->nodes[0]); | |
3278 | } | |
3279 | ||
3280 | btrfs_free_path(path); | |
3281 | return -EUCLEAN; | |
a28ec197 CM |
3282 | } |
3283 | ||
1887be66 | 3284 | /* |
f0486c68 | 3285 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3286 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3287 | * a given extent, and if there are no other delayed refs to be processed, it | |
3288 | * removes it from the tree. | |
3289 | */ | |
3290 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3291 | u64 bytenr) |
1887be66 CM |
3292 | { |
3293 | struct btrfs_delayed_ref_head *head; | |
3294 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3295 | int ret = 0; |
1887be66 CM |
3296 | |
3297 | delayed_refs = &trans->transaction->delayed_refs; | |
3298 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3299 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3300 | if (!head) |
cf93da7b | 3301 | goto out_delayed_unlock; |
1887be66 | 3302 | |
d7df2c79 | 3303 | spin_lock(&head->lock); |
e3d03965 | 3304 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3305 | goto out; |
3306 | ||
bedc6617 JB |
3307 | if (cleanup_extent_op(head) != NULL) |
3308 | goto out; | |
5d4f98a2 | 3309 | |
1887be66 CM |
3310 | /* |
3311 | * waiting for the lock here would deadlock. If someone else has it | |
3312 | * locked they are already in the process of dropping it anyway | |
3313 | */ | |
3314 | if (!mutex_trylock(&head->mutex)) | |
3315 | goto out; | |
3316 | ||
d7baffda | 3317 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3318 | head->processing = 0; |
d7baffda | 3319 | |
d7df2c79 | 3320 | spin_unlock(&head->lock); |
1887be66 CM |
3321 | spin_unlock(&delayed_refs->lock); |
3322 | ||
f0486c68 YZ |
3323 | BUG_ON(head->extent_op); |
3324 | if (head->must_insert_reserved) | |
3325 | ret = 1; | |
3326 | ||
31890da0 | 3327 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3328 | mutex_unlock(&head->mutex); |
d278850e | 3329 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3330 | return ret; |
1887be66 | 3331 | out: |
d7df2c79 | 3332 | spin_unlock(&head->lock); |
cf93da7b CM |
3333 | |
3334 | out_delayed_unlock: | |
1887be66 CM |
3335 | spin_unlock(&delayed_refs->lock); |
3336 | return 0; | |
3337 | } | |
3338 | ||
f0486c68 YZ |
3339 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3340 | struct btrfs_root *root, | |
3341 | struct extent_buffer *buf, | |
5581a51a | 3342 | u64 parent, int last_ref) |
f0486c68 | 3343 | { |
0b246afa | 3344 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3345 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 3346 | int pin = 1; |
f0486c68 YZ |
3347 | int ret; |
3348 | ||
ed4f255b QW |
3349 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3350 | buf->start, buf->len, parent); | |
3351 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3352 | root->root_key.objectid); | |
3353 | ||
f0486c68 | 3354 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
3355 | int old_ref_mod, new_ref_mod; |
3356 | ||
8a5040f7 | 3357 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 3358 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 3359 | &old_ref_mod, &new_ref_mod); |
79787eaa | 3360 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 3361 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
3362 | } |
3363 | ||
0a16c7d7 | 3364 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3365 | struct btrfs_block_group *cache; |
6219872d | 3366 | |
f0486c68 | 3367 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3368 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 3369 | if (!ret) |
37be25bc | 3370 | goto out; |
f0486c68 YZ |
3371 | } |
3372 | ||
4da8b76d | 3373 | pin = 0; |
0b246afa | 3374 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3375 | |
f0486c68 | 3376 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3377 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3378 | btrfs_put_block_group(cache); |
37be25bc | 3379 | goto out; |
f0486c68 YZ |
3380 | } |
3381 | ||
3382 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3383 | ||
3384 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3385 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3386 | btrfs_put_block_group(cache); |
71ff6437 | 3387 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3388 | } |
3389 | out: | |
b150a4f1 | 3390 | if (pin) |
78192442 | 3391 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 3392 | |
0a16c7d7 OS |
3393 | if (last_ref) { |
3394 | /* | |
3395 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3396 | * matter anymore. | |
3397 | */ | |
3398 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3399 | } | |
f0486c68 YZ |
3400 | } |
3401 | ||
79787eaa | 3402 | /* Can return -ENOMEM */ |
ffd4bb2a | 3403 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3404 | { |
ffd4bb2a | 3405 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 3406 | int old_ref_mod, new_ref_mod; |
925baedd CM |
3407 | int ret; |
3408 | ||
f5ee5c9a | 3409 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3410 | return 0; |
fccb84c9 | 3411 | |
56bec294 CM |
3412 | /* |
3413 | * tree log blocks never actually go into the extent allocation | |
3414 | * tree, just update pinning info and exit early. | |
56bec294 | 3415 | */ |
ffd4bb2a QW |
3416 | if ((ref->type == BTRFS_REF_METADATA && |
3417 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3418 | (ref->type == BTRFS_REF_DATA && | |
3419 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3420 | /* unlocks the pinned mutex */ |
b25c36f8 | 3421 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
d7eae340 | 3422 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 3423 | ret = 0; |
ffd4bb2a QW |
3424 | } else if (ref->type == BTRFS_REF_METADATA) { |
3425 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 3426 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 3427 | } else { |
ffd4bb2a | 3428 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 3429 | &old_ref_mod, &new_ref_mod); |
56bec294 | 3430 | } |
d7eae340 | 3431 | |
ffd4bb2a QW |
3432 | if (!((ref->type == BTRFS_REF_METADATA && |
3433 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3434 | (ref->type == BTRFS_REF_DATA && | |
3435 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3436 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3437 | |
ddf30cf0 | 3438 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 3439 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 3440 | |
925baedd CM |
3441 | return ret; |
3442 | } | |
3443 | ||
817d52f8 | 3444 | enum btrfs_loop_type { |
f262fa8d DS |
3445 | LOOP_CACHING_NOWAIT, |
3446 | LOOP_CACHING_WAIT, | |
3447 | LOOP_ALLOC_CHUNK, | |
3448 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3449 | }; |
3450 | ||
e570fd27 | 3451 | static inline void |
32da5386 | 3452 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3453 | int delalloc) |
3454 | { | |
3455 | if (delalloc) | |
3456 | down_read(&cache->data_rwsem); | |
3457 | } | |
3458 | ||
32da5386 | 3459 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3460 | int delalloc) |
3461 | { | |
3462 | btrfs_get_block_group(cache); | |
3463 | if (delalloc) | |
3464 | down_read(&cache->data_rwsem); | |
3465 | } | |
3466 | ||
32da5386 DS |
3467 | static struct btrfs_block_group *btrfs_lock_cluster( |
3468 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3469 | struct btrfs_free_cluster *cluster, |
3470 | int delalloc) | |
c142c6a4 | 3471 | __acquires(&cluster->refill_lock) |
e570fd27 | 3472 | { |
32da5386 | 3473 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3474 | |
e570fd27 | 3475 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3476 | while (1) { |
3477 | used_bg = cluster->block_group; | |
3478 | if (!used_bg) | |
3479 | return NULL; | |
3480 | ||
3481 | if (used_bg == block_group) | |
e570fd27 MX |
3482 | return used_bg; |
3483 | ||
6719afdc | 3484 | btrfs_get_block_group(used_bg); |
e570fd27 | 3485 | |
6719afdc GU |
3486 | if (!delalloc) |
3487 | return used_bg; | |
e570fd27 | 3488 | |
6719afdc GU |
3489 | if (down_read_trylock(&used_bg->data_rwsem)) |
3490 | return used_bg; | |
e570fd27 | 3491 | |
6719afdc | 3492 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3493 | |
e321f8a8 LB |
3494 | /* We should only have one-level nested. */ |
3495 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3496 | |
6719afdc GU |
3497 | spin_lock(&cluster->refill_lock); |
3498 | if (used_bg == cluster->block_group) | |
3499 | return used_bg; | |
e570fd27 | 3500 | |
6719afdc GU |
3501 | up_read(&used_bg->data_rwsem); |
3502 | btrfs_put_block_group(used_bg); | |
3503 | } | |
e570fd27 MX |
3504 | } |
3505 | ||
3506 | static inline void | |
32da5386 | 3507 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3508 | int delalloc) |
3509 | { | |
3510 | if (delalloc) | |
3511 | up_read(&cache->data_rwsem); | |
3512 | btrfs_put_block_group(cache); | |
3513 | } | |
3514 | ||
cb2f96f8 NA |
3515 | enum btrfs_extent_allocation_policy { |
3516 | BTRFS_EXTENT_ALLOC_CLUSTERED, | |
3517 | }; | |
3518 | ||
b4bd745d QW |
3519 | /* |
3520 | * Structure used internally for find_free_extent() function. Wraps needed | |
3521 | * parameters. | |
3522 | */ | |
3523 | struct find_free_extent_ctl { | |
3524 | /* Basic allocation info */ | |
b4bd745d QW |
3525 | u64 num_bytes; |
3526 | u64 empty_size; | |
3527 | u64 flags; | |
3528 | int delalloc; | |
3529 | ||
3530 | /* Where to start the search inside the bg */ | |
3531 | u64 search_start; | |
3532 | ||
3533 | /* For clustered allocation */ | |
3534 | u64 empty_cluster; | |
c10859be NA |
3535 | struct btrfs_free_cluster *last_ptr; |
3536 | bool use_cluster; | |
b4bd745d QW |
3537 | |
3538 | bool have_caching_bg; | |
3539 | bool orig_have_caching_bg; | |
3540 | ||
3541 | /* RAID index, converted from flags */ | |
3542 | int index; | |
3543 | ||
e72d79d6 QW |
3544 | /* |
3545 | * Current loop number, check find_free_extent_update_loop() for details | |
3546 | */ | |
b4bd745d QW |
3547 | int loop; |
3548 | ||
3549 | /* | |
3550 | * Whether we're refilling a cluster, if true we need to re-search | |
3551 | * current block group but don't try to refill the cluster again. | |
3552 | */ | |
3553 | bool retry_clustered; | |
3554 | ||
3555 | /* | |
3556 | * Whether we're updating free space cache, if true we need to re-search | |
3557 | * current block group but don't try updating free space cache again. | |
3558 | */ | |
3559 | bool retry_unclustered; | |
3560 | ||
3561 | /* If current block group is cached */ | |
3562 | int cached; | |
3563 | ||
3564 | /* Max contiguous hole found */ | |
3565 | u64 max_extent_size; | |
3566 | ||
3567 | /* Total free space from free space cache, not always contiguous */ | |
3568 | u64 total_free_space; | |
3569 | ||
3570 | /* Found result */ | |
3571 | u64 found_offset; | |
cb2f96f8 | 3572 | |
ea544149 NA |
3573 | /* Hint where to start looking for an empty space */ |
3574 | u64 hint_byte; | |
3575 | ||
cb2f96f8 NA |
3576 | /* Allocation policy */ |
3577 | enum btrfs_extent_allocation_policy policy; | |
b4bd745d QW |
3578 | }; |
3579 | ||
d06e3bb6 QW |
3580 | |
3581 | /* | |
3582 | * Helper function for find_free_extent(). | |
3583 | * | |
3584 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3585 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3586 | * Return >0 to inform caller that we find nothing | |
3587 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3588 | */ | |
32da5386 | 3589 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3590 | struct find_free_extent_ctl *ffe_ctl, |
3591 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3592 | { |
32da5386 | 3593 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3594 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3595 | u64 aligned_cluster; |
3596 | u64 offset; | |
3597 | int ret; | |
3598 | ||
3599 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3600 | if (!cluster_bg) | |
3601 | goto refill_cluster; | |
3602 | if (cluster_bg != bg && (cluster_bg->ro || | |
3603 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3604 | goto release_cluster; | |
3605 | ||
3606 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3607 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3608 | &ffe_ctl->max_extent_size); |
3609 | if (offset) { | |
3610 | /* We have a block, we're done */ | |
3611 | spin_unlock(&last_ptr->refill_lock); | |
3612 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3613 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3614 | *cluster_bg_ret = cluster_bg; | |
3615 | ffe_ctl->found_offset = offset; | |
3616 | return 0; | |
3617 | } | |
3618 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3619 | ||
3620 | release_cluster: | |
3621 | /* | |
3622 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3623 | * lets just skip it and let the allocator find whatever block it can | |
3624 | * find. If we reach this point, we will have tried the cluster | |
