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" |
169e0da9 | 37 | #include "zoned.h" |
fec577fb | 38 | |
709c0486 AJ |
39 | #undef SCRAMBLE_DELAYED_REFS |
40 | ||
9f9b8e8d | 41 | |
5d4f98a2 | 42 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
43 | struct btrfs_delayed_ref_node *node, u64 parent, |
44 | u64 root_objectid, u64 owner_objectid, | |
45 | u64 owner_offset, int refs_to_drop, | |
46 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
47 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
48 | struct extent_buffer *leaf, | |
49 | struct btrfs_extent_item *ei); | |
50 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
51 | u64 parent, u64 root_objectid, |
52 | u64 flags, u64 owner, u64 offset, | |
53 | struct btrfs_key *ins, int ref_mod); | |
54 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 55 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 56 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
57 | static int find_next_key(struct btrfs_path *path, int level, |
58 | struct btrfs_key *key); | |
6a63209f | 59 | |
32da5386 | 60 | static int block_group_bits(struct btrfs_block_group *cache, u64 bits) |
0f9dd46c JB |
61 | { |
62 | return (cache->flags & bits) == bits; | |
63 | } | |
64 | ||
6f410d1b JB |
65 | int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info, |
66 | u64 start, u64 num_bytes) | |
817d52f8 | 67 | { |
11833d66 | 68 | u64 end = start + num_bytes - 1; |
fe119a6e NB |
69 | set_extent_bits(&fs_info->excluded_extents, start, end, |
70 | EXTENT_UPTODATE); | |
11833d66 YZ |
71 | return 0; |
72 | } | |
817d52f8 | 73 | |
32da5386 | 74 | void btrfs_free_excluded_extents(struct btrfs_block_group *cache) |
11833d66 | 75 | { |
9e715da8 | 76 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 77 | u64 start, end; |
817d52f8 | 78 | |
b3470b5d DS |
79 | start = cache->start; |
80 | end = start + cache->length - 1; | |
11833d66 | 81 | |
fe119a6e NB |
82 | clear_extent_bits(&fs_info->excluded_extents, start, end, |
83 | EXTENT_UPTODATE); | |
817d52f8 JB |
84 | } |
85 | ||
1a4ed8fd | 86 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 87 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
88 | { |
89 | int ret; | |
90 | struct btrfs_key key; | |
31840ae1 | 91 | struct btrfs_path *path; |
e02119d5 | 92 | |
31840ae1 | 93 | path = btrfs_alloc_path(); |
d8926bb3 MF |
94 | if (!path) |
95 | return -ENOMEM; | |
96 | ||
e02119d5 CM |
97 | key.objectid = start; |
98 | key.offset = len; | |
3173a18f | 99 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 100 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 101 | btrfs_free_path(path); |
7bb86316 CM |
102 | return ret; |
103 | } | |
104 | ||
a22285a6 | 105 | /* |
3173a18f | 106 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
107 | * |
108 | * the head node for delayed ref is used to store the sum of all the | |
109 | * reference count modifications queued up in the rbtree. the head | |
110 | * node may also store the extent flags to set. This way you can check | |
111 | * to see what the reference count and extent flags would be if all of | |
112 | * the delayed refs are not processed. | |
113 | */ | |
114 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 115 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 116 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
117 | { |
118 | struct btrfs_delayed_ref_head *head; | |
119 | struct btrfs_delayed_ref_root *delayed_refs; | |
120 | struct btrfs_path *path; | |
121 | struct btrfs_extent_item *ei; | |
122 | struct extent_buffer *leaf; | |
123 | struct btrfs_key key; | |
124 | u32 item_size; | |
125 | u64 num_refs; | |
126 | u64 extent_flags; | |
127 | int ret; | |
128 | ||
3173a18f JB |
129 | /* |
130 | * If we don't have skinny metadata, don't bother doing anything | |
131 | * different | |
132 | */ | |
0b246afa JM |
133 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
134 | offset = fs_info->nodesize; | |
3173a18f JB |
135 | metadata = 0; |
136 | } | |
137 | ||
a22285a6 YZ |
138 | path = btrfs_alloc_path(); |
139 | if (!path) | |
140 | return -ENOMEM; | |
141 | ||
a22285a6 YZ |
142 | if (!trans) { |
143 | path->skip_locking = 1; | |
144 | path->search_commit_root = 1; | |
145 | } | |
639eefc8 FDBM |
146 | |
147 | search_again: | |
148 | key.objectid = bytenr; | |
149 | key.offset = offset; | |
150 | if (metadata) | |
151 | key.type = BTRFS_METADATA_ITEM_KEY; | |
152 | else | |
153 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
154 | ||
0b246afa | 155 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
156 | if (ret < 0) |
157 | goto out_free; | |
158 | ||
3173a18f | 159 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
160 | if (path->slots[0]) { |
161 | path->slots[0]--; | |
162 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
163 | path->slots[0]); | |
164 | if (key.objectid == bytenr && | |
165 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 166 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
167 | ret = 0; |
168 | } | |
3173a18f JB |
169 | } |
170 | ||
a22285a6 YZ |
171 | if (ret == 0) { |
172 | leaf = path->nodes[0]; | |
173 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
174 | if (item_size >= sizeof(*ei)) { | |
175 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
176 | struct btrfs_extent_item); | |
177 | num_refs = btrfs_extent_refs(leaf, ei); | |
178 | extent_flags = btrfs_extent_flags(leaf, ei); | |
179 | } else { | |
ba3c2b19 NB |
180 | ret = -EINVAL; |
181 | btrfs_print_v0_err(fs_info); | |
182 | if (trans) | |
183 | btrfs_abort_transaction(trans, ret); | |
184 | else | |
185 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
186 | ||
187 | goto out_free; | |
a22285a6 | 188 | } |
ba3c2b19 | 189 | |
a22285a6 YZ |
190 | BUG_ON(num_refs == 0); |
191 | } else { | |
192 | num_refs = 0; | |
193 | extent_flags = 0; | |
194 | ret = 0; | |
195 | } | |
196 | ||
197 | if (!trans) | |
198 | goto out; | |
199 | ||
200 | delayed_refs = &trans->transaction->delayed_refs; | |
201 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 202 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
203 | if (head) { |
204 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 205 | refcount_inc(&head->refs); |
a22285a6 YZ |
206 | spin_unlock(&delayed_refs->lock); |
207 | ||
b3b4aa74 | 208 | btrfs_release_path(path); |
a22285a6 | 209 | |
8cc33e5c DS |
210 | /* |
211 | * Mutex was contended, block until it's released and try | |
212 | * again | |
213 | */ | |
a22285a6 YZ |
214 | mutex_lock(&head->mutex); |
215 | mutex_unlock(&head->mutex); | |
d278850e | 216 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 217 | goto search_again; |
a22285a6 | 218 | } |
d7df2c79 | 219 | spin_lock(&head->lock); |
a22285a6 YZ |
220 | if (head->extent_op && head->extent_op->update_flags) |
221 | extent_flags |= head->extent_op->flags_to_set; | |
222 | else | |
223 | BUG_ON(num_refs == 0); | |
224 | ||
d278850e | 225 | num_refs += head->ref_mod; |
d7df2c79 | 226 | spin_unlock(&head->lock); |
a22285a6 YZ |
227 | mutex_unlock(&head->mutex); |
228 | } | |
229 | spin_unlock(&delayed_refs->lock); | |
230 | out: | |
231 | WARN_ON(num_refs == 0); | |
232 | if (refs) | |
233 | *refs = num_refs; | |
234 | if (flags) | |
235 | *flags = extent_flags; | |
236 | out_free: | |
237 | btrfs_free_path(path); | |
238 | return ret; | |
239 | } | |
240 | ||
d8d5f3e1 CM |
241 | /* |
242 | * Back reference rules. Back refs have three main goals: | |
243 | * | |
244 | * 1) differentiate between all holders of references to an extent so that | |
245 | * when a reference is dropped we can make sure it was a valid reference | |
246 | * before freeing the extent. | |
247 | * | |
248 | * 2) Provide enough information to quickly find the holders of an extent | |
249 | * if we notice a given block is corrupted or bad. | |
250 | * | |
251 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
252 | * maintenance. This is actually the same as #2, but with a slightly | |
253 | * different use case. | |
254 | * | |
5d4f98a2 YZ |
255 | * There are two kinds of back refs. The implicit back refs is optimized |
256 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
257 | * back refs of this kind provide information about the block's owner tree | |
258 | * and the pointer's key. These information allow us to find the block by | |
259 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
260 | * referenced by their owner trees. The location of tree block is recorded | |
261 | * in the back refs. Actually the full back refs is generic, and can be | |
262 | * used in all cases the implicit back refs is used. The major shortcoming | |
263 | * of the full back refs is its overhead. Every time a tree block gets | |
264 | * COWed, we have to update back refs entry for all pointers in it. | |
265 | * | |
266 | * For a newly allocated tree block, we use implicit back refs for | |
267 | * pointers in it. This means most tree related operations only involve | |
268 | * implicit back refs. For a tree block created in old transaction, the | |
269 | * only way to drop a reference to it is COW it. So we can detect the | |
270 | * event that tree block loses its owner tree's reference and do the | |
271 | * back refs conversion. | |
272 | * | |
01327610 | 273 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
274 | * |
275 | * The reference count of the block is one and the tree is the block's | |
276 | * owner tree. Nothing to do in this case. | |
277 | * | |
278 | * The reference count of the block is one and the tree is not the | |
279 | * block's owner tree. In this case, full back refs is used for pointers | |
280 | * in the block. Remove these full back refs, add implicit back refs for | |
281 | * every pointers in the new block. | |
282 | * | |
283 | * The reference count of the block is greater than one and the tree is | |
284 | * the block's owner tree. In this case, implicit back refs is used for | |
285 | * pointers in the block. Add full back refs for every pointers in the | |
286 | * block, increase lower level extents' reference counts. The original | |
287 | * implicit back refs are entailed to the new block. | |
288 | * | |
289 | * The reference count of the block is greater than one and the tree is | |
290 | * not the block's owner tree. Add implicit back refs for every pointer in | |
291 | * the new block, increase lower level extents' reference count. | |
292 | * | |
293 | * Back Reference Key composing: | |
294 | * | |
295 | * The key objectid corresponds to the first byte in the extent, | |
296 | * The key type is used to differentiate between types of back refs. | |
297 | * There are different meanings of the key offset for different types | |
298 | * of back refs. | |
299 | * | |
d8d5f3e1 CM |
300 | * File extents can be referenced by: |
301 | * | |
302 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 303 | * - different files inside a single subvolume |
d8d5f3e1 CM |
304 | * - different offsets inside a file (bookend extents in file.c) |
305 | * | |
5d4f98a2 | 306 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
307 | * |
308 | * - Objectid of the subvolume root | |
d8d5f3e1 | 309 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
310 | * - original offset in the file |
311 | * - how many bookend extents | |
d8d5f3e1 | 312 | * |
5d4f98a2 YZ |
313 | * The key offset for the implicit back refs is hash of the first |
314 | * three fields. | |
d8d5f3e1 | 315 | * |
5d4f98a2 | 316 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 317 | * |
5d4f98a2 | 318 | * - number of pointers in the tree leaf |
d8d5f3e1 | 319 | * |
5d4f98a2 YZ |
320 | * The key offset for the implicit back refs is the first byte of |
321 | * the tree leaf | |
d8d5f3e1 | 322 | * |
5d4f98a2 YZ |
323 | * When a file extent is allocated, The implicit back refs is used. |
324 | * the fields are filled in: | |
d8d5f3e1 | 325 | * |
5d4f98a2 | 326 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 327 | * |
5d4f98a2 YZ |
328 | * When a file extent is removed file truncation, we find the |
329 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 330 | * |
5d4f98a2 | 331 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 332 | * |
5d4f98a2 | 333 | * Btree extents can be referenced by: |
d8d5f3e1 | 334 | * |
5d4f98a2 | 335 | * - Different subvolumes |
d8d5f3e1 | 336 | * |
5d4f98a2 YZ |
337 | * Both the implicit back refs and the full back refs for tree blocks |
338 | * only consist of key. The key offset for the implicit back refs is | |
339 | * objectid of block's owner tree. The key offset for the full back refs | |
340 | * is the first byte of parent block. | |
d8d5f3e1 | 341 | * |
5d4f98a2 YZ |
342 | * When implicit back refs is used, information about the lowest key and |
343 | * level of the tree block are required. These information are stored in | |
344 | * tree block info structure. | |
d8d5f3e1 | 345 | */ |
31840ae1 | 346 | |
167ce953 LB |
347 | /* |
348 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 349 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
350 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
351 | */ | |
352 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
353 | struct btrfs_extent_inline_ref *iref, | |
354 | enum btrfs_inline_ref_type is_data) | |
355 | { | |
356 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 357 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
358 | |
359 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
360 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
361 | type == BTRFS_SHARED_DATA_REF_KEY || | |
362 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
363 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 364 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 365 | return type; |
64ecdb64 LB |
366 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
367 | ASSERT(eb->fs_info); | |
368 | /* | |
ea57788e QW |
369 | * Every shared one has parent tree block, |
370 | * which must be aligned to sector size. | |
64ecdb64 LB |
371 | */ |
372 | if (offset && | |
ea57788e | 373 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
374 | return type; |
375 | } | |
167ce953 | 376 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 377 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 378 | return type; |
64ecdb64 LB |
379 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
380 | ASSERT(eb->fs_info); | |
381 | /* | |
ea57788e QW |
382 | * Every shared one has parent tree block, |
383 | * which must be aligned to sector size. | |
64ecdb64 LB |
384 | */ |
385 | if (offset && | |
ea57788e | 386 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
387 | return type; |
388 | } | |
167ce953 LB |
389 | } else { |
390 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
391 | return type; | |
392 | } | |
393 | } | |
394 | ||
395 | btrfs_print_leaf((struct extent_buffer *)eb); | |
ea57788e QW |
396 | btrfs_err(eb->fs_info, |
397 | "eb %llu iref 0x%lx invalid extent inline ref type %d", | |
398 | eb->start, (unsigned long)iref, type); | |
167ce953 LB |
399 | WARN_ON(1); |
400 | ||
401 | return BTRFS_REF_TYPE_INVALID; | |
402 | } | |
403 | ||
0785a9aa | 404 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
405 | { |
406 | u32 high_crc = ~(u32)0; | |
407 | u32 low_crc = ~(u32)0; | |
408 | __le64 lenum; | |
409 | ||
410 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 411 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 412 | lenum = cpu_to_le64(owner); |
65019df8 | 413 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 414 | lenum = cpu_to_le64(offset); |
65019df8 | 415 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
416 | |
417 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
418 | } | |
419 | ||
420 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
421 | struct btrfs_extent_data_ref *ref) | |
422 | { | |
423 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
424 | btrfs_extent_data_ref_objectid(leaf, ref), | |
425 | btrfs_extent_data_ref_offset(leaf, ref)); | |
426 | } | |
427 | ||
428 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
429 | struct btrfs_extent_data_ref *ref, | |
430 | u64 root_objectid, u64 owner, u64 offset) | |
431 | { | |
432 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
433 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
434 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
435 | return 0; | |
436 | return 1; | |
437 | } | |
438 | ||
439 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
440 | struct btrfs_path *path, |
441 | u64 bytenr, u64 parent, | |
442 | u64 root_objectid, | |
443 | u64 owner, u64 offset) | |
444 | { | |
bd1d53ef | 445 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
446 | struct btrfs_key key; |
447 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 448 | struct extent_buffer *leaf; |
5d4f98a2 | 449 | u32 nritems; |
74493f7a | 450 | int ret; |
5d4f98a2 YZ |
451 | int recow; |
452 | int err = -ENOENT; | |
74493f7a | 453 | |
31840ae1 | 454 | key.objectid = bytenr; |
5d4f98a2 YZ |
455 | if (parent) { |
456 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
457 | key.offset = parent; | |
458 | } else { | |
459 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
460 | key.offset = hash_extent_data_ref(root_objectid, | |
461 | owner, offset); | |
462 | } | |
463 | again: | |
464 | recow = 0; | |
465 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
466 | if (ret < 0) { | |
467 | err = ret; | |
468 | goto fail; | |
469 | } | |
31840ae1 | 470 | |
5d4f98a2 YZ |
471 | if (parent) { |
472 | if (!ret) | |
473 | return 0; | |
5d4f98a2 | 474 | goto fail; |
31840ae1 ZY |
475 | } |
476 | ||
477 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
478 | nritems = btrfs_header_nritems(leaf); |
479 | while (1) { | |
480 | if (path->slots[0] >= nritems) { | |
481 | ret = btrfs_next_leaf(root, path); | |
482 | if (ret < 0) | |
483 | err = ret; | |
484 | if (ret) | |
485 | goto fail; | |
486 | ||
487 | leaf = path->nodes[0]; | |
488 | nritems = btrfs_header_nritems(leaf); | |
489 | recow = 1; | |
490 | } | |
491 | ||
492 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
493 | if (key.objectid != bytenr || | |
494 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
495 | goto fail; | |
496 | ||
497 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
498 | struct btrfs_extent_data_ref); | |
499 | ||
500 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
501 | owner, offset)) { | |
502 | if (recow) { | |
b3b4aa74 | 503 | btrfs_release_path(path); |
5d4f98a2 YZ |
504 | goto again; |
505 | } | |
506 | err = 0; | |
507 | break; | |
508 | } | |
509 | path->slots[0]++; | |
31840ae1 | 510 | } |
5d4f98a2 YZ |
511 | fail: |
512 | return err; | |
31840ae1 ZY |
513 | } |
514 | ||
5d4f98a2 | 515 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
516 | struct btrfs_path *path, |
517 | u64 bytenr, u64 parent, | |
518 | u64 root_objectid, u64 owner, | |
519 | u64 offset, int refs_to_add) | |
31840ae1 | 520 | { |
62b895af | 521 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
522 | struct btrfs_key key; |
523 | struct extent_buffer *leaf; | |
5d4f98a2 | 524 | u32 size; |
31840ae1 ZY |
525 | u32 num_refs; |
526 | int ret; | |
74493f7a | 527 | |
74493f7a | 528 | key.objectid = bytenr; |
5d4f98a2 YZ |
529 | if (parent) { |
530 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
531 | key.offset = parent; | |
532 | size = sizeof(struct btrfs_shared_data_ref); | |
533 | } else { | |
534 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
535 | key.offset = hash_extent_data_ref(root_objectid, | |
536 | owner, offset); | |
537 | size = sizeof(struct btrfs_extent_data_ref); | |
538 | } | |
74493f7a | 539 | |
5d4f98a2 YZ |
540 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
541 | if (ret && ret != -EEXIST) | |
542 | goto fail; | |
543 | ||
544 | leaf = path->nodes[0]; | |
545 | if (parent) { | |
546 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 547 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
548 | struct btrfs_shared_data_ref); |
549 | if (ret == 0) { | |
550 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
551 | } else { | |
552 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
553 | num_refs += refs_to_add; | |
554 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 555 | } |
5d4f98a2 YZ |
556 | } else { |
557 | struct btrfs_extent_data_ref *ref; | |
558 | while (ret == -EEXIST) { | |
559 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
560 | struct btrfs_extent_data_ref); | |
561 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
562 | owner, offset)) | |
563 | break; | |
b3b4aa74 | 564 | btrfs_release_path(path); |
5d4f98a2 YZ |
565 | key.offset++; |
566 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
567 | size); | |
568 | if (ret && ret != -EEXIST) | |
569 | goto fail; | |
31840ae1 | 570 | |
5d4f98a2 YZ |
571 | leaf = path->nodes[0]; |
572 | } | |
573 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
574 | struct btrfs_extent_data_ref); | |
575 | if (ret == 0) { | |
576 | btrfs_set_extent_data_ref_root(leaf, ref, | |
577 | root_objectid); | |
578 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
579 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
580 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
581 | } else { | |
582 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
583 | num_refs += refs_to_add; | |
584 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 585 | } |
31840ae1 | 586 | } |
5d4f98a2 YZ |
587 | btrfs_mark_buffer_dirty(leaf); |
588 | ret = 0; | |
589 | fail: | |
b3b4aa74 | 590 | btrfs_release_path(path); |
7bb86316 | 591 | return ret; |
74493f7a CM |
592 | } |
593 | ||
5d4f98a2 | 594 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 595 | struct btrfs_path *path, |
fcebe456 | 596 | int refs_to_drop, int *last_ref) |
31840ae1 | 597 | { |
5d4f98a2 YZ |
598 | struct btrfs_key key; |
599 | struct btrfs_extent_data_ref *ref1 = NULL; | |
600 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 601 | struct extent_buffer *leaf; |
5d4f98a2 | 602 | u32 num_refs = 0; |
31840ae1 ZY |
603 | int ret = 0; |
604 | ||
605 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
606 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
607 | ||
608 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
609 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
610 | struct btrfs_extent_data_ref); | |
611 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
612 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
613 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
614 | struct btrfs_shared_data_ref); | |
615 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 616 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
617 | btrfs_print_v0_err(trans->fs_info); |
618 | btrfs_abort_transaction(trans, -EINVAL); | |
619 | return -EINVAL; | |
5d4f98a2 YZ |
620 | } else { |
621 | BUG(); | |
622 | } | |
623 | ||
56bec294 CM |
624 | BUG_ON(num_refs < refs_to_drop); |
625 | num_refs -= refs_to_drop; | |
5d4f98a2 | 626 | |
31840ae1 | 627 | if (num_refs == 0) { |
e9f6290d | 628 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 629 | *last_ref = 1; |
31840ae1 | 630 | } else { |
5d4f98a2 YZ |
631 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
632 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
633 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
634 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
635 | btrfs_mark_buffer_dirty(leaf); |
636 | } | |
31840ae1 ZY |
637 | return ret; |
638 | } | |
639 | ||
9ed0dea0 | 640 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 641 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 642 | { |
5d4f98a2 YZ |
643 | struct btrfs_key key; |
644 | struct extent_buffer *leaf; | |
645 | struct btrfs_extent_data_ref *ref1; | |
646 | struct btrfs_shared_data_ref *ref2; | |
647 | u32 num_refs = 0; | |
3de28d57 | 648 | int type; |
5d4f98a2 YZ |
649 | |
650 | leaf = path->nodes[0]; | |
651 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
652 | |
653 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 654 | if (iref) { |
3de28d57 LB |
655 | /* |
656 | * If type is invalid, we should have bailed out earlier than | |
657 | * this call. | |
658 | */ | |
659 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
660 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
661 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
662 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
663 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
664 | } else { | |
665 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
666 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
667 | } | |
668 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
669 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
670 | struct btrfs_extent_data_ref); | |
671 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
672 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
673 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
674 | struct btrfs_shared_data_ref); | |
675 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
676 | } else { |
677 | WARN_ON(1); | |
678 | } | |
679 | return num_refs; | |
680 | } | |
15916de8 | 681 | |
5d4f98a2 | 682 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
683 | struct btrfs_path *path, |
684 | u64 bytenr, u64 parent, | |
685 | u64 root_objectid) | |
1f3c79a2 | 686 | { |
b8582eea | 687 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 688 | struct btrfs_key key; |
1f3c79a2 | 689 | int ret; |
1f3c79a2 | 690 | |
5d4f98a2 YZ |
691 | key.objectid = bytenr; |
692 | if (parent) { | |
693 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
694 | key.offset = parent; | |
695 | } else { | |
696 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
697 | key.offset = root_objectid; | |
1f3c79a2 LH |
698 | } |
699 | ||
5d4f98a2 YZ |
700 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
701 | if (ret > 0) | |
702 | ret = -ENOENT; | |
5d4f98a2 | 703 | return ret; |
1f3c79a2 LH |
704 | } |
705 | ||
5d4f98a2 | 706 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
707 | struct btrfs_path *path, |
708 | u64 bytenr, u64 parent, | |
709 | u64 root_objectid) | |
31840ae1 | 710 | { |
5d4f98a2 | 711 | struct btrfs_key key; |
31840ae1 | 712 | int ret; |
31840ae1 | 713 | |
5d4f98a2 YZ |
714 | key.objectid = bytenr; |
715 | if (parent) { | |
716 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
717 | key.offset = parent; | |
718 | } else { | |
719 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
720 | key.offset = root_objectid; | |
721 | } | |
722 | ||
10728404 | 723 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 724 | path, &key, 0); |
b3b4aa74 | 725 | btrfs_release_path(path); |
31840ae1 ZY |
726 | return ret; |
727 | } | |
728 | ||
5d4f98a2 | 729 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 730 | { |
5d4f98a2 YZ |
731 | int type; |
732 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
733 | if (parent > 0) | |
734 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
735 | else | |
736 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
737 | } else { | |
738 | if (parent > 0) | |
739 | type = BTRFS_SHARED_DATA_REF_KEY; | |
740 | else | |
741 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
742 | } | |
743 | return type; | |
31840ae1 | 744 | } |
56bec294 | 745 | |
2c47e605 YZ |
746 | static int find_next_key(struct btrfs_path *path, int level, |
747 | struct btrfs_key *key) | |
56bec294 | 748 | |
02217ed2 | 749 | { |
2c47e605 | 750 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
751 | if (!path->nodes[level]) |
752 | break; | |
5d4f98a2 YZ |
753 | if (path->slots[level] + 1 >= |
754 | btrfs_header_nritems(path->nodes[level])) | |
755 | continue; | |
756 | if (level == 0) | |
757 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
758 | path->slots[level] + 1); | |
759 | else | |
760 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
761 | path->slots[level] + 1); | |
762 | return 0; | |
763 | } | |
764 | return 1; | |
765 | } | |
037e6390 | 766 | |
5d4f98a2 YZ |
767 | /* |
768 | * look for inline back ref. if back ref is found, *ref_ret is set | |