3625 | * allocator plenty of times and not have found anything, so we are | |
3626 | * likely way too fragmented for the clustering stuff to find anything. | |
3627 | * | |
3628 | * However, if the cluster is taken from the current block group, | |
3629 | * release the cluster first, so that we stand a better chance of | |
3630 | * succeeding in the unclustered allocation. | |
3631 | */ | |
3632 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3633 | spin_unlock(&last_ptr->refill_lock); | |
3634 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3635 | return -ENOENT; | |
3636 | } | |
3637 | ||
3638 | /* This cluster didn't work out, free it and start over */ | |
3639 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3640 | ||
3641 | if (cluster_bg != bg) | |
3642 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3643 | ||
3644 | refill_cluster: | |
3645 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3646 | spin_unlock(&last_ptr->refill_lock); | |
3647 | return -ENOENT; | |
3648 | } | |
3649 | ||
3650 | aligned_cluster = max_t(u64, | |
3651 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3652 | bg->full_stripe_len); | |
2ceeae2e DS |
3653 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3654 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3655 | if (ret == 0) { |
3656 | /* Now pull our allocation out of this cluster */ | |
3657 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3658 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3659 | &ffe_ctl->max_extent_size); | |
3660 | if (offset) { | |
3661 | /* We found one, proceed */ | |
3662 | spin_unlock(&last_ptr->refill_lock); | |
3663 | trace_btrfs_reserve_extent_cluster(bg, | |
3664 | ffe_ctl->search_start, | |
3665 | ffe_ctl->num_bytes); | |
3666 | ffe_ctl->found_offset = offset; | |
3667 | return 0; | |
3668 | } | |
3669 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3670 | !ffe_ctl->retry_clustered) { | |
3671 | spin_unlock(&last_ptr->refill_lock); | |
3672 | ||
3673 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3674 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3675 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3676 | return -EAGAIN; | |
3677 | } | |
3678 | /* | |
3679 | * At this point we either didn't find a cluster or we weren't able to | |
3680 | * allocate a block from our cluster. Free the cluster we've been | |
3681 | * trying to use, and go to the next block group. | |
3682 | */ | |
3683 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3684 | spin_unlock(&last_ptr->refill_lock); | |
3685 | return 1; | |
3686 | } | |
3687 | ||
e1a41848 QW |
3688 | /* |
3689 | * Return >0 to inform caller that we find nothing | |
3690 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3691 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3692 | */ | |
32da5386 | 3693 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3694 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3695 | { |
897cae79 | 3696 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3697 | u64 offset; |
3698 | ||
3699 | /* | |
3700 | * We are doing an unclustered allocation, set the fragmented flag so | |
3701 | * we don't bother trying to setup a cluster again until we get more | |
3702 | * space. | |
3703 | */ | |
3704 | if (unlikely(last_ptr)) { | |
3705 | spin_lock(&last_ptr->lock); | |
3706 | last_ptr->fragmented = 1; | |
3707 | spin_unlock(&last_ptr->lock); | |
3708 | } | |
3709 | if (ffe_ctl->cached) { | |
3710 | struct btrfs_free_space_ctl *free_space_ctl; | |
3711 | ||
3712 | free_space_ctl = bg->free_space_ctl; | |
3713 | spin_lock(&free_space_ctl->tree_lock); | |
3714 | if (free_space_ctl->free_space < | |
3715 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3716 | ffe_ctl->empty_size) { | |
3717 | ffe_ctl->total_free_space = max_t(u64, | |
3718 | ffe_ctl->total_free_space, | |
3719 | free_space_ctl->free_space); | |
3720 | spin_unlock(&free_space_ctl->tree_lock); | |
3721 | return 1; | |
3722 | } | |
3723 | spin_unlock(&free_space_ctl->tree_lock); | |
3724 | } | |
3725 | ||
3726 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3727 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3728 | &ffe_ctl->max_extent_size); | |
3729 | ||
3730 | /* | |
3731 | * If we didn't find a chunk, and we haven't failed on this block group | |
3732 | * before, and this block group is in the middle of caching and we are | |
3733 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3734 | * set @retry_unclustered to true. | |
3735 | * | |
3736 | * If @retry_unclustered is true then we've already waited on this | |
3737 | * block group once and should move on to the next block group. | |
3738 | */ | |
3739 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3740 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3741 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3742 | ffe_ctl->empty_size); | |
e1a41848 QW |
3743 | ffe_ctl->retry_unclustered = true; |
3744 | return -EAGAIN; | |
3745 | } else if (!offset) { | |
3746 | return 1; | |
3747 | } | |
3748 | ffe_ctl->found_offset = offset; | |
3749 | return 0; | |
3750 | } | |
3751 | ||
c668690d NA |
3752 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3753 | struct find_free_extent_ctl *ffe_ctl, | |
3754 | struct btrfs_block_group **bg_ret) | |
3755 | { | |
3756 | int ret; | |
3757 | ||
3758 | /* We want to try and use the cluster allocator, so lets look there */ | |
3759 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3760 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
c668690d NA |
3761 | if (ret >= 0 || ret == -EAGAIN) |
3762 | return ret; | |
3763 | /* ret == -ENOENT case falls through */ | |
3764 | } | |
3765 | ||
897cae79 | 3766 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3767 | } |
3768 | ||
3769 | static int do_allocation(struct btrfs_block_group *block_group, | |
3770 | struct find_free_extent_ctl *ffe_ctl, | |
3771 | struct btrfs_block_group **bg_ret) | |
3772 | { | |
3773 | switch (ffe_ctl->policy) { | |
3774 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3775 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
3776 | default: | |
3777 | BUG(); | |
3778 | } | |
3779 | } | |
3780 | ||
baba5062 NA |
3781 | static void release_block_group(struct btrfs_block_group *block_group, |
3782 | struct find_free_extent_ctl *ffe_ctl, | |
3783 | int delalloc) | |
3784 | { | |
3785 | switch (ffe_ctl->policy) { | |
3786 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3787 | ffe_ctl->retry_clustered = false; | |
3788 | ffe_ctl->retry_unclustered = false; | |
3789 | break; | |
3790 | default: | |
3791 | BUG(); | |
3792 | } | |
3793 | ||
3794 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
3795 | ffe_ctl->index); | |
3796 | btrfs_release_block_group(block_group, delalloc); | |
3797 | } | |
3798 | ||
0ab9724b NA |
3799 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
3800 | struct btrfs_key *ins) | |
3801 | { | |
3802 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3803 | ||
3804 | if (!ffe_ctl->use_cluster && last_ptr) { | |
3805 | spin_lock(&last_ptr->lock); | |
3806 | last_ptr->window_start = ins->objectid; | |
3807 | spin_unlock(&last_ptr->lock); | |
3808 | } | |
3809 | } | |
3810 | ||
3811 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
3812 | struct btrfs_key *ins) | |
3813 | { | |
3814 | switch (ffe_ctl->policy) { | |
3815 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3816 | found_extent_clustered(ffe_ctl, ins); | |
3817 | break; | |
3818 | default: | |
3819 | BUG(); | |
3820 | } | |
3821 | } | |
3822 | ||
c70e2139 NA |
3823 | static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl) |
3824 | { | |
3825 | switch (ffe_ctl->policy) { | |
3826 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3827 | /* | |
3828 | * If we can't allocate a new chunk we've already looped through | |
3829 | * at least once, move on to the NO_EMPTY_SIZE case. | |
3830 | */ | |
3831 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3832 | return 0; | |
3833 | default: | |
3834 | BUG(); | |
3835 | } | |
3836 | } | |
3837 | ||
e72d79d6 QW |
3838 | /* |
3839 | * Return >0 means caller needs to re-search for free extent | |
3840 | * Return 0 means we have the needed free extent. | |
3841 | * Return <0 means we failed to locate any free extent. | |
3842 | */ | |
3843 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
3844 | struct btrfs_key *ins, |
3845 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 3846 | bool full_search) |
e72d79d6 QW |
3847 | { |
3848 | struct btrfs_root *root = fs_info->extent_root; | |
3849 | int ret; | |
3850 | ||
3851 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3852 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3853 | ffe_ctl->orig_have_caching_bg = true; | |
3854 | ||
3855 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3856 | ffe_ctl->have_caching_bg) | |
3857 | return 1; | |
3858 | ||
3859 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3860 | return 1; | |
3861 | ||
3862 | if (ins->objectid) { | |
0ab9724b | 3863 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
3864 | return 0; |
3865 | } | |
3866 | ||
3867 | /* | |
3868 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3869 | * caching kthreads as we move along | |
3870 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3871 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3872 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3873 | * again | |
3874 | */ | |
3875 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3876 | ffe_ctl->index = 0; | |
3877 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3878 | /* | |
3879 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3880 | * don't have any uncached bgs and we've already done a | |
3881 | * full search through. | |
3882 | */ | |
3883 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3884 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3885 | else | |
3886 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3887 | } else { | |
3888 | ffe_ctl->loop++; | |
3889 | } | |
3890 | ||
3891 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3892 | struct btrfs_trans_handle *trans; | |
3893 | int exist = 0; | |
3894 | ||
3895 | trans = current->journal_info; | |
3896 | if (trans) | |
3897 | exist = 1; | |
3898 | else | |
3899 | trans = btrfs_join_transaction(root); | |
3900 | ||
3901 | if (IS_ERR(trans)) { | |
3902 | ret = PTR_ERR(trans); | |
3903 | return ret; | |
3904 | } | |
3905 | ||
fc471cb0 JB |
3906 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3907 | CHUNK_ALLOC_FORCE); | |
e72d79d6 | 3908 | |
e72d79d6 | 3909 | /* Do not bail out on ENOSPC since we can do more. */ |
c70e2139 NA |
3910 | if (ret == -ENOSPC) |
3911 | ret = chunk_allocation_failed(ffe_ctl); | |
3912 | else if (ret < 0) | |
e72d79d6 QW |
3913 | btrfs_abort_transaction(trans, ret); |
3914 | else | |
3915 | ret = 0; | |
3916 | if (!exist) | |
3917 | btrfs_end_transaction(trans); | |
3918 | if (ret) | |
3919 | return ret; | |
3920 | } | |
3921 | ||
3922 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
3923 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
3924 | return -ENOSPC; | |
3925 | ||
e72d79d6 QW |
3926 | /* |
3927 | * Don't loop again if we already have no empty_size and | |
3928 | * no empty_cluster. | |
3929 | */ | |
3930 | if (ffe_ctl->empty_size == 0 && | |
3931 | ffe_ctl->empty_cluster == 0) | |
3932 | return -ENOSPC; | |
3933 | ffe_ctl->empty_size = 0; | |
3934 | ffe_ctl->empty_cluster = 0; | |
3935 | } | |
3936 | return 1; | |
3937 | } | |
3938 | return -ENOSPC; | |
3939 | } | |
3940 | ||
7e895409 NA |
3941 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
3942 | struct find_free_extent_ctl *ffe_ctl, | |
3943 | struct btrfs_space_info *space_info, | |
3944 | struct btrfs_key *ins) | |
3945 | { | |
3946 | /* | |
3947 | * If our free space is heavily fragmented we may not be able to make | |
3948 | * big contiguous allocations, so instead of doing the expensive search | |
3949 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3950 | * can go ahead and search for a more manageable chunk. | |
3951 | * | |
3952 | * If our max_extent_size is large enough for our allocation simply | |
3953 | * disable clustering since we will likely not be able to find enough | |
3954 | * space to create a cluster and induce latency trying. | |
3955 | */ | |
3956 | if (space_info->max_extent_size) { | |
3957 | spin_lock(&space_info->lock); | |
3958 | if (space_info->max_extent_size && | |
3959 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
3960 | ins->offset = space_info->max_extent_size; | |
3961 | spin_unlock(&space_info->lock); | |
3962 | return -ENOSPC; | |
3963 | } else if (space_info->max_extent_size) { | |
3964 | ffe_ctl->use_cluster = false; | |
3965 | } | |
3966 | spin_unlock(&space_info->lock); | |
3967 | } | |
3968 | ||
3969 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
3970 | &ffe_ctl->empty_cluster); | |
3971 | if (ffe_ctl->last_ptr) { | |
3972 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3973 | ||
3974 | spin_lock(&last_ptr->lock); | |
3975 | if (last_ptr->block_group) | |
3976 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3977 | if (last_ptr->fragmented) { | |
3978 | /* | |
3979 | * We still set window_start so we can keep track of the | |
3980 | * last place we found an allocation to try and save | |
3981 | * some time. | |
3982 | */ | |
3983 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3984 | ffe_ctl->use_cluster = false; | |
3985 | } | |
3986 | spin_unlock(&last_ptr->lock); | |
3987 | } | |
3988 | ||
3989 | return 0; | |
3990 | } | |
3991 | ||
3992 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
3993 | struct find_free_extent_ctl *ffe_ctl, | |
3994 | struct btrfs_space_info *space_info, | |
3995 | struct btrfs_key *ins) | |
3996 | { | |
3997 | switch (ffe_ctl->policy) { | |
3998 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3999 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
4000 | space_info, ins); | |
4001 | default: | |
4002 | BUG(); | |
4003 | } | |
4004 | } | |
4005 | ||
fec577fb CM |
4006 | /* |
4007 | * walks the btree of allocated extents and find a hole of a given size. | |
4008 | * The key ins is changed to record the hole: | |
a4820398 | 4009 | * ins->objectid == start position |
62e2749e | 4010 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4011 | * ins->offset == the size of the hole. |
fec577fb | 4012 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4013 | * |
4014 | * If there is no suitable free space, we will record the max size of | |
4015 | * the free space extent currently. | |
e72d79d6 QW |
4016 | * |
4017 | * The overall logic and call chain: | |
4018 | * | |
4019 | * find_free_extent() | |
4020 | * |- Iterate through all block groups | |
4021 | * | |- Get a valid block group | |
4022 | * | |- Try to do clustered allocation in that block group | |
4023 | * | |- Try to do unclustered allocation in that block group | |
4024 | * | |- Check if the result is valid | |
4025 | * | | |- If valid, then exit | |
4026 | * | |- Jump to next block group | |
4027 | * | | |
4028 | * |- Push harder to find free extents | |
4029 | * |- If not found, re-iterate all block groups | |
fec577fb | 4030 | */ |
437490fe | 4031 | static noinline int find_free_extent(struct btrfs_root *root, |
18513091 | 4032 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
ea544149 | 4033 | u64 hint_byte_orig, struct btrfs_key *ins, |
18513091 | 4034 | u64 flags, int delalloc) |
fec577fb | 4035 | { |
437490fe | 4036 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4037 | int ret = 0; |
db8fe64f | 4038 | int cache_block_group_error = 0; |
32da5386 | 4039 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 4040 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 4041 | struct btrfs_space_info *space_info; |
a5e681d9 | 4042 | bool full_search = false; |
fec577fb | 4043 | |
0b246afa | 4044 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 4045 | |
b4bd745d QW |
4046 | ffe_ctl.num_bytes = num_bytes; |
4047 | ffe_ctl.empty_size = empty_size; | |
4048 | ffe_ctl.flags = flags; | |
4049 | ffe_ctl.search_start = 0; | |
b4bd745d QW |
4050 | ffe_ctl.delalloc = delalloc; |
4051 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
4052 | ffe_ctl.have_caching_bg = false; | |
4053 | ffe_ctl.orig_have_caching_bg = false; | |
4054 | ffe_ctl.found_offset = 0; | |
ea544149 | 4055 | ffe_ctl.hint_byte = hint_byte_orig; |
cb2f96f8 | 4056 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED; |
b4bd745d | 4057 | |
c10859be NA |
4058 | /* For clustered allocation */ |
4059 | ffe_ctl.retry_clustered = false; | |
4060 | ffe_ctl.retry_unclustered = false; | |
4061 | ffe_ctl.last_ptr = NULL; | |
4062 | ffe_ctl.use_cluster = true; | |
4063 | ||
962a298f | 4064 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4065 | ins->objectid = 0; |
4066 | ins->offset = 0; | |
b1a4d965 | 4067 | |
437490fe | 4068 | trace_find_free_extent(root, num_bytes, empty_size, flags); |
3f7de037 | 4069 | |
280c2908 | 4070 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 4071 | if (!space_info) { |
0b246afa | 4072 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
4073 | return -ENOSPC; |
4074 | } | |
2552d17e | 4075 | |
7e895409 NA |
4076 | ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins); |
4077 | if (ret < 0) | |
4078 | return ret; | |
fa9c0d79 | 4079 | |
b4bd745d QW |
4080 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
4081 | first_logical_byte(fs_info, 0)); | |
ea544149 NA |
4082 | ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte); |
4083 | if (ffe_ctl.search_start == ffe_ctl.hint_byte) { | |
b4bd745d QW |
4084 | block_group = btrfs_lookup_block_group(fs_info, |
4085 | ffe_ctl.search_start); | |
817d52f8 JB |
4086 | /* |
4087 | * we don't want to use the block group if it doesn't match our | |
4088 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4089 | * |
4090 | * However if we are re-searching with an ideal block group | |
4091 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4092 | */ |
b6919a58 | 4093 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 4094 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4095 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4096 | if (list_empty(&block_group->list) || |
4097 | block_group->ro) { | |
4098 | /* | |
4099 | * someone is removing this block group, | |
4100 | * we can't jump into the have_block_group | |
4101 | * target because our list pointers are not | |
4102 | * valid | |
4103 | */ | |
4104 | btrfs_put_block_group(block_group); | |
4105 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4106 | } else { |
b4bd745d | 4107 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 4108 | block_group->flags); |
e570fd27 | 4109 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 4110 | goto have_block_group; |
ccf0e725 | 4111 | } |
2552d17e | 4112 | } else if (block_group) { |
fa9c0d79 | 4113 | btrfs_put_block_group(block_group); |
2552d17e | 4114 | } |
42e70e7a | 4115 | } |
2552d17e | 4116 | search: |
b4bd745d QW |
4117 | ffe_ctl.have_caching_bg = false; |
4118 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
4119 | ffe_ctl.index == 0) | |
a5e681d9 | 4120 | full_search = true; |
80eb234a | 4121 | down_read(&space_info->groups_sem); |
b4bd745d QW |
4122 | list_for_each_entry(block_group, |
4123 | &space_info->block_groups[ffe_ctl.index], list) { | |
c668690d NA |
4124 | struct btrfs_block_group *bg_ret; |
4125 | ||
14443937 JM |
4126 | /* If the block group is read-only, we can skip it entirely. */ |
4127 | if (unlikely(block_group->ro)) | |
4128 | continue; | |
4129 | ||
e570fd27 | 4130 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 4131 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 4132 | |
83a50de9 CM |
4133 | /* |
4134 | * this can happen if we end up cycling through all the | |
4135 | * raid types, but we want to make sure we only allocate | |
4136 | * for the proper type. | |
4137 | */ | |
b6919a58 | 4138 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 4139 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4140 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4141 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4142 | BTRFS_BLOCK_GROUP_RAID10; |
4143 | ||
4144 | /* | |
4145 | * if they asked for extra copies and this block group | |
4146 | * doesn't provide them, bail. This does allow us to | |
4147 | * fill raid0 from raid1. | |
4148 | */ | |
b6919a58 | 4149 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4150 | goto loop; |
2a28468e QW |
4151 | |
4152 | /* | |
4153 | * This block group has different flags than we want. | |
4154 | * It's possible that we have MIXED_GROUP flag but no | |
4155 | * block group is mixed. Just skip such block group. | |
4156 | */ | |
4157 | btrfs_release_block_group(block_group, delalloc); | |
4158 | continue; | |
83a50de9 CM |
4159 | } |
4160 | ||
2552d17e | 4161 | have_block_group: |
32da5386 | 4162 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
4163 | if (unlikely(!ffe_ctl.cached)) { |
4164 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 4165 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
4166 | |
4167 | /* | |
4168 | * If we get ENOMEM here or something else we want to | |
4169 | * try other block groups, because it may not be fatal. | |
4170 | * However if we can't find anything else we need to | |
4171 | * save our return here so that we return the actual | |
4172 | * error that caused problems, not ENOSPC. | |
4173 | */ | |
4174 | if (ret < 0) { | |
4175 | if (!cache_block_group_error) | |
4176 | cache_block_group_error = ret; | |
4177 | ret = 0; | |
4178 | goto loop; | |
4179 | } | |
1d4284bd | 4180 | ret = 0; |
817d52f8 JB |
4181 | } |
4182 | ||
36cce922 JB |
4183 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
4184 | goto loop; | |
0f9dd46c | 4185 | |
c668690d NA |
4186 | bg_ret = NULL; |
4187 | ret = do_allocation(block_group, &ffe_ctl, &bg_ret); | |
4188 | if (ret == 0) { | |
4189 | if (bg_ret && bg_ret != block_group) { | |
4190 | btrfs_release_block_group(block_group, delalloc); | |
4191 | block_group = bg_ret; | |
fa9c0d79 | 4192 | } |
c668690d | 4193 | } else if (ret == -EAGAIN) { |
817d52f8 | 4194 | goto have_block_group; |
c668690d | 4195 | } else if (ret > 0) { |
1cdda9b8 | 4196 | goto loop; |
c668690d NA |
4197 | } |
4198 | ||
4199 | /* Checks */ | |
b4bd745d QW |
4200 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4201 | fs_info->stripesize); | |
25179201 | 4202 | |
2552d17e | 4203 | /* move on to the next group */ |
b4bd745d | 4204 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4205 | block_group->start + block_group->length) { |
b4bd745d QW |
4206 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4207 | num_bytes); | |
2552d17e | 4208 | goto loop; |
6226cb0a | 4209 | } |
f5a31e16 | 4210 | |
b4bd745d QW |
4211 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
4212 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
4213 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4214 | |
18513091 WX |
4215 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4216 | num_bytes, delalloc); | |
f0486c68 | 4217 | if (ret == -EAGAIN) { |
b4bd745d QW |
4218 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4219 | num_bytes); | |
2552d17e | 4220 | goto loop; |
0f9dd46c | 4221 | } |
9cfa3e34 | 4222 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4223 | |
f0486c68 | 4224 | /* we are all good, lets return */ |
b4bd745d | 4225 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4226 | ins->offset = num_bytes; |
d2fb3437 | 4227 | |
b4bd745d QW |
4228 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4229 | num_bytes); | |
e570fd27 | 4230 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4231 | break; |
4232 | loop: | |
baba5062 | 4233 | release_block_group(block_group, &ffe_ctl, delalloc); |
14443937 | 4234 | cond_resched(); |
2552d17e JB |
4235 | } |
4236 | up_read(&space_info->groups_sem); | |
4237 | ||
15b7ee65 | 4238 | ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search); |
e72d79d6 | 4239 | if (ret > 0) |
b742bb82 YZ |
4240 | goto search; |
4241 | ||
db8fe64f | 4242 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4243 | /* |
4244 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4245 | * any contiguous hole. | |
4246 | */ | |
4247 | if (!ffe_ctl.max_extent_size) | |
4248 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4249 | spin_lock(&space_info->lock); |
b4bd745d | 4250 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4251 | spin_unlock(&space_info->lock); |
b4bd745d | 4252 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4253 | } else if (ret == -ENOSPC) { |
4254 | ret = cache_block_group_error; | |
4f4db217 | 4255 | } |
0f70abe2 | 4256 | return ret; |
fec577fb | 4257 | } |
ec44a35c | 4258 | |
6f47c706 NB |
4259 | /* |
4260 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4261 | * hole that is at least as big as @num_bytes. | |
4262 | * | |
4263 | * @root - The root that will contain this extent | |
4264 | * | |
4265 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4266 | * is used for accounting purposes. This value differs | |
4267 | * from @num_bytes only in the case of compressed extents. | |
4268 | * | |
4269 | * @num_bytes - Number of bytes to allocate on-disk. | |
4270 | * | |
4271 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4272 | * allocator should try to satisfy. In some cases | |
4273 | * @num_bytes may be larger than what is required and if | |
4274 | * the filesystem is fragmented then allocation fails. | |
4275 | * However, the presence of @min_alloc_size gives a | |