769 | * to the address of inline back ref, and 0 is returned. | |
770 | * | |
771 | * if back ref isn't found, *ref_ret is set to the address where it | |
772 | * should be inserted, and -ENOENT is returned. | |
773 | * | |
774 | * if insert is true and there are too many inline back refs, the path | |
775 | * points to the extent item, and -EAGAIN is returned. | |
776 | * | |
777 | * NOTE: inline back refs are ordered in the same way that back ref | |
778 | * items in the tree are ordered. | |
779 | */ | |
780 | static noinline_for_stack | |
781 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
782 | struct btrfs_path *path, |
783 | struct btrfs_extent_inline_ref **ref_ret, | |
784 | u64 bytenr, u64 num_bytes, | |
785 | u64 parent, u64 root_objectid, | |
786 | u64 owner, u64 offset, int insert) | |
787 | { | |
867cc1fb | 788 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 789 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
790 | struct btrfs_key key; |
791 | struct extent_buffer *leaf; | |
792 | struct btrfs_extent_item *ei; | |
793 | struct btrfs_extent_inline_ref *iref; | |
794 | u64 flags; | |
795 | u64 item_size; | |
796 | unsigned long ptr; | |
797 | unsigned long end; | |
798 | int extra_size; | |
799 | int type; | |
800 | int want; | |
801 | int ret; | |
802 | int err = 0; | |
0b246afa | 803 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 804 | int needed; |
26b8003f | 805 | |
db94535d | 806 | key.objectid = bytenr; |
31840ae1 | 807 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 808 | key.offset = num_bytes; |
31840ae1 | 809 | |
5d4f98a2 YZ |
810 | want = extent_ref_type(parent, owner); |
811 | if (insert) { | |
812 | extra_size = btrfs_extent_inline_ref_size(want); | |
9a664971 | 813 | path->search_for_extension = 1; |
85d4198e | 814 | path->keep_locks = 1; |
5d4f98a2 YZ |
815 | } else |
816 | extra_size = -1; | |
3173a18f JB |
817 | |
818 | /* | |
16d1c062 NB |
819 | * Owner is our level, so we can just add one to get the level for the |
820 | * block we are interested in. | |
3173a18f JB |
821 | */ |
822 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
823 | key.type = BTRFS_METADATA_ITEM_KEY; | |
824 | key.offset = owner; | |
825 | } | |
826 | ||
827 | again: | |
5d4f98a2 | 828 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 829 | if (ret < 0) { |
5d4f98a2 YZ |
830 | err = ret; |
831 | goto out; | |
832 | } | |
3173a18f JB |
833 | |
834 | /* | |
835 | * We may be a newly converted file system which still has the old fat | |
836 | * extent entries for metadata, so try and see if we have one of those. | |
837 | */ | |
838 | if (ret > 0 && skinny_metadata) { | |
839 | skinny_metadata = false; | |
840 | if (path->slots[0]) { | |
841 | path->slots[0]--; | |
842 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
843 | path->slots[0]); | |
844 | if (key.objectid == bytenr && | |
845 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
846 | key.offset == num_bytes) | |
847 | ret = 0; | |
848 | } | |
849 | if (ret) { | |
9ce49a0b | 850 | key.objectid = bytenr; |
3173a18f JB |
851 | key.type = BTRFS_EXTENT_ITEM_KEY; |
852 | key.offset = num_bytes; | |
853 | btrfs_release_path(path); | |
854 | goto again; | |
855 | } | |
856 | } | |
857 | ||
79787eaa JM |
858 | if (ret && !insert) { |
859 | err = -ENOENT; | |
860 | goto out; | |
fae7f21c | 861 | } else if (WARN_ON(ret)) { |
492104c8 | 862 | err = -EIO; |
492104c8 | 863 | goto out; |
79787eaa | 864 | } |
5d4f98a2 YZ |
865 | |
866 | leaf = path->nodes[0]; | |
867 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 868 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
869 | err = -EINVAL; |
870 | btrfs_print_v0_err(fs_info); | |
871 | btrfs_abort_transaction(trans, err); | |
872 | goto out; | |
873 | } | |
5d4f98a2 | 874 | |
5d4f98a2 YZ |
875 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
876 | flags = btrfs_extent_flags(leaf, ei); | |
877 | ||
878 | ptr = (unsigned long)(ei + 1); | |
879 | end = (unsigned long)ei + item_size; | |
880 | ||
3173a18f | 881 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
882 | ptr += sizeof(struct btrfs_tree_block_info); |
883 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
884 | } |
885 | ||
3de28d57 LB |
886 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
887 | needed = BTRFS_REF_TYPE_DATA; | |
888 | else | |
889 | needed = BTRFS_REF_TYPE_BLOCK; | |
890 | ||
5d4f98a2 YZ |
891 | err = -ENOENT; |
892 | while (1) { | |
893 | if (ptr >= end) { | |
894 | WARN_ON(ptr > end); | |
895 | break; | |
896 | } | |
897 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
898 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
899 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 900 | err = -EUCLEAN; |
3de28d57 LB |
901 | goto out; |
902 | } | |
903 | ||
5d4f98a2 YZ |
904 | if (want < type) |
905 | break; | |
906 | if (want > type) { | |
907 | ptr += btrfs_extent_inline_ref_size(type); | |
908 | continue; | |
909 | } | |
910 | ||
911 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
912 | struct btrfs_extent_data_ref *dref; | |
913 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
914 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
915 | owner, offset)) { | |
916 | err = 0; | |
917 | break; | |
918 | } | |
919 | if (hash_extent_data_ref_item(leaf, dref) < | |
920 | hash_extent_data_ref(root_objectid, owner, offset)) | |
921 | break; | |
922 | } else { | |
923 | u64 ref_offset; | |
924 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
925 | if (parent > 0) { | |
926 | if (parent == ref_offset) { | |
927 | err = 0; | |
928 | break; | |
929 | } | |
930 | if (ref_offset < parent) | |
931 | break; | |
932 | } else { | |
933 | if (root_objectid == ref_offset) { | |
934 | err = 0; | |
935 | break; | |
936 | } | |
937 | if (ref_offset < root_objectid) | |
938 | break; | |
939 | } | |
940 | } | |
941 | ptr += btrfs_extent_inline_ref_size(type); | |
942 | } | |
943 | if (err == -ENOENT && insert) { | |
944 | if (item_size + extra_size >= | |
945 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
946 | err = -EAGAIN; | |
947 | goto out; | |
948 | } | |
949 | /* | |
950 | * To add new inline back ref, we have to make sure | |
951 | * there is no corresponding back ref item. | |
952 | * For simplicity, we just do not add new inline back | |
953 | * ref if there is any kind of item for this block | |
954 | */ | |
2c47e605 YZ |
955 | if (find_next_key(path, 0, &key) == 0 && |
956 | key.objectid == bytenr && | |
85d4198e | 957 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
958 | err = -EAGAIN; |
959 | goto out; | |
960 | } | |
961 | } | |
962 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
963 | out: | |
85d4198e | 964 | if (insert) { |
5d4f98a2 | 965 | path->keep_locks = 0; |
9a664971 | 966 | path->search_for_extension = 0; |
5d4f98a2 YZ |
967 | btrfs_unlock_up_safe(path, 1); |
968 | } | |
969 | return err; | |
970 | } | |
971 | ||
972 | /* | |
973 | * helper to add new inline back ref | |
974 | */ | |
975 | static noinline_for_stack | |
87bde3cd | 976 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
977 | struct btrfs_path *path, |
978 | struct btrfs_extent_inline_ref *iref, | |
979 | u64 parent, u64 root_objectid, | |
980 | u64 owner, u64 offset, int refs_to_add, | |
981 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
982 | { |
983 | struct extent_buffer *leaf; | |
984 | struct btrfs_extent_item *ei; | |
985 | unsigned long ptr; | |
986 | unsigned long end; | |
987 | unsigned long item_offset; | |
988 | u64 refs; | |
989 | int size; | |
990 | int type; | |
5d4f98a2 YZ |
991 | |
992 | leaf = path->nodes[0]; | |
993 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
994 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
995 | ||
996 | type = extent_ref_type(parent, owner); | |
997 | size = btrfs_extent_inline_ref_size(type); | |
998 | ||
c71dd880 | 999 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1000 | |
1001 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1002 | refs = btrfs_extent_refs(leaf, ei); | |
1003 | refs += refs_to_add; | |
1004 | btrfs_set_extent_refs(leaf, ei, refs); | |
1005 | if (extent_op) | |
1006 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1007 | ||
1008 | ptr = (unsigned long)ei + item_offset; | |
1009 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1010 | if (ptr < end - size) | |
1011 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1012 | end - size - ptr); | |
1013 | ||
1014 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1015 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1016 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1017 | struct btrfs_extent_data_ref *dref; | |
1018 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1019 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1020 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1021 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1022 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1023 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1024 | struct btrfs_shared_data_ref *sref; | |
1025 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1026 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1027 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1028 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1029 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1030 | } else { | |
1031 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1032 | } | |
1033 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1034 | } |
1035 | ||
1036 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1037 | struct btrfs_path *path, |
1038 | struct btrfs_extent_inline_ref **ref_ret, | |
1039 | u64 bytenr, u64 num_bytes, u64 parent, | |
1040 | u64 root_objectid, u64 owner, u64 offset) | |
1041 | { | |
1042 | int ret; | |
1043 | ||
867cc1fb NB |
1044 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1045 | num_bytes, parent, root_objectid, | |
1046 | owner, offset, 0); | |
5d4f98a2 | 1047 | if (ret != -ENOENT) |
54aa1f4d | 1048 | return ret; |
5d4f98a2 | 1049 | |
b3b4aa74 | 1050 | btrfs_release_path(path); |
5d4f98a2 YZ |
1051 | *ref_ret = NULL; |
1052 | ||
1053 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1054 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1055 | root_objectid); | |
5d4f98a2 | 1056 | } else { |
bd1d53ef NB |
1057 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1058 | root_objectid, owner, offset); | |
b9473439 | 1059 | } |
5d4f98a2 YZ |
1060 | return ret; |
1061 | } | |
31840ae1 | 1062 | |
5d4f98a2 YZ |
1063 | /* |
1064 | * helper to update/remove inline back ref | |
1065 | */ | |
1066 | static noinline_for_stack | |
61a18f1c | 1067 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1068 | struct btrfs_extent_inline_ref *iref, |
1069 | int refs_to_mod, | |
fcebe456 JB |
1070 | struct btrfs_delayed_extent_op *extent_op, |
1071 | int *last_ref) | |
5d4f98a2 | 1072 | { |
61a18f1c | 1073 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1074 | struct btrfs_extent_item *ei; |
1075 | struct btrfs_extent_data_ref *dref = NULL; | |
1076 | struct btrfs_shared_data_ref *sref = NULL; | |
1077 | unsigned long ptr; | |
1078 | unsigned long end; | |
1079 | u32 item_size; | |
1080 | int size; | |
1081 | int type; | |
5d4f98a2 YZ |
1082 | u64 refs; |
1083 | ||
5d4f98a2 YZ |
1084 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1085 | refs = btrfs_extent_refs(leaf, ei); | |
1086 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1087 | refs += refs_to_mod; | |
1088 | btrfs_set_extent_refs(leaf, ei, refs); | |
1089 | if (extent_op) | |
1090 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1091 | ||
3de28d57 LB |
1092 | /* |
1093 | * If type is invalid, we should have bailed out after | |
1094 | * lookup_inline_extent_backref(). | |
1095 | */ | |
1096 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1097 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1098 | |
1099 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1100 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1101 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1102 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1103 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1104 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1105 | } else { | |
1106 | refs = 1; | |
1107 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1108 | } |
31840ae1 | 1109 | |
5d4f98a2 YZ |
1110 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1111 | refs += refs_to_mod; | |
1112 | ||
1113 | if (refs > 0) { | |
1114 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1115 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1116 | else | |
1117 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1118 | } else { | |
fcebe456 | 1119 | *last_ref = 1; |
5d4f98a2 YZ |
1120 | size = btrfs_extent_inline_ref_size(type); |
1121 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1122 | ptr = (unsigned long)iref; | |
1123 | end = (unsigned long)ei + item_size; | |
1124 | if (ptr + size < end) | |
1125 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1126 | end - ptr - size); | |
1127 | item_size -= size; | |
78ac4f9e | 1128 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1129 | } |
1130 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1131 | } |
1132 | ||
1133 | static noinline_for_stack | |
1134 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1135 | struct btrfs_path *path, |
1136 | u64 bytenr, u64 num_bytes, u64 parent, | |
1137 | u64 root_objectid, u64 owner, | |
1138 | u64 offset, int refs_to_add, | |
1139 | struct btrfs_delayed_extent_op *extent_op) | |
1140 | { | |
1141 | struct btrfs_extent_inline_ref *iref; | |
1142 | int ret; | |
1143 | ||
867cc1fb NB |
1144 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1145 | num_bytes, parent, root_objectid, | |
1146 | owner, offset, 1); | |
5d4f98a2 | 1147 | if (ret == 0) { |
07cce5cf QW |
1148 | /* |
1149 | * We're adding refs to a tree block we already own, this | |
1150 | * should not happen at all. | |
1151 | */ | |
1152 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1153 | btrfs_crit(trans->fs_info, | |
1154 | "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu", | |
1155 | bytenr, num_bytes, root_objectid); | |
1156 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
1157 | WARN_ON(1); | |
1158 | btrfs_crit(trans->fs_info, | |
1159 | "path->slots[0]=%d path->nodes[0]:", path->slots[0]); | |
1160 | btrfs_print_leaf(path->nodes[0]); | |
1161 | } | |
1162 | return -EUCLEAN; | |
1163 | } | |
61a18f1c NB |
1164 | update_inline_extent_backref(path, iref, refs_to_add, |
1165 | extent_op, NULL); | |
5d4f98a2 | 1166 | } else if (ret == -ENOENT) { |
a639cdeb | 1167 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1168 | root_objectid, owner, offset, |
1169 | refs_to_add, extent_op); | |
1170 | ret = 0; | |
771ed689 | 1171 | } |
5d4f98a2 YZ |
1172 | return ret; |
1173 | } | |
31840ae1 | 1174 | |
5d4f98a2 | 1175 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1176 | struct btrfs_path *path, |
1177 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1178 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1179 | { |
143bede5 | 1180 | int ret = 0; |
b9473439 | 1181 | |
5d4f98a2 YZ |
1182 | BUG_ON(!is_data && refs_to_drop != 1); |
1183 | if (iref) { | |
61a18f1c NB |
1184 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1185 | last_ref); | |
5d4f98a2 | 1186 | } else if (is_data) { |
e9f6290d | 1187 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1188 | last_ref); |
5d4f98a2 | 1189 | } else { |
fcebe456 | 1190 | *last_ref = 1; |
87cc7a8a | 1191 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1192 | } |
1193 | return ret; | |
1194 | } | |
1195 | ||
d04c6b88 JM |
1196 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1197 | u64 *discarded_bytes) | |
5d4f98a2 | 1198 | { |
86557861 JM |
1199 | int j, ret = 0; |
1200 | u64 bytes_left, end; | |
4d89d377 | 1201 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1202 | |
4d89d377 JM |
1203 | if (WARN_ON(start != aligned_start)) { |
1204 | len -= aligned_start - start; | |
1205 | len = round_down(len, 1 << 9); | |
1206 | start = aligned_start; | |
1207 | } | |
d04c6b88 | 1208 | |
4d89d377 | 1209 | *discarded_bytes = 0; |
86557861 JM |
1210 | |
1211 | if (!len) | |
1212 | return 0; | |
1213 | ||
1214 | end = start + len; | |
1215 | bytes_left = len; | |
1216 | ||
1217 | /* Skip any superblocks on this device. */ | |
1218 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1219 | u64 sb_start = btrfs_sb_offset(j); | |
1220 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1221 | u64 size = sb_start - start; | |
1222 | ||
1223 | if (!in_range(sb_start, start, bytes_left) && | |
1224 | !in_range(sb_end, start, bytes_left) && | |
1225 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1226 | continue; | |
1227 | ||
1228 | /* | |
1229 | * Superblock spans beginning of range. Adjust start and | |
1230 | * try again. | |
1231 | */ | |
1232 | if (sb_start <= start) { | |
1233 | start += sb_end - start; | |
1234 | if (start > end) { | |
1235 | bytes_left = 0; | |
1236 | break; | |
1237 | } | |
1238 | bytes_left = end - start; | |
1239 | continue; | |
1240 | } | |
1241 | ||
1242 | if (size) { | |
1243 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1244 | GFP_NOFS, 0); | |
1245 | if (!ret) | |
1246 | *discarded_bytes += size; | |
1247 | else if (ret != -EOPNOTSUPP) | |
1248 | return ret; | |
1249 | } | |
1250 | ||
1251 | start = sb_end; | |
1252 | if (start > end) { | |
1253 | bytes_left = 0; | |
1254 | break; | |
1255 | } | |
1256 | bytes_left = end - start; | |
1257 | } | |
1258 | ||
1259 | if (bytes_left) { | |
1260 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1261 | GFP_NOFS, 0); |
1262 | if (!ret) | |
86557861 | 1263 | *discarded_bytes += bytes_left; |
4d89d377 | 1264 | } |
d04c6b88 | 1265 | return ret; |
5d4f98a2 | 1266 | } |
5d4f98a2 | 1267 | |
2ff7e61e | 1268 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1269 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1270 | { |
6b7faadd | 1271 | int ret = 0; |
5378e607 | 1272 | u64 discarded_bytes = 0; |
6b7faadd QW |
1273 | u64 end = bytenr + num_bytes; |
1274 | u64 cur = bytenr; | |
a1d3c478 | 1275 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1276 | |
e244a0ae | 1277 | |
2999241d FM |
1278 | /* |
1279 | * Avoid races with device replace and make sure our bbio has devices | |
1280 | * associated to its stripes that don't go away while we are discarding. | |
1281 | */ | |
0b246afa | 1282 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd QW |
1283 | while (cur < end) { |
1284 | struct btrfs_bio_stripe *stripe; | |
5d4f98a2 YZ |
1285 | int i; |
1286 | ||
6b7faadd QW |
1287 | num_bytes = end - cur; |
1288 | /* Tell the block device(s) that the sectors can be discarded */ | |
1289 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
1290 | &num_bytes, &bbio, 0); | |
1291 | /* | |
1292 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1293 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1294 | * thus we can't continue anyway. | |
1295 | */ | |
1296 | if (ret < 0) | |
1297 | goto out; | |
5d4f98a2 | 1298 | |
6b7faadd | 1299 | stripe = bbio->stripes; |
a1d3c478 | 1300 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1301 | u64 bytes; |
38b5f68e AJ |
1302 | struct request_queue *req_q; |
1303 | ||
627e0873 FM |
1304 | if (!stripe->dev->bdev) { |
1305 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1306 | continue; | |
1307 | } | |
38b5f68e AJ |
1308 | req_q = bdev_get_queue(stripe->dev->bdev); |
1309 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1310 | continue; |
1311 | ||
5378e607 LD |
1312 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1313 | stripe->physical, | |
d04c6b88 JM |
1314 | stripe->length, |
1315 | &bytes); | |
6b7faadd | 1316 | if (!ret) { |
d04c6b88 | 1317 | discarded_bytes += bytes; |
6b7faadd QW |
1318 | } else if (ret != -EOPNOTSUPP) { |
1319 | /* | |
1320 | * Logic errors or -ENOMEM, or -EIO, but | |
1321 | * unlikely to happen. | |
1322 | * | |
1323 | * And since there are two loops, explicitly | |
1324 | * go to out to avoid confusion. | |
1325 | */ | |
1326 | btrfs_put_bbio(bbio); | |
1327 | goto out; | |
1328 | } | |
d5e2003c JB |
1329 | |
1330 | /* | |
1331 | * Just in case we get back EOPNOTSUPP for some reason, | |
1332 | * just ignore the return value so we don't screw up | |
1333 | * people calling discard_extent. | |
1334 | */ | |
1335 | ret = 0; | |
5d4f98a2 | 1336 | } |
6e9606d2 | 1337 | btrfs_put_bbio(bbio); |
6b7faadd | 1338 | cur += num_bytes; |
5d4f98a2 | 1339 | } |
6b7faadd | 1340 | out: |
0b246afa | 1341 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1342 | |
1343 | if (actual_bytes) | |
1344 | *actual_bytes = discarded_bytes; | |
1345 | ||
5d4f98a2 | 1346 | |
53b381b3 DW |
1347 | if (ret == -EOPNOTSUPP) |
1348 | ret = 0; | |
5d4f98a2 | 1349 | return ret; |
5d4f98a2 YZ |
1350 | } |
1351 | ||
79787eaa | 1352 | /* Can return -ENOMEM */ |
5d4f98a2 | 1353 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1354 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1355 | { |
82fa113f | 1356 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 1357 | int ret; |
66d7e7f0 | 1358 | |
82fa113f QW |
1359 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1360 | generic_ref->action); | |
1361 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1362 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1363 | |
82fa113f | 1364 | if (generic_ref->type == BTRFS_REF_METADATA) |
2187374f | 1365 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL); |
82fa113f | 1366 | else |
2187374f | 1367 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0); |
d7eae340 | 1368 | |
82fa113f | 1369 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1370 | |
5d4f98a2 YZ |
1371 | return ret; |
1372 | } | |
1373 | ||
bd3c685e NB |
1374 | /* |
1375 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1376 | * | |
07cce5cf QW |
1377 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1378 | * how it works. | |
1379 | * | |
bd3c685e NB |
1380 | * @trans: Handle of transaction |
1381 | * | |
1382 | * @node: The delayed ref node used to get the bytenr/length for | |
1383 | * extent whose references are incremented. | |
1384 | * | |
1385 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1386 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1387 | * bytenr of the parent block. Since new extents are always | |
1388 | * created with indirect references, this will only be the case | |
1389 | * when relocating a shared extent. In that case, root_objectid | |
1390 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1391 | * be 0 | |
1392 | * | |
1393 | * @root_objectid: The id of the root where this modification has originated, | |
1394 | * this can be either one of the well-known metadata trees or | |
1395 | * the subvolume id which references this extent. | |
1396 | * | |
1397 | * @owner: For data extents it is the inode number of the owning file. | |
1398 | * For metadata extents this parameter holds the level in the | |
1399 | * tree of the extent. | |
1400 | * | |
1401 | * @offset: For metadata extents the offset is ignored and is currently | |
1402 | * always passed as 0. For data extents it is the fileoffset | |
1403 | * this extent belongs to. | |
1404 | * | |
1405 | * @refs_to_add Number of references to add | |
1406 | * | |
1407 | * @extent_op Pointer to a structure, holding information necessary when | |
1408 | * updating a tree block's flags | |
1409 | * | |
1410 | */ | |
5d4f98a2 | 1411 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1412 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1413 | u64 parent, u64 root_objectid, |
1414 | u64 owner, u64 offset, int refs_to_add, | |
1415 | struct btrfs_delayed_extent_op *extent_op) | |
1416 | { | |
1417 | struct btrfs_path *path; | |
1418 | struct extent_buffer *leaf; | |
1419 | struct btrfs_extent_item *item; | |
fcebe456 | 1420 | struct btrfs_key key; |
c682f9b3 QW |
1421 | u64 bytenr = node->bytenr; |
1422 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1423 | u64 refs; |
1424 | int ret; | |
5d4f98a2 YZ |
1425 | |
1426 | path = btrfs_alloc_path(); | |
1427 | if (!path) | |
1428 | return -ENOMEM; | |
1429 | ||
5d4f98a2 | 1430 | /* this will setup the path even if it fails to insert the back ref */ |
a639cdeb NB |
1431 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1432 | parent, root_objectid, owner, | |
1433 | offset, refs_to_add, extent_op); | |
0ed4792a | 1434 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1435 | goto out; |
fcebe456 JB |
1436 | |
1437 | /* | |
1438 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1439 | * inline extent ref, so just update the reference count and add a | |
1440 | * normal backref. | |
1441 | */ | |
5d4f98a2 | 1442 | leaf = path->nodes[0]; |
fcebe456 | 1443 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1444 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1445 | refs = btrfs_extent_refs(leaf, item); | |
1446 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1447 | if (extent_op) | |
1448 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1449 | |
5d4f98a2 | 1450 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1451 | btrfs_release_path(path); |
56bec294 | 1452 | |
56bec294 | 1453 | /* now insert the actual backref */ |
65cd6d9e NB |
1454 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
1455 | BUG_ON(refs_to_add != 1); | |
1456 | ret = insert_tree_block_ref(trans, path, bytenr, parent, | |
1457 | root_objectid); | |
1458 | } else { | |
1459 | ret = insert_extent_data_ref(trans, path, bytenr, parent, | |
1460 | root_objectid, owner, offset, | |
1461 | refs_to_add); | |
1462 | } | |
79787eaa | 1463 | if (ret) |
66642832 | 1464 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1465 | out: |
56bec294 | 1466 | btrfs_free_path(path); |
30d133fc | 1467 | return ret; |
56bec294 CM |
1468 | } |
1469 | ||
5d4f98a2 | 1470 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1471 | struct btrfs_delayed_ref_node *node, |
1472 | struct btrfs_delayed_extent_op *extent_op, | |
1473 | int insert_reserved) | |
56bec294 | 1474 | { |
5d4f98a2 YZ |
1475 | int ret = 0; |
1476 | struct btrfs_delayed_data_ref *ref; | |
1477 | struct btrfs_key ins; | |
1478 | u64 parent = 0; | |
1479 | u64 ref_root = 0; | |
1480 | u64 flags = 0; | |
1481 | ||
1482 | ins.objectid = node->bytenr; | |
1483 | ins.offset = node->num_bytes; | |
1484 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1485 | ||
1486 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1487 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1488 | |
5d4f98a2 YZ |
1489 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1490 | parent = ref->parent; | |
fcebe456 | 1491 | ref_root = ref->root; |
5d4f98a2 YZ |
1492 | |
1493 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1494 | if (extent_op) |
5d4f98a2 | 1495 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1496 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1497 | flags, ref->objectid, | |
1498 | ref->offset, &ins, | |
1499 | node->ref_mod); | |
5d4f98a2 | 1500 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1501 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1502 | ref->objectid, ref->offset, | |
1503 | node->ref_mod, extent_op); | |
5d4f98a2 | 1504 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1505 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1506 | ref_root, ref->objectid, |
1507 | ref->offset, node->ref_mod, | |
c682f9b3 | 1508 | extent_op); |
5d4f98a2 YZ |
1509 | } else { |
1510 | BUG(); | |
1511 | } | |
1512 | return ret; | |
1513 | } | |
1514 | ||
1515 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1516 | struct extent_buffer *leaf, | |
1517 | struct btrfs_extent_item *ei) | |
1518 | { | |
1519 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1520 | if (extent_op->update_flags) { | |
1521 | flags |= extent_op->flags_to_set; | |
1522 | btrfs_set_extent_flags(leaf, ei, flags); | |
1523 | } | |
1524 | ||
1525 | if (extent_op->update_key) { | |
1526 | struct btrfs_tree_block_info *bi; | |