4276 | * chance to try and satisfy the smaller allocation. | |
4277 | * | |
4278 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4279 | * size in bytes the allocator should try to find free | |
4280 | * next to the block it returns. This is just a hint and | |
4281 | * may be ignored by the allocator. | |
4282 | * | |
4283 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4284 | * address passed. It might be ignored. | |
4285 | * | |
4286 | * @ins - This key is modified to record the found hole. It will | |
4287 | * have the following values: | |
4288 | * ins->objectid == start position | |
4289 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4290 | * ins->offset == the size of the hole. | |
4291 | * | |
4292 | * @is_data - Boolean flag indicating whether an extent is | |
4293 | * allocated for data (true) or metadata (false) | |
4294 | * | |
4295 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4296 | * delalloc or not. If 'true' data_rwsem of block groups | |
4297 | * is going to be acquired. | |
4298 | * | |
4299 | * | |
4300 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4301 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4302 | * largest available hole the allocator managed to find. | |
4303 | */ | |
18513091 | 4304 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4305 | u64 num_bytes, u64 min_alloc_size, |
4306 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4307 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4308 | { |
ab8d0fc4 | 4309 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4310 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4311 | u64 flags; |
fec577fb | 4312 | int ret; |
925baedd | 4313 | |
1b86826d | 4314 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4315 | again: |
0b246afa | 4316 | WARN_ON(num_bytes < fs_info->sectorsize); |
437490fe | 4317 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
18513091 | 4318 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4319 | if (!ret && !is_data) { |
ab8d0fc4 | 4320 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4321 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4322 | if (!final_tried && ins->offset) { |
4323 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4324 | num_bytes = round_down(num_bytes, |
0b246afa | 4325 | fs_info->sectorsize); |
9e622d6b | 4326 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4327 | ram_bytes = num_bytes; |
9e622d6b MX |
4328 | if (num_bytes == min_alloc_size) |
4329 | final_tried = true; | |
4330 | goto again; | |
ab8d0fc4 | 4331 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4332 | struct btrfs_space_info *sinfo; |
4333 | ||
280c2908 | 4334 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4335 | btrfs_err(fs_info, |
5d163e0e JM |
4336 | "allocation failed flags %llu, wanted %llu", |
4337 | flags, num_bytes); | |
53804280 | 4338 | if (sinfo) |
5da6afeb JB |
4339 | btrfs_dump_space_info(fs_info, sinfo, |
4340 | num_bytes, 1); | |
9e622d6b | 4341 | } |
925baedd | 4342 | } |
0f9dd46c JB |
4343 | |
4344 | return ret; | |
e6dcd2dc CM |
4345 | } |
4346 | ||
a0fbf736 NB |
4347 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4348 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4349 | { |
32da5386 | 4350 | struct btrfs_block_group *cache; |
0f9dd46c | 4351 | |
0b246afa | 4352 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4353 | if (!cache) { |
0b246afa JM |
4354 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4355 | start); | |
0f9dd46c JB |
4356 | return -ENOSPC; |
4357 | } | |
1f3c79a2 | 4358 | |
a0fbf736 NB |
4359 | btrfs_add_free_space(cache, start, len); |
4360 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4361 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4362 | ||
fa9c0d79 | 4363 | btrfs_put_block_group(cache); |
a0fbf736 | 4364 | return 0; |
e6dcd2dc CM |
4365 | } |
4366 | ||
7bfc1007 NB |
4367 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, |
4368 | u64 len) | |
e688b725 | 4369 | { |
7ef54d54 | 4370 | struct btrfs_block_group *cache; |
a0fbf736 | 4371 | int ret = 0; |
7ef54d54 | 4372 | |
7bfc1007 | 4373 | cache = btrfs_lookup_block_group(trans->fs_info, start); |
7ef54d54 | 4374 | if (!cache) { |
7bfc1007 NB |
4375 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
4376 | start); | |
7ef54d54 NB |
4377 | return -ENOSPC; |
4378 | } | |
4379 | ||
6690d071 | 4380 | ret = pin_down_extent(trans, cache, start, len, 1); |
7ef54d54 | 4381 | btrfs_put_block_group(cache); |
a0fbf736 | 4382 | return ret; |
e688b725 CM |
4383 | } |
4384 | ||
5d4f98a2 | 4385 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4386 | u64 parent, u64 root_objectid, |
4387 | u64 flags, u64 owner, u64 offset, | |
4388 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4389 | { |
ef89b824 | 4390 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4391 | int ret; |
e6dcd2dc | 4392 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4393 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4394 | struct btrfs_path *path; |
5d4f98a2 YZ |
4395 | struct extent_buffer *leaf; |
4396 | int type; | |
4397 | u32 size; | |
26b8003f | 4398 | |
5d4f98a2 YZ |
4399 | if (parent > 0) |
4400 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4401 | else | |
4402 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4403 | |
5d4f98a2 | 4404 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4405 | |
4406 | path = btrfs_alloc_path(); | |
db5b493a TI |
4407 | if (!path) |
4408 | return -ENOMEM; | |
47e4bb98 | 4409 | |
5d4f98a2 YZ |
4410 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4411 | ins, size); | |
79787eaa JM |
4412 | if (ret) { |
4413 | btrfs_free_path(path); | |
4414 | return ret; | |
4415 | } | |
0f9dd46c | 4416 | |
5d4f98a2 YZ |
4417 | leaf = path->nodes[0]; |
4418 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4419 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4420 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4421 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4422 | btrfs_set_extent_flags(leaf, extent_item, | |
4423 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4424 | ||
4425 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4426 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4427 | if (parent > 0) { | |
4428 | struct btrfs_shared_data_ref *ref; | |
4429 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4430 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4431 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4432 | } else { | |
4433 | struct btrfs_extent_data_ref *ref; | |
4434 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4435 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4436 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4437 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4438 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4439 | } | |
47e4bb98 CM |
4440 | |
4441 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4442 | btrfs_free_path(path); |
f510cfec | 4443 | |
25a356d3 | 4444 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4445 | if (ret) |
4446 | return ret; | |
4447 | ||
ade4b516 | 4448 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4449 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4450 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4451 | ins->objectid, ins->offset); |
f5947066 CM |
4452 | BUG(); |
4453 | } | |
71ff6437 | 4454 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4455 | return ret; |
4456 | } | |
4457 | ||
5d4f98a2 | 4458 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4459 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4460 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4461 | { |
9dcdbe01 | 4462 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4463 | int ret; |
5d4f98a2 | 4464 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4465 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4466 | struct btrfs_tree_block_info *block_info; |
4467 | struct btrfs_extent_inline_ref *iref; | |
4468 | struct btrfs_path *path; | |
4469 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4470 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4471 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4472 | u64 num_bytes; |
21ebfbe7 | 4473 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4474 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4475 | |
4e6bd4e0 NB |
4476 | ref = btrfs_delayed_node_to_tree_ref(node); |
4477 | ||
4e6bd4e0 NB |
4478 | extent_key.objectid = node->bytenr; |
4479 | if (skinny_metadata) { | |
4480 | extent_key.offset = ref->level; | |
4481 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4482 | num_bytes = fs_info->nodesize; | |
4483 | } else { | |
4484 | extent_key.offset = node->num_bytes; | |
4485 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4486 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4487 | num_bytes = node->num_bytes; |
4488 | } | |
1c2308f8 | 4489 | |
5d4f98a2 | 4490 | path = btrfs_alloc_path(); |
80ee54bf | 4491 | if (!path) |
d8926bb3 | 4492 | return -ENOMEM; |
56bec294 | 4493 | |
5d4f98a2 | 4494 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4e6bd4e0 | 4495 | &extent_key, size); |
79787eaa | 4496 | if (ret) { |
dd825259 | 4497 | btrfs_free_path(path); |
79787eaa JM |
4498 | return ret; |
4499 | } | |
5d4f98a2 YZ |
4500 | |
4501 | leaf = path->nodes[0]; | |
4502 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4503 | struct btrfs_extent_item); | |
4504 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4505 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4506 | btrfs_set_extent_flags(leaf, extent_item, | |
4507 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4508 | |
3173a18f JB |
4509 | if (skinny_metadata) { |
4510 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4511 | } else { | |
4512 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4513 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4514 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4515 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4516 | } | |
5d4f98a2 | 4517 | |
d4b20733 | 4518 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4519 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4520 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4521 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4522 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4523 | } else { |
4524 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4525 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4526 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4527 | } |
4528 | ||
4529 | btrfs_mark_buffer_dirty(leaf); | |
4530 | btrfs_free_path(path); | |
4531 | ||
4e6bd4e0 NB |
4532 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4533 | num_bytes); | |
1e144fb8 OS |
4534 | if (ret) |
4535 | return ret; | |
4536 | ||
ade4b516 JB |
4537 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4538 | fs_info->nodesize, 1); | |
79787eaa | 4539 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4540 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4541 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4542 | BUG(); |
4543 | } | |
0be5dc67 | 4544 | |
4e6bd4e0 | 4545 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4546 | fs_info->nodesize); |
5d4f98a2 YZ |
4547 | return ret; |
4548 | } | |
4549 | ||
4550 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4551 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4552 | u64 offset, u64 ram_bytes, |
4553 | struct btrfs_key *ins) | |
5d4f98a2 | 4554 | { |
76675593 | 4555 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
4556 | int ret; |
4557 | ||
84f7d8e6 | 4558 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4559 | |
76675593 QW |
4560 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4561 | ins->objectid, ins->offset, 0); | |
4562 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4563 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
4564 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
4565 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
4566 | return ret; |
4567 | } | |
e02119d5 CM |
4568 | |
4569 | /* | |
4570 | * this is used by the tree logging recovery code. It records that | |
4571 | * an extent has been allocated and makes sure to clear the free | |
4572 | * space cache bits as well | |
4573 | */ | |
5d4f98a2 | 4574 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4575 | u64 root_objectid, u64 owner, u64 offset, |
4576 | struct btrfs_key *ins) | |
e02119d5 | 4577 | { |
61da2abf | 4578 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4579 | int ret; |
32da5386 | 4580 | struct btrfs_block_group *block_group; |
ed7a6948 | 4581 | struct btrfs_space_info *space_info; |
11833d66 | 4582 | |
8c2a1a30 JB |
4583 | /* |
4584 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4585 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4586 | */ |
0b246afa | 4587 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4588 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4589 | ins->offset); | |
b50c6e25 | 4590 | if (ret) |
8c2a1a30 | 4591 | return ret; |
11833d66 YZ |
4592 | } |
4593 | ||
0b246afa | 4594 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4595 | if (!block_group) |
4596 | return -EINVAL; | |
4597 | ||
ed7a6948 WX |
4598 | space_info = block_group->space_info; |
4599 | spin_lock(&space_info->lock); | |
4600 | spin_lock(&block_group->lock); | |
4601 | space_info->bytes_reserved += ins->offset; | |
4602 | block_group->reserved += ins->offset; | |
4603 | spin_unlock(&block_group->lock); | |
4604 | spin_unlock(&space_info->lock); | |
4605 | ||
ef89b824 NB |
4606 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4607 | offset, ins, 1); | |
bd727173 | 4608 | if (ret) |
ab9b2c7b | 4609 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
b50c6e25 | 4610 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4611 | return ret; |
4612 | } | |
4613 | ||
48a3b636 ES |
4614 | static struct extent_buffer * |
4615 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
9631e4cc JB |
4616 | u64 bytenr, int level, u64 owner, |
4617 | enum btrfs_lock_nesting nest) | |
65b51a00 | 4618 | { |
0b246afa | 4619 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4620 | struct extent_buffer *buf; |
4621 | ||
3fbaf258 | 4622 | buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level); |
c871b0f2 LB |
4623 | if (IS_ERR(buf)) |
4624 | return buf; | |
4625 | ||
b72c3aba QW |
4626 | /* |
4627 | * Extra safety check in case the extent tree is corrupted and extent | |
4628 | * allocator chooses to use a tree block which is already used and | |
4629 | * locked. | |
4630 | */ | |
4631 | if (buf->lock_owner == current->pid) { | |
4632 | btrfs_err_rl(fs_info, | |
4633 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4634 | buf->start, btrfs_header_owner(buf), current->pid); | |
4635 | free_extent_buffer(buf); | |
4636 | return ERR_PTR(-EUCLEAN); | |
4637 | } | |
4638 | ||
e114c545 JB |
4639 | /* |
4640 | * This needs to stay, because we could allocate a freed block from an | |
4641 | * old tree into a new tree, so we need to make sure this new block is | |
4642 | * set to the appropriate level and owner. | |
4643 | */ | |
ad244665 | 4644 | btrfs_set_buffer_lockdep_class(owner, buf, level); |
9631e4cc | 4645 | __btrfs_tree_lock(buf, nest); |
6a884d7d | 4646 | btrfs_clean_tree_block(buf); |
3083ee2e | 4647 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 4648 | |
4db8c528 | 4649 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4650 | |
bc877d28 NB |
4651 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4652 | btrfs_set_header_level(buf, level); | |
4653 | btrfs_set_header_bytenr(buf, buf->start); | |
4654 | btrfs_set_header_generation(buf, trans->transid); | |
4655 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4656 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4657 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4658 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4659 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4660 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4661 | /* |
4662 | * we allow two log transactions at a time, use different | |
52042d8e | 4663 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4664 | */ |
656f30db | 4665 | if (buf->log_index == 0) |
8cef4e16 YZ |
4666 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4667 | buf->start + buf->len - 1, GFP_NOFS); | |
4668 | else | |
4669 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4670 | buf->start + buf->len - 1); |
d0c803c4 | 4671 | } else { |
656f30db | 4672 | buf->log_index = -1; |
d0c803c4 | 4673 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4674 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4675 | } |
64c12921 | 4676 | trans->dirty = true; |
b4ce94de | 4677 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4678 | return buf; |
4679 | } | |
4680 | ||
fec577fb | 4681 | /* |
f0486c68 | 4682 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4683 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4684 | */ |
4d75f8a9 | 4685 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4686 | struct btrfs_root *root, |
4687 | u64 parent, u64 root_objectid, | |
4688 | const struct btrfs_disk_key *key, | |
4689 | int level, u64 hint, | |
9631e4cc JB |
4690 | u64 empty_size, |
4691 | enum btrfs_lock_nesting nest) | |
fec577fb | 4692 | { |
0b246afa | 4693 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4694 | struct btrfs_key ins; |
f0486c68 | 4695 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4696 | struct extent_buffer *buf; |
67b7859e | 4697 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4698 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4699 | u64 flags = 0; |
4700 | int ret; | |
0b246afa JM |
4701 | u32 blocksize = fs_info->nodesize; |
4702 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4703 | |
05653ef3 | 4704 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4705 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4706 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
9631e4cc | 4707 | level, root_objectid, nest); |
faa2dbf0 JB |
4708 | if (!IS_ERR(buf)) |
4709 | root->alloc_bytenr += blocksize; | |
4710 | return buf; | |
4711 | } | |
05653ef3 | 4712 | #endif |
fccb84c9 | 4713 | |
67f9c220 | 4714 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4715 | if (IS_ERR(block_rsv)) |
4716 | return ERR_CAST(block_rsv); | |
4717 | ||
18513091 | 4718 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4719 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4720 | if (ret) |
4721 | goto out_unuse; | |
55c69072 | 4722 | |
bc877d28 | 4723 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
9631e4cc | 4724 | root_objectid, nest); |
67b7859e OS |
4725 | if (IS_ERR(buf)) { |
4726 | ret = PTR_ERR(buf); | |
4727 | goto out_free_reserved; | |
4728 | } | |
f0486c68 YZ |
4729 | |
4730 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4731 | if (parent == 0) | |
4732 | parent = ins.objectid; | |
4733 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4734 | } else | |
4735 | BUG_ON(parent > 0); | |
4736 | ||
4737 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4738 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4739 | if (!extent_op) { |
4740 | ret = -ENOMEM; | |
4741 | goto out_free_buf; | |
4742 | } | |
f0486c68 YZ |
4743 | if (key) |
4744 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4745 | else | |
4746 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4747 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4748 | extent_op->update_key = skinny_metadata ? false : true; |
4749 | extent_op->update_flags = true; | |
4750 | extent_op->is_data = false; | |
b1c79e09 | 4751 | extent_op->level = level; |
f0486c68 | 4752 | |
ed4f255b QW |
4753 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4754 | ins.objectid, ins.offset, parent); | |
4755 | generic_ref.real_root = root->root_key.objectid; | |
4756 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4757 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 4758 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 4759 | extent_op, NULL, NULL); |
67b7859e OS |
4760 | if (ret) |
4761 | goto out_free_delayed; | |
f0486c68 | 4762 | } |
fec577fb | 4763 | return buf; |
67b7859e OS |
4764 | |
4765 | out_free_delayed: | |
4766 | btrfs_free_delayed_extent_op(extent_op); | |
4767 | out_free_buf: | |
4768 | free_extent_buffer(buf); | |
4769 | out_free_reserved: | |
2ff7e61e | 4770 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4771 | out_unuse: |
67f9c220 | 4772 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4773 | return ERR_PTR(ret); |
fec577fb | 4774 | } |
a28ec197 | 4775 | |
2c47e605 YZ |
4776 | struct walk_control { |
4777 | u64 refs[BTRFS_MAX_LEVEL]; | |
4778 | u64 flags[BTRFS_MAX_LEVEL]; | |
4779 | struct btrfs_key update_progress; | |
aea6f028 JB |
4780 | struct btrfs_key drop_progress; |
4781 | int drop_level; | |
2c47e605 YZ |
4782 | int stage; |
4783 | int level; | |
4784 | int shared_level; | |
4785 | int update_ref; | |
4786 | int keep_locks; | |
1c4850e2 YZ |
4787 | int reada_slot; |
4788 | int reada_count; | |
78c52d9e | 4789 | int restarted; |
2c47e605 YZ |
4790 | }; |
4791 | ||
4792 | #define DROP_REFERENCE 1 | |
4793 | #define UPDATE_BACKREF 2 | |
4794 | ||
1c4850e2 YZ |
4795 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4796 | struct btrfs_root *root, | |
4797 | struct walk_control *wc, | |
4798 | struct btrfs_path *path) | |
6407bf6d | 4799 | { |
0b246afa | 4800 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4801 | u64 bytenr; |
4802 | u64 generation; | |
4803 | u64 refs; | |
94fcca9f | 4804 | u64 flags; |
5d4f98a2 | 4805 | u32 nritems; |
1c4850e2 YZ |
4806 | struct btrfs_key key; |
4807 | struct extent_buffer *eb; | |
6407bf6d | 4808 | int ret; |
1c4850e2 YZ |
4809 | int slot; |
4810 | int nread = 0; | |
6407bf6d | 4811 | |
1c4850e2 YZ |
4812 | if (path->slots[wc->level] < wc->reada_slot) { |
4813 | wc->reada_count = wc->reada_count * 2 / 3; | |
4814 | wc->reada_count = max(wc->reada_count, 2); | |
4815 | } else { | |
4816 | wc->reada_count = wc->reada_count * 3 / 2; | |
4817 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4818 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4819 | } |
7bb86316 | 4820 | |
1c4850e2 YZ |
4821 | eb = path->nodes[wc->level]; |
4822 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4823 | |
1c4850e2 YZ |
4824 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4825 | if (nread >= wc->reada_count) | |
4826 | break; | |
bd56b302 | 4827 | |
2dd3e67b | 4828 | cond_resched(); |
1c4850e2 YZ |
4829 | bytenr = btrfs_node_blockptr(eb, slot); |
4830 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4831 | |
1c4850e2 YZ |
4832 | if (slot == path->slots[wc->level]) |
4833 | goto reada; | |
5d4f98a2 | 4834 | |
1c4850e2 YZ |
4835 | if (wc->stage == UPDATE_BACKREF && |
4836 | generation <= root->root_key.offset) | |
bd56b302 CM |
4837 | continue; |
4838 | ||
94fcca9f | 4839 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4840 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4841 | wc->level - 1, 1, &refs, |
4842 | &flags); | |
79787eaa JM |
4843 | /* We don't care about errors in readahead. */ |
4844 | if (ret < 0) | |
4845 | continue; | |
94fcca9f YZ |
4846 | BUG_ON(refs == 0); |
4847 | ||
1c4850e2 | 4848 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4849 | if (refs == 1) |
4850 | goto reada; | |
bd56b302 | 4851 | |
94fcca9f YZ |
4852 | if (wc->level == 1 && |
4853 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4854 | continue; | |
1c4850e2 YZ |
4855 | if (!wc->update_ref || |
4856 | generation <= root->root_key.offset) | |
4857 | continue; | |
4858 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4859 | ret = btrfs_comp_cpu_keys(&key, | |
4860 | &wc->update_progress); | |
4861 | if (ret < 0) | |
4862 | continue; | |
94fcca9f YZ |
4863 | } else { |
4864 | if (wc->level == 1 && | |
4865 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4866 | continue; | |
6407bf6d | 4867 | } |
1c4850e2 | 4868 | reada: |
bfb484d9 | 4869 | btrfs_readahead_node_child(eb, slot); |
1c4850e2 | 4870 | nread++; |
20524f02 | 4871 | } |
1c4850e2 | 4872 | wc->reada_slot = slot; |
20524f02 | 4873 | } |
2c47e605 | 4874 | |
f82d02d9 | 4875 | /* |
2c016dc2 | 4876 | * helper to process tree block while walking down the tree. |
2c47e605 | 4877 | * |
2c47e605 YZ |
4878 | * when wc->stage == UPDATE_BACKREF, this function updates |
4879 | * back refs for pointers in the block. | |
4880 | * | |
4881 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4882 | */ |
2c47e605 | 4883 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4884 | struct btrfs_root *root, |
2c47e605 | 4885 | struct btrfs_path *path, |
94fcca9f | 4886 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4887 | { |
2ff7e61e | 4888 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4889 | int level = wc->level; |
4890 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4891 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4892 | int ret; |
4893 | ||
2c47e605 YZ |
4894 | if (wc->stage == UPDATE_BACKREF && |
4895 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4896 | return 1; | |
f82d02d9 | 4897 | |
2c47e605 YZ |
4898 | /* |
4899 | * when reference count of tree block is 1, it won't increase | |
4900 | * again. once full backref flag is set, we never clear it. | |
4901 | */ | |
94fcca9f YZ |
4902 | if (lookup_info && |
4903 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4904 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4905 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4906 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4907 | eb->start, level, 1, |