1527 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1528 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1529 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1530 | } | |
1531 | } | |
1532 | ||
1533 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1534 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1535 | struct btrfs_delayed_extent_op *extent_op) |
1536 | { | |
20b9a2d6 | 1537 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1538 | struct btrfs_key key; |
1539 | struct btrfs_path *path; | |
1540 | struct btrfs_extent_item *ei; | |
1541 | struct extent_buffer *leaf; | |
1542 | u32 item_size; | |
56bec294 | 1543 | int ret; |
5d4f98a2 | 1544 | int err = 0; |
b1c79e09 | 1545 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1546 | |
bf31f87f | 1547 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1548 | return 0; |
1549 | ||
0b246afa | 1550 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1551 | metadata = 0; |
1552 | ||
5d4f98a2 YZ |
1553 | path = btrfs_alloc_path(); |
1554 | if (!path) | |
1555 | return -ENOMEM; | |
1556 | ||
d278850e | 1557 | key.objectid = head->bytenr; |
5d4f98a2 | 1558 | |
3173a18f | 1559 | if (metadata) { |
3173a18f | 1560 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1561 | key.offset = extent_op->level; |
3173a18f JB |
1562 | } else { |
1563 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1564 | key.offset = head->num_bytes; |
3173a18f JB |
1565 | } |
1566 | ||
1567 | again: | |
0b246afa | 1568 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1569 | if (ret < 0) { |
1570 | err = ret; | |
1571 | goto out; | |
1572 | } | |
1573 | if (ret > 0) { | |
3173a18f | 1574 | if (metadata) { |
55994887 FDBM |
1575 | if (path->slots[0] > 0) { |
1576 | path->slots[0]--; | |
1577 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1578 | path->slots[0]); | |
d278850e | 1579 | if (key.objectid == head->bytenr && |
55994887 | 1580 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1581 | key.offset == head->num_bytes) |
55994887 FDBM |
1582 | ret = 0; |
1583 | } | |
1584 | if (ret > 0) { | |
1585 | btrfs_release_path(path); | |
1586 | metadata = 0; | |
3173a18f | 1587 | |
d278850e JB |
1588 | key.objectid = head->bytenr; |
1589 | key.offset = head->num_bytes; | |
55994887 FDBM |
1590 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1591 | goto again; | |
1592 | } | |
1593 | } else { | |
1594 | err = -EIO; | |
1595 | goto out; | |
3173a18f | 1596 | } |
5d4f98a2 YZ |
1597 | } |
1598 | ||
1599 | leaf = path->nodes[0]; | |
1600 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1601 | |
6d8ff4e4 | 1602 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1603 | err = -EINVAL; |
1604 | btrfs_print_v0_err(fs_info); | |
1605 | btrfs_abort_transaction(trans, err); | |
1606 | goto out; | |
1607 | } | |
1608 | ||
5d4f98a2 YZ |
1609 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1610 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1611 | |
5d4f98a2 YZ |
1612 | btrfs_mark_buffer_dirty(leaf); |
1613 | out: | |
1614 | btrfs_free_path(path); | |
1615 | return err; | |
56bec294 CM |
1616 | } |
1617 | ||
5d4f98a2 | 1618 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1619 | struct btrfs_delayed_ref_node *node, |
1620 | struct btrfs_delayed_extent_op *extent_op, | |
1621 | int insert_reserved) | |
56bec294 CM |
1622 | { |
1623 | int ret = 0; | |
5d4f98a2 | 1624 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1625 | u64 parent = 0; |
1626 | u64 ref_root = 0; | |
56bec294 | 1627 | |
5d4f98a2 | 1628 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1629 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1630 | |
5d4f98a2 YZ |
1631 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1632 | parent = ref->parent; | |
fcebe456 | 1633 | ref_root = ref->root; |
5d4f98a2 | 1634 | |
02794222 | 1635 | if (node->ref_mod != 1) { |
f97806f2 | 1636 | btrfs_err(trans->fs_info, |
02794222 LB |
1637 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1638 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1639 | parent); | |
1640 | return -EIO; | |
1641 | } | |
5d4f98a2 | 1642 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1643 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1644 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1645 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1646 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1647 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1648 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1649 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1650 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1651 | } else { |
1652 | BUG(); | |
1653 | } | |
56bec294 CM |
1654 | return ret; |
1655 | } | |
1656 | ||
1657 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1658 | static int run_one_delayed_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 | 1662 | { |
79787eaa JM |
1663 | int ret = 0; |
1664 | ||
bf31f87f | 1665 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1666 | if (insert_reserved) |
b25c36f8 | 1667 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1668 | return 0; |
857cc2fc | 1669 | } |
79787eaa | 1670 | |
5d4f98a2 YZ |
1671 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1672 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1673 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1674 | insert_reserved); |
1675 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1676 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1677 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1678 | insert_reserved); |
1679 | else | |
1680 | BUG(); | |
80ee54bf | 1681 | if (ret && insert_reserved) |
b25c36f8 | 1682 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
5d4f98a2 | 1683 | return ret; |
56bec294 CM |
1684 | } |
1685 | ||
c6fc2454 | 1686 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1687 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1688 | { | |
cffc3374 FM |
1689 | struct btrfs_delayed_ref_node *ref; |
1690 | ||
e3d03965 | 1691 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1692 | return NULL; |
d7df2c79 | 1693 | |
cffc3374 FM |
1694 | /* |
1695 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1696 | * This is to prevent a ref count from going down to zero, which deletes | |
1697 | * the extent item from the extent tree, when there still are references | |
1698 | * to add, which would fail because they would not find the extent item. | |
1699 | */ | |
1d57ee94 WX |
1700 | if (!list_empty(&head->ref_add_list)) |
1701 | return list_first_entry(&head->ref_add_list, | |
1702 | struct btrfs_delayed_ref_node, add_list); | |
1703 | ||
e3d03965 | 1704 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1705 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1706 | ASSERT(list_empty(&ref->add_list)); |
1707 | return ref; | |
56bec294 CM |
1708 | } |
1709 | ||
2eadaa22 JB |
1710 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1711 | struct btrfs_delayed_ref_head *head) | |
1712 | { | |
1713 | spin_lock(&delayed_refs->lock); | |
1714 | head->processing = 0; | |
1715 | delayed_refs->num_heads_ready++; | |
1716 | spin_unlock(&delayed_refs->lock); | |
1717 | btrfs_delayed_ref_unlock(head); | |
1718 | } | |
1719 | ||
bedc6617 JB |
1720 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1721 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1722 | { |
1723 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1724 | |
1725 | if (!extent_op) | |
bedc6617 JB |
1726 | return NULL; |
1727 | ||
b00e6250 | 1728 | if (head->must_insert_reserved) { |
bedc6617 | 1729 | head->extent_op = NULL; |
b00e6250 | 1730 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1731 | return NULL; |
b00e6250 | 1732 | } |
bedc6617 JB |
1733 | return extent_op; |
1734 | } | |
1735 | ||
1736 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1737 | struct btrfs_delayed_ref_head *head) | |
1738 | { | |
1739 | struct btrfs_delayed_extent_op *extent_op; | |
1740 | int ret; | |
1741 | ||
1742 | extent_op = cleanup_extent_op(head); | |
1743 | if (!extent_op) | |
1744 | return 0; | |
1745 | head->extent_op = NULL; | |
b00e6250 | 1746 | spin_unlock(&head->lock); |
20b9a2d6 | 1747 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1748 | btrfs_free_delayed_extent_op(extent_op); |
1749 | return ret ? ret : 1; | |
1750 | } | |
1751 | ||
31890da0 JB |
1752 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1753 | struct btrfs_delayed_ref_root *delayed_refs, | |
1754 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1755 | { |
ba2c4d4e | 1756 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 | 1757 | |
81e75ac7 JB |
1758 | /* |
1759 | * We had csum deletions accounted for in our delayed refs rsv, we need | |
1760 | * to drop the csum leaves for this update from our delayed_refs_rsv. | |
1761 | */ | |
1762 | if (head->total_ref_mod < 0 && head->is_data) { | |
1763 | spin_lock(&delayed_refs->lock); | |
1764 | delayed_refs->pending_csums -= head->num_bytes; | |
1765 | spin_unlock(&delayed_refs->lock); | |
1766 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes); | |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * We were dropping refs, or had a new ref and dropped it, and thus must | |
1771 | * adjust down our total_bytes_pinned, the space may or may not have | |
1772 | * been pinned and so is accounted for properly in the pinned space by | |
1773 | * now. | |
1774 | */ | |
1775 | if (head->total_ref_mod < 0 || | |
1776 | (head->total_ref_mod == 0 && head->must_insert_reserved)) { | |
2187374f | 1777 | u64 flags = btrfs_ref_head_to_space_flags(head); |
07c47775 | 1778 | |
2187374f | 1779 | btrfs_mod_total_bytes_pinned(fs_info, flags, -head->num_bytes); |
07c47775 JB |
1780 | } |
1781 | ||
ba2c4d4e | 1782 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1783 | } |
1784 | ||
194ab0bc | 1785 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1786 | struct btrfs_delayed_ref_head *head) |
1787 | { | |
f9871edd NB |
1788 | |
1789 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1790 | struct btrfs_delayed_ref_root *delayed_refs; |
1791 | int ret; | |
1792 | ||
1793 | delayed_refs = &trans->transaction->delayed_refs; | |
1794 | ||
bedc6617 | 1795 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1796 | if (ret < 0) { |
1797 | unselect_delayed_ref_head(delayed_refs, head); | |
1798 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1799 | return ret; | |
1800 | } else if (ret) { | |
1801 | return ret; | |
1802 | } | |
1803 | ||
1804 | /* | |
1805 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1806 | * and then re-check to make sure nobody got added. | |
1807 | */ | |
1808 | spin_unlock(&head->lock); | |
1809 | spin_lock(&delayed_refs->lock); | |
1810 | spin_lock(&head->lock); | |
e3d03965 | 1811 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1812 | spin_unlock(&head->lock); |
1813 | spin_unlock(&delayed_refs->lock); | |
1814 | return 1; | |
1815 | } | |
d7baffda | 1816 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1817 | spin_unlock(&head->lock); |
1e7a1421 | 1818 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1819 | |
c1103f7a | 1820 | if (head->must_insert_reserved) { |
b25c36f8 | 1821 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1822 | if (head->is_data) { |
40e046ac FM |
1823 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1824 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1825 | } |
1826 | } | |
1827 | ||
31890da0 | 1828 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1829 | |
1830 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1831 | btrfs_delayed_ref_unlock(head); |
d278850e | 1832 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1833 | return 0; |
1834 | } | |
1835 | ||
b1cdbcb5 NB |
1836 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1837 | struct btrfs_trans_handle *trans) | |
1838 | { | |
1839 | struct btrfs_delayed_ref_root *delayed_refs = | |
1840 | &trans->transaction->delayed_refs; | |
1841 | struct btrfs_delayed_ref_head *head = NULL; | |
1842 | int ret; | |
1843 | ||
1844 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1845 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1846 | if (!head) { |
1847 | spin_unlock(&delayed_refs->lock); | |
1848 | return head; | |
1849 | } | |
1850 | ||
1851 | /* | |
1852 | * Grab the lock that says we are going to process all the refs for | |
1853 | * this head | |
1854 | */ | |
9e920a6f | 1855 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1856 | spin_unlock(&delayed_refs->lock); |
1857 | ||
1858 | /* | |
1859 | * We may have dropped the spin lock to get the head mutex lock, and | |
1860 | * that might have given someone else time to free the head. If that's | |
1861 | * true, it has been removed from our list and we can move on. | |
1862 | */ | |
1863 | if (ret == -EAGAIN) | |
1864 | head = ERR_PTR(-EAGAIN); | |
1865 | ||
1866 | return head; | |
1867 | } | |
1868 | ||
e7261386 NB |
1869 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1870 | struct btrfs_delayed_ref_head *locked_ref, | |
1871 | unsigned long *run_refs) | |
1872 | { | |
1873 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1874 | struct btrfs_delayed_ref_root *delayed_refs; | |
1875 | struct btrfs_delayed_extent_op *extent_op; | |
1876 | struct btrfs_delayed_ref_node *ref; | |
1877 | int must_insert_reserved = 0; | |
1878 | int ret; | |
1879 | ||
1880 | delayed_refs = &trans->transaction->delayed_refs; | |
1881 | ||
0110a4c4 NB |
1882 | lockdep_assert_held(&locked_ref->mutex); |
1883 | lockdep_assert_held(&locked_ref->lock); | |
1884 | ||
e7261386 NB |
1885 | while ((ref = select_delayed_ref(locked_ref))) { |
1886 | if (ref->seq && | |
1887 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1888 | spin_unlock(&locked_ref->lock); | |
1889 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1890 | return -EAGAIN; | |
1891 | } | |
1892 | ||
1893 | (*run_refs)++; | |
1894 | ref->in_tree = 0; | |
1895 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1896 | RB_CLEAR_NODE(&ref->ref_node); | |
1897 | if (!list_empty(&ref->add_list)) | |
1898 | list_del(&ref->add_list); | |
1899 | /* | |
1900 | * When we play the delayed ref, also correct the ref_mod on | |
1901 | * head | |
1902 | */ | |
1903 | switch (ref->action) { | |
1904 | case BTRFS_ADD_DELAYED_REF: | |
1905 | case BTRFS_ADD_DELAYED_EXTENT: | |
1906 | locked_ref->ref_mod -= ref->ref_mod; | |
1907 | break; | |
1908 | case BTRFS_DROP_DELAYED_REF: | |
1909 | locked_ref->ref_mod += ref->ref_mod; | |
1910 | break; | |
1911 | default: | |
1912 | WARN_ON(1); | |
1913 | } | |
1914 | atomic_dec(&delayed_refs->num_entries); | |
1915 | ||
1916 | /* | |
1917 | * Record the must_insert_reserved flag before we drop the | |
1918 | * spin lock. | |
1919 | */ | |
1920 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1921 | locked_ref->must_insert_reserved = 0; | |
1922 | ||
1923 | extent_op = locked_ref->extent_op; | |
1924 | locked_ref->extent_op = NULL; | |
1925 | spin_unlock(&locked_ref->lock); | |
1926 | ||
1927 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1928 | must_insert_reserved); | |
1929 | ||
1930 | btrfs_free_delayed_extent_op(extent_op); | |
1931 | if (ret) { | |
1932 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1933 | btrfs_put_delayed_ref(ref); | |
1934 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1935 | ret); | |
1936 | return ret; | |
1937 | } | |
1938 | ||
1939 | btrfs_put_delayed_ref(ref); | |
1940 | cond_resched(); | |
1941 | ||
1942 | spin_lock(&locked_ref->lock); | |
1943 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1944 | } | |
1945 | ||
1946 | return 0; | |
1947 | } | |
1948 | ||
79787eaa JM |
1949 | /* |
1950 | * Returns 0 on success or if called with an already aborted transaction. | |
1951 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1952 | */ | |
d7df2c79 | 1953 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 1954 | unsigned long nr) |
56bec294 | 1955 | { |
0a1e458a | 1956 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 1957 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 1958 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 1959 | ktime_t start = ktime_get(); |
56bec294 | 1960 | int ret; |
d7df2c79 | 1961 | unsigned long count = 0; |
0a2b2a84 | 1962 | unsigned long actual_count = 0; |
56bec294 CM |
1963 | |
1964 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 1965 | do { |
56bec294 | 1966 | if (!locked_ref) { |
b1cdbcb5 | 1967 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
1968 | if (IS_ERR_OR_NULL(locked_ref)) { |
1969 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
1970 | continue; | |
1971 | } else { | |
1972 | break; | |
1973 | } | |
56bec294 | 1974 | } |
0110a4c4 | 1975 | count++; |
56bec294 | 1976 | } |
2c3cf7d5 FM |
1977 | /* |
1978 | * We need to try and merge add/drops of the same ref since we | |
1979 | * can run into issues with relocate dropping the implicit ref | |
1980 | * and then it being added back again before the drop can | |
1981 | * finish. If we merged anything we need to re-loop so we can | |
1982 | * get a good ref. | |
1983 | * Or we can get node references of the same type that weren't | |
1984 | * merged when created due to bumps in the tree mod seq, and | |
1985 | * we need to merge them to prevent adding an inline extent | |
1986 | * backref before dropping it (triggering a BUG_ON at | |
1987 | * insert_inline_extent_backref()). | |
1988 | */ | |
d7df2c79 | 1989 | spin_lock(&locked_ref->lock); |
be97f133 | 1990 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 1991 | |
0110a4c4 NB |
1992 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
1993 | &actual_count); | |
1994 | if (ret < 0 && ret != -EAGAIN) { | |
1995 | /* | |
1996 | * Error, btrfs_run_delayed_refs_for_head already | |
1997 | * unlocked everything so just bail out | |
1998 | */ | |
1999 | return ret; | |
2000 | } else if (!ret) { | |
2001 | /* | |
2002 | * Success, perform the usual cleanup of a processed | |
2003 | * head | |
2004 | */ | |
f9871edd | 2005 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2006 | if (ret > 0 ) { |
b00e6250 JB |
2007 | /* We dropped our lock, we need to loop. */ |
2008 | ret = 0; | |
d7df2c79 | 2009 | continue; |
194ab0bc JB |
2010 | } else if (ret) { |
2011 | return ret; | |
5d4f98a2 | 2012 | } |
22cd2e7d | 2013 | } |
1ce7a5ec | 2014 | |
b00e6250 | 2015 | /* |
0110a4c4 NB |
2016 | * Either success case or btrfs_run_delayed_refs_for_head |
2017 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2018 | */ |
b00e6250 | 2019 | |
0110a4c4 | 2020 | locked_ref = NULL; |
c3e69d58 | 2021 | cond_resched(); |
0110a4c4 | 2022 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2023 | |
2024 | /* | |
2025 | * We don't want to include ref heads since we can have empty ref heads | |
2026 | * and those will drastically skew our runtime down since we just do | |
2027 | * accounting, no actual extent tree updates. | |
2028 | */ | |
2029 | if (actual_count > 0) { | |
2030 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2031 | u64 avg; | |
2032 | ||
2033 | /* | |
2034 | * We weigh the current average higher than our current runtime | |
2035 | * to avoid large swings in the average. | |
2036 | */ | |
2037 | spin_lock(&delayed_refs->lock); | |
2038 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2039 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2040 | spin_unlock(&delayed_refs->lock); |
2041 | } | |
d7df2c79 | 2042 | return 0; |
c3e69d58 CM |
2043 | } |
2044 | ||
709c0486 AJ |
2045 | #ifdef SCRAMBLE_DELAYED_REFS |
2046 | /* | |
2047 | * Normally delayed refs get processed in ascending bytenr order. This | |
2048 | * correlates in most cases to the order added. To expose dependencies on this | |
2049 | * order, we start to process the tree in the middle instead of the beginning | |
2050 | */ | |
2051 | static u64 find_middle(struct rb_root *root) | |
2052 | { | |
2053 | struct rb_node *n = root->rb_node; | |
2054 | struct btrfs_delayed_ref_node *entry; | |
2055 | int alt = 1; | |
2056 | u64 middle; | |
2057 | u64 first = 0, last = 0; | |
2058 | ||
2059 | n = rb_first(root); | |
2060 | if (n) { | |
2061 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2062 | first = entry->bytenr; | |
2063 | } | |
2064 | n = rb_last(root); | |
2065 | if (n) { | |
2066 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2067 | last = entry->bytenr; | |
2068 | } | |
2069 | n = root->rb_node; | |
2070 | ||
2071 | while (n) { | |
2072 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2073 | WARN_ON(!entry->in_tree); | |
2074 | ||
2075 | middle = entry->bytenr; | |
2076 | ||
2077 | if (alt) | |
2078 | n = n->rb_left; | |
2079 | else | |
2080 | n = n->rb_right; | |
2081 | ||
2082 | alt = 1 - alt; | |
2083 | } | |
2084 | return middle; | |
2085 | } | |
2086 | #endif | |
2087 | ||
c3e69d58 CM |
2088 | /* |
2089 | * this starts processing the delayed reference count updates and | |
2090 | * extent insertions we have queued up so far. count can be | |
2091 | * 0, which means to process everything in the tree at the start | |
2092 | * of the run (but not newly added entries), or it can be some target | |
2093 | * number you'd like to process. | |
79787eaa JM |
2094 | * |
2095 | * Returns 0 on success or if called with an aborted transaction | |
2096 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2097 | */ |
2098 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2099 | unsigned long count) |
c3e69d58 | 2100 | { |
c79a70b1 | 2101 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2102 | struct rb_node *node; |
2103 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2104 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2105 | int ret; |
2106 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2107 | |
79787eaa | 2108 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2109 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2110 | return 0; |
2111 | ||
0b246afa | 2112 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2113 | return 0; |
2114 | ||
c3e69d58 | 2115 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2116 | if (count == 0) |
61a56a99 | 2117 | count = delayed_refs->num_heads_ready; |
bb721703 | 2118 | |
c3e69d58 | 2119 | again: |
709c0486 AJ |
2120 | #ifdef SCRAMBLE_DELAYED_REFS |
2121 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2122 | #endif | |
0a1e458a | 2123 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2124 | if (ret < 0) { |
66642832 | 2125 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2126 | return ret; |
eb099670 | 2127 | } |
c3e69d58 | 2128 | |
56bec294 | 2129 | if (run_all) { |
119e80df | 2130 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2131 | |
d7df2c79 | 2132 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2133 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2134 | if (!node) { |
2135 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2136 | goto out; |
d7df2c79 | 2137 | } |
d278850e JB |
2138 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2139 | href_node); | |
2140 | refcount_inc(&head->refs); | |
2141 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2142 | |
d278850e JB |
2143 | /* Mutex was contended, block until it's released and retry. */ |
2144 | mutex_lock(&head->mutex); | |
2145 | mutex_unlock(&head->mutex); | |
56bec294 | 2146 | |
d278850e | 2147 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2148 | cond_resched(); |
56bec294 | 2149 | goto again; |
5f39d397 | 2150 | } |
54aa1f4d | 2151 | out: |
a28ec197 CM |
2152 | return 0; |
2153 | } | |
2154 | ||
5d4f98a2 | 2155 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
42c9d0b5 | 2156 | struct extent_buffer *eb, u64 flags, |
b1c79e09 | 2157 | int level, int is_data) |
5d4f98a2 YZ |
2158 | { |
2159 | struct btrfs_delayed_extent_op *extent_op; | |
2160 | int ret; | |
2161 | ||
78a6184a | 2162 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2163 | if (!extent_op) |
2164 | return -ENOMEM; | |
2165 | ||
2166 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2167 | extent_op->update_flags = true; |
2168 | extent_op->update_key = false; | |
2169 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2170 | extent_op->level = level; |
5d4f98a2 | 2171 | |
42c9d0b5 | 2172 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2173 | if (ret) |
78a6184a | 2174 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2175 | return ret; |
2176 | } | |
2177 | ||
e4c3b2dc | 2178 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2179 | struct btrfs_path *path, |
2180 | u64 objectid, u64 offset, u64 bytenr) | |
2181 | { | |
2182 | struct btrfs_delayed_ref_head *head; | |
2183 | struct btrfs_delayed_ref_node *ref; | |
2184 | struct btrfs_delayed_data_ref *data_ref; | |
2185 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2186 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2187 | struct rb_node *node; |
5d4f98a2 YZ |
2188 | int ret = 0; |
2189 | ||
998ac6d2 | 2190 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2191 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2192 | if (cur_trans) |
2193 | refcount_inc(&cur_trans->use_count); | |
2194 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2195 | if (!cur_trans) |
2196 | return 0; | |
2197 | ||
2198 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2199 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2200 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2201 | if (!head) { |
2202 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2203 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2204 | return 0; |
2205 | } | |
5d4f98a2 YZ |
2206 | |
2207 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2208 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2209 | spin_unlock(&delayed_refs->lock); |
2210 | ||
b3b4aa74 | 2211 | btrfs_release_path(path); |
5d4f98a2 | 2212 | |
8cc33e5c DS |
2213 | /* |
2214 | * Mutex was contended, block until it's released and let | |
2215 | * caller try again | |
2216 | */ | |
5d4f98a2 YZ |
2217 | mutex_lock(&head->mutex); |
2218 | mutex_unlock(&head->mutex); | |
d278850e | 2219 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2220 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2221 | return -EAGAIN; |
2222 | } | |
d7df2c79 | 2223 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2224 | |
d7df2c79 | 2225 | spin_lock(&head->lock); |
0e0adbcf JB |
2226 | /* |
2227 | * XXX: We should replace this with a proper search function in the | |
2228 | * future. | |
2229 | */ | |
e3d03965 LB |
2230 | for (node = rb_first_cached(&head->ref_tree); node; |
2231 | node = rb_next(node)) { | |
0e0adbcf | 2232 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2233 | /* If it's a shared ref we know a cross reference exists */ |
2234 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2235 | ret = 1; | |
2236 | break; | |
2237 | } | |
5d4f98a2 | 2238 | |
d7df2c79 | 2239 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2240 | |
d7df2c79 JB |
2241 | /* |
2242 | * If our ref doesn't match the one we're currently looking at | |
2243 | * then we have a cross reference. | |
2244 | */ | |
2245 | if (data_ref->root != root->root_key.objectid || | |
2246 | data_ref->objectid != objectid || | |
2247 | data_ref->offset != offset) { | |
2248 | ret = 1; | |
2249 | break; | |
2250 | } | |
5d4f98a2 | 2251 | } |
d7df2c79 | 2252 | spin_unlock(&head->lock); |
5d4f98a2 | 2253 | mutex_unlock(&head->mutex); |
998ac6d2 | 2254 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2255 | return ret; |
2256 | } | |
2257 | ||
e4c3b2dc | 2258 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2259 | struct btrfs_path *path, |
a84d5d42 BB |
2260 | u64 objectid, u64 offset, u64 bytenr, |
2261 | bool strict) | |
be20aa9d | 2262 | { |
0b246afa JM |
2263 | struct btrfs_fs_info *fs_info = root->fs_info; |
2264 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2265 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2266 | struct btrfs_extent_data_ref *ref; |
2267 | struct btrfs_extent_inline_ref *iref; | |
2268 | struct btrfs_extent_item *ei; | |
f321e491 | 2269 | struct btrfs_key key; |
5d4f98a2 | 2270 | u32 item_size; |
3de28d57 | 2271 | int type; |