2c47e605 YZ |
4908 | &wc->refs[level], |
4909 | &wc->flags[level]); | |
79787eaa JM |
4910 | BUG_ON(ret == -ENOMEM); |
4911 | if (ret) | |
4912 | return ret; | |
2c47e605 YZ |
4913 | BUG_ON(wc->refs[level] == 0); |
4914 | } | |
5d4f98a2 | 4915 | |
2c47e605 YZ |
4916 | if (wc->stage == DROP_REFERENCE) { |
4917 | if (wc->refs[level] > 1) | |
4918 | return 1; | |
f82d02d9 | 4919 | |
2c47e605 | 4920 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4921 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4922 | path->locks[level] = 0; |
4923 | } | |
4924 | return 0; | |
4925 | } | |
f82d02d9 | 4926 | |
2c47e605 YZ |
4927 | /* wc->stage == UPDATE_BACKREF */ |
4928 | if (!(wc->flags[level] & flag)) { | |
4929 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4930 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4931 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4932 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4933 | BUG_ON(ret); /* -ENOMEM */ |
42c9d0b5 | 4934 | ret = btrfs_set_disk_extent_flags(trans, eb, flag, |
b1c79e09 | 4935 | btrfs_header_level(eb), 0); |
79787eaa | 4936 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4937 | wc->flags[level] |= flag; |
4938 | } | |
4939 | ||
4940 | /* | |
4941 | * the block is shared by multiple trees, so it's not good to | |
4942 | * keep the tree lock | |
4943 | */ | |
4944 | if (path->locks[level] && level > 0) { | |
bd681513 | 4945 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4946 | path->locks[level] = 0; |
4947 | } | |
4948 | return 0; | |
4949 | } | |
4950 | ||
78c52d9e JB |
4951 | /* |
4952 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4953 | * would have a drop_progress key that hadn't been updated properly. | |
4954 | */ | |
4955 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4956 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4957 | int level) | |
4958 | { | |
4959 | struct btrfs_path *path; | |
4960 | struct btrfs_extent_inline_ref *iref; | |
4961 | int ret; | |
4962 | ||
4963 | path = btrfs_alloc_path(); | |
4964 | if (!path) | |
4965 | return -ENOMEM; | |
4966 | ||
4967 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
4968 | root->fs_info->nodesize, parent, | |
4969 | root->root_key.objectid, level, 0); | |
4970 | btrfs_free_path(path); | |
4971 | if (ret == -ENOENT) | |
4972 | return 0; | |
4973 | if (ret < 0) | |
4974 | return ret; | |
4975 | return 1; | |
4976 | } | |
4977 | ||
1c4850e2 | 4978 | /* |
2c016dc2 | 4979 | * helper to process tree block pointer. |
1c4850e2 YZ |
4980 | * |
4981 | * when wc->stage == DROP_REFERENCE, this function checks | |
4982 | * reference count of the block pointed to. if the block | |
4983 | * is shared and we need update back refs for the subtree | |
4984 | * rooted at the block, this function changes wc->stage to | |
4985 | * UPDATE_BACKREF. if the block is shared and there is no | |
4986 | * need to update back, this function drops the reference | |
4987 | * to the block. | |
4988 | * | |
4989 | * NOTE: return value 1 means we should stop walking down. | |
4990 | */ | |
4991 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
4992 | struct btrfs_root *root, | |
4993 | struct btrfs_path *path, | |
94fcca9f | 4994 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 4995 | { |
0b246afa | 4996 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4997 | u64 bytenr; |
4998 | u64 generation; | |
4999 | u64 parent; | |
1c4850e2 | 5000 | struct btrfs_key key; |
581c1760 | 5001 | struct btrfs_key first_key; |
ffd4bb2a | 5002 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
5003 | struct extent_buffer *next; |
5004 | int level = wc->level; | |
5005 | int reada = 0; | |
5006 | int ret = 0; | |
1152651a | 5007 | bool need_account = false; |
1c4850e2 YZ |
5008 | |
5009 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5010 | path->slots[level]); | |
5011 | /* | |
5012 | * if the lower level block was created before the snapshot | |
5013 | * was created, we know there is no need to update back refs | |
5014 | * for the subtree | |
5015 | */ | |
5016 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5017 | generation <= root->root_key.offset) { |
5018 | *lookup_info = 1; | |
1c4850e2 | 5019 | return 1; |
94fcca9f | 5020 | } |
1c4850e2 YZ |
5021 | |
5022 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
5023 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
5024 | path->slots[level]); | |
1c4850e2 | 5025 | |
0b246afa | 5026 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5027 | if (!next) { |
3fbaf258 JB |
5028 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
5029 | root->root_key.objectid, level - 1); | |
c871b0f2 LB |
5030 | if (IS_ERR(next)) |
5031 | return PTR_ERR(next); | |
1c4850e2 YZ |
5032 | reada = 1; |
5033 | } | |
5034 | btrfs_tree_lock(next); | |
1c4850e2 | 5035 | |
2ff7e61e | 5036 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
5037 | &wc->refs[level - 1], |
5038 | &wc->flags[level - 1]); | |
4867268c JB |
5039 | if (ret < 0) |
5040 | goto out_unlock; | |
79787eaa | 5041 | |
c2cf52eb | 5042 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5043 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5044 | ret = -EIO; |
5045 | goto out_unlock; | |
c2cf52eb | 5046 | } |
94fcca9f | 5047 | *lookup_info = 0; |
1c4850e2 | 5048 | |
94fcca9f | 5049 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5050 | if (wc->refs[level - 1] > 1) { |
1152651a | 5051 | need_account = true; |
94fcca9f YZ |
5052 | if (level == 1 && |
5053 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5054 | goto skip; | |
5055 | ||
1c4850e2 YZ |
5056 | if (!wc->update_ref || |
5057 | generation <= root->root_key.offset) | |
5058 | goto skip; | |
5059 | ||
5060 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5061 | path->slots[level]); | |
5062 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5063 | if (ret < 0) | |
5064 | goto skip; | |
5065 | ||
5066 | wc->stage = UPDATE_BACKREF; | |
5067 | wc->shared_level = level - 1; | |
5068 | } | |
94fcca9f YZ |
5069 | } else { |
5070 | if (level == 1 && | |
5071 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5072 | goto skip; | |
1c4850e2 YZ |
5073 | } |
5074 | ||
b9fab919 | 5075 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5076 | btrfs_tree_unlock(next); |
5077 | free_extent_buffer(next); | |
5078 | next = NULL; | |
94fcca9f | 5079 | *lookup_info = 1; |
1c4850e2 YZ |
5080 | } |
5081 | ||
5082 | if (!next) { | |
5083 | if (reada && level == 1) | |
5084 | reada_walk_down(trans, root, wc, path); | |
1b7ec85e JB |
5085 | next = read_tree_block(fs_info, bytenr, root->root_key.objectid, |
5086 | generation, level - 1, &first_key); | |
64c043de LB |
5087 | if (IS_ERR(next)) { |
5088 | return PTR_ERR(next); | |
5089 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5090 | free_extent_buffer(next); |
97d9a8a4 | 5091 | return -EIO; |
416bc658 | 5092 | } |
1c4850e2 | 5093 | btrfs_tree_lock(next); |
1c4850e2 YZ |
5094 | } |
5095 | ||
5096 | level--; | |
4867268c JB |
5097 | ASSERT(level == btrfs_header_level(next)); |
5098 | if (level != btrfs_header_level(next)) { | |
5099 | btrfs_err(root->fs_info, "mismatched level"); | |
5100 | ret = -EIO; | |
5101 | goto out_unlock; | |
5102 | } | |
1c4850e2 YZ |
5103 | path->nodes[level] = next; |
5104 | path->slots[level] = 0; | |
ac5887c8 | 5105 | path->locks[level] = BTRFS_WRITE_LOCK; |
1c4850e2 YZ |
5106 | wc->level = level; |
5107 | if (wc->level == 1) | |
5108 | wc->reada_slot = 0; | |
5109 | return 0; | |
5110 | skip: | |
5111 | wc->refs[level - 1] = 0; | |
5112 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5113 | if (wc->stage == DROP_REFERENCE) { |
5114 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5115 | parent = path->nodes[level]->start; | |
5116 | } else { | |
4867268c | 5117 | ASSERT(root->root_key.objectid == |
94fcca9f | 5118 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5119 | if (root->root_key.objectid != |
5120 | btrfs_header_owner(path->nodes[level])) { | |
5121 | btrfs_err(root->fs_info, | |
5122 | "mismatched block owner"); | |
5123 | ret = -EIO; | |
5124 | goto out_unlock; | |
5125 | } | |
94fcca9f YZ |
5126 | parent = 0; |
5127 | } | |
1c4850e2 | 5128 | |
78c52d9e JB |
5129 | /* |
5130 | * If we had a drop_progress we need to verify the refs are set | |
5131 | * as expected. If we find our ref then we know that from here | |
5132 | * on out everything should be correct, and we can clear the | |
5133 | * ->restarted flag. | |
5134 | */ | |
5135 | if (wc->restarted) { | |
5136 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5137 | level - 1); | |
5138 | if (ret < 0) | |
5139 | goto out_unlock; | |
5140 | if (ret == 0) | |
5141 | goto no_delete; | |
5142 | ret = 0; | |
5143 | wc->restarted = 0; | |
5144 | } | |
5145 | ||
2cd86d30 QW |
5146 | /* |
5147 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5148 | * already accounted them at merge time (replace_path), | |
5149 | * thus we could skip expensive subtree trace here. | |
5150 | */ | |
5151 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5152 | need_account) { | |
deb40627 | 5153 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5154 | generation, level - 1); |
1152651a | 5155 | if (ret) { |
0b246afa | 5156 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5157 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5158 | ret); | |
1152651a MF |
5159 | } |
5160 | } | |
aea6f028 JB |
5161 | |
5162 | /* | |
5163 | * We need to update the next key in our walk control so we can | |
5164 | * update the drop_progress key accordingly. We don't care if | |
5165 | * find_next_key doesn't find a key because that means we're at | |
5166 | * the end and are going to clean up now. | |
5167 | */ | |
5168 | wc->drop_level = level; | |
5169 | find_next_key(path, level, &wc->drop_progress); | |
5170 | ||
ffd4bb2a QW |
5171 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
5172 | fs_info->nodesize, parent); | |
5173 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
5174 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
5175 | if (ret) |
5176 | goto out_unlock; | |
1c4850e2 | 5177 | } |
78c52d9e | 5178 | no_delete: |
4867268c JB |
5179 | *lookup_info = 1; |
5180 | ret = 1; | |
5181 | ||
5182 | out_unlock: | |
1c4850e2 YZ |
5183 | btrfs_tree_unlock(next); |
5184 | free_extent_buffer(next); | |
4867268c JB |
5185 | |
5186 | return ret; | |
1c4850e2 YZ |
5187 | } |
5188 | ||
2c47e605 | 5189 | /* |
2c016dc2 | 5190 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5191 | * |
5192 | * when wc->stage == DROP_REFERENCE, this function drops | |
5193 | * reference count on the block. | |
5194 | * | |
5195 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5196 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5197 | * to UPDATE_BACKREF previously while processing the block. | |
5198 | * | |
5199 | * NOTE: return value 1 means we should stop walking up. | |
5200 | */ | |
5201 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5202 | struct btrfs_root *root, | |
5203 | struct btrfs_path *path, | |
5204 | struct walk_control *wc) | |
5205 | { | |
0b246afa | 5206 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5207 | int ret; |
2c47e605 YZ |
5208 | int level = wc->level; |
5209 | struct extent_buffer *eb = path->nodes[level]; | |
5210 | u64 parent = 0; | |
5211 | ||
5212 | if (wc->stage == UPDATE_BACKREF) { | |
5213 | BUG_ON(wc->shared_level < level); | |
5214 | if (level < wc->shared_level) | |
5215 | goto out; | |
5216 | ||
2c47e605 YZ |
5217 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5218 | if (ret > 0) | |
5219 | wc->update_ref = 0; | |
5220 | ||
5221 | wc->stage = DROP_REFERENCE; | |
5222 | wc->shared_level = -1; | |
5223 | path->slots[level] = 0; | |
5224 | ||
5225 | /* | |
5226 | * check reference count again if the block isn't locked. | |
5227 | * we should start walking down the tree again if reference | |
5228 | * count is one. | |
5229 | */ | |
5230 | if (!path->locks[level]) { | |
5231 | BUG_ON(level == 0); | |
5232 | btrfs_tree_lock(eb); | |
ac5887c8 | 5233 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5234 | |
2ff7e61e | 5235 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5236 | eb->start, level, 1, |
2c47e605 YZ |
5237 | &wc->refs[level], |
5238 | &wc->flags[level]); | |
79787eaa JM |
5239 | if (ret < 0) { |
5240 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5241 | path->locks[level] = 0; |
79787eaa JM |
5242 | return ret; |
5243 | } | |
2c47e605 YZ |
5244 | BUG_ON(wc->refs[level] == 0); |
5245 | if (wc->refs[level] == 1) { | |
bd681513 | 5246 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5247 | path->locks[level] = 0; |
2c47e605 YZ |
5248 | return 1; |
5249 | } | |
f82d02d9 | 5250 | } |
2c47e605 | 5251 | } |
f82d02d9 | 5252 | |
2c47e605 YZ |
5253 | /* wc->stage == DROP_REFERENCE */ |
5254 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5255 | |
2c47e605 YZ |
5256 | if (wc->refs[level] == 1) { |
5257 | if (level == 0) { | |
5258 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5259 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5260 | else |
e339a6b0 | 5261 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5262 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5263 | if (is_fstree(root->root_key.objectid)) { |
5264 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5265 | if (ret) { | |
5266 | btrfs_err_rl(fs_info, | |
5267 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5268 | ret); |
c4140cbf | 5269 | } |
1152651a | 5270 | } |
2c47e605 | 5271 | } |
6a884d7d | 5272 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5273 | if (!path->locks[level] && |
5274 | btrfs_header_generation(eb) == trans->transid) { | |
5275 | btrfs_tree_lock(eb); | |
ac5887c8 | 5276 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5277 | } |
6a884d7d | 5278 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5279 | } |
5280 | ||
5281 | if (eb == root->node) { | |
5282 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5283 | parent = eb->start; | |
65c6e82b QW |
5284 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5285 | goto owner_mismatch; | |
2c47e605 YZ |
5286 | } else { |
5287 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5288 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5289 | else if (root->root_key.objectid != |
5290 | btrfs_header_owner(path->nodes[level + 1])) | |
5291 | goto owner_mismatch; | |
f82d02d9 | 5292 | } |
f82d02d9 | 5293 | |
5581a51a | 5294 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5295 | out: |
5296 | wc->refs[level] = 0; | |
5297 | wc->flags[level] = 0; | |
f0486c68 | 5298 | return 0; |
65c6e82b QW |
5299 | |
5300 | owner_mismatch: | |
5301 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5302 | btrfs_header_owner(eb), root->root_key.objectid); | |
5303 | return -EUCLEAN; | |
2c47e605 YZ |
5304 | } |
5305 | ||
5306 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5307 | struct btrfs_root *root, | |
5308 | struct btrfs_path *path, | |
5309 | struct walk_control *wc) | |
5310 | { | |
2c47e605 | 5311 | int level = wc->level; |
94fcca9f | 5312 | int lookup_info = 1; |
2c47e605 YZ |
5313 | int ret; |
5314 | ||
5315 | while (level >= 0) { | |
94fcca9f | 5316 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5317 | if (ret > 0) |
5318 | break; | |
5319 | ||
5320 | if (level == 0) | |
5321 | break; | |
5322 | ||
7a7965f8 YZ |
5323 | if (path->slots[level] >= |
5324 | btrfs_header_nritems(path->nodes[level])) | |
5325 | break; | |
5326 | ||
94fcca9f | 5327 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5328 | if (ret > 0) { |
5329 | path->slots[level]++; | |
5330 | continue; | |
90d2c51d MX |
5331 | } else if (ret < 0) |
5332 | return ret; | |
1c4850e2 | 5333 | level = wc->level; |
f82d02d9 | 5334 | } |
f82d02d9 YZ |
5335 | return 0; |
5336 | } | |
5337 | ||
d397712b | 5338 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5339 | struct btrfs_root *root, |
f82d02d9 | 5340 | struct btrfs_path *path, |
2c47e605 | 5341 | struct walk_control *wc, int max_level) |
20524f02 | 5342 | { |
2c47e605 | 5343 | int level = wc->level; |
20524f02 | 5344 | int ret; |
9f3a7427 | 5345 | |
2c47e605 YZ |
5346 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5347 | while (level < max_level && path->nodes[level]) { | |
5348 | wc->level = level; | |
5349 | if (path->slots[level] + 1 < | |
5350 | btrfs_header_nritems(path->nodes[level])) { | |
5351 | path->slots[level]++; | |
20524f02 CM |
5352 | return 0; |
5353 | } else { | |
2c47e605 YZ |
5354 | ret = walk_up_proc(trans, root, path, wc); |
5355 | if (ret > 0) | |
5356 | return 0; | |
65c6e82b QW |
5357 | if (ret < 0) |
5358 | return ret; | |
bd56b302 | 5359 | |
2c47e605 | 5360 | if (path->locks[level]) { |
bd681513 CM |
5361 | btrfs_tree_unlock_rw(path->nodes[level], |
5362 | path->locks[level]); | |
2c47e605 | 5363 | path->locks[level] = 0; |
f82d02d9 | 5364 | } |
2c47e605 YZ |
5365 | free_extent_buffer(path->nodes[level]); |
5366 | path->nodes[level] = NULL; | |
5367 | level++; | |
20524f02 CM |
5368 | } |
5369 | } | |
5370 | return 1; | |
5371 | } | |
5372 | ||
9aca1d51 | 5373 | /* |
2c47e605 YZ |
5374 | * drop a subvolume tree. |
5375 | * | |
5376 | * this function traverses the tree freeing any blocks that only | |
5377 | * referenced by the tree. | |
5378 | * | |
5379 | * when a shared tree block is found. this function decreases its | |
5380 | * reference count by one. if update_ref is true, this function | |
5381 | * also make sure backrefs for the shared block and all lower level | |
5382 | * blocks are properly updated. | |
9d1a2a3a DS |
5383 | * |
5384 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5385 | */ |
0078a9f9 | 5386 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5387 | { |
ab8d0fc4 | 5388 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5389 | struct btrfs_path *path; |
2c47e605 | 5390 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5391 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5392 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5393 | struct walk_control *wc; |
5394 | struct btrfs_key key; | |
5395 | int err = 0; | |
5396 | int ret; | |
5397 | int level; | |
d29a9f62 | 5398 | bool root_dropped = false; |
20524f02 | 5399 | |
4fd786e6 | 5400 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5401 | |
5caf2a00 | 5402 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5403 | if (!path) { |
5404 | err = -ENOMEM; | |
5405 | goto out; | |
5406 | } | |
20524f02 | 5407 | |
2c47e605 | 5408 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5409 | if (!wc) { |
5410 | btrfs_free_path(path); | |
cb1b69f4 TI |
5411 | err = -ENOMEM; |
5412 | goto out; | |
38a1a919 | 5413 | } |
2c47e605 | 5414 | |
f3e3d9cc QW |
5415 | /* |
5416 | * Use join to avoid potential EINTR from transaction start. See | |
5417 | * wait_reserve_ticket and the whole reservation callchain. | |
5418 | */ | |
5419 | if (for_reloc) | |
5420 | trans = btrfs_join_transaction(tree_root); | |
5421 | else | |
5422 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5423 | if (IS_ERR(trans)) { |
5424 | err = PTR_ERR(trans); | |
5425 | goto out_free; | |
5426 | } | |
98d5dc13 | 5427 | |
0568e82d JB |
5428 | err = btrfs_run_delayed_items(trans); |
5429 | if (err) | |
5430 | goto out_end_trans; | |
5431 | ||
83354f07 JB |
5432 | /* |
5433 | * This will help us catch people modifying the fs tree while we're | |
5434 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5435 | * dropped as we unlock the root node and parent nodes as we walk down | |
5436 | * the tree, assuming nothing will change. If something does change | |
5437 | * then we'll have stale information and drop references to blocks we've | |
5438 | * already dropped. | |
5439 | */ | |
5440 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5441 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5442 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5443 | path->nodes[level] = btrfs_lock_root_node(root); |
9f3a7427 | 5444 | path->slots[level] = 0; |
ac5887c8 | 5445 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5446 | memset(&wc->update_progress, 0, |
5447 | sizeof(wc->update_progress)); | |
9f3a7427 | 5448 | } else { |
9f3a7427 | 5449 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5450 | memcpy(&wc->update_progress, &key, |
5451 | sizeof(wc->update_progress)); | |
5452 | ||
c8422684 | 5453 | level = btrfs_root_drop_level(root_item); |
2c47e605 | 5454 | BUG_ON(level == 0); |
6702ed49 | 5455 | path->lowest_level = level; |
2c47e605 YZ |
5456 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5457 | path->lowest_level = 0; | |
5458 | if (ret < 0) { | |
5459 | err = ret; | |
79787eaa | 5460 | goto out_end_trans; |
9f3a7427 | 5461 | } |
1c4850e2 | 5462 | WARN_ON(ret > 0); |
2c47e605 | 5463 | |
7d9eb12c CM |
5464 | /* |
5465 | * unlock our path, this is safe because only this | |
5466 | * function is allowed to delete this snapshot | |
5467 | */ | |
5d4f98a2 | 5468 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5469 | |
5470 | level = btrfs_header_level(root->node); | |
5471 | while (1) { | |
5472 | btrfs_tree_lock(path->nodes[level]); | |
ac5887c8 | 5473 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5474 | |
2ff7e61e | 5475 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5476 | path->nodes[level]->start, |
3173a18f | 5477 | level, 1, &wc->refs[level], |
2c47e605 | 5478 | &wc->flags[level]); |
79787eaa JM |
5479 | if (ret < 0) { |
5480 | err = ret; | |
5481 | goto out_end_trans; | |
5482 | } | |
2c47e605 YZ |
5483 | BUG_ON(wc->refs[level] == 0); |
5484 | ||
c8422684 | 5485 | if (level == btrfs_root_drop_level(root_item)) |
2c47e605 YZ |
5486 | break; |
5487 | ||
5488 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5489 | path->locks[level] = 0; |
2c47e605 YZ |
5490 | WARN_ON(wc->refs[level] != 1); |
5491 | level--; | |
5492 | } | |
9f3a7427 | 5493 | } |
2c47e605 | 5494 | |
78c52d9e | 5495 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5496 | wc->level = level; |
5497 | wc->shared_level = -1; | |
5498 | wc->stage = DROP_REFERENCE; | |
5499 | wc->update_ref = update_ref; | |
5500 | wc->keep_locks = 0; | |
0b246afa | 5501 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5502 | |
d397712b | 5503 | while (1) { |
9d1a2a3a | 5504 | |
2c47e605 YZ |
5505 | ret = walk_down_tree(trans, root, path, wc); |
5506 | if (ret < 0) { | |
5507 | err = ret; | |
20524f02 | 5508 | break; |
2c47e605 | 5509 | } |
9aca1d51 | 5510 | |
2c47e605 YZ |
5511 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5512 | if (ret < 0) { | |
5513 | err = ret; | |
20524f02 | 5514 | break; |
2c47e605 YZ |
5515 | } |
5516 | ||
5517 | if (ret > 0) { | |
5518 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5519 | break; |
5520 | } | |
2c47e605 YZ |
5521 | |
5522 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5523 | wc->drop_level = wc->level; |
5524 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5525 | &wc->drop_progress, | |
5526 | path->slots[wc->drop_level]); | |
5527 | } | |
5528 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5529 | &wc->drop_progress); | |
c8422684 | 5530 | btrfs_set_root_drop_level(root_item, wc->drop_level); |
2c47e605 YZ |
5531 | |
5532 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5533 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5534 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5535 | ret = btrfs_update_root(trans, tree_root, |
5536 | &root->root_key, | |
5537 | root_item); | |
79787eaa | 5538 | if (ret) { |
66642832 | 5539 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5540 | err = ret; |
5541 | goto out_end_trans; | |
5542 | } | |
2c47e605 | 5543 | |
3a45bb20 | 5544 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5545 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5546 | btrfs_debug(fs_info, |
5547 | "drop snapshot early exit"); | |
3c8f2422 JB |
5548 | err = -EAGAIN; |
5549 | goto out_free; | |
5550 | } | |
5551 | ||
18d3bff4 JB |
5552 | /* |
5553 | * Use join to avoid potential EINTR from transaction | |
5554 | * start. See wait_reserve_ticket and the whole | |
5555 | * reservation callchain. | |
5556 | */ | |
5557 | if (for_reloc) | |
5558 | trans = btrfs_join_transaction(tree_root); | |
5559 | else | |
5560 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5561 | if (IS_ERR(trans)) { |
5562 | err = PTR_ERR(trans); | |
5563 | goto out_free; | |
5564 | } | |
c3e69d58 | 5565 | } |
20524f02 | 5566 | } |
b3b4aa74 | 5567 | btrfs_release_path(path); |
79787eaa JM |
5568 | if (err) |
5569 | goto out_end_trans; | |
2c47e605 | 5570 | |
ab9ce7d4 | 5571 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5572 | if (ret) { |
66642832 | 5573 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5574 | err = ret; |
79787eaa JM |
5575 | goto out_end_trans; |
5576 | } | |
2c47e605 | 5577 | |
76dda93c | 5578 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5579 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5580 | NULL, NULL); | |
79787eaa | 5581 | if (ret < 0) { |
66642832 | 5582 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5583 | err = ret; |
5584 | goto out_end_trans; | |
5585 | } else if (ret > 0) { | |
84cd948c JB |
5586 | /* if we fail to delete the orphan item this time |
5587 | * around, it'll get picked up the next time. | |
5588 | * | |
5589 | * The most common failure here is just -ENOENT. | |
5590 | */ | |
5591 | btrfs_del_orphan_item(trans, tree_root, | |
5592 | root->root_key.objectid); | |
76dda93c YZ |
5593 | } |
5594 | } | |
5595 | ||
a3cf0e43 QW |
5596 | /* |
5597 | * This subvolume is going to be completely dropped, and won't be | |