be20aa9d | 2272 | int ret; |
925baedd | 2273 | |
be20aa9d | 2274 | key.objectid = bytenr; |
31840ae1 | 2275 | key.offset = (u64)-1; |
f321e491 | 2276 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2277 | |
be20aa9d CM |
2278 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2279 | if (ret < 0) | |
2280 | goto out; | |
79787eaa | 2281 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2282 | |
2283 | ret = -ENOENT; | |
2284 | if (path->slots[0] == 0) | |
31840ae1 | 2285 | goto out; |
be20aa9d | 2286 | |
31840ae1 | 2287 | path->slots[0]--; |
f321e491 | 2288 | leaf = path->nodes[0]; |
5d4f98a2 | 2289 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2290 | |
5d4f98a2 | 2291 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2292 | goto out; |
f321e491 | 2293 | |
5d4f98a2 YZ |
2294 | ret = 1; |
2295 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2296 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2297 | |
a6bd9cd1 | 2298 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2299 | if (item_size != sizeof(*ei) + |
2300 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2301 | goto out; | |
be20aa9d | 2302 | |
a84d5d42 BB |
2303 | /* |
2304 | * If extent created before last snapshot => it's shared unless the | |
2305 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2306 | */ | |
2307 | if (!strict && | |
2308 | (btrfs_extent_generation(leaf, ei) <= | |
2309 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2310 | goto out; |
2311 | ||
2312 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2313 | |
a6bd9cd1 | 2314 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2315 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2316 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2317 | goto out; |
2318 | ||
2319 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2320 | if (btrfs_extent_refs(leaf, ei) != | |
2321 | btrfs_extent_data_ref_count(leaf, ref) || | |
2322 | btrfs_extent_data_ref_root(leaf, ref) != | |
2323 | root->root_key.objectid || | |
2324 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2325 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2326 | goto out; | |
2327 | ||
2328 | ret = 0; | |
2329 | out: | |
2330 | return ret; | |
2331 | } | |
2332 | ||
e4c3b2dc | 2333 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
a84d5d42 | 2334 | u64 bytenr, bool strict) |
5d4f98a2 YZ |
2335 | { |
2336 | struct btrfs_path *path; | |
2337 | int ret; | |
5d4f98a2 YZ |
2338 | |
2339 | path = btrfs_alloc_path(); | |
2340 | if (!path) | |
9132c4ff | 2341 | return -ENOMEM; |
5d4f98a2 YZ |
2342 | |
2343 | do { | |
e4c3b2dc | 2344 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2345 | offset, bytenr, strict); |
5d4f98a2 | 2346 | if (ret && ret != -ENOENT) |
f321e491 | 2347 | goto out; |
80ff3856 | 2348 | |
380fd066 MT |
2349 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2350 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2351 | |
be20aa9d | 2352 | out: |
80ff3856 | 2353 | btrfs_free_path(path); |
f0486c68 YZ |
2354 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2355 | WARN_ON(ret > 0); | |
f321e491 | 2356 | return ret; |
be20aa9d | 2357 | } |
c5739bba | 2358 | |
5d4f98a2 | 2359 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2360 | struct btrfs_root *root, |
5d4f98a2 | 2361 | struct extent_buffer *buf, |
e339a6b0 | 2362 | int full_backref, int inc) |
31840ae1 | 2363 | { |
0b246afa | 2364 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2365 | u64 bytenr; |
5d4f98a2 YZ |
2366 | u64 num_bytes; |
2367 | u64 parent; | |
31840ae1 | 2368 | u64 ref_root; |
31840ae1 | 2369 | u32 nritems; |
31840ae1 ZY |
2370 | struct btrfs_key key; |
2371 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2372 | struct btrfs_ref generic_ref = { 0 }; |
2373 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2374 | int i; |
82fa113f | 2375 | int action; |
31840ae1 ZY |
2376 | int level; |
2377 | int ret = 0; | |
fccb84c9 | 2378 | |
0b246afa | 2379 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2380 | return 0; |
fccb84c9 | 2381 | |
31840ae1 | 2382 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2383 | nritems = btrfs_header_nritems(buf); |
2384 | level = btrfs_header_level(buf); | |
2385 | ||
92a7cc42 | 2386 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2387 | return 0; |
31840ae1 | 2388 | |
5d4f98a2 YZ |
2389 | if (full_backref) |
2390 | parent = buf->start; | |
2391 | else | |
2392 | parent = 0; | |
82fa113f QW |
2393 | if (inc) |
2394 | action = BTRFS_ADD_DELAYED_REF; | |
2395 | else | |
2396 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2397 | |
2398 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2399 | if (level == 0) { |
5d4f98a2 | 2400 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2401 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2402 | continue; |
5d4f98a2 | 2403 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2404 | struct btrfs_file_extent_item); |
2405 | if (btrfs_file_extent_type(buf, fi) == | |
2406 | BTRFS_FILE_EXTENT_INLINE) | |
2407 | continue; | |
2408 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2409 | if (bytenr == 0) | |
2410 | continue; | |
5d4f98a2 YZ |
2411 | |
2412 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2413 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2414 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2415 | num_bytes, parent); | |
2416 | generic_ref.real_root = root->root_key.objectid; | |
2417 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2418 | key.offset); | |
2419 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2420 | if (inc) |
82fa113f | 2421 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2422 | else |
ffd4bb2a | 2423 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2424 | if (ret) |
2425 | goto fail; | |
2426 | } else { | |
5d4f98a2 | 2427 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2428 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2429 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2430 | num_bytes, parent); | |
2431 | generic_ref.real_root = root->root_key.objectid; | |
2432 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2433 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2434 | if (inc) |
82fa113f | 2435 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2436 | else |
ffd4bb2a | 2437 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2438 | if (ret) |
2439 | goto fail; | |
2440 | } | |
2441 | } | |
2442 | return 0; | |
2443 | fail: | |
5d4f98a2 YZ |
2444 | return ret; |
2445 | } | |
2446 | ||
2447 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2448 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2449 | { |
e339a6b0 | 2450 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2451 | } |
2452 | ||
2453 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2454 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2455 | { |
e339a6b0 | 2456 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2457 | } |
2458 | ||
2ff7e61e | 2459 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 | 2460 | { |
32da5386 | 2461 | struct btrfs_block_group *block_group; |
d2fb3437 YZ |
2462 | int readonly = 0; |
2463 | ||
0b246afa | 2464 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2465 | if (!block_group || block_group->ro) |
2466 | readonly = 1; | |
2467 | if (block_group) | |
fa9c0d79 | 2468 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2469 | return readonly; |
2470 | } | |
2471 | ||
1b86826d | 2472 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2473 | { |
0b246afa | 2474 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2475 | u64 flags; |
53b381b3 | 2476 | u64 ret; |
9ed74f2d | 2477 | |
b742bb82 YZ |
2478 | if (data) |
2479 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2480 | else if (root == fs_info->chunk_root) |
b742bb82 | 2481 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2482 | else |
b742bb82 | 2483 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2484 | |
878d7b67 | 2485 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2486 | return ret; |
6a63209f | 2487 | } |
9ed74f2d | 2488 | |
2ff7e61e | 2489 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2490 | { |
32da5386 | 2491 | struct btrfs_block_group *cache; |
d2fb3437 | 2492 | u64 bytenr; |
0f9dd46c | 2493 | |
0b246afa JM |
2494 | spin_lock(&fs_info->block_group_cache_lock); |
2495 | bytenr = fs_info->first_logical_byte; | |
2496 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2497 | |
2498 | if (bytenr < (u64)-1) | |
2499 | return bytenr; | |
2500 | ||
0b246afa | 2501 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2502 | if (!cache) |
a061fc8d | 2503 | return 0; |
0f9dd46c | 2504 | |
b3470b5d | 2505 | bytenr = cache->start; |
fa9c0d79 | 2506 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2507 | |
2508 | return bytenr; | |
a061fc8d CM |
2509 | } |
2510 | ||
6690d071 NB |
2511 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2512 | struct btrfs_block_group *cache, | |
f0486c68 | 2513 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2514 | { |
fdf08605 DS |
2515 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2516 | ||
11833d66 YZ |
2517 | spin_lock(&cache->space_info->lock); |
2518 | spin_lock(&cache->lock); | |
2519 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2520 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2521 | num_bytes); | |
11833d66 YZ |
2522 | if (reserved) { |
2523 | cache->reserved -= num_bytes; | |
2524 | cache->space_info->bytes_reserved -= num_bytes; | |
2525 | } | |
2526 | spin_unlock(&cache->lock); | |
2527 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2528 | |
2187374f | 2529 | __btrfs_mod_total_bytes_pinned(cache->space_info, num_bytes); |
fe119a6e | 2530 | set_extent_dirty(&trans->transaction->pinned_extents, bytenr, |
f0486c68 YZ |
2531 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2532 | return 0; | |
2533 | } | |
68b38550 | 2534 | |
b25c36f8 | 2535 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2536 | u64 bytenr, u64 num_bytes, int reserved) |
2537 | { | |
32da5386 | 2538 | struct btrfs_block_group *cache; |
68b38550 | 2539 | |
b25c36f8 | 2540 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2541 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2542 | |
6690d071 | 2543 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2544 | |
2545 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2546 | return 0; |
2547 | } | |
2548 | ||
f0486c68 | 2549 | /* |
e688b725 CM |
2550 | * this function must be called within transaction |
2551 | */ | |
9fce5704 | 2552 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
e688b725 CM |
2553 | u64 bytenr, u64 num_bytes) |
2554 | { | |
32da5386 | 2555 | struct btrfs_block_group *cache; |
b50c6e25 | 2556 | int ret; |
e688b725 | 2557 | |
9fce5704 | 2558 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
b50c6e25 JB |
2559 | if (!cache) |
2560 | return -EINVAL; | |
e688b725 CM |
2561 | |
2562 | /* | |
2563 | * pull in the free space cache (if any) so that our pin | |
2564 | * removes the free space from the cache. We have load_only set | |
2565 | * to one because the slow code to read in the free extents does check | |
2566 | * the pinned extents. | |
2567 | */ | |
676f1f75 | 2568 | btrfs_cache_block_group(cache, 1); |
9ad6d91f FM |
2569 | /* |
2570 | * Make sure we wait until the cache is completely built in case it is | |
2571 | * missing or is invalid and therefore needs to be rebuilt. | |
2572 | */ | |
2573 | ret = btrfs_wait_block_group_cache_done(cache); | |
2574 | if (ret) | |
2575 | goto out; | |
e688b725 | 2576 | |
6690d071 | 2577 | pin_down_extent(trans, cache, bytenr, num_bytes, 0); |
e688b725 CM |
2578 | |
2579 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2580 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
9ad6d91f | 2581 | out: |
e688b725 | 2582 | btrfs_put_block_group(cache); |
b50c6e25 | 2583 | return ret; |
e688b725 CM |
2584 | } |
2585 | ||
2ff7e61e JM |
2586 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2587 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2588 | { |
2589 | int ret; | |
32da5386 | 2590 | struct btrfs_block_group *block_group; |
8c2a1a30 | 2591 | |
0b246afa | 2592 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2593 | if (!block_group) |
2594 | return -EINVAL; | |
2595 | ||
9ad6d91f FM |
2596 | btrfs_cache_block_group(block_group, 1); |
2597 | /* | |
2598 | * Make sure we wait until the cache is completely built in case it is | |
2599 | * missing or is invalid and therefore needs to be rebuilt. | |
2600 | */ | |
2601 | ret = btrfs_wait_block_group_cache_done(block_group); | |
2602 | if (ret) | |
2603 | goto out; | |
8c2a1a30 | 2604 | |
9ad6d91f FM |
2605 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2606 | out: | |
8c2a1a30 JB |
2607 | btrfs_put_block_group(block_group); |
2608 | return ret; | |
2609 | } | |
2610 | ||
bcdc428c | 2611 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2612 | { |
bcdc428c | 2613 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2614 | struct btrfs_file_extent_item *item; |
2615 | struct btrfs_key key; | |
2616 | int found_type; | |
2617 | int i; | |
b89311ef | 2618 | int ret = 0; |
8c2a1a30 | 2619 | |
2ff7e61e | 2620 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2621 | return 0; |
2622 | ||
2623 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2624 | btrfs_item_key_to_cpu(eb, &key, i); | |
2625 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2626 | continue; | |
2627 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2628 | found_type = btrfs_file_extent_type(eb, item); | |
2629 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2630 | continue; | |
2631 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2632 | continue; | |
2633 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2634 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2635 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2636 | if (ret) | |
2637 | break; | |
8c2a1a30 JB |
2638 | } |
2639 | ||
b89311ef | 2640 | return ret; |
8c2a1a30 JB |
2641 | } |
2642 | ||
9cfa3e34 | 2643 | static void |
32da5386 | 2644 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2645 | { |
2646 | atomic_inc(&bg->reservations); | |
2647 | } | |
2648 | ||
c759c4e1 JB |
2649 | /* |
2650 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2651 | * what it should be based on the mount options. | |
2652 | */ | |
2653 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2654 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2655 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2656 | { |
2657 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2658 | |
2659 | *empty_cluster = 0; | |
2660 | if (btrfs_mixed_space_info(space_info)) | |
2661 | return ret; | |
2662 | ||
c759c4e1 | 2663 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2664 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2665 | if (btrfs_test_opt(fs_info, SSD)) |
2666 | *empty_cluster = SZ_2M; | |
2667 | else | |
ee22184b | 2668 | *empty_cluster = SZ_64K; |
583b7231 HK |
2669 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2670 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2671 | *empty_cluster = SZ_2M; | |
0b246afa | 2672 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2673 | } |
2674 | ||
2675 | return ret; | |
2676 | } | |
2677 | ||
2ff7e61e JM |
2678 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2679 | u64 start, u64 end, | |
678886bd | 2680 | const bool return_free_space) |
ccd467d6 | 2681 | { |
32da5386 | 2682 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2683 | struct btrfs_space_info *space_info; |
2684 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2685 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2686 | u64 len; |
c759c4e1 JB |
2687 | u64 total_unpinned = 0; |
2688 | u64 empty_cluster = 0; | |
7b398f8e | 2689 | bool readonly; |
ccd467d6 | 2690 | |
11833d66 | 2691 | while (start <= end) { |
7b398f8e | 2692 | readonly = false; |
11833d66 | 2693 | if (!cache || |
b3470b5d | 2694 | start >= cache->start + cache->length) { |
11833d66 YZ |
2695 | if (cache) |
2696 | btrfs_put_block_group(cache); | |
c759c4e1 | 2697 | total_unpinned = 0; |
11833d66 | 2698 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2699 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2700 | |
2ff7e61e | 2701 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2702 | cache->space_info, |
2703 | &empty_cluster); | |
2704 | empty_cluster <<= 1; | |
11833d66 YZ |
2705 | } |
2706 | ||
b3470b5d | 2707 | len = cache->start + cache->length - start; |
11833d66 YZ |
2708 | len = min(len, end + 1 - start); |
2709 | ||
2ca08c56 | 2710 | down_read(&fs_info->commit_root_sem); |
9076dbd5 JB |
2711 | if (start < cache->last_byte_to_unpin && return_free_space) { |
2712 | u64 add_len = min(len, cache->last_byte_to_unpin - start); | |
2713 | ||
2714 | btrfs_add_free_space(cache, start, add_len); | |
11833d66 | 2715 | } |
2ca08c56 | 2716 | up_read(&fs_info->commit_root_sem); |
11833d66 | 2717 | |
f0486c68 | 2718 | start += len; |
c759c4e1 | 2719 | total_unpinned += len; |
7b398f8e | 2720 | space_info = cache->space_info; |
f0486c68 | 2721 | |
c759c4e1 JB |
2722 | /* |
2723 | * If this space cluster has been marked as fragmented and we've | |
2724 | * unpinned enough in this block group to potentially allow a | |
2725 | * cluster to be created inside of it go ahead and clear the | |
2726 | * fragmented check. | |
2727 | */ | |
2728 | if (cluster && cluster->fragmented && | |
2729 | total_unpinned > empty_cluster) { | |
2730 | spin_lock(&cluster->lock); | |
2731 | cluster->fragmented = 0; | |
2732 | spin_unlock(&cluster->lock); | |
2733 | } | |
2734 | ||
7b398f8e | 2735 | spin_lock(&space_info->lock); |
11833d66 YZ |
2736 | spin_lock(&cache->lock); |
2737 | cache->pinned -= len; | |
bb96c4e5 | 2738 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2739 | space_info->max_extent_size = 0; |
2187374f | 2740 | __btrfs_mod_total_bytes_pinned(space_info, -len); |
7b398f8e JB |
2741 | if (cache->ro) { |
2742 | space_info->bytes_readonly += len; | |
2743 | readonly = true; | |
169e0da9 NA |
2744 | } else if (btrfs_is_zoned(fs_info)) { |
2745 | /* Need reset before reusing in a zoned block group */ | |
2746 | space_info->bytes_zone_unusable += len; | |
2747 | readonly = true; | |
7b398f8e | 2748 | } |
11833d66 | 2749 | spin_unlock(&cache->lock); |
957780eb JB |
2750 | if (!readonly && return_free_space && |
2751 | global_rsv->space_info == space_info) { | |
2752 | u64 to_add = len; | |
92ac58ec | 2753 | |
7b398f8e JB |
2754 | spin_lock(&global_rsv->lock); |
2755 | if (!global_rsv->full) { | |
957780eb JB |
2756 | to_add = min(len, global_rsv->size - |
2757 | global_rsv->reserved); | |
2758 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2759 | btrfs_space_info_update_bytes_may_use(fs_info, |
2760 | space_info, to_add); | |
7b398f8e JB |
2761 | if (global_rsv->reserved >= global_rsv->size) |
2762 | global_rsv->full = 1; | |
957780eb | 2763 | len -= to_add; |
7b398f8e JB |
2764 | } |
2765 | spin_unlock(&global_rsv->lock); | |
2766 | } | |
2732798c JB |
2767 | /* Add to any tickets we may have */ |
2768 | if (!readonly && return_free_space && len) | |
2769 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2770 | spin_unlock(&space_info->lock); |
ccd467d6 | 2771 | } |
11833d66 YZ |
2772 | |
2773 | if (cache) | |
2774 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2775 | return 0; |
2776 | } | |
2777 | ||
5ead2dd0 | 2778 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2779 | { |
5ead2dd0 | 2780 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2781 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2782 | struct list_head *deleted_bgs; |
11833d66 | 2783 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2784 | u64 start; |
2785 | u64 end; | |
a28ec197 | 2786 | int ret; |
a28ec197 | 2787 | |
fe119a6e | 2788 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2789 | |
bf31f87f | 2790 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2791 | struct extent_state *cached_state = NULL; |
2792 | ||
d4b450cd | 2793 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2794 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2795 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2796 | if (ret) { |
2797 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2798 | break; |
d4b450cd | 2799 | } |
1f3c79a2 | 2800 | |
46b27f50 | 2801 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2802 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2803 | end + 1 - start, NULL); |
1f3c79a2 | 2804 | |
0e6ec385 | 2805 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2806 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2807 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2808 | free_extent_state(cached_state); |
b9473439 | 2809 | cond_resched(); |
a28ec197 | 2810 | } |
817d52f8 | 2811 | |
a2309300 DZ |
2812 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2813 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2814 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2815 | } |
b0643e59 | 2816 | |
e33e17ee JM |
2817 | /* |
2818 | * Transaction is finished. We don't need the lock anymore. We | |
2819 | * do need to clean up the block groups in case of a transaction | |
2820 | * abort. | |
2821 | */ | |
2822 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2823 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2824 | u64 trimmed = 0; | |
2825 | ||
2826 | ret = -EROFS; | |
bf31f87f | 2827 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2828 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2829 | block_group->start, |
2830 | block_group->length, | |
e33e17ee JM |
2831 | &trimmed); |
2832 | ||
2833 | list_del_init(&block_group->bg_list); | |
6b7304af | 2834 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2835 | btrfs_put_block_group(block_group); |
2836 | ||
2837 | if (ret) { | |
2838 | const char *errstr = btrfs_decode_error(ret); | |
2839 | btrfs_warn(fs_info, | |
913e1535 | 2840 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2841 | ret, errstr); |
2842 | } | |
2843 | } | |
2844 | ||
e20d96d6 CM |
2845 | return 0; |
2846 | } | |
2847 | ||
1c2a07f5 QW |
2848 | /* |
2849 | * Drop one or more refs of @node. | |
2850 | * | |
2851 | * 1. Locate the extent refs. | |
2852 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
2853 | * Locate it, then reduce the refs number or remove the ref line completely. | |
2854 | * | |
2855 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
2856 | * | |
2857 | * Inline backref case: | |
2858 | * | |
2859 | * in extent tree we have: | |
2860 | * | |
2861 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2862 | * refs 2 gen 6 flags DATA | |
2863 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2864 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
2865 | * | |
2866 | * This function gets called with: | |
2867 | * | |
2868 | * node->bytenr = 13631488 | |
2869 | * node->num_bytes = 1048576 | |
2870 | * root_objectid = FS_TREE | |
2871 | * owner_objectid = 257 | |
2872 | * owner_offset = 0 | |
2873 | * refs_to_drop = 1 | |
2874 | * | |
2875 | * Then we should get some like: | |
2876 | * | |
2877 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2878 | * refs 1 gen 6 flags DATA | |
2879 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2880 | * | |
2881 | * Keyed backref case: | |
2882 | * | |
2883 | * in extent tree we have: | |
2884 | * | |
2885 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2886 | * refs 754 gen 6 flags DATA | |
2887 | * [...] | |
2888 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
2889 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
2890 | * | |
2891 | * This function get called with: | |
2892 | * | |
2893 | * node->bytenr = 13631488 | |
2894 | * node->num_bytes = 1048576 | |
2895 | * root_objectid = FS_TREE | |
2896 | * owner_objectid = 866 | |
2897 | * owner_offset = 0 | |
2898 | * refs_to_drop = 1 | |
2899 | * | |
2900 | * Then we should get some like: | |
2901 | * | |
2902 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2903 | * refs 753 gen 6 flags DATA | |
2904 | * | |
2905 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
2906 | */ | |
5d4f98a2 | 2907 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2908 | struct btrfs_delayed_ref_node *node, u64 parent, |
2909 | u64 root_objectid, u64 owner_objectid, | |
2910 | u64 owner_offset, int refs_to_drop, | |
2911 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2912 | { |
e72cb923 | 2913 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2914 | struct btrfs_key key; |
5d4f98a2 | 2915 | struct btrfs_path *path; |
1261ec42 | 2916 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2917 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2918 | struct btrfs_extent_item *ei; |
2919 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2920 | int ret; |
5d4f98a2 | 2921 | int is_data; |
952fccac CM |
2922 | int extent_slot = 0; |
2923 | int found_extent = 0; | |
2924 | int num_to_del = 1; | |
5d4f98a2 YZ |
2925 | u32 item_size; |
2926 | u64 refs; | |
c682f9b3 QW |
2927 | u64 bytenr = node->bytenr; |
2928 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2929 | int last_ref = 0; |
0b246afa | 2930 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 2931 | |
5caf2a00 | 2932 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2933 | if (!path) |
2934 | return -ENOMEM; | |
5f26f772 | 2935 | |
5d4f98a2 | 2936 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
1c2a07f5 QW |
2937 | |
2938 | if (!is_data && refs_to_drop != 1) { | |
2939 | btrfs_crit(info, | |
2940 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
2941 | node->bytenr, refs_to_drop); | |
2942 | ret = -EINVAL; | |
2943 | btrfs_abort_transaction(trans, ret); | |
2944 | goto out; | |
2945 | } | |
5d4f98a2 | 2946 | |
3173a18f | 2947 | if (is_data) |
897ca819 | 2948 | skinny_metadata = false; |
3173a18f | 2949 | |
fbe4801b NB |
2950 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
2951 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 2952 | owner_offset); |
7bb86316 | 2953 | if (ret == 0) { |
1c2a07f5 QW |
2954 | /* |
2955 | * Either the inline backref or the SHARED_DATA_REF/ | |
2956 | * SHARED_BLOCK_REF is found | |
2957 | * | |
2958 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
2959 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
2960 | */ | |
952fccac | 2961 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
2962 | while (extent_slot >= 0) { |
2963 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 2964 | extent_slot); |
5d4f98a2 | 2965 | if (key.objectid != bytenr) |
952fccac | 2966 | break; |
5d4f98a2 YZ |
2967 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
2968 | key.offset == num_bytes) { | |
952fccac CM |
2969 | found_extent = 1; |
2970 | break; | |
2971 | } | |
3173a18f JB |
2972 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
2973 | key.offset == owner_objectid) { | |
2974 | found_extent = 1; | |
2975 | break; | |
2976 | } | |
1c2a07f5 QW |
2977 | |
2978 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
2979 | if (path->slots[0] - extent_slot > 5) |
2980 | break; | |
5d4f98a2 | 2981 | extent_slot--; |
952fccac | 2982 | } |
a79865c6 | 2983 | |
31840ae1 | 2984 | if (!found_extent) { |
1c2a07f5 QW |
2985 | if (iref) { |
2986 | btrfs_crit(info, | |
2987 | "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref"); | |
2988 | btrfs_abort_transaction(trans, -EUCLEAN); | |
2989 | goto err_dump; | |
2990 | } | |
2991 | /* Must be SHARED_* item, remove the backref first */ | |
87cc7a8a | 2992 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 2993 | refs_to_drop, |
fcebe456 | 2994 | is_data, &last_ref); |
005d6427 | 2995 | if (ret) { |
66642832 | 2996 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
2997 | goto out; |
2998 | } | |
b3b4aa74 | 2999 | btrfs_release_path(path); |
5d4f98a2 | 3000 | |
1c2a07f5 | 3001 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3002 | key.objectid = bytenr; |
3003 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3004 | key.offset = num_bytes; | |
3005 | ||
3173a18f JB |
3006 | if (!is_data && skinny_metadata) { |
3007 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3008 | key.offset = owner_objectid; | |
3009 | } | |
3010 | ||
31840ae1 ZY |
3011 | ret = btrfs_search_slot(trans, extent_root, |
3012 | &key, path, -1, 1); | |
3173a18f JB |
3013 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3014 | /* | |
3015 | * Couldn't find our skinny metadata item, | |
3016 | * see if we have ye olde extent item. | |
3017 | */ | |
3018 | path->slots[0]--; | |
3019 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3020 | path->slots[0]); | |
3021 | if (key.objectid == bytenr && | |
3022 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3023 | key.offset == num_bytes) | |
3024 | ret = 0; | |
3025 | } | |
3026 | ||
3027 | if (ret > 0 && skinny_metadata) { | |
3028 | skinny_metadata = false; | |
9ce49a0b | 3029 | key.objectid = bytenr; |
3173a18f JB |
3030 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3031 | key.offset = num_bytes; | |
3032 | btrfs_release_path(path); | |
3033 | ret = btrfs_search_slot(trans, extent_root, | |
3034 | &key, path, -1, 1); | |
3035 | } | |
3036 | ||
f3465ca4 | 3037 | if (ret) { |
5d163e0e JM |
3038 | btrfs_err(info, |
3039 | "umm, got %d back from search, was looking for %llu", | |
3040 | ret, bytenr); | |
b783e62d | 3041 | if (ret > 0) |
a4f78750 | 3042 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3043 | } |
005d6427 | 3044 | if (ret < 0) { |
66642832 | 3045 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3046 | goto out; |
3047 | } | |
31840ae1 ZY |
3048 | extent_slot = path->slots[0]; |
3049 | } | |
fae7f21c | 3050 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3051 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3052 | btrfs_err(info, |
3053 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3054 | bytenr, parent, root_objectid, owner_objectid, |
3055 | owner_offset); | |
66642832 | 3056 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3057 | goto out; |
79787eaa | 3058 | } else { |
66642832 | 3059 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3060 | goto out; |
7bb86316 | 3061 | } |
5f39d397 CM |
3062 | |
3063 | leaf = path->nodes[0]; | |
5d4f98a2 | 3064 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3065 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3066 | ret = -EINVAL; |
3067 | btrfs_print_v0_err(info); | |
3068 | btrfs_abort_transaction(trans, ret); | |
3069 | goto out; | |
3070 | } | |
952fccac | 3071 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3072 | struct btrfs_extent_item); |
3173a18f JB |
3073 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3074 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3075 | struct btrfs_tree_block_info *bi; |
1c2a07f5 QW |
3076 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
3077 | btrfs_crit(info, | |
cad69d13 | 3078 | "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu", |
1c2a07f5 QW |
3079 | key.objectid, key.type, key.offset, |
3080 | owner_objectid, item_size, | |
3081 | sizeof(*ei) + sizeof(*bi)); | |
3082 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3083 | goto err_dump; | |
3084 | } | |
5d4f98a2 YZ |
3085 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3086 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3087 | } | |
56bec294 | 3088 | |
5d4f98a2 | 3089 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3090 | if (refs < refs_to_drop) { |
1c2a07f5 QW |
3091 | btrfs_crit(info, |
3092 | "trying to drop %d refs but we only have %llu for bytenr %llu", | |
5d163e0e | 3093 | refs_to_drop, refs, bytenr); |
1c2a07f5 QW |
3094 | btrfs_abort_transaction(trans, -EUCLEAN); |
3095 | goto err_dump; | |
32b02538 | 3096 | } |
56bec294 | 3097 | refs -= refs_to_drop; |
5f39d397 | 3098 | |
5d4f98a2 YZ |
3099 | if (refs > 0) { |
3100 | if (extent_op) | |
3101 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3102 | /* | |
3103 | * In the case of inline back ref, reference count will | |
3104 | * be updated by remove_extent_backref | |
952fccac | 3105 | */ |
5d4f98a2 | 3106 | if (iref) { |
1c2a07f5 QW |
3107 | if (!found_extent) { |
3108 | btrfs_crit(info, | |
3109 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found"); | |
3110 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3111 | goto err_dump; | |
3112 | } | |
5d4f98a2 YZ |
3113 | } else { |
3114 | btrfs_set_extent_refs(leaf, ei, refs); | |
3115 | btrfs_mark_buffer_dirty(leaf); | |
3116 | } | |
3117 | if (found_extent) { | |
87cc7a8a NB |
3118 | ret = remove_extent_backref(trans, path, iref, |
3119 | refs_to_drop, is_data, | |
3120 | &last_ref); | |
005d6427 | 3121 | if (ret) { |
66642832 | 3122 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3123 | goto out; |
3124 | } | |
952fccac | 3125 | } |
5d4f98a2 | 3126 | } else { |
1c2a07f5 | 3127 | /* In this branch refs == 1 */ |
5d4f98a2 | 3128 | if (found_extent) { |
1c2a07f5 QW |
3129 | if (is_data && refs_to_drop != |
3130 | extent_data_ref_count(path, iref)) { | |
3131 | btrfs_crit(info, | |
3132 | "invalid refs_to_drop, current refs %u refs_to_drop %u", | |
3133 | extent_data_ref_count(path, iref), | |
3134 | refs_to_drop); | |
3135 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3136 | goto err_dump; | |
3137 | } | |
5d4f98a2 | 3138 | if (iref) { |
1c2a07f5 QW |
3139 | if (path->slots[0] != extent_slot) { |
3140 | btrfs_crit(info, | |
3141 | "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref", | |
3142 | key.objectid, key.type, | |
3143 | key.offset); | |
3144 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3145 | goto err_dump; | |
3146 | } | |
5d4f98a2 | 3147 | } else { |
1c2a07f5 QW |
3148 | /* |
3149 | * No inline ref, we must be at SHARED_* item, | |
3150 | * And it's single ref, it must be: | |
3151 | * | extent_slot ||extent_slot + 1| | |
3152 | * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ] | |
3153 | */ | |
3154 | if (path->slots[0] != extent_slot + 1) { | |
3155 | btrfs_crit(info, | |
3156 | "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM"); | |
3157 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3158 | goto err_dump; | |
3159 | } | |
5d4f98a2 YZ |
3160 | path->slots[0] = extent_slot; |
3161 | num_to_del = 2; | |
3162 | } | |
78fae27e | 3163 | } |
b9473439 | 3164 | |
fcebe456 | 3165 | last_ref = 1; |
952fccac CM |
3166 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3167 | num_to_del); | |
005d6427 | 3168 | if (ret) { |
66642832 | 3169 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3170 | goto out; |
3171 | } | |
b3b4aa74 | 3172 | btrfs_release_path(path); |
21af804c | 3173 | |
5d4f98a2 | 3174 | if (is_data) { |
40e046ac FM |
3175 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3176 | num_bytes); | |
005d6427 | 3177 | if (ret) { |
66642832 | 3178 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3179 | goto out; |
3180 | } | |
459931ec CM |
3181 | } |
3182 | ||
e7355e50 | 3183 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3184 | if (ret) { |
66642832 | 3185 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3186 | goto out; |
3187 | } | |
3188 | ||
ade4b516 | 3189 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3190 | if (ret) { |
66642832 | 3191 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3192 | goto out; |
3193 | } | |
a28ec197 | 3194 | } |
fcebe456 JB |
3195 | btrfs_release_path(path); |
3196 | ||
79787eaa | 3197 | out: |
5caf2a00 | 3198 | btrfs_free_path(path); |
a28ec197 | 3199 | return ret; |
1c2a07f5 QW |
3200 | err_dump: |
3201 | /* | |
3202 | * Leaf dump can take up a lot of log buffer, so we only do full leaf | |
3203 | * dump for debug build. | |
3204 | */ | |
3205 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
3206 | btrfs_crit(info, "path->slots[0]=%d extent_slot=%d", | |
3207 | path->slots[0], extent_slot); | |
3208 | btrfs_print_leaf(path->nodes[0]); | |
3209 | } | |
3210 | ||
3211 | btrfs_free_path(path); | |
3212 | return -EUCLEAN; | |
a28ec197 CM |
3213 | } |
3214 | ||
1887be66 | 3215 | /* |
f0486c68 | 3216 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3217 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3218 | * a given extent, and if there are no other delayed refs to be processed, it | |
3219 | * removes it from the tree. | |
3220 | */ | |
3221 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3222 | u64 bytenr) |
1887be66 CM |
3223 | { |
3224 | struct btrfs_delayed_ref_head *head; | |
3225 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3226 | int ret = 0; |
1887be66 CM |
3227 | |
3228 | delayed_refs = &trans->transaction->delayed_refs; | |
3229 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3230 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3231 | if (!head) |
cf93da7b | 3232 | goto out_delayed_unlock; |
1887be66 | 3233 | |
d7df2c79 | 3234 | spin_lock(&head->lock); |
e3d03965 | 3235 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3236 | goto out; |
3237 | ||
bedc6617 JB |
3238 | if (cleanup_extent_op(head) != NULL) |
3239 | goto out; | |
5d4f98a2 | 3240 | |
1887be66 CM |
3241 | /* |
3242 | * waiting for the lock here would deadlock. If someone else has it | |
3243 | * locked they are already in the process of dropping it anyway | |
3244 | */ | |
3245 | if (!mutex_trylock(&head->mutex)) | |
3246 | goto out; | |
3247 | ||
d7baffda | 3248 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3249 | head->processing = 0; |
d7baffda | 3250 | |
d7df2c79 | 3251 | spin_unlock(&head->lock); |
1887be66 CM |
3252 | spin_unlock(&delayed_refs->lock); |
3253 | ||
f0486c68 YZ |
3254 | BUG_ON(head->extent_op); |
3255 | if (head->must_insert_reserved) | |
3256 | ret = 1; | |
3257 | ||
31890da0 | 3258 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3259 | mutex_unlock(&head->mutex); |
d278850e | 3260 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3261 | return ret; |
1887be66 | 3262 | out: |
d7df2c79 | 3263 | spin_unlock(&head->lock); |
cf93da7b CM |
3264 | |
3265 | out_delayed_unlock: | |
1887be66 CM |
3266 | spin_unlock(&delayed_refs->lock); |
3267 | return 0; | |
3268 | } | |
3269 | ||
f0486c68 YZ |
3270 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3271 | struct btrfs_root *root, | |
3272 | struct extent_buffer *buf, | |
5581a51a | 3273 | u64 parent, int last_ref) |
f0486c68 | 3274 | { |
0b246afa | 3275 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3276 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
3277 | int ret; |
3278 | ||
ed4f255b QW |
3279 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3280 | buf->start, buf->len, parent); | |
3281 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3282 | root->root_key.objectid); | |
3283 | ||
f0486c68 | 3284 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
8a5040f7 | 3285 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 3286 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL); |
79787eaa | 3287 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
3288 | } |
3289 | ||
0a16c7d7 | 3290 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3291 | struct btrfs_block_group *cache; |
6219872d | 3292 | |
f0486c68 | 3293 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3294 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 3295 | if (!ret) |
37be25bc | 3296 | goto out; |
f0486c68 YZ |
3297 | } |
3298 | ||
0b246afa | 3299 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3300 | |
f0486c68 | 3301 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3302 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3303 | btrfs_put_block_group(cache); |
37be25bc | 3304 | goto out; |
f0486c68 YZ |
3305 | } |
3306 | ||
3307 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3308 | ||
3309 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3310 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3311 | btrfs_put_block_group(cache); |
71ff6437 | 3312 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3313 | } |
3314 | out: | |
0a16c7d7 OS |
3315 | if (last_ref) { |
3316 | /* | |
3317 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3318 | * matter anymore. | |
3319 | */ | |
3320 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3321 | } | |
f0486c68 YZ |
3322 | } |
3323 | ||
79787eaa | 3324 | /* Can return -ENOMEM */ |
ffd4bb2a | 3325 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3326 | { |
ffd4bb2a | 3327 | struct btrfs_fs_info *fs_info = trans->fs_info; |
925baedd CM |
3328 | int ret; |
3329 | ||
f5ee5c9a | 3330 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3331 | return 0; |
fccb84c9 | 3332 | |
56bec294 CM |
3333 | /* |
3334 | * tree log blocks never actually go into the extent allocation | |
3335 | * tree, just update pinning info and exit early. | |
56bec294 | 3336 | */ |
ffd4bb2a QW |
3337 | if ((ref->type == BTRFS_REF_METADATA && |
3338 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3339 | (ref->type == BTRFS_REF_DATA && | |
3340 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3341 | /* unlocks the pinned mutex */ |
b25c36f8 | 3342 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
56bec294 | 3343 | ret = 0; |
ffd4bb2a | 3344 | } else if (ref->type == BTRFS_REF_METADATA) { |
2187374f | 3345 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL); |
5d4f98a2 | 3346 | } else { |
2187374f | 3347 | ret = btrfs_add_delayed_data_ref(trans, ref, 0); |
56bec294 | 3348 | } |
d7eae340 | 3349 | |
ffd4bb2a QW |
3350 | if (!((ref->type == BTRFS_REF_METADATA && |
3351 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3352 | (ref->type == BTRFS_REF_DATA && | |
3353 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3354 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3355 | |
925baedd CM |
3356 | return ret; |
3357 | } | |
3358 | ||
817d52f8 | 3359 | enum btrfs_loop_type { |
f262fa8d DS |
3360 | LOOP_CACHING_NOWAIT, |
3361 | LOOP_CACHING_WAIT, | |
3362 | LOOP_ALLOC_CHUNK, | |
3363 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3364 | }; |
3365 | ||
e570fd27 | 3366 | static inline void |
32da5386 | 3367 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3368 | int delalloc) |
3369 | { | |
3370 | if (delalloc) | |
3371 | down_read(&cache->data_rwsem); | |
3372 | } | |
3373 | ||
32da5386 | 3374 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3375 | int delalloc) |
3376 | { | |
3377 | btrfs_get_block_group(cache); | |
3378 | if (delalloc) | |
3379 | down_read(&cache->data_rwsem); | |
3380 | } | |
3381 | ||
32da5386 DS |
3382 | static struct btrfs_block_group *btrfs_lock_cluster( |
3383 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3384 | struct btrfs_free_cluster *cluster, |
3385 | int delalloc) | |
c142c6a4 | 3386 | __acquires(&cluster->refill_lock) |
e570fd27 | 3387 | { |
32da5386 | 3388 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3389 | |
e570fd27 | 3390 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3391 | while (1) { |
3392 | used_bg = cluster->block_group; | |
3393 | if (!used_bg) | |
3394 | return NULL; | |
3395 | ||
3396 | if (used_bg == block_group) | |
e570fd27 MX |
3397 | return used_bg; |
3398 | ||
6719afdc | 3399 | btrfs_get_block_group(used_bg); |
e570fd27 | 3400 | |
6719afdc GU |
3401 | if (!delalloc) |
3402 | return used_bg; | |
e570fd27 | 3403 | |
6719afdc GU |
3404 | if (down_read_trylock(&used_bg->data_rwsem)) |
3405 | return used_bg; | |
e570fd27 | 3406 | |
6719afdc | 3407 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3408 | |
e321f8a8 LB |
3409 | /* We should only have one-level nested. */ |
3410 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3411 | |
6719afdc GU |
3412 | spin_lock(&cluster->refill_lock); |
3413 | if (used_bg == cluster->block_group) | |
3414 | return used_bg; | |
e570fd27 | 3415 | |
6719afdc GU |
3416 | up_read(&used_bg->data_rwsem); |
3417 | btrfs_put_block_group(used_bg); | |
3418 | } | |
e570fd27 MX |
3419 | } |
3420 | ||
3421 | static inline void | |
32da5386 | 3422 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3423 | int delalloc) |
3424 | { | |
3425 | if (delalloc) | |
3426 | up_read(&cache->data_rwsem); | |
3427 | btrfs_put_block_group(cache); | |
3428 | } | |
3429 | ||
cb2f96f8 NA |
3430 | enum btrfs_extent_allocation_policy { |
3431 | BTRFS_EXTENT_ALLOC_CLUSTERED, | |
2eda5708 | 3432 | BTRFS_EXTENT_ALLOC_ZONED, |
cb2f96f8 NA |
3433 | }; |
3434 | ||
b4bd745d QW |
3435 | /* |
3436 | * Structure used internally for find_free_extent() function. Wraps needed | |
3437 | * parameters. | |
3438 | */ | |
3439 | struct find_free_extent_ctl { | |
3440 | /* Basic allocation info */ | |
b4bd745d QW |
3441 | u64 num_bytes; |
3442 | u64 empty_size; | |
3443 | u64 flags; | |
3444 | int delalloc; | |
3445 | ||
3446 | /* Where to start the search inside the bg */ | |
3447 | u64 search_start; | |
3448 | ||
3449 | /* For clustered allocation */ | |
3450 | u64 empty_cluster; | |
c10859be NA |
3451 | struct btrfs_free_cluster *last_ptr; |
3452 | bool use_cluster; | |
b4bd745d QW |
3453 | |
3454 | bool have_caching_bg; | |
3455 | bool orig_have_caching_bg; | |
3456 | ||
3457 | /* RAID index, converted from flags */ | |
3458 | int index; | |
3459 | ||
e72d79d6 QW |
3460 | /* |
3461 | * Current loop number, check find_free_extent_update_loop() for details | |
3462 | */ | |
b4bd745d QW |
3463 | int loop; |
3464 | ||
3465 | /* | |
3466 | * Whether we're refilling a cluster, if true we need to re-search | |
3467 | * current block group but don't try to refill the cluster again. | |
3468 | */ | |
3469 | bool retry_clustered; | |
3470 | ||
3471 | /* | |
3472 | * Whether we're updating free space cache, if true we need to re-search | |
3473 | * current block group but don't try updating free space cache again. | |
3474 | */ | |
3475 | bool retry_unclustered; | |
3476 | ||
3477 | /* If current block group is cached */ | |
3478 | int cached; | |
3479 | ||
3480 | /* Max contiguous hole found */ | |
3481 | u64 max_extent_size; | |
3482 | ||
3483 | /* Total free space from free space cache, not always contiguous */ | |
3484 | u64 total_free_space; | |
3485 | ||
3486 | /* Found result */ | |
3487 | u64 found_offset; | |
cb2f96f8 | 3488 | |
ea544149 NA |
3489 | /* Hint where to start looking for an empty space */ |
3490 | u64 hint_byte; | |
3491 | ||
cb2f96f8 NA |
3492 | /* Allocation policy */ |
3493 | enum btrfs_extent_allocation_policy policy; | |
b4bd745d QW |
3494 | }; |
3495 | ||
d06e3bb6 QW |
3496 | |
3497 | /* | |
3498 | * Helper function for find_free_extent(). | |
3499 | * | |
3500 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3501 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3502 | * Return >0 to inform caller that we find nothing | |
3503 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3504 | */ | |
32da5386 | 3505 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3506 | struct find_free_extent_ctl *ffe_ctl, |
3507 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3508 | { |
32da5386 | 3509 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3510 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3511 | u64 aligned_cluster; |
3512 | u64 offset; | |
3513 | int ret; | |
3514 | ||
3515 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3516 | if (!cluster_bg) | |
3517 | goto refill_cluster; | |
3518 | if (cluster_bg != bg && (cluster_bg->ro || | |
3519 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3520 | goto release_cluster; | |
3521 | ||
3522 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3523 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3524 | &ffe_ctl->max_extent_size); |
3525 | if (offset) { | |
3526 | /* We have a block, we're done */ | |
3527 | spin_unlock(&last_ptr->refill_lock); | |
3528 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3529 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3530 | *cluster_bg_ret = cluster_bg; | |
3531 | ffe_ctl->found_offset = offset; | |
3532 | return 0; | |
3533 | } | |
3534 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3535 | ||
3536 | release_cluster: | |
3537 | /* | |
3538 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3539 | * lets just skip it and let the allocator find whatever block it can | |
3540 | * find. If we reach this point, we will have tried the cluster | |
3541 | * allocator plenty of times and not have found anything, so we are | |
3542 | * likely way too fragmented for the clustering stuff to find anything. | |
3543 | * | |
3544 | * However, if the cluster is taken from the current block group, | |
3545 | * release the cluster first, so that we stand a better chance of | |
3546 | * succeeding in the unclustered allocation. | |
3547 | */ | |
3548 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3549 | spin_unlock(&last_ptr->refill_lock); | |
3550 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3551 | return -ENOENT; | |
3552 | } | |
3553 | ||
3554 | /* This cluster didn't work out, free it and start over */ | |
3555 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3556 | ||
3557 | if (cluster_bg != bg) | |
3558 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3559 | ||
3560 | refill_cluster: | |
3561 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3562 | spin_unlock(&last_ptr->refill_lock); | |
3563 | return -ENOENT; | |
3564 | } | |
3565 | ||
3566 | aligned_cluster = max_t(u64, | |
3567 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3568 | bg->full_stripe_len); | |
2ceeae2e DS |
3569 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3570 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3571 | if (ret == 0) { |
3572 | /* Now pull our allocation out of this cluster */ | |
3573 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3574 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3575 | &ffe_ctl->max_extent_size); | |
3576 | if (offset) { | |
3577 | /* We found one, proceed */ | |
3578 | spin_unlock(&last_ptr->refill_lock); | |
3579 | trace_btrfs_reserve_extent_cluster(bg, | |
3580 | ffe_ctl->search_start, | |
3581 | ffe_ctl->num_bytes); | |
3582 | ffe_ctl->found_offset = offset; | |
3583 | return 0; | |
3584 | } | |
3585 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3586 | !ffe_ctl->retry_clustered) { | |
3587 | spin_unlock(&last_ptr->refill_lock); | |
3588 | ||
3589 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3590 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3591 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3592 | return -EAGAIN; | |
3593 | } | |
3594 | /* | |
3595 | * At this point we either didn't find a cluster or we weren't able to | |
3596 | * allocate a block from our cluster. Free the cluster we've been | |
3597 | * trying to use, and go to the next block group. | |
3598 | */ | |
3599 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3600 | spin_unlock(&last_ptr->refill_lock); | |
3601 | return 1; | |
3602 | } | |
3603 | ||
e1a41848 QW |
3604 | /* |
3605 | * Return >0 to inform caller that we find nothing | |
3606 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3607 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3608 | */ | |
32da5386 | 3609 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3610 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3611 | { |
897cae79 | 3612 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3613 | u64 offset; |
3614 | ||
3615 | /* | |
3616 | * We are doing an unclustered allocation, set the fragmented flag so | |
3617 | * we don't bother trying to setup a cluster again until we get more | |
3618 | * space. | |
3619 | */ | |
3620 | if (unlikely(last_ptr)) { | |
3621 | spin_lock(&last_ptr->lock); | |
3622 | last_ptr->fragmented = 1; | |
3623 | spin_unlock(&last_ptr->lock); | |
3624 | } | |
3625 | if (ffe_ctl->cached) { | |
3626 | struct btrfs_free_space_ctl *free_space_ctl; | |
3627 | ||
3628 | free_space_ctl = bg->free_space_ctl; | |
3629 | spin_lock(&free_space_ctl->tree_lock); | |
3630 | if (free_space_ctl->free_space < | |
3631 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3632 | ffe_ctl->empty_size) { | |
3633 | ffe_ctl->total_free_space = max_t(u64, | |
3634 | ffe_ctl->total_free_space, | |
3635 | free_space_ctl->free_space); | |
3636 | spin_unlock(&free_space_ctl->tree_lock); | |
3637 | return 1; | |
3638 | } | |
3639 | spin_unlock(&free_space_ctl->tree_lock); | |
3640 | } | |
3641 | ||
3642 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3643 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3644 | &ffe_ctl->max_extent_size); | |
3645 | ||
3646 | /* | |
3647 | * If we didn't find a chunk, and we haven't failed on this block group | |
3648 | * before, and this block group is in the middle of caching and we are | |
3649 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3650 | * set @retry_unclustered to true. | |
3651 | * | |
3652 | * If @retry_unclustered is true then we've already waited on this | |
3653 | * block group once and should move on to the next block group. | |
3654 | */ | |
3655 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3656 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3657 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3658 | ffe_ctl->empty_size); | |
e1a41848 QW |
3659 | ffe_ctl->retry_unclustered = true; |
3660 | return -EAGAIN; | |
3661 | } else if (!offset) { | |
3662 | return 1; | |
3663 | } | |
3664 | ffe_ctl->found_offset = offset; | |
3665 | return 0; | |
3666 | } | |
3667 | ||
c668690d NA |
3668 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3669 | struct find_free_extent_ctl *ffe_ctl, | |
3670 | struct btrfs_block_group **bg_ret) | |
3671 | { | |
3672 | int ret; | |
3673 | ||
3674 | /* We want to try and use the cluster allocator, so lets look there */ | |
3675 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3676 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
c668690d NA |
3677 | if (ret >= 0 || ret == -EAGAIN) |
3678 | return ret; | |
3679 | /* ret == -ENOENT case falls through */ | |
3680 | } | |
3681 | ||
897cae79 | 3682 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3683 | } |
3684 | ||
2eda5708 NA |
3685 | /* |
3686 | * Simple allocator for sequential-only block group. It only allows sequential | |
3687 | * allocation. No need to play with trees. This function also reserves the | |
3688 | * bytes as in btrfs_add_reserved_bytes. | |
3689 | */ | |
3690 | static int do_allocation_zoned(struct btrfs_block_group *block_group, | |
3691 | struct find_free_extent_ctl *ffe_ctl, | |
3692 | struct btrfs_block_group **bg_ret) | |
3693 | { | |
3694 | struct btrfs_space_info *space_info = block_group->space_info; | |
3695 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3696 | u64 start = block_group->start; | |
3697 | u64 num_bytes = ffe_ctl->num_bytes; | |
3698 | u64 avail; | |
3699 | int ret = 0; | |
3700 | ||
3701 | ASSERT(btrfs_is_zoned(block_group->fs_info)); | |
3702 | ||
3703 | spin_lock(&space_info->lock); | |
3704 | spin_lock(&block_group->lock); | |
3705 | ||
3706 | if (block_group->ro) { | |
3707 | ret = 1; | |
3708 | goto out; | |
3709 | } | |
3710 | ||
3711 | avail = block_group->length - block_group->alloc_offset; | |
3712 | if (avail < num_bytes) { | |
3713 | if (ffe_ctl->max_extent_size < avail) { | |
3714 | /* | |
3715 | * With sequential allocator, free space is always | |
3716 | * contiguous | |
3717 | */ | |
3718 | ffe_ctl->max_extent_size = avail; | |
3719 | ffe_ctl->total_free_space = avail; | |
3720 | } | |
3721 | ret = 1; | |
3722 | goto out; | |
3723 | } | |
3724 | ||
3725 | ffe_ctl->found_offset = start + block_group->alloc_offset; | |
3726 | block_group->alloc_offset += num_bytes; | |
3727 | spin_lock(&ctl->tree_lock); | |
3728 | ctl->free_space -= num_bytes; | |
3729 | spin_unlock(&ctl->tree_lock); | |
3730 | ||
3731 | /* | |
3732 | * We do not check if found_offset is aligned to stripesize. The | |
3733 | * address is anyway rewritten when using zone append writing. | |
3734 | */ | |
3735 | ||
3736 | ffe_ctl->search_start = ffe_ctl->found_offset; | |
3737 | ||
3738 | out: | |
3739 | spin_unlock(&block_group->lock); | |
3740 | spin_unlock(&space_info->lock); | |
3741 | return ret; | |
3742 | } | |
3743 | ||
c668690d NA |
3744 | static int do_allocation(struct btrfs_block_group *block_group, |
3745 | struct find_free_extent_ctl *ffe_ctl, | |
3746 | struct btrfs_block_group **bg_ret) | |
3747 | { | |
3748 | switch (ffe_ctl->policy) { | |