5598 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
5599 | * commit transaction time. So free it here manually. | |
5600 | */ | |
5601 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
5602 | btrfs_qgroup_free_meta_all_pertrans(root); | |
5603 | ||
8c38938c | 5604 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 5605 | btrfs_add_dropped_root(trans, root); |
8c38938c | 5606 | else |
00246528 | 5607 | btrfs_put_root(root); |
d29a9f62 | 5608 | root_dropped = true; |
79787eaa | 5609 | out_end_trans: |
3a45bb20 | 5610 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5611 | out_free: |
2c47e605 | 5612 | kfree(wc); |
5caf2a00 | 5613 | btrfs_free_path(path); |
cb1b69f4 | 5614 | out: |
d29a9f62 JB |
5615 | /* |
5616 | * So if we need to stop dropping the snapshot for whatever reason we | |
5617 | * need to make sure to add it back to the dead root list so that we | |
5618 | * keep trying to do the work later. This also cleans up roots if we | |
5619 | * don't have it in the radix (like when we recover after a power fail | |
5620 | * or unmount) so we don't leak memory. | |
5621 | */ | |
897ca819 | 5622 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5623 | btrfs_add_dead_root(root); |
2c536799 | 5624 | return err; |
20524f02 | 5625 | } |
9078a3e1 | 5626 | |
2c47e605 YZ |
5627 | /* |
5628 | * drop subtree rooted at tree block 'node'. | |
5629 | * | |
5630 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5631 | * only used by relocation code |
2c47e605 | 5632 | */ |
f82d02d9 YZ |
5633 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5634 | struct btrfs_root *root, | |
5635 | struct extent_buffer *node, | |
5636 | struct extent_buffer *parent) | |
5637 | { | |
0b246afa | 5638 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5639 | struct btrfs_path *path; |
2c47e605 | 5640 | struct walk_control *wc; |
f82d02d9 YZ |
5641 | int level; |
5642 | int parent_level; | |
5643 | int ret = 0; | |
5644 | int wret; | |
5645 | ||
2c47e605 YZ |
5646 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5647 | ||
f82d02d9 | 5648 | path = btrfs_alloc_path(); |
db5b493a TI |
5649 | if (!path) |
5650 | return -ENOMEM; | |
f82d02d9 | 5651 | |
2c47e605 | 5652 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5653 | if (!wc) { |
5654 | btrfs_free_path(path); | |
5655 | return -ENOMEM; | |
5656 | } | |
2c47e605 | 5657 | |
b9447ef8 | 5658 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5659 | parent_level = btrfs_header_level(parent); |
67439dad | 5660 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5661 | path->nodes[parent_level] = parent; |
5662 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5663 | ||
b9447ef8 | 5664 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5665 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5666 | path->nodes[level] = node; |
5667 | path->slots[level] = 0; | |
ac5887c8 | 5668 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5669 | |
5670 | wc->refs[parent_level] = 1; | |
5671 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5672 | wc->level = level; | |
5673 | wc->shared_level = -1; | |
5674 | wc->stage = DROP_REFERENCE; | |
5675 | wc->update_ref = 0; | |
5676 | wc->keep_locks = 1; | |
0b246afa | 5677 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5678 | |
5679 | while (1) { | |
2c47e605 YZ |
5680 | wret = walk_down_tree(trans, root, path, wc); |
5681 | if (wret < 0) { | |
f82d02d9 | 5682 | ret = wret; |
f82d02d9 | 5683 | break; |
2c47e605 | 5684 | } |
f82d02d9 | 5685 | |
2c47e605 | 5686 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5687 | if (wret < 0) |
5688 | ret = wret; | |
5689 | if (wret != 0) | |
5690 | break; | |
5691 | } | |
5692 | ||
2c47e605 | 5693 | kfree(wc); |
f82d02d9 YZ |
5694 | btrfs_free_path(path); |
5695 | return ret; | |
5696 | } | |
5697 | ||
6d07bcec MX |
5698 | /* |
5699 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5700 | * space_info. takes mirrors into account. |
6d07bcec | 5701 | */ |
633c0aad | 5702 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5703 | { |
32da5386 | 5704 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5705 | u64 free_bytes = 0; |
5706 | int factor; | |
5707 | ||
01327610 | 5708 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5709 | if (list_empty(&sinfo->ro_bgs)) |
5710 | return 0; | |
5711 | ||
5712 | spin_lock(&sinfo->lock); | |
5713 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5714 | spin_lock(&block_group->lock); |
5715 | ||
5716 | if (!block_group->ro) { | |
5717 | spin_unlock(&block_group->lock); | |
5718 | continue; | |
5719 | } | |
5720 | ||
46df06b8 | 5721 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5722 | free_bytes += (block_group->length - |
bf38be65 | 5723 | block_group->used) * factor; |
6d07bcec MX |
5724 | |
5725 | spin_unlock(&block_group->lock); | |
5726 | } | |
6d07bcec MX |
5727 | spin_unlock(&sinfo->lock); |
5728 | ||
5729 | return free_bytes; | |
5730 | } | |
5731 | ||
2ff7e61e JM |
5732 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5733 | u64 start, u64 end) | |
acce952b | 5734 | { |
2ff7e61e | 5735 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5736 | } |
5737 | ||
499f377f JM |
5738 | /* |
5739 | * It used to be that old block groups would be left around forever. | |
5740 | * Iterating over them would be enough to trim unused space. Since we | |
5741 | * now automatically remove them, we also need to iterate over unallocated | |
5742 | * space. | |
5743 | * | |
5744 | * We don't want a transaction for this since the discard may take a | |
5745 | * substantial amount of time. We don't require that a transaction be | |
5746 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5747 | * to ensure that we're not discarding chunks that were released or |
5748 | * allocated in the current transaction. | |
499f377f JM |
5749 | * |
5750 | * Holding the chunks lock will prevent other threads from allocating | |
5751 | * or releasing chunks, but it won't prevent a running transaction | |
5752 | * from committing and releasing the memory that the pending chunks | |
5753 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5754 | * transaction and hold the commit root sem. We only need to hold |
5755 | * it while performing the free space search since we have already | |
5756 | * held back allocations. | |
499f377f | 5757 | */ |
8103d10b | 5758 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5759 | { |
8103d10b | 5760 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5761 | int ret; |
5762 | ||
5763 | *trimmed = 0; | |
5764 | ||
0be88e36 JM |
5765 | /* Discard not supported = nothing to do. */ |
5766 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5767 | return 0; | |
5768 | ||
52042d8e | 5769 | /* Not writable = nothing to do. */ |
ebbede42 | 5770 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5771 | return 0; |
5772 | ||
5773 | /* No free space = nothing to do. */ | |
5774 | if (device->total_bytes <= device->bytes_used) | |
5775 | return 0; | |
5776 | ||
5777 | ret = 0; | |
5778 | ||
5779 | while (1) { | |
fb456252 | 5780 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5781 | u64 bytes; |
5782 | ||
5783 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5784 | if (ret) | |
fee7acc3 | 5785 | break; |
499f377f | 5786 | |
929be17a NB |
5787 | find_first_clear_extent_bit(&device->alloc_state, start, |
5788 | &start, &end, | |
5789 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 5790 | |
c57dd1f2 QW |
5791 | /* Check if there are any CHUNK_* bits left */ |
5792 | if (start > device->total_bytes) { | |
5793 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5794 | btrfs_warn_in_rcu(fs_info, | |
5795 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
5796 | start, end - start + 1, | |
5797 | rcu_str_deref(device->name), | |
5798 | device->total_bytes); | |
5799 | mutex_unlock(&fs_info->chunk_mutex); | |
5800 | ret = 0; | |
5801 | break; | |
5802 | } | |
5803 | ||
53460a45 NB |
5804 | /* Ensure we skip the reserved area in the first 1M */ |
5805 | start = max_t(u64, start, SZ_1M); | |
5806 | ||
929be17a NB |
5807 | /* |
5808 | * If find_first_clear_extent_bit find a range that spans the | |
5809 | * end of the device it will set end to -1, in this case it's up | |
5810 | * to the caller to trim the value to the size of the device. | |
5811 | */ | |
5812 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5813 | |
929be17a | 5814 | len = end - start + 1; |
499f377f | 5815 | |
929be17a NB |
5816 | /* We didn't find any extents */ |
5817 | if (!len) { | |
499f377f | 5818 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5819 | ret = 0; |
499f377f JM |
5820 | break; |
5821 | } | |
5822 | ||
929be17a NB |
5823 | ret = btrfs_issue_discard(device->bdev, start, len, |
5824 | &bytes); | |
5825 | if (!ret) | |
5826 | set_extent_bits(&device->alloc_state, start, | |
5827 | start + bytes - 1, | |
5828 | CHUNK_TRIMMED); | |
499f377f JM |
5829 | mutex_unlock(&fs_info->chunk_mutex); |
5830 | ||
5831 | if (ret) | |
5832 | break; | |
5833 | ||
5834 | start += len; | |
5835 | *trimmed += bytes; | |
5836 | ||
5837 | if (fatal_signal_pending(current)) { | |
5838 | ret = -ERESTARTSYS; | |
5839 | break; | |
5840 | } | |
5841 | ||
5842 | cond_resched(); | |
5843 | } | |
5844 | ||
5845 | return ret; | |
5846 | } | |
5847 | ||
93bba24d QW |
5848 | /* |
5849 | * Trim the whole filesystem by: | |
5850 | * 1) trimming the free space in each block group | |
5851 | * 2) trimming the unallocated space on each device | |
5852 | * | |
5853 | * This will also continue trimming even if a block group or device encounters | |
5854 | * an error. The return value will be the last error, or 0 if nothing bad | |
5855 | * happens. | |
5856 | */ | |
2ff7e61e | 5857 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5858 | { |
32da5386 | 5859 | struct btrfs_block_group *cache = NULL; |
499f377f JM |
5860 | struct btrfs_device *device; |
5861 | struct list_head *devices; | |
f7039b1d | 5862 | u64 group_trimmed; |
07301df7 | 5863 | u64 range_end = U64_MAX; |
f7039b1d LD |
5864 | u64 start; |
5865 | u64 end; | |
5866 | u64 trimmed = 0; | |
93bba24d QW |
5867 | u64 bg_failed = 0; |
5868 | u64 dev_failed = 0; | |
5869 | int bg_ret = 0; | |
5870 | int dev_ret = 0; | |
f7039b1d LD |
5871 | int ret = 0; |
5872 | ||
07301df7 QW |
5873 | /* |
5874 | * Check range overflow if range->len is set. | |
5875 | * The default range->len is U64_MAX. | |
5876 | */ | |
5877 | if (range->len != U64_MAX && | |
5878 | check_add_overflow(range->start, range->len, &range_end)) | |
5879 | return -EINVAL; | |
5880 | ||
6ba9fc8e | 5881 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5882 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 5883 | if (cache->start >= range_end) { |
f7039b1d LD |
5884 | btrfs_put_block_group(cache); |
5885 | break; | |
5886 | } | |
5887 | ||
b3470b5d DS |
5888 | start = max(range->start, cache->start); |
5889 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
5890 | |
5891 | if (end - start >= range->minlen) { | |
32da5386 | 5892 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 5893 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 5894 | if (ret) { |
93bba24d QW |
5895 | bg_failed++; |
5896 | bg_ret = ret; | |
5897 | continue; | |
1be41b78 | 5898 | } |
676f1f75 | 5899 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5900 | if (ret) { |
93bba24d QW |
5901 | bg_failed++; |
5902 | bg_ret = ret; | |
5903 | continue; | |
1be41b78 | 5904 | } |
f7039b1d LD |
5905 | } |
5906 | ret = btrfs_trim_block_group(cache, | |
5907 | &group_trimmed, | |
5908 | start, | |
5909 | end, | |
5910 | range->minlen); | |
5911 | ||
5912 | trimmed += group_trimmed; | |
5913 | if (ret) { | |
93bba24d QW |
5914 | bg_failed++; |
5915 | bg_ret = ret; | |
5916 | continue; | |
f7039b1d LD |
5917 | } |
5918 | } | |
f7039b1d LD |
5919 | } |
5920 | ||
93bba24d QW |
5921 | if (bg_failed) |
5922 | btrfs_warn(fs_info, | |
5923 | "failed to trim %llu block group(s), last error %d", | |
5924 | bg_failed, bg_ret); | |
0b246afa | 5925 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5926 | devices = &fs_info->fs_devices->devices; |
5927 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5928 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5929 | if (ret) { |
5930 | dev_failed++; | |
5931 | dev_ret = ret; | |
499f377f | 5932 | break; |
93bba24d | 5933 | } |
499f377f JM |
5934 | |
5935 | trimmed += group_trimmed; | |
5936 | } | |
0b246afa | 5937 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5938 | |
93bba24d QW |
5939 | if (dev_failed) |
5940 | btrfs_warn(fs_info, | |
5941 | "failed to trim %llu device(s), last error %d", | |
5942 | dev_failed, dev_ret); | |
f7039b1d | 5943 | range->len = trimmed; |
93bba24d QW |
5944 | if (bg_ret) |
5945 | return bg_ret; | |
5946 | return dev_ret; | |
f7039b1d | 5947 | } |