3749 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3750 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
2eda5708 NA |
3751 | case BTRFS_EXTENT_ALLOC_ZONED: |
3752 | return do_allocation_zoned(block_group, ffe_ctl, bg_ret); | |
c668690d NA |
3753 | default: |
3754 | BUG(); | |
3755 | } | |
3756 | } | |
3757 | ||
baba5062 NA |
3758 | static void release_block_group(struct btrfs_block_group *block_group, |
3759 | struct find_free_extent_ctl *ffe_ctl, | |
3760 | int delalloc) | |
3761 | { | |
3762 | switch (ffe_ctl->policy) { | |
3763 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3764 | ffe_ctl->retry_clustered = false; | |
3765 | ffe_ctl->retry_unclustered = false; | |
3766 | break; | |
2eda5708 NA |
3767 | case BTRFS_EXTENT_ALLOC_ZONED: |
3768 | /* Nothing to do */ | |
3769 | break; | |
baba5062 NA |
3770 | default: |
3771 | BUG(); | |
3772 | } | |
3773 | ||
3774 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
3775 | ffe_ctl->index); | |
3776 | btrfs_release_block_group(block_group, delalloc); | |
3777 | } | |
3778 | ||
0ab9724b NA |
3779 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
3780 | struct btrfs_key *ins) | |
3781 | { | |
3782 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3783 | ||
3784 | if (!ffe_ctl->use_cluster && last_ptr) { | |
3785 | spin_lock(&last_ptr->lock); | |
3786 | last_ptr->window_start = ins->objectid; | |
3787 | spin_unlock(&last_ptr->lock); | |
3788 | } | |
3789 | } | |
3790 | ||
3791 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
3792 | struct btrfs_key *ins) | |
3793 | { | |
3794 | switch (ffe_ctl->policy) { | |
3795 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3796 | found_extent_clustered(ffe_ctl, ins); | |
3797 | break; | |
2eda5708 NA |
3798 | case BTRFS_EXTENT_ALLOC_ZONED: |
3799 | /* Nothing to do */ | |
3800 | break; | |
0ab9724b NA |
3801 | default: |
3802 | BUG(); | |
3803 | } | |
3804 | } | |
3805 | ||
c70e2139 NA |
3806 | static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl) |
3807 | { | |
3808 | switch (ffe_ctl->policy) { | |
3809 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3810 | /* | |
3811 | * If we can't allocate a new chunk we've already looped through | |
3812 | * at least once, move on to the NO_EMPTY_SIZE case. | |
3813 | */ | |
3814 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3815 | return 0; | |
2eda5708 NA |
3816 | case BTRFS_EXTENT_ALLOC_ZONED: |
3817 | /* Give up here */ | |
3818 | return -ENOSPC; | |
c70e2139 NA |
3819 | default: |
3820 | BUG(); | |
3821 | } | |
3822 | } | |
3823 | ||
e72d79d6 QW |
3824 | /* |
3825 | * Return >0 means caller needs to re-search for free extent | |
3826 | * Return 0 means we have the needed free extent. | |
3827 | * Return <0 means we failed to locate any free extent. | |
3828 | */ | |
3829 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
3830 | struct btrfs_key *ins, |
3831 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 3832 | bool full_search) |
e72d79d6 QW |
3833 | { |
3834 | struct btrfs_root *root = fs_info->extent_root; | |
3835 | int ret; | |
3836 | ||
3837 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3838 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3839 | ffe_ctl->orig_have_caching_bg = true; | |
3840 | ||
3841 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3842 | ffe_ctl->have_caching_bg) | |
3843 | return 1; | |
3844 | ||
3845 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3846 | return 1; | |
3847 | ||
3848 | if (ins->objectid) { | |
0ab9724b | 3849 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
3850 | return 0; |
3851 | } | |
3852 | ||
3853 | /* | |
3854 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3855 | * caching kthreads as we move along | |
3856 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3857 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3858 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3859 | * again | |
3860 | */ | |
3861 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3862 | ffe_ctl->index = 0; | |
3863 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3864 | /* | |
3865 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3866 | * don't have any uncached bgs and we've already done a | |
3867 | * full search through. | |
3868 | */ | |
3869 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3870 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3871 | else | |
3872 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3873 | } else { | |
3874 | ffe_ctl->loop++; | |
3875 | } | |
3876 | ||
3877 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3878 | struct btrfs_trans_handle *trans; | |
3879 | int exist = 0; | |
3880 | ||
3881 | trans = current->journal_info; | |
3882 | if (trans) | |
3883 | exist = 1; | |
3884 | else | |
3885 | trans = btrfs_join_transaction(root); | |
3886 | ||
3887 | if (IS_ERR(trans)) { | |
3888 | ret = PTR_ERR(trans); | |
3889 | return ret; | |
3890 | } | |
3891 | ||
fc471cb0 JB |
3892 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3893 | CHUNK_ALLOC_FORCE); | |
e72d79d6 | 3894 | |
e72d79d6 | 3895 | /* Do not bail out on ENOSPC since we can do more. */ |
c70e2139 NA |
3896 | if (ret == -ENOSPC) |
3897 | ret = chunk_allocation_failed(ffe_ctl); | |
3898 | else if (ret < 0) | |
e72d79d6 QW |
3899 | btrfs_abort_transaction(trans, ret); |
3900 | else | |
3901 | ret = 0; | |
3902 | if (!exist) | |
3903 | btrfs_end_transaction(trans); | |
3904 | if (ret) | |
3905 | return ret; | |
3906 | } | |
3907 | ||
3908 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
3909 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
3910 | return -ENOSPC; | |
3911 | ||
e72d79d6 QW |
3912 | /* |
3913 | * Don't loop again if we already have no empty_size and | |
3914 | * no empty_cluster. | |
3915 | */ | |
3916 | if (ffe_ctl->empty_size == 0 && | |
3917 | ffe_ctl->empty_cluster == 0) | |
3918 | return -ENOSPC; | |
3919 | ffe_ctl->empty_size = 0; | |
3920 | ffe_ctl->empty_cluster = 0; | |
3921 | } | |
3922 | return 1; | |
3923 | } | |
3924 | return -ENOSPC; | |
3925 | } | |
3926 | ||
7e895409 NA |
3927 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
3928 | struct find_free_extent_ctl *ffe_ctl, | |
3929 | struct btrfs_space_info *space_info, | |
3930 | struct btrfs_key *ins) | |
3931 | { | |
3932 | /* | |
3933 | * If our free space is heavily fragmented we may not be able to make | |
3934 | * big contiguous allocations, so instead of doing the expensive search | |
3935 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3936 | * can go ahead and search for a more manageable chunk. | |
3937 | * | |
3938 | * If our max_extent_size is large enough for our allocation simply | |
3939 | * disable clustering since we will likely not be able to find enough | |
3940 | * space to create a cluster and induce latency trying. | |
3941 | */ | |
3942 | if (space_info->max_extent_size) { | |
3943 | spin_lock(&space_info->lock); | |
3944 | if (space_info->max_extent_size && | |
3945 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
3946 | ins->offset = space_info->max_extent_size; | |
3947 | spin_unlock(&space_info->lock); | |
3948 | return -ENOSPC; | |
3949 | } else if (space_info->max_extent_size) { | |
3950 | ffe_ctl->use_cluster = false; | |
3951 | } | |
3952 | spin_unlock(&space_info->lock); | |
3953 | } | |
3954 | ||
3955 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
3956 | &ffe_ctl->empty_cluster); | |
3957 | if (ffe_ctl->last_ptr) { | |
3958 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3959 | ||
3960 | spin_lock(&last_ptr->lock); | |
3961 | if (last_ptr->block_group) | |
3962 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3963 | if (last_ptr->fragmented) { | |
3964 | /* | |
3965 | * We still set window_start so we can keep track of the | |
3966 | * last place we found an allocation to try and save | |
3967 | * some time. | |
3968 | */ | |
3969 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3970 | ffe_ctl->use_cluster = false; | |
3971 | } | |
3972 | spin_unlock(&last_ptr->lock); | |
3973 | } | |
3974 | ||
3975 | return 0; | |
3976 | } | |
3977 | ||
3978 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
3979 | struct find_free_extent_ctl *ffe_ctl, | |
3980 | struct btrfs_space_info *space_info, | |
3981 | struct btrfs_key *ins) | |
3982 | { | |
3983 | switch (ffe_ctl->policy) { | |
3984 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3985 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
3986 | space_info, ins); | |
2eda5708 NA |
3987 | case BTRFS_EXTENT_ALLOC_ZONED: |
3988 | /* Nothing to do */ | |
3989 | return 0; | |
7e895409 NA |
3990 | default: |
3991 | BUG(); | |
3992 | } | |
3993 | } | |
3994 | ||
fec577fb CM |
3995 | /* |
3996 | * walks the btree of allocated extents and find a hole of a given size. | |
3997 | * The key ins is changed to record the hole: | |
a4820398 | 3998 | * ins->objectid == start position |
62e2749e | 3999 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4000 | * ins->offset == the size of the hole. |
fec577fb | 4001 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4002 | * |
4003 | * If there is no suitable free space, we will record the max size of | |
4004 | * the free space extent currently. | |
e72d79d6 QW |
4005 | * |
4006 | * The overall logic and call chain: | |
4007 | * | |
4008 | * find_free_extent() | |
4009 | * |- Iterate through all block groups | |
4010 | * | |- Get a valid block group | |
4011 | * | |- Try to do clustered allocation in that block group | |
4012 | * | |- Try to do unclustered allocation in that block group | |
4013 | * | |- Check if the result is valid | |
4014 | * | | |- If valid, then exit | |
4015 | * | |- Jump to next block group | |
4016 | * | | |
4017 | * |- Push harder to find free extents | |
4018 | * |- If not found, re-iterate all block groups | |
fec577fb | 4019 | */ |
437490fe | 4020 | static noinline int find_free_extent(struct btrfs_root *root, |
18513091 | 4021 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
ea544149 | 4022 | u64 hint_byte_orig, struct btrfs_key *ins, |
18513091 | 4023 | u64 flags, int delalloc) |
fec577fb | 4024 | { |
437490fe | 4025 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4026 | int ret = 0; |
db8fe64f | 4027 | int cache_block_group_error = 0; |
32da5386 | 4028 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 4029 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 4030 | struct btrfs_space_info *space_info; |
a5e681d9 | 4031 | bool full_search = false; |
fec577fb | 4032 | |
0b246afa | 4033 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 4034 | |
b4bd745d QW |
4035 | ffe_ctl.num_bytes = num_bytes; |
4036 | ffe_ctl.empty_size = empty_size; | |
4037 | ffe_ctl.flags = flags; | |
4038 | ffe_ctl.search_start = 0; | |
b4bd745d QW |
4039 | ffe_ctl.delalloc = delalloc; |
4040 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
4041 | ffe_ctl.have_caching_bg = false; | |
4042 | ffe_ctl.orig_have_caching_bg = false; | |
4043 | ffe_ctl.found_offset = 0; | |
ea544149 | 4044 | ffe_ctl.hint_byte = hint_byte_orig; |
cb2f96f8 | 4045 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED; |
b4bd745d | 4046 | |
c10859be NA |
4047 | /* For clustered allocation */ |
4048 | ffe_ctl.retry_clustered = false; | |
4049 | ffe_ctl.retry_unclustered = false; | |
4050 | ffe_ctl.last_ptr = NULL; | |
4051 | ffe_ctl.use_cluster = true; | |
4052 | ||
2eda5708 NA |
4053 | if (btrfs_is_zoned(fs_info)) |
4054 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_ZONED; | |
4055 | ||
962a298f | 4056 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4057 | ins->objectid = 0; |
4058 | ins->offset = 0; | |
b1a4d965 | 4059 | |
437490fe | 4060 | trace_find_free_extent(root, num_bytes, empty_size, flags); |
3f7de037 | 4061 | |
280c2908 | 4062 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 4063 | if (!space_info) { |
0b246afa | 4064 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
4065 | return -ENOSPC; |
4066 | } | |
2552d17e | 4067 | |
7e895409 NA |
4068 | ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins); |
4069 | if (ret < 0) | |
4070 | return ret; | |
fa9c0d79 | 4071 | |
b4bd745d QW |
4072 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
4073 | first_logical_byte(fs_info, 0)); | |
ea544149 NA |
4074 | ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte); |
4075 | if (ffe_ctl.search_start == ffe_ctl.hint_byte) { | |
b4bd745d QW |
4076 | block_group = btrfs_lookup_block_group(fs_info, |
4077 | ffe_ctl.search_start); | |
817d52f8 JB |
4078 | /* |
4079 | * we don't want to use the block group if it doesn't match our | |
4080 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4081 | * |
4082 | * However if we are re-searching with an ideal block group | |
4083 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4084 | */ |
b6919a58 | 4085 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 4086 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4087 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4088 | if (list_empty(&block_group->list) || |
4089 | block_group->ro) { | |
4090 | /* | |
4091 | * someone is removing this block group, | |
4092 | * we can't jump into the have_block_group | |
4093 | * target because our list pointers are not | |
4094 | * valid | |
4095 | */ | |
4096 | btrfs_put_block_group(block_group); | |
4097 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4098 | } else { |
b4bd745d | 4099 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 4100 | block_group->flags); |
e570fd27 | 4101 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 4102 | goto have_block_group; |
ccf0e725 | 4103 | } |
2552d17e | 4104 | } else if (block_group) { |
fa9c0d79 | 4105 | btrfs_put_block_group(block_group); |
2552d17e | 4106 | } |
42e70e7a | 4107 | } |
2552d17e | 4108 | search: |
b4bd745d QW |
4109 | ffe_ctl.have_caching_bg = false; |
4110 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
4111 | ffe_ctl.index == 0) | |
a5e681d9 | 4112 | full_search = true; |
80eb234a | 4113 | down_read(&space_info->groups_sem); |
b4bd745d QW |
4114 | list_for_each_entry(block_group, |
4115 | &space_info->block_groups[ffe_ctl.index], list) { | |
c668690d NA |
4116 | struct btrfs_block_group *bg_ret; |
4117 | ||
14443937 JM |
4118 | /* If the block group is read-only, we can skip it entirely. */ |
4119 | if (unlikely(block_group->ro)) | |
4120 | continue; | |
4121 | ||
e570fd27 | 4122 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 4123 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 4124 | |
83a50de9 CM |
4125 | /* |
4126 | * this can happen if we end up cycling through all the | |
4127 | * raid types, but we want to make sure we only allocate | |
4128 | * for the proper type. | |
4129 | */ | |
b6919a58 | 4130 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 4131 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4132 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4133 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4134 | BTRFS_BLOCK_GROUP_RAID10; |
4135 | ||
4136 | /* | |
4137 | * if they asked for extra copies and this block group | |
4138 | * doesn't provide them, bail. This does allow us to | |
4139 | * fill raid0 from raid1. | |
4140 | */ | |
b6919a58 | 4141 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4142 | goto loop; |
2a28468e QW |
4143 | |
4144 | /* | |
4145 | * This block group has different flags than we want. | |
4146 | * It's possible that we have MIXED_GROUP flag but no | |
4147 | * block group is mixed. Just skip such block group. | |
4148 | */ | |
4149 | btrfs_release_block_group(block_group, delalloc); | |
4150 | continue; | |
83a50de9 CM |
4151 | } |
4152 | ||
2552d17e | 4153 | have_block_group: |
32da5386 | 4154 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
4155 | if (unlikely(!ffe_ctl.cached)) { |
4156 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 4157 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
4158 | |
4159 | /* | |
4160 | * If we get ENOMEM here or something else we want to | |
4161 | * try other block groups, because it may not be fatal. | |
4162 | * However if we can't find anything else we need to | |
4163 | * save our return here so that we return the actual | |
4164 | * error that caused problems, not ENOSPC. | |
4165 | */ | |
4166 | if (ret < 0) { | |
4167 | if (!cache_block_group_error) | |
4168 | cache_block_group_error = ret; | |
4169 | ret = 0; | |
4170 | goto loop; | |
4171 | } | |
1d4284bd | 4172 | ret = 0; |
817d52f8 JB |
4173 | } |
4174 | ||
36cce922 JB |
4175 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
4176 | goto loop; | |
0f9dd46c | 4177 | |
c668690d NA |
4178 | bg_ret = NULL; |
4179 | ret = do_allocation(block_group, &ffe_ctl, &bg_ret); | |
4180 | if (ret == 0) { | |
4181 | if (bg_ret && bg_ret != block_group) { | |
4182 | btrfs_release_block_group(block_group, delalloc); | |
4183 | block_group = bg_ret; | |
fa9c0d79 | 4184 | } |
c668690d | 4185 | } else if (ret == -EAGAIN) { |
817d52f8 | 4186 | goto have_block_group; |
c668690d | 4187 | } else if (ret > 0) { |
1cdda9b8 | 4188 | goto loop; |
c668690d NA |
4189 | } |
4190 | ||
4191 | /* Checks */ | |
b4bd745d QW |
4192 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4193 | fs_info->stripesize); | |
25179201 | 4194 | |
2552d17e | 4195 | /* move on to the next group */ |
b4bd745d | 4196 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4197 | block_group->start + block_group->length) { |
2eda5708 NA |
4198 | btrfs_add_free_space_unused(block_group, |
4199 | ffe_ctl.found_offset, num_bytes); | |
2552d17e | 4200 | goto loop; |
6226cb0a | 4201 | } |
f5a31e16 | 4202 | |
b4bd745d | 4203 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
2eda5708 NA |
4204 | btrfs_add_free_space_unused(block_group, |
4205 | ffe_ctl.found_offset, | |
4206 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4207 | |
18513091 WX |
4208 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4209 | num_bytes, delalloc); | |
f0486c68 | 4210 | if (ret == -EAGAIN) { |
2eda5708 NA |
4211 | btrfs_add_free_space_unused(block_group, |
4212 | ffe_ctl.found_offset, num_bytes); | |
2552d17e | 4213 | goto loop; |
0f9dd46c | 4214 | } |
9cfa3e34 | 4215 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4216 | |
f0486c68 | 4217 | /* we are all good, lets return */ |
b4bd745d | 4218 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4219 | ins->offset = num_bytes; |
d2fb3437 | 4220 | |
b4bd745d QW |
4221 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4222 | num_bytes); | |
e570fd27 | 4223 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4224 | break; |
4225 | loop: | |
baba5062 | 4226 | release_block_group(block_group, &ffe_ctl, delalloc); |
14443937 | 4227 | cond_resched(); |
2552d17e JB |
4228 | } |
4229 | up_read(&space_info->groups_sem); | |
4230 | ||
15b7ee65 | 4231 | ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search); |
e72d79d6 | 4232 | if (ret > 0) |
b742bb82 YZ |
4233 | goto search; |
4234 | ||
db8fe64f | 4235 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4236 | /* |
4237 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4238 | * any contiguous hole. | |
4239 | */ | |
4240 | if (!ffe_ctl.max_extent_size) | |
4241 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4242 | spin_lock(&space_info->lock); |
b4bd745d | 4243 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4244 | spin_unlock(&space_info->lock); |
b4bd745d | 4245 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4246 | } else if (ret == -ENOSPC) { |
4247 | ret = cache_block_group_error; | |
4f4db217 | 4248 | } |
0f70abe2 | 4249 | return ret; |
fec577fb | 4250 | } |
ec44a35c | 4251 | |
6f47c706 NB |
4252 | /* |
4253 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4254 | * hole that is at least as big as @num_bytes. | |
4255 | * | |
4256 | * @root - The root that will contain this extent | |
4257 | * | |
4258 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4259 | * is used for accounting purposes. This value differs | |
4260 | * from @num_bytes only in the case of compressed extents. | |
4261 | * | |
4262 | * @num_bytes - Number of bytes to allocate on-disk. | |
4263 | * | |
4264 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4265 | * allocator should try to satisfy. In some cases | |
4266 | * @num_bytes may be larger than what is required and if | |
4267 | * the filesystem is fragmented then allocation fails. | |
4268 | * However, the presence of @min_alloc_size gives a | |
4269 | * chance to try and satisfy the smaller allocation. | |
4270 | * | |
4271 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4272 | * size in bytes the allocator should try to find free | |
4273 | * next to the block it returns. This is just a hint and | |
4274 | * may be ignored by the allocator. | |
4275 | * | |
4276 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4277 | * address passed. It might be ignored. | |
4278 | * | |
4279 | * @ins - This key is modified to record the found hole. It will | |
4280 | * have the following values: | |
4281 | * ins->objectid == start position | |
4282 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4283 | * ins->offset == the size of the hole. | |
4284 | * | |
4285 | * @is_data - Boolean flag indicating whether an extent is | |
4286 | * allocated for data (true) or metadata (false) | |
4287 | * | |
4288 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4289 | * delalloc or not. If 'true' data_rwsem of block groups | |
4290 | * is going to be acquired. | |
4291 | * | |
4292 | * | |
4293 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4294 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4295 | * largest available hole the allocator managed to find. | |
4296 | */ | |
18513091 | 4297 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4298 | u64 num_bytes, u64 min_alloc_size, |
4299 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4300 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4301 | { |
ab8d0fc4 | 4302 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4303 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4304 | u64 flags; |
fec577fb | 4305 | int ret; |
925baedd | 4306 | |
1b86826d | 4307 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4308 | again: |
0b246afa | 4309 | WARN_ON(num_bytes < fs_info->sectorsize); |
437490fe | 4310 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
18513091 | 4311 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4312 | if (!ret && !is_data) { |
ab8d0fc4 | 4313 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4314 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4315 | if (!final_tried && ins->offset) { |
4316 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4317 | num_bytes = round_down(num_bytes, |
0b246afa | 4318 | fs_info->sectorsize); |
9e622d6b | 4319 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4320 | ram_bytes = num_bytes; |
9e622d6b MX |
4321 | if (num_bytes == min_alloc_size) |
4322 | final_tried = true; | |
4323 | goto again; | |
ab8d0fc4 | 4324 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4325 | struct btrfs_space_info *sinfo; |
4326 | ||
280c2908 | 4327 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4328 | btrfs_err(fs_info, |
5d163e0e JM |
4329 | "allocation failed flags %llu, wanted %llu", |
4330 | flags, num_bytes); | |
53804280 | 4331 | if (sinfo) |
5da6afeb JB |
4332 | btrfs_dump_space_info(fs_info, sinfo, |
4333 | num_bytes, 1); | |
9e622d6b | 4334 | } |
925baedd | 4335 | } |
0f9dd46c JB |
4336 | |
4337 | return ret; | |
e6dcd2dc CM |
4338 | } |
4339 | ||
a0fbf736 NB |
4340 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4341 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4342 | { |
32da5386 | 4343 | struct btrfs_block_group *cache; |
0f9dd46c | 4344 | |
0b246afa | 4345 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4346 | if (!cache) { |
0b246afa JM |
4347 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4348 | start); | |
0f9dd46c JB |
4349 | return -ENOSPC; |
4350 | } | |
1f3c79a2 | 4351 | |
a0fbf736 NB |
4352 | btrfs_add_free_space(cache, start, len); |
4353 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4354 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4355 | ||
fa9c0d79 | 4356 | btrfs_put_block_group(cache); |
a0fbf736 | 4357 | return 0; |
e6dcd2dc CM |
4358 | } |
4359 | ||
7bfc1007 NB |
4360 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, |
4361 | u64 len) | |
e688b725 | 4362 | { |
7ef54d54 | 4363 | struct btrfs_block_group *cache; |
a0fbf736 | 4364 | int ret = 0; |
7ef54d54 | 4365 | |
7bfc1007 | 4366 | cache = btrfs_lookup_block_group(trans->fs_info, start); |
7ef54d54 | 4367 | if (!cache) { |
7bfc1007 NB |
4368 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
4369 | start); | |
7ef54d54 NB |
4370 | return -ENOSPC; |
4371 | } | |
4372 | ||
6690d071 | 4373 | ret = pin_down_extent(trans, cache, start, len, 1); |
7ef54d54 | 4374 | btrfs_put_block_group(cache); |
a0fbf736 | 4375 | return ret; |
e688b725 CM |
4376 | } |
4377 | ||
5d4f98a2 | 4378 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4379 | u64 parent, u64 root_objectid, |
4380 | u64 flags, u64 owner, u64 offset, | |
4381 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4382 | { |
ef89b824 | 4383 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4384 | int ret; |
e6dcd2dc | 4385 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4386 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4387 | struct btrfs_path *path; |
5d4f98a2 YZ |
4388 | struct extent_buffer *leaf; |
4389 | int type; | |
4390 | u32 size; | |
26b8003f | 4391 | |
5d4f98a2 YZ |
4392 | if (parent > 0) |
4393 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4394 | else | |
4395 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4396 | |
5d4f98a2 | 4397 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4398 | |
4399 | path = btrfs_alloc_path(); | |
db5b493a TI |
4400 | if (!path) |
4401 | return -ENOMEM; | |
47e4bb98 | 4402 | |
5d4f98a2 YZ |
4403 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4404 | ins, size); | |
79787eaa JM |
4405 | if (ret) { |
4406 | btrfs_free_path(path); | |
4407 | return ret; | |
4408 | } | |
0f9dd46c | 4409 | |
5d4f98a2 YZ |
4410 | leaf = path->nodes[0]; |
4411 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4412 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4413 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4414 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4415 | btrfs_set_extent_flags(leaf, extent_item, | |
4416 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4417 | ||
4418 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4419 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4420 | if (parent > 0) { | |
4421 | struct btrfs_shared_data_ref *ref; | |
4422 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4423 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4424 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4425 | } else { | |
4426 | struct btrfs_extent_data_ref *ref; | |
4427 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4428 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4429 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4430 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4431 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4432 | } | |
47e4bb98 CM |
4433 | |
4434 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4435 | btrfs_free_path(path); |
f510cfec | 4436 | |
25a356d3 | 4437 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4438 | if (ret) |
4439 | return ret; | |
4440 | ||
ade4b516 | 4441 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4442 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4443 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4444 | ins->objectid, ins->offset); |
f5947066 CM |
4445 | BUG(); |
4446 | } | |
71ff6437 | 4447 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4448 | return ret; |
4449 | } | |
4450 | ||
5d4f98a2 | 4451 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4452 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4453 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4454 | { |
9dcdbe01 | 4455 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4456 | int ret; |
5d4f98a2 | 4457 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4458 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4459 | struct btrfs_tree_block_info *block_info; |
4460 | struct btrfs_extent_inline_ref *iref; | |
4461 | struct btrfs_path *path; | |
4462 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4463 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4464 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4465 | u64 num_bytes; |
21ebfbe7 | 4466 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4467 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4468 | |
4e6bd4e0 NB |
4469 | ref = btrfs_delayed_node_to_tree_ref(node); |
4470 | ||
4e6bd4e0 NB |
4471 | extent_key.objectid = node->bytenr; |
4472 | if (skinny_metadata) { | |
4473 | extent_key.offset = ref->level; | |
4474 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4475 | num_bytes = fs_info->nodesize; | |
4476 | } else { | |
4477 | extent_key.offset = node->num_bytes; | |
4478 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4479 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4480 | num_bytes = node->num_bytes; |
4481 | } | |
1c2308f8 | 4482 | |
5d4f98a2 | 4483 | path = btrfs_alloc_path(); |
80ee54bf | 4484 | if (!path) |
d8926bb3 | 4485 | return -ENOMEM; |
56bec294 | 4486 | |
5d4f98a2 | 4487 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4e6bd4e0 | 4488 | &extent_key, size); |
79787eaa | 4489 | if (ret) { |
dd825259 | 4490 | btrfs_free_path(path); |
79787eaa JM |
4491 | return ret; |
4492 | } | |
5d4f98a2 YZ |
4493 | |
4494 | leaf = path->nodes[0]; | |
4495 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4496 | struct btrfs_extent_item); | |
4497 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4498 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4499 | btrfs_set_extent_flags(leaf, extent_item, | |
4500 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4501 | |
3173a18f JB |
4502 | if (skinny_metadata) { |
4503 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4504 | } else { | |
4505 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4506 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4507 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4508 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4509 | } | |
5d4f98a2 | 4510 | |
d4b20733 | 4511 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4512 | btrfs_set_extent_inline_ref_type(leaf, iref, |
4513 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4514 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4515 | } else { |
4516 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4517 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4518 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4519 | } |
4520 | ||
4521 | btrfs_mark_buffer_dirty(leaf); | |
4522 | btrfs_free_path(path); | |
4523 | ||
4e6bd4e0 NB |
4524 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4525 | num_bytes); | |
1e144fb8 OS |
4526 | if (ret) |
4527 | return ret; | |
4528 | ||
ade4b516 JB |
4529 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4530 | fs_info->nodesize, 1); | |
79787eaa | 4531 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4532 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4533 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4534 | BUG(); |
4535 | } | |
0be5dc67 | 4536 | |
4e6bd4e0 | 4537 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4538 | fs_info->nodesize); |
5d4f98a2 YZ |
4539 | return ret; |
4540 | } | |
4541 | ||
4542 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4543 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4544 | u64 offset, u64 ram_bytes, |
4545 | struct btrfs_key *ins) | |
5d4f98a2 | 4546 | { |
76675593 | 4547 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 | 4548 | |
84f7d8e6 | 4549 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4550 | |
76675593 QW |
4551 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4552 | ins->objectid, ins->offset, 0); | |
4553 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4554 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
2187374f JB |
4555 | |
4556 | return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes); | |
e6dcd2dc | 4557 | } |
e02119d5 CM |
4558 | |
4559 | /* | |
4560 | * this is used by the tree logging recovery code. It records that | |
4561 | * an extent has been allocated and makes sure to clear the free | |
4562 | * space cache bits as well | |
4563 | */ | |
5d4f98a2 | 4564 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4565 | u64 root_objectid, u64 owner, u64 offset, |
4566 | struct btrfs_key *ins) | |
e02119d5 | 4567 | { |
61da2abf | 4568 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4569 | int ret; |
32da5386 | 4570 | struct btrfs_block_group *block_group; |
ed7a6948 | 4571 | struct btrfs_space_info *space_info; |
11833d66 | 4572 | |
8c2a1a30 JB |
4573 | /* |
4574 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4575 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4576 | */ |
0b246afa | 4577 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4578 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4579 | ins->offset); | |
b50c6e25 | 4580 | if (ret) |
8c2a1a30 | 4581 | return ret; |
11833d66 YZ |
4582 | } |
4583 | ||
0b246afa | 4584 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4585 | if (!block_group) |
4586 | return -EINVAL; | |
4587 | ||
ed7a6948 WX |
4588 | space_info = block_group->space_info; |
4589 | spin_lock(&space_info->lock); | |
4590 | spin_lock(&block_group->lock); | |
4591 | space_info->bytes_reserved += ins->offset; | |
4592 | block_group->reserved += ins->offset; | |
4593 | spin_unlock(&block_group->lock); | |
4594 | spin_unlock(&space_info->lock); | |
4595 | ||
ef89b824 NB |
4596 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4597 | offset, ins, 1); | |
bd727173 | 4598 | if (ret) |
ab9b2c7b | 4599 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
b50c6e25 | 4600 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4601 | return ret; |
4602 | } | |
4603 | ||
48a3b636 ES |
4604 | static struct extent_buffer * |
4605 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
9631e4cc JB |
4606 | u64 bytenr, int level, u64 owner, |
4607 | enum btrfs_lock_nesting nest) | |
65b51a00 | 4608 | { |
0b246afa | 4609 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4610 | struct extent_buffer *buf; |
4611 | ||
3fbaf258 | 4612 | buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level); |
c871b0f2 LB |
4613 | if (IS_ERR(buf)) |
4614 | return buf; | |
4615 | ||
b72c3aba QW |
4616 | /* |
4617 | * Extra safety check in case the extent tree is corrupted and extent | |
4618 | * allocator chooses to use a tree block which is already used and | |
4619 | * locked. | |
4620 | */ | |
4621 | if (buf->lock_owner == current->pid) { | |
4622 | btrfs_err_rl(fs_info, | |
4623 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4624 | buf->start, btrfs_header_owner(buf), current->pid); | |
4625 | free_extent_buffer(buf); | |
4626 | return ERR_PTR(-EUCLEAN); | |
4627 | } | |
4628 | ||
e114c545 JB |
4629 | /* |
4630 | * This needs to stay, because we could allocate a freed block from an | |
4631 | * old tree into a new tree, so we need to make sure this new block is | |
4632 | * set to the appropriate level and owner. | |
4633 | */ | |
ad244665 | 4634 | btrfs_set_buffer_lockdep_class(owner, buf, level); |
9631e4cc | 4635 | __btrfs_tree_lock(buf, nest); |
6a884d7d | 4636 | btrfs_clean_tree_block(buf); |
3083ee2e | 4637 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 4638 | |
4db8c528 | 4639 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4640 | |
bc877d28 NB |
4641 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4642 | btrfs_set_header_level(buf, level); | |
4643 | btrfs_set_header_bytenr(buf, buf->start); | |
4644 | btrfs_set_header_generation(buf, trans->transid); | |
4645 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4646 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4647 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4648 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4649 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4650 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4651 | /* |
4652 | * we allow two log transactions at a time, use different | |
52042d8e | 4653 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4654 | */ |
656f30db | 4655 | if (buf->log_index == 0) |
8cef4e16 YZ |
4656 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4657 | buf->start + buf->len - 1, GFP_NOFS); | |
4658 | else | |
4659 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4660 | buf->start + buf->len - 1); |
d0c803c4 | 4661 | } else { |
656f30db | 4662 | buf->log_index = -1; |
d0c803c4 | 4663 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4664 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4665 | } |
64c12921 | 4666 | trans->dirty = true; |
b4ce94de | 4667 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4668 | return buf; |
4669 | } | |
4670 | ||
fec577fb | 4671 | /* |
f0486c68 | 4672 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4673 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4674 | */ |
4d75f8a9 | 4675 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4676 | struct btrfs_root *root, |
4677 | u64 parent, u64 root_objectid, | |
4678 | const struct btrfs_disk_key *key, | |
4679 | int level, u64 hint, | |
9631e4cc JB |
4680 | u64 empty_size, |
4681 | enum btrfs_lock_nesting nest) | |
fec577fb | 4682 | { |
0b246afa | 4683 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4684 | struct btrfs_key ins; |
f0486c68 | 4685 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4686 | struct extent_buffer *buf; |
67b7859e | 4687 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4688 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4689 | u64 flags = 0; |
4690 | int ret; | |
0b246afa JM |
4691 | u32 blocksize = fs_info->nodesize; |
4692 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4693 | |
05653ef3 | 4694 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4695 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4696 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
9631e4cc | 4697 | level, root_objectid, nest); |
faa2dbf0 JB |
4698 | if (!IS_ERR(buf)) |
4699 | root->alloc_bytenr += blocksize; | |
4700 | return buf; | |
4701 | } | |
05653ef3 | 4702 | #endif |
fccb84c9 | 4703 | |
67f9c220 | 4704 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4705 | if (IS_ERR(block_rsv)) |
4706 | return ERR_CAST(block_rsv); | |
4707 | ||
18513091 | 4708 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4709 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4710 | if (ret) |
4711 | goto out_unuse; | |
55c69072 | 4712 | |
bc877d28 | 4713 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
9631e4cc | 4714 | root_objectid, nest); |
67b7859e OS |
4715 | if (IS_ERR(buf)) { |
4716 | ret = PTR_ERR(buf); | |
4717 | goto out_free_reserved; | |
4718 | } | |
f0486c68 YZ |
4719 | |
4720 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4721 | if (parent == 0) | |
4722 | parent = ins.objectid; | |
4723 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4724 | } else | |
4725 | BUG_ON(parent > 0); | |
4726 | ||
4727 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4728 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4729 | if (!extent_op) { |
4730 | ret = -ENOMEM; | |
4731 | goto out_free_buf; | |
4732 | } | |
f0486c68 YZ |
4733 | if (key) |
4734 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4735 | else | |
4736 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4737 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4738 | extent_op->update_key = skinny_metadata ? false : true; |
4739 | extent_op->update_flags = true; | |
4740 | extent_op->is_data = false; | |
b1c79e09 | 4741 | extent_op->level = level; |
f0486c68 | 4742 | |
ed4f255b QW |
4743 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4744 | ins.objectid, ins.offset, parent); | |
4745 | generic_ref.real_root = root->root_key.objectid; | |
4746 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4747 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 4748 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op); |
67b7859e OS |
4749 | if (ret) |
4750 | goto out_free_delayed; | |
f0486c68 | 4751 | } |
fec577fb | 4752 | return buf; |
67b7859e OS |
4753 | |
4754 | out_free_delayed: | |
4755 | btrfs_free_delayed_extent_op(extent_op); | |
4756 | out_free_buf: | |
4757 | free_extent_buffer(buf); | |
4758 | out_free_reserved: | |
2ff7e61e | 4759 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4760 | out_unuse: |
67f9c220 | 4761 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4762 | return ERR_PTR(ret); |
fec577fb | 4763 | } |
a28ec197 | 4764 | |
2c47e605 YZ |
4765 | struct walk_control { |
4766 | u64 refs[BTRFS_MAX_LEVEL]; | |
4767 | u64 flags[BTRFS_MAX_LEVEL]; | |
4768 | struct btrfs_key update_progress; | |
aea6f028 JB |
4769 | struct btrfs_key drop_progress; |
4770 | int drop_level; | |
2c47e605 YZ |
4771 | int stage; |
4772 | int level; | |
4773 | int shared_level; | |
4774 | int update_ref; | |
4775 | int keep_locks; | |
1c4850e2 YZ |
4776 | int reada_slot; |
4777 | int reada_count; | |
78c52d9e | 4778 | int restarted; |
2c47e605 YZ |
4779 | }; |
4780 | ||
4781 | #define DROP_REFERENCE 1 | |
4782 | #define UPDATE_BACKREF 2 | |
4783 | ||
1c4850e2 YZ |
4784 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4785 | struct btrfs_root *root, | |
4786 | struct walk_control *wc, | |
4787 | struct btrfs_path *path) | |
6407bf6d | 4788 | { |
0b246afa | 4789 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4790 | u64 bytenr; |
4791 | u64 generation; | |
4792 | u64 refs; | |
94fcca9f | 4793 | u64 flags; |
5d4f98a2 | 4794 | u32 nritems; |
1c4850e2 YZ |
4795 | struct btrfs_key key; |
4796 | struct extent_buffer *eb; | |
6407bf6d | 4797 | int ret; |
1c4850e2 YZ |
4798 | int slot; |
4799 | int nread = 0; | |
6407bf6d | 4800 | |
1c4850e2 YZ |
4801 | if (path->slots[wc->level] < wc->reada_slot) { |
4802 | wc->reada_count = wc->reada_count * 2 / 3; | |
4803 | wc->reada_count = max(wc->reada_count, 2); | |
4804 | } else { | |
4805 | wc->reada_count = wc->reada_count * 3 / 2; | |
4806 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4807 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4808 | } |
7bb86316 | 4809 | |
1c4850e2 YZ |
4810 | eb = path->nodes[wc->level]; |
4811 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4812 | |
1c4850e2 YZ |
4813 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4814 | if (nread >= wc->reada_count) | |
4815 | break; | |
bd56b302 | 4816 | |
2dd3e67b | 4817 | cond_resched(); |
1c4850e2 YZ |
4818 | bytenr = btrfs_node_blockptr(eb, slot); |
4819 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4820 | |
1c4850e2 YZ |
4821 | if (slot == path->slots[wc->level]) |
4822 | goto reada; | |
5d4f98a2 | 4823 | |
1c4850e2 YZ |
4824 | if (wc->stage == UPDATE_BACKREF && |
4825 | generation <= root->root_key.offset) | |
bd56b302 CM |
4826 | continue; |
4827 | ||
94fcca9f | 4828 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4829 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4830 | wc->level - 1, 1, &refs, |
4831 | &flags); | |
79787eaa JM |
4832 | /* We don't care about errors in readahead. */ |
4833 | if (ret < 0) | |
4834 | continue; | |
94fcca9f YZ |
4835 | BUG_ON(refs == 0); |
4836 | ||
1c4850e2 | 4837 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4838 | if (refs == 1) |
4839 | goto reada; | |
bd56b302 | 4840 | |
94fcca9f YZ |
4841 | if (wc->level == 1 && |
4842 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4843 | continue; | |
1c4850e2 YZ |
4844 | if (!wc->update_ref || |
4845 | generation <= root->root_key.offset) | |
4846 | continue; | |
4847 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4848 | ret = btrfs_comp_cpu_keys(&key, | |
4849 | &wc->update_progress); | |
4850 | if (ret < 0) | |
4851 | continue; | |
94fcca9f YZ |
4852 | } else { |
4853 | if (wc->level == 1 && | |
4854 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4855 | continue; | |
6407bf6d | 4856 | } |
1c4850e2 | 4857 | reada: |
bfb484d9 | 4858 | btrfs_readahead_node_child(eb, slot); |
1c4850e2 | 4859 | nread++; |
20524f02 | 4860 | } |
1c4850e2 | 4861 | wc->reada_slot = slot; |
20524f02 | 4862 | } |
2c47e605 | 4863 | |
f82d02d9 | 4864 | /* |
2c016dc2 | 4865 | * helper to process tree block while walking down the tree. |
2c47e605 | 4866 | * |
2c47e605 YZ |
4867 | * when wc->stage == UPDATE_BACKREF, this function updates |
4868 | * back refs for pointers in the block. | |
4869 | * | |
4870 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4871 | */ |
2c47e605 | 4872 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4873 | struct btrfs_root *root, |
2c47e605 | 4874 | struct btrfs_path *path, |
94fcca9f | 4875 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4876 | { |
2ff7e61e | 4877 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4878 | int level = wc->level; |
4879 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4880 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4881 | int ret; |
4882 | ||
2c47e605 YZ |
4883 | if (wc->stage == UPDATE_BACKREF && |
4884 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4885 | return 1; | |
f82d02d9 | 4886 | |
2c47e605 YZ |
4887 | /* |
4888 | * when reference count of tree block is 1, it won't increase | |
4889 | * again. once full backref flag is set, we never clear it. | |
4890 | */ | |
94fcca9f YZ |
4891 | if (lookup_info && |
4892 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4893 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4894 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4895 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4896 | eb->start, level, 1, |
2c47e605 YZ |
4897 | &wc->refs[level], |
4898 | &wc->flags[level]); | |
79787eaa JM |
4899 | BUG_ON(ret == -ENOMEM); |
4900 | if (ret) | |
4901 | return ret; | |
2c47e605 YZ |
4902 | BUG_ON(wc->refs[level] == 0); |
4903 | } | |
5d4f98a2 | 4904 | |
2c47e605 YZ |
4905 | if (wc->stage == DROP_REFERENCE) { |
4906 | if (wc->refs[level] > 1) | |
4907 | return 1; | |
f82d02d9 | 4908 | |
2c47e605 | 4909 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4910 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4911 | path->locks[level] = 0; |
4912 | } | |
4913 | return 0; | |
4914 | } | |
f82d02d9 | 4915 | |
2c47e605 YZ |
4916 | /* wc->stage == UPDATE_BACKREF */ |
4917 | if (!(wc->flags[level] & flag)) { | |
4918 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4919 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4920 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4921 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4922 | BUG_ON(ret); /* -ENOMEM */ |
42c9d0b5 | 4923 | ret = btrfs_set_disk_extent_flags(trans, eb, flag, |
b1c79e09 | 4924 | btrfs_header_level(eb), 0); |
79787eaa | 4925 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4926 | wc->flags[level] |= flag; |
4927 | } | |
4928 | ||
4929 | /* | |
4930 | * the block is shared by multiple trees, so it's not good to | |
4931 | * keep the tree lock | |
4932 | */ | |
4933 | if (path->locks[level] && level > 0) { | |
bd681513 | 4934 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4935 | path->locks[level] = 0; |
4936 | } | |
4937 | return 0; | |
4938 | } | |
4939 | ||
78c52d9e JB |
4940 | /* |
4941 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4942 | * would have a drop_progress key that hadn't been updated properly. | |
4943 | */ | |
4944 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4945 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4946 | int level) | |
4947 | { | |
4948 | struct btrfs_path *path; | |
4949 | struct btrfs_extent_inline_ref *iref; | |
4950 | int ret; | |
4951 | ||
4952 | path = btrfs_alloc_path(); | |
4953 | if (!path) | |
4954 | return -ENOMEM; | |
4955 | ||
4956 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
4957 | root->fs_info->nodesize, parent, | |
4958 | root->root_key.objectid, level, 0); | |
4959 | btrfs_free_path(path); | |
4960 | if (ret == -ENOENT) | |
4961 | return 0; | |
4962 | if (ret < 0) | |
4963 | return ret; | |
4964 | return 1; | |
4965 | } | |
4966 | ||
1c4850e2 | 4967 | /* |
2c016dc2 | 4968 | * helper to process tree block pointer. |
1c4850e2 YZ |
4969 | * |
4970 | * when wc->stage == DROP_REFERENCE, this function checks | |
4971 | * reference count of the block pointed to. if the block | |
4972 | * is shared and we need update back refs for the subtree | |
4973 | * rooted at the block, this function changes wc->stage to | |
4974 | * UPDATE_BACKREF. if the block is shared and there is no | |
4975 | * need to update back, this function drops the reference | |
4976 | * to the block. | |
4977 | * | |
4978 | * NOTE: return value 1 means we should stop walking down. | |
4979 | */ | |
4980 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
4981 | struct btrfs_root *root, | |
4982 | struct btrfs_path *path, | |
94fcca9f | 4983 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 4984 | { |
0b246afa | 4985 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4986 | u64 bytenr; |
4987 | u64 generation; | |
4988 | u64 parent; | |
1c4850e2 | 4989 | struct btrfs_key key; |
581c1760 | 4990 | struct btrfs_key first_key; |
ffd4bb2a | 4991 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
4992 | struct extent_buffer *next; |
4993 | int level = wc->level; | |
4994 | int reada = 0; | |
4995 | int ret = 0; | |
1152651a | 4996 | bool need_account = false; |
1c4850e2 YZ |
4997 | |
4998 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
4999 | path->slots[level]); | |
5000 | /* | |
5001 | * if the lower level block was created before the snapshot | |
5002 | * was created, we know there is no need to update back refs | |
5003 | * for the subtree | |
5004 | */ | |
5005 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5006 | generation <= root->root_key.offset) { |
5007 | *lookup_info = 1; | |
1c4850e2 | 5008 | return 1; |
94fcca9f | 5009 | } |
1c4850e2 YZ |
5010 | |
5011 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
5012 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
5013 | path->slots[level]); | |
1c4850e2 | 5014 | |
0b246afa | 5015 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5016 | if (!next) { |
3fbaf258 JB |
5017 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
5018 | root->root_key.objectid, level - 1); | |
c871b0f2 LB |
5019 | if (IS_ERR(next)) |
5020 | return PTR_ERR(next); | |
1c4850e2 YZ |
5021 | reada = 1; |
5022 | } | |
5023 | btrfs_tree_lock(next); | |
1c4850e2 | 5024 | |
2ff7e61e | 5025 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
5026 | &wc->refs[level - 1], |
5027 | &wc->flags[level - 1]); | |
4867268c JB |
5028 | if (ret < 0) |
5029 | goto out_unlock; | |
79787eaa | 5030 | |
c2cf52eb | 5031 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5032 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5033 | ret = -EIO; |
5034 | goto out_unlock; | |
c2cf52eb | 5035 | } |
94fcca9f | 5036 | *lookup_info = 0; |
1c4850e2 | 5037 | |
94fcca9f | 5038 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5039 | if (wc->refs[level - 1] > 1) { |
1152651a | 5040 | need_account = true; |
94fcca9f YZ |
5041 | if (level == 1 && |
5042 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5043 | goto skip; | |
5044 | ||
1c4850e2 YZ |
5045 | if (!wc->update_ref || |
5046 | generation <= root->root_key.offset) | |
5047 | goto skip; | |
5048 | ||
5049 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5050 | path->slots[level]); | |
5051 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5052 | if (ret < 0) | |
5053 | goto skip; | |
5054 | ||
5055 | wc->stage = UPDATE_BACKREF; | |
5056 | wc->shared_level = level - 1; | |
5057 | } | |
94fcca9f YZ |
5058 | } else { |
5059 | if (level == 1 && | |
5060 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5061 | goto skip; | |
1c4850e2 YZ |
5062 | } |
5063 | ||
b9fab919 | 5064 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5065 | btrfs_tree_unlock(next); |
5066 | free_extent_buffer(next); | |
5067 | next = NULL; | |
94fcca9f | 5068 | *lookup_info = 1; |
1c4850e2 YZ |
5069 | } |
5070 | ||
5071 | if (!next) { | |
5072 | if (reada && level == 1) | |
5073 | reada_walk_down(trans, root, wc, path); | |
1b7ec85e JB |
5074 | next = read_tree_block(fs_info, bytenr, root->root_key.objectid, |
5075 | generation, level - 1, &first_key); | |
64c043de LB |
5076 | if (IS_ERR(next)) { |
5077 | return PTR_ERR(next); | |
5078 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5079 | free_extent_buffer(next); |
97d9a8a4 | 5080 | return -EIO; |
416bc658 | 5081 | } |
1c4850e2 | 5082 | btrfs_tree_lock(next); |
1c4850e2 YZ |
5083 | } |
5084 | ||
5085 | level--; | |
4867268c JB |
5086 | ASSERT(level == btrfs_header_level(next)); |
5087 | if (level != btrfs_header_level(next)) { | |
5088 | btrfs_err(root->fs_info, "mismatched level"); | |
5089 | ret = -EIO; | |
5090 | goto out_unlock; | |
5091 | } | |
1c4850e2 YZ |
5092 | path->nodes[level] = next; |
5093 | path->slots[level] = 0; | |
ac5887c8 | 5094 | path->locks[level] = BTRFS_WRITE_LOCK; |
1c4850e2 YZ |
5095 | wc->level = level; |
5096 | if (wc->level == 1) | |
5097 | wc->reada_slot = 0; | |
5098 | return 0; | |
5099 | skip: | |
5100 | wc->refs[level - 1] = 0; | |
5101 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5102 | if (wc->stage == DROP_REFERENCE) { |
5103 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5104 | parent = path->nodes[level]->start; | |
5105 | } else { | |
4867268c | 5106 | ASSERT(root->root_key.objectid == |
94fcca9f | 5107 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5108 | if (root->root_key.objectid != |
5109 | btrfs_header_owner(path->nodes[level])) { | |
5110 | btrfs_err(root->fs_info, | |
5111 | "mismatched block owner"); | |
5112 | ret = -EIO; | |
5113 | goto out_unlock; | |
5114 | } | |
94fcca9f YZ |
5115 | parent = 0; |
5116 | } | |
1c4850e2 | 5117 | |
78c52d9e JB |
5118 | /* |
5119 | * If we had a drop_progress we need to verify the refs are set | |
5120 | * as expected. If we find our ref then we know that from here | |
5121 | * on out everything should be correct, and we can clear the | |
5122 | * ->restarted flag. | |
5123 | */ | |
5124 | if (wc->restarted) { | |
5125 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5126 | level - 1); | |
5127 | if (ret < 0) | |
5128 | goto out_unlock; | |
5129 | if (ret == 0) | |
5130 | goto no_delete; | |
5131 | ret = 0; | |
5132 | wc->restarted = 0; | |
5133 | } | |
5134 | ||
2cd86d30 QW |
5135 | /* |
5136 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5137 | * already accounted them at merge time (replace_path), | |
5138 | * thus we could skip expensive subtree trace here. | |
5139 | */ | |
5140 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5141 | need_account) { | |
deb40627 | 5142 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5143 | generation, level - 1); |
1152651a | 5144 | if (ret) { |
0b246afa | 5145 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5146 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5147 | ret); | |
1152651a MF |
5148 | } |
5149 | } | |
aea6f028 JB |
5150 | |
5151 | /* | |
5152 | * We need to update the next key in our walk control so we can | |
5153 | * update the drop_progress key accordingly. We don't care if | |
5154 | * find_next_key doesn't find a key because that means we're at | |
5155 | * the end and are going to clean up now. | |
5156 | */ | |
5157 | wc->drop_level = level; | |
5158 | find_next_key(path, level, &wc->drop_progress); | |
5159 | ||
ffd4bb2a QW |
5160 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
5161 | fs_info->nodesize, parent); | |
5162 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
5163 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
5164 | if (ret) |
5165 | goto out_unlock; | |
1c4850e2 | 5166 | } |
78c52d9e | 5167 | no_delete: |
4867268c JB |
5168 | *lookup_info = 1; |
5169 | ret = 1; | |
5170 | ||
5171 | out_unlock: | |
1c4850e2 YZ |
5172 | btrfs_tree_unlock(next); |
5173 | free_extent_buffer(next); | |
4867268c JB |
5174 | |
5175 | return ret; | |
1c4850e2 YZ |
5176 | } |
5177 | ||
2c47e605 | 5178 | /* |
2c016dc2 | 5179 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5180 | * |
5181 | * when wc->stage == DROP_REFERENCE, this function drops | |
5182 | * reference count on the block. | |
5183 | * | |
5184 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5185 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5186 | * to UPDATE_BACKREF previously while processing the block. | |
5187 | * | |
5188 | * NOTE: return value 1 means we should stop walking up. | |
5189 | */ | |
5190 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5191 | struct btrfs_root *root, | |
5192 | struct btrfs_path *path, | |
5193 | struct walk_control *wc) | |
5194 | { | |
0b246afa | 5195 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5196 | int ret; |
2c47e605 YZ |
5197 | int level = wc->level; |
5198 | struct extent_buffer *eb = path->nodes[level]; | |
5199 | u64 parent = 0; | |
5200 | ||
5201 | if (wc->stage == UPDATE_BACKREF) { | |
5202 | BUG_ON(wc->shared_level < level); | |
5203 | if (level < wc->shared_level) | |
5204 | goto out; | |
5205 | ||
2c47e605 YZ |
5206 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5207 | if (ret > 0) | |
5208 | wc->update_ref = 0; | |
5209 | ||
5210 | wc->stage = DROP_REFERENCE; | |
5211 | wc->shared_level = -1; | |
5212 | path->slots[level] = 0; | |
5213 | ||
5214 | /* | |
5215 | * check reference count again if the block isn't locked. | |
5216 | * we should start walking down the tree again if reference | |
5217 | * count is one. | |
5218 | */ | |
5219 | if (!path->locks[level]) { | |
5220 | BUG_ON(level == 0); | |
5221 | btrfs_tree_lock(eb); | |
ac5887c8 | 5222 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5223 | |
2ff7e61e | 5224 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5225 | eb->start, level, 1, |
2c47e605 YZ |
5226 | &wc->refs[level], |
5227 | &wc->flags[level]); | |
79787eaa JM |
5228 | if (ret < 0) { |
5229 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5230 | path->locks[level] = 0; |
79787eaa JM |
5231 | return ret; |
5232 | } | |
2c47e605 YZ |
5233 | BUG_ON(wc->refs[level] == 0); |
5234 | if (wc->refs[level] == 1) { | |
bd681513 | 5235 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5236 | path->locks[level] = 0; |
2c47e605 YZ |
5237 | return 1; |
5238 | } | |
f82d02d9 | 5239 | } |
2c47e605 | 5240 | } |
f82d02d9 | 5241 | |
2c47e605 YZ |
5242 | /* wc->stage == DROP_REFERENCE */ |
5243 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5244 | |
2c47e605 YZ |
5245 | if (wc->refs[level] == 1) { |
5246 | if (level == 0) { | |
5247 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5248 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5249 | else |
e339a6b0 | 5250 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5251 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5252 | if (is_fstree(root->root_key.objectid)) { |
5253 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5254 | if (ret) { | |
5255 | btrfs_err_rl(fs_info, | |
5256 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5257 | ret); |
c4140cbf | 5258 | } |
1152651a | 5259 | } |
2c47e605 | 5260 | } |
6a884d7d | 5261 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5262 | if (!path->locks[level] && |
5263 | btrfs_header_generation(eb) == trans->transid) { | |
5264 | btrfs_tree_lock(eb); | |
ac5887c8 | 5265 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5266 | } |
6a884d7d | 5267 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5268 | } |
5269 | ||
5270 | if (eb == root->node) { | |
5271 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5272 | parent = eb->start; | |
65c6e82b QW |
5273 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5274 | goto owner_mismatch; | |
2c47e605 YZ |
5275 | } else { |
5276 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5277 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5278 | else if (root->root_key.objectid != |
5279 | btrfs_header_owner(path->nodes[level + 1])) | |
5280 | goto owner_mismatch; | |
f82d02d9 | 5281 | } |
f82d02d9 | 5282 | |
5581a51a | 5283 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5284 | out: |
5285 | wc->refs[level] = 0; | |
5286 | wc->flags[level] = 0; | |
f0486c68 | 5287 | return 0; |
65c6e82b QW |
5288 | |
5289 | owner_mismatch: | |
5290 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5291 | btrfs_header_owner(eb), root->root_key.objectid); | |
5292 | return -EUCLEAN; | |
2c47e605 YZ |
5293 | } |
5294 | ||
5295 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5296 | struct btrfs_root *root, | |
5297 | struct btrfs_path *path, | |
5298 | struct walk_control *wc) | |
5299 | { | |
2c47e605 | 5300 | int level = wc->level; |
94fcca9f | 5301 | int lookup_info = 1; |
2c47e605 YZ |
5302 | int ret; |
5303 | ||
5304 | while (level >= 0) { | |
94fcca9f | 5305 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5306 | if (ret > 0) |
5307 | break; | |
5308 | ||
5309 | if (level == 0) | |
5310 | break; | |
5311 | ||
7a7965f8 YZ |
5312 | if (path->slots[level] >= |
5313 | btrfs_header_nritems(path->nodes[level])) | |
5314 | break; | |
5315 | ||
94fcca9f | 5316 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5317 | if (ret > 0) { |
5318 | path->slots[level]++; | |
5319 | continue; | |
90d2c51d MX |
5320 | } else if (ret < 0) |
5321 | return ret; | |
1c4850e2 | 5322 | level = wc->level; |
f82d02d9 | 5323 | } |
f82d02d9 YZ |
5324 | return 0; |
5325 | } | |
5326 | ||
d397712b | 5327 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5328 | struct btrfs_root *root, |
f82d02d9 | 5329 | struct btrfs_path *path, |
2c47e605 | 5330 | struct walk_control *wc, int max_level) |
20524f02 | 5331 | { |
2c47e605 | 5332 | int level = wc->level; |
20524f02 | 5333 | int ret; |
9f3a7427 | 5334 | |
2c47e605 YZ |
5335 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5336 | while (level < max_level && path->nodes[level]) { | |
5337 | wc->level = level; | |
5338 | if (path->slots[level] + 1 < | |
5339 | btrfs_header_nritems(path->nodes[level])) { | |
5340 | path->slots[level]++; | |
20524f02 CM |
5341 | return 0; |
5342 | } else { | |
2c47e605 YZ |
5343 | ret = walk_up_proc(trans, root, path, wc); |
5344 | if (ret > 0) | |
5345 | return 0; | |
65c6e82b QW |
5346 | if (ret < 0) |
5347 | return ret; | |
bd56b302 | 5348 | |
2c47e605 | 5349 | if (path->locks[level]) { |
bd681513 CM |
5350 | btrfs_tree_unlock_rw(path->nodes[level], |
5351 | path->locks[level]); | |
2c47e605 | 5352 | path->locks[level] = 0; |
f82d02d9 | 5353 | } |
2c47e605 YZ |
5354 | free_extent_buffer(path->nodes[level]); |
5355 | path->nodes[level] = NULL; | |
5356 | level++; | |
20524f02 CM |
5357 | } |
5358 | } | |
5359 | return 1; | |
5360 | } | |
5361 | ||
9aca1d51 | 5362 | /* |
2c47e605 YZ |
5363 | * drop a subvolume tree. |
5364 | * | |
5365 | * this function traverses the tree freeing any blocks that only | |
5366 | * referenced by the tree. | |
5367 | * | |
5368 | * when a shared tree block is found. this function decreases its | |
5369 | * reference count by one. if update_ref is true, this function | |
5370 | * also make sure backrefs for the shared block and all lower level | |
5371 | * blocks are properly updated. | |
9d1a2a3a DS |
5372 | * |
5373 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5374 | */ |
0078a9f9 | 5375 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5376 | { |
ab8d0fc4 | 5377 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5378 | struct btrfs_path *path; |
2c47e605 | 5379 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5380 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5381 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5382 | struct walk_control *wc; |
5383 | struct btrfs_key key; | |
5384 | int err = 0; | |
5385 | int ret; | |
5386 | int level; | |
d29a9f62 | 5387 | bool root_dropped = false; |
20524f02 | 5388 | |
4fd786e6 | 5389 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5390 | |
5caf2a00 | 5391 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5392 | if (!path) { |
5393 | err = -ENOMEM; | |
5394 | goto out; | |
5395 | } | |
20524f02 | 5396 | |
2c47e605 | 5397 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5398 | if (!wc) { |
5399 | btrfs_free_path(path); | |
cb1b69f4 TI |
5400 | err = -ENOMEM; |
5401 | goto out; | |
38a1a919 | 5402 | } |
2c47e605 | 5403 | |
f3e3d9cc QW |
5404 | /* |
5405 | * Use join to avoid potential EINTR from transaction start. See | |
5406 | * wait_reserve_ticket and the whole reservation callchain. | |
5407 | */ | |
5408 | if (for_reloc) | |
5409 | trans = btrfs_join_transaction(tree_root); | |
5410 | else | |
5411 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5412 | if (IS_ERR(trans)) { |
5413 | err = PTR_ERR(trans); | |
5414 | goto out_free; | |
5415 | } | |
98d5dc13 | 5416 | |
0568e82d JB |
5417 | err = btrfs_run_delayed_items(trans); |
5418 | if (err) | |
5419 | goto out_end_trans; | |
5420 | ||
83354f07 JB |
5421 | /* |
5422 | * This will help us catch people modifying the fs tree while we're | |
5423 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5424 | * dropped as we unlock the root node and parent nodes as we walk down | |
5425 | * the tree, assuming nothing will change. If something does change | |
5426 | * then we'll have stale information and drop references to blocks we've | |
5427 | * already dropped. | |
5428 | */ | |
5429 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5430 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5431 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5432 | path->nodes[level] = btrfs_lock_root_node(root); |
9f3a7427 | 5433 | path->slots[level] = 0; |
ac5887c8 | 5434 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5435 | memset(&wc->update_progress, 0, |
5436 | sizeof(wc->update_progress)); | |
9f3a7427 | 5437 | } else { |
9f3a7427 | 5438 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5439 | memcpy(&wc->update_progress, &key, |
5440 | sizeof(wc->update_progress)); | |
5441 | ||
c8422684 | 5442 | level = btrfs_root_drop_level(root_item); |
2c47e605 | 5443 | BUG_ON(level == 0); |
6702ed49 | 5444 | path->lowest_level = level; |
2c47e605 YZ |
5445 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5446 | path->lowest_level = 0; | |
5447 | if (ret < 0) { | |
5448 | err = ret; | |
79787eaa | 5449 | goto out_end_trans; |
9f3a7427 | 5450 | } |
1c4850e2 | 5451 | WARN_ON(ret > 0); |
2c47e605 | 5452 | |
7d9eb12c CM |
5453 | /* |
5454 | * unlock our path, this is safe because only this | |
5455 | * function is allowed to delete this snapshot | |
5456 | */ | |
5d4f98a2 | 5457 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5458 | |
5459 | level = btrfs_header_level(root->node); | |
5460 | while (1) { | |
5461 | btrfs_tree_lock(path->nodes[level]); | |
ac5887c8 | 5462 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5463 | |
2ff7e61e | 5464 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5465 | path->nodes[level]->start, |
3173a18f | 5466 | level, 1, &wc->refs[level], |
2c47e605 | 5467 | &wc->flags[level]); |
79787eaa JM |
5468 | if (ret < 0) { |
5469 | err = ret; | |
5470 | goto out_end_trans; | |
5471 | } | |
2c47e605 YZ |
5472 | BUG_ON(wc->refs[level] == 0); |
5473 | ||
c8422684 | 5474 | if (level == btrfs_root_drop_level(root_item)) |
2c47e605 YZ |
5475 | break; |
5476 | ||
5477 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5478 | path->locks[level] = 0; |
2c47e605 YZ |
5479 | WARN_ON(wc->refs[level] != 1); |
5480 | level--; | |
5481 | } | |
9f3a7427 | 5482 | } |
2c47e605 | 5483 | |
78c52d9e | 5484 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5485 | wc->level = level; |
5486 | wc->shared_level = -1; | |
5487 | wc->stage = DROP_REFERENCE; | |
5488 | wc->update_ref = update_ref; | |
5489 | wc->keep_locks = 0; | |
0b246afa | 5490 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5491 | |
d397712b | 5492 | while (1) { |
9d1a2a3a | 5493 | |
2c47e605 YZ |
5494 | ret = walk_down_tree(trans, root, path, wc); |
5495 | if (ret < 0) { | |
5496 | err = ret; | |
20524f02 | 5497 | break; |
2c47e605 | 5498 | } |
9aca1d51 | 5499 | |
2c47e605 YZ |
5500 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5501 | if (ret < 0) { | |
5502 | err = ret; | |
20524f02 | 5503 | break; |
2c47e605 YZ |
5504 | } |
5505 | ||
5506 | if (ret > 0) { | |
5507 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5508 | break; |
5509 | } | |
2c47e605 YZ |
5510 | |
5511 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5512 | wc->drop_level = wc->level; |
5513 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5514 | &wc->drop_progress, | |
5515 | path->slots[wc->drop_level]); | |
5516 | } | |
5517 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5518 | &wc->drop_progress); | |
c8422684 | 5519 | btrfs_set_root_drop_level(root_item, wc->drop_level); |
2c47e605 YZ |
5520 | |
5521 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5522 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5523 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5524 | ret = btrfs_update_root(trans, tree_root, |
5525 | &root->root_key, | |
5526 | root_item); | |
79787eaa | 5527 | if (ret) { |
66642832 | 5528 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5529 | err = ret; |
5530 | goto out_end_trans; | |
5531 | } | |
2c47e605 | 5532 | |
3a45bb20 | 5533 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5534 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5535 | btrfs_debug(fs_info, |
5536 | "drop snapshot early exit"); | |
3c8f2422 JB |
5537 | err = -EAGAIN; |
5538 | goto out_free; | |
5539 | } | |
5540 | ||
18d3bff4 JB |
5541 | /* |
5542 | * Use join to avoid potential EINTR from transaction | |
5543 | * start. See wait_reserve_ticket and the whole | |
5544 | * reservation callchain. | |
5545 | */ | |
5546 | if (for_reloc) | |
5547 | trans = btrfs_join_transaction(tree_root); | |
5548 | else | |
5549 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5550 | if (IS_ERR(trans)) { |
5551 | err = PTR_ERR(trans); | |
5552 | goto out_free; | |
5553 | } | |
c3e69d58 | 5554 | } |
20524f02 | 5555 | } |
b3b4aa74 | 5556 | btrfs_release_path(path); |
79787eaa JM |
5557 | if (err) |
5558 | goto out_end_trans; | |
2c47e605 | 5559 | |
ab9ce7d4 | 5560 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5561 | if (ret) { |
66642832 | 5562 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5563 | err = ret; |
79787eaa JM |
5564 | goto out_end_trans; |
5565 | } | |
2c47e605 | 5566 | |
76dda93c | 5567 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5568 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5569 | NULL, NULL); | |
79787eaa | 5570 | if (ret < 0) { |
66642832 | 5571 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5572 | err = ret; |
5573 | goto out_end_trans; | |
5574 | } else if (ret > 0) { | |
84cd948c JB |
5575 | /* if we fail to delete the orphan item this time |
5576 | * around, it'll get picked up the next time. | |
5577 | * | |
5578 | * The most common failure here is just -ENOENT. | |
5579 | */ | |
5580 | btrfs_del_orphan_item(trans, tree_root, | |
5581 | root->root_key.objectid); | |
76dda93c YZ |
5582 | } |
5583 | } | |
5584 | ||
a3cf0e43 QW |
5585 | /* |
5586 | * This subvolume is going to be completely dropped, and won't be | |
5587 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
5588 | * commit transaction time. So free it here manually. | |
5589 | */ | |
5590 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
5591 | btrfs_qgroup_free_meta_all_pertrans(root); | |
5592 | ||
8c38938c | 5593 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 5594 | btrfs_add_dropped_root(trans, root); |
8c38938c | 5595 | else |
00246528 | 5596 | btrfs_put_root(root); |
d29a9f62 | 5597 | root_dropped = true; |
79787eaa | 5598 | out_end_trans: |
3a45bb20 | 5599 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5600 | out_free: |
2c47e605 | 5601 | kfree(wc); |
5caf2a00 | 5602 | btrfs_free_path(path); |
cb1b69f4 | 5603 | out: |
d29a9f62 JB |
5604 | /* |
5605 | * So if we need to stop dropping the snapshot for whatever reason we | |
5606 | * need to make sure to add it back to the dead root list so that we | |
5607 | * keep trying to do the work later. This also cleans up roots if we | |
5608 | * don't have it in the radix (like when we recover after a power fail | |
5609 | * or unmount) so we don't leak memory. | |
5610 | */ | |
897ca819 | 5611 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5612 | btrfs_add_dead_root(root); |
2c536799 | 5613 | return err; |
20524f02 | 5614 | } |
9078a3e1 | 5615 | |
2c47e605 YZ |
5616 | /* |
5617 | * drop subtree rooted at tree block 'node'. | |
5618 | * | |
5619 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5620 | * only used by relocation code |
2c47e605 | 5621 | */ |
f82d02d9 YZ |
5622 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5623 | struct btrfs_root *root, | |
5624 | struct extent_buffer *node, | |
5625 | struct extent_buffer *parent) | |
5626 | { | |
0b246afa | 5627 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5628 | struct btrfs_path *path; |
2c47e605 | 5629 | struct walk_control *wc; |
f82d02d9 YZ |
5630 | int level; |
5631 | int parent_level; | |
5632 | int ret = 0; | |
5633 | int wret; | |
5634 | ||
2c47e605 YZ |
5635 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5636 | ||
f82d02d9 | 5637 | path = btrfs_alloc_path(); |
db5b493a TI |
5638 | if (!path) |
5639 | return -ENOMEM; | |
f82d02d9 | 5640 | |
2c47e605 | 5641 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5642 | if (!wc) { |
5643 | btrfs_free_path(path); | |
5644 | return -ENOMEM; | |
5645 | } | |
2c47e605 | 5646 | |
b9447ef8 | 5647 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5648 | parent_level = btrfs_header_level(parent); |
67439dad | 5649 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5650 | path->nodes[parent_level] = parent; |
5651 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5652 | ||
b9447ef8 | 5653 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5654 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5655 | path->nodes[level] = node; |
5656 | path->slots[level] = 0; | |
ac5887c8 | 5657 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5658 | |
5659 | wc->refs[parent_level] = 1; | |
5660 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5661 | wc->level = level; | |
5662 | wc->shared_level = -1; | |
5663 | wc->stage = DROP_REFERENCE; | |
5664 | wc->update_ref = 0; | |
5665 | wc->keep_locks = 1; | |
0b246afa | 5666 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5667 | |
5668 | while (1) { | |
2c47e605 YZ |
5669 | wret = walk_down_tree(trans, root, path, wc); |
5670 | if (wret < 0) { | |
f82d02d9 | 5671 | ret = wret; |
f82d02d9 | 5672 | break; |
2c47e605 | 5673 | } |
f82d02d9 | 5674 | |
2c47e605 | 5675 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5676 | if (wret < 0) |
5677 | ret = wret; | |
5678 | if (wret != 0) | |
5679 | break; | |
5680 | } | |
5681 | ||
2c47e605 | 5682 | kfree(wc); |
f82d02d9 YZ |
5683 | btrfs_free_path(path); |
5684 | return ret; | |
5685 | } | |
5686 | ||
6d07bcec MX |
5687 | /* |
5688 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5689 | * space_info. takes mirrors into account. |
6d07bcec | 5690 | */ |
633c0aad | 5691 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5692 | { |
32da5386 | 5693 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5694 | u64 free_bytes = 0; |
5695 | int factor; | |
5696 | ||
01327610 | 5697 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5698 | if (list_empty(&sinfo->ro_bgs)) |
5699 | return 0; | |
5700 | ||
5701 | spin_lock(&sinfo->lock); | |
5702 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5703 | spin_lock(&block_group->lock); |
5704 | ||
5705 | if (!block_group->ro) { | |
5706 | spin_unlock(&block_group->lock); | |
5707 | continue; | |
5708 | } | |
5709 | ||
46df06b8 | 5710 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5711 | free_bytes += (block_group->length - |
bf38be65 | 5712 | block_group->used) * factor; |
6d07bcec MX |
5713 | |
5714 | spin_unlock(&block_group->lock); | |
5715 | } | |
6d07bcec MX |
5716 | spin_unlock(&sinfo->lock); |
5717 | ||
5718 | return free_bytes; | |
5719 | } | |
5720 | ||
2ff7e61e JM |
5721 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5722 | u64 start, u64 end) | |
acce952b | 5723 | { |
2ff7e61e | 5724 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5725 | } |
5726 | ||
499f377f JM |
5727 | /* |
5728 | * It used to be that old block groups would be left around forever. | |
5729 | * Iterating over them would be enough to trim unused space. Since we | |
5730 | * now automatically remove them, we also need to iterate over unallocated | |
5731 | * space. | |
5732 | * | |
5733 | * We don't want a transaction for this since the discard may take a | |
5734 | * substantial amount of time. We don't require that a transaction be | |
5735 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5736 | * to ensure that we're not discarding chunks that were released or |
5737 | * allocated in the current transaction. | |
499f377f JM |
5738 | * |
5739 | * Holding the chunks lock will prevent other threads from allocating | |
5740 | * or releasing chunks, but it won't prevent a running transaction | |
5741 | * from committing and releasing the memory that the pending chunks | |
5742 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5743 | * transaction and hold the commit root sem. We only need to hold |
5744 | * it while performing the free space search since we have already | |
5745 | * held back allocations. | |
499f377f | 5746 | */ |
8103d10b | 5747 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5748 | { |
8103d10b | 5749 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5750 | int ret; |
5751 | ||
5752 | *trimmed = 0; | |
5753 | ||
0be88e36 JM |
5754 | /* Discard not supported = nothing to do. */ |
5755 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5756 | return 0; | |
5757 | ||
52042d8e | 5758 | /* Not writable = nothing to do. */ |
ebbede42 | 5759 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5760 | return 0; |
5761 | ||
5762 | /* No free space = nothing to do. */ | |
5763 | if (device->total_bytes <= device->bytes_used) | |
5764 | return 0; | |
5765 | ||
5766 | ret = 0; | |
5767 | ||
5768 | while (1) { | |
fb456252 | 5769 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5770 | u64 bytes; |
5771 | ||
5772 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5773 | if (ret) | |
fee7acc3 | 5774 | break; |
499f377f | 5775 | |
929be17a NB |
5776 | find_first_clear_extent_bit(&device->alloc_state, start, |
5777 | &start, &end, | |
5778 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 5779 | |
c57dd1f2 QW |
5780 | /* Check if there are any CHUNK_* bits left */ |
5781 | if (start > device->total_bytes) { | |
5782 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5783 | btrfs_warn_in_rcu(fs_info, | |
5784 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
5785 | start, end - start + 1, | |
5786 | rcu_str_deref(device->name), | |
5787 | device->total_bytes); | |
5788 | mutex_unlock(&fs_info->chunk_mutex); | |
5789 | ret = 0; | |
5790 | break; | |
5791 | } | |
5792 | ||
53460a45 NB |
5793 | /* Ensure we skip the reserved area in the first 1M */ |
5794 | start = max_t(u64, start, SZ_1M); | |
5795 | ||
929be17a NB |
5796 | /* |
5797 | * If find_first_clear_extent_bit find a range that spans the | |
5798 | * end of the device it will set end to -1, in this case it's up | |
5799 | * to the caller to trim the value to the size of the device. | |
5800 | */ | |
5801 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5802 | |
929be17a | 5803 | len = end - start + 1; |
499f377f | 5804 | |
929be17a NB |
5805 | /* We didn't find any extents */ |
5806 | if (!len) { | |
499f377f | 5807 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5808 | ret = 0; |
499f377f JM |
5809 | break; |
5810 | } | |
5811 | ||
929be17a NB |
5812 | ret = btrfs_issue_discard(device->bdev, start, len, |
5813 | &bytes); | |
5814 | if (!ret) | |
5815 | set_extent_bits(&device->alloc_state, start, | |
5816 | start + bytes - 1, | |
5817 | CHUNK_TRIMMED); | |
499f377f JM |
5818 | mutex_unlock(&fs_info->chunk_mutex); |
5819 | ||
5820 | if (ret) | |
5821 | break; | |
5822 | ||
5823 | start += len; | |
5824 | *trimmed += bytes; | |
5825 | ||
5826 | if (fatal_signal_pending(current)) { | |
5827 | ret = -ERESTARTSYS; | |
5828 | break; | |
5829 | } | |
5830 | ||
5831 | cond_resched(); | |
5832 | } | |
5833 | ||
5834 | return ret; | |
5835 | } | |
5836 | ||
93bba24d QW |
5837 | /* |
5838 | * Trim the whole filesystem by: | |
5839 | * 1) trimming the free space in each block group | |
5840 | * 2) trimming the unallocated space on each device | |
5841 | * | |
5842 | * This will also continue trimming even if a block group or device encounters | |
5843 | * an error. The return value will be the last error, or 0 if nothing bad | |
5844 | * happens. | |
5845 | */ | |
2ff7e61e | 5846 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5847 | { |
32da5386 | 5848 | struct btrfs_block_group *cache = NULL; |
499f377f JM |
5849 | struct btrfs_device *device; |
5850 | struct list_head *devices; | |
f7039b1d | 5851 | u64 group_trimmed; |
07301df7 | 5852 | u64 range_end = U64_MAX; |
f7039b1d LD |
5853 | u64 start; |
5854 | u64 end; | |
5855 | u64 trimmed = 0; | |
93bba24d QW |
5856 | u64 bg_failed = 0; |
5857 | u64 dev_failed = 0; | |
5858 | int bg_ret = 0; | |
5859 | int dev_ret = 0; | |
f7039b1d LD |
5860 | int ret = 0; |
5861 | ||
07301df7 QW |
5862 | /* |
5863 | * Check range overflow if range->len is set. | |
5864 | * The default range->len is U64_MAX. | |
5865 | */ | |
5866 | if (range->len != U64_MAX && | |
5867 | check_add_overflow(range->start, range->len, &range_end)) | |
5868 | return -EINVAL; | |
5869 | ||
6ba9fc8e | 5870 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5871 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 5872 | if (cache->start >= range_end) { |
f7039b1d LD |
5873 | btrfs_put_block_group(cache); |
5874 | break; | |
5875 | } | |
5876 | ||
b3470b5d DS |
5877 | start = max(range->start, cache->start); |
5878 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
5879 | |
5880 | if (end - start >= range->minlen) { | |
32da5386 | 5881 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 5882 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 5883 | if (ret) { |
93bba24d QW |
5884 | bg_failed++; |
5885 | bg_ret = ret; | |
5886 | continue; | |
1be41b78 | 5887 | } |
676f1f75 | 5888 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5889 | if (ret) { |
93bba24d QW |
5890 | bg_failed++; |
5891 | bg_ret = ret; | |
5892 | continue; | |
1be41b78 | 5893 | } |
f7039b1d LD |
5894 | } |
5895 | ret = btrfs_trim_block_group(cache, | |
5896 | &group_trimmed, | |
5897 | start, | |
5898 | end, | |
5899 | range->minlen); | |
5900 | ||
5901 | trimmed += group_trimmed; | |
5902 | if (ret) { | |
93bba24d QW |
5903 | bg_failed++; |
5904 | bg_ret = ret; | |
5905 | continue; | |
f7039b1d LD |
5906 | } |
5907 | } | |
f7039b1d LD |
5908 | } |
5909 | ||
93bba24d QW |
5910 | if (bg_failed) |
5911 | btrfs_warn(fs_info, | |
5912 | "failed to trim %llu block group(s), last error %d", | |
5913 | bg_failed, bg_ret); | |
0b246afa | 5914 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5915 | devices = &fs_info->fs_devices->devices; |
5916 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5917 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5918 | if (ret) { |
5919 | dev_failed++; | |
5920 | dev_ret = ret; | |
499f377f | 5921 | break; |
93bba24d | 5922 | } |
499f377f JM |
5923 | |
5924 | trimmed += group_trimmed; | |
5925 | } | |
0b246afa | 5926 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5927 | |
93bba24d QW |
5928 | if (dev_failed) |
5929 | btrfs_warn(fs_info, | |
5930 | "failed to trim %llu device(s), last error %d", | |
5931 | dev_failed, dev_ret); | |
f7039b1d | 5932 | range->len = trimmed; |
93bba24d QW |
5933 | if (bg_ret) |
5934 | return bg_ret; | |
5935 | return dev_ret; | |
f7039b1d | 5936 | } |