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> |
cfc2de0f BB |
19 | #include "ctree.h" |
20 | #include "extent-tree.h" | |
2b712e3b | 21 | #include "transaction.h" |
fec577fb CM |
22 | #include "disk-io.h" |
23 | #include "print-tree.h" | |
0b86a832 | 24 | #include "volumes.h" |
53b381b3 | 25 | #include "raid56.h" |
925baedd | 26 | #include "locking.h" |
fa9c0d79 | 27 | #include "free-space-cache.h" |
1e144fb8 | 28 | #include "free-space-tree.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
8719aaae | 31 | #include "space-info.h" |
d12ffdd1 | 32 | #include "block-rsv.h" |
b0643e59 | 33 | #include "discard.h" |
169e0da9 | 34 | #include "zoned.h" |
6143c23c | 35 | #include "dev-replace.h" |
c7f13d42 | 36 | #include "fs.h" |
07e81dc9 | 37 | #include "accessors.h" |
45c40c8f | 38 | #include "root-tree.h" |
7c8ede16 | 39 | #include "file-item.h" |
aa5d3003 | 40 | #include "orphan.h" |
103c1972 | 41 | #include "tree-checker.h" |
02c372e1 | 42 | #include "raid-stripe-tree.h" |
fec577fb | 43 | |
709c0486 AJ |
44 | #undef SCRAMBLE_DELAYED_REFS |
45 | ||
9f9b8e8d | 46 | |
5d4f98a2 | 47 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
cecbb533 | 48 | struct btrfs_delayed_ref_head *href, |
85bb9f54 | 49 | struct btrfs_delayed_ref_node *node, |
e72cb923 | 50 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
51 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
52 | struct extent_buffer *leaf, | |
53 | struct btrfs_extent_item *ei); | |
54 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
55 | u64 parent, u64 root_objectid, |
56 | u64 flags, u64 owner, u64 offset, | |
2672a051 | 57 | struct btrfs_key *ins, int ref_mod, u64 oref_root); |
5d4f98a2 | 58 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 59 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 60 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
61 | static int find_next_key(struct btrfs_path *path, int level, |
62 | struct btrfs_key *key); | |
6a63209f | 63 | |
32da5386 | 64 | static int block_group_bits(struct btrfs_block_group *cache, u64 bits) |
0f9dd46c JB |
65 | { |
66 | return (cache->flags & bits) == bits; | |
67 | } | |
68 | ||
1a4ed8fd | 69 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 70 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 | 71 | { |
29cbcf40 | 72 | struct btrfs_root *root = btrfs_extent_root(fs_info, start); |
e02119d5 CM |
73 | int ret; |
74 | struct btrfs_key key; | |
31840ae1 | 75 | struct btrfs_path *path; |
e02119d5 | 76 | |
31840ae1 | 77 | path = btrfs_alloc_path(); |
d8926bb3 MF |
78 | if (!path) |
79 | return -ENOMEM; | |
80 | ||
e02119d5 CM |
81 | key.objectid = start; |
82 | key.offset = len; | |
3173a18f | 83 | key.type = BTRFS_EXTENT_ITEM_KEY; |
29cbcf40 | 84 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
31840ae1 | 85 | btrfs_free_path(path); |
7bb86316 CM |
86 | return ret; |
87 | } | |
88 | ||
a22285a6 | 89 | /* |
3173a18f | 90 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
91 | * |
92 | * the head node for delayed ref is used to store the sum of all the | |
93 | * reference count modifications queued up in the rbtree. the head | |
94 | * node may also store the extent flags to set. This way you can check | |
95 | * to see what the reference count and extent flags would be if all of | |
96 | * the delayed refs are not processed. | |
97 | */ | |
98 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 99 | struct btrfs_fs_info *fs_info, u64 bytenr, |
d8ba2a91 JB |
100 | u64 offset, int metadata, u64 *refs, u64 *flags, |
101 | u64 *owning_root) | |
a22285a6 | 102 | { |
29cbcf40 | 103 | struct btrfs_root *extent_root; |
a22285a6 YZ |
104 | struct btrfs_delayed_ref_head *head; |
105 | struct btrfs_delayed_ref_root *delayed_refs; | |
106 | struct btrfs_path *path; | |
107 | struct btrfs_extent_item *ei; | |
108 | struct extent_buffer *leaf; | |
109 | struct btrfs_key key; | |
110 | u32 item_size; | |
111 | u64 num_refs; | |
112 | u64 extent_flags; | |
d8ba2a91 | 113 | u64 owner = 0; |
a22285a6 YZ |
114 | int ret; |
115 | ||
3173a18f JB |
116 | /* |
117 | * If we don't have skinny metadata, don't bother doing anything | |
118 | * different | |
119 | */ | |
0b246afa JM |
120 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
121 | offset = fs_info->nodesize; | |
3173a18f JB |
122 | metadata = 0; |
123 | } | |
124 | ||
a22285a6 YZ |
125 | path = btrfs_alloc_path(); |
126 | if (!path) | |
127 | return -ENOMEM; | |
128 | ||
a22285a6 YZ |
129 | if (!trans) { |
130 | path->skip_locking = 1; | |
131 | path->search_commit_root = 1; | |
132 | } | |
639eefc8 FDBM |
133 | |
134 | search_again: | |
135 | key.objectid = bytenr; | |
136 | key.offset = offset; | |
137 | if (metadata) | |
138 | key.type = BTRFS_METADATA_ITEM_KEY; | |
139 | else | |
140 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
141 | ||
29cbcf40 JB |
142 | extent_root = btrfs_extent_root(fs_info, bytenr); |
143 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
a22285a6 YZ |
144 | if (ret < 0) |
145 | goto out_free; | |
146 | ||
3173a18f | 147 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
148 | if (path->slots[0]) { |
149 | path->slots[0]--; | |
150 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
151 | path->slots[0]); | |
152 | if (key.objectid == bytenr && | |
153 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 154 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
155 | ret = 0; |
156 | } | |
3173a18f JB |
157 | } |
158 | ||
a22285a6 YZ |
159 | if (ret == 0) { |
160 | leaf = path->nodes[0]; | |
3212fa14 | 161 | item_size = btrfs_item_size(leaf, path->slots[0]); |
a22285a6 YZ |
162 | if (item_size >= sizeof(*ei)) { |
163 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
164 | struct btrfs_extent_item); | |
165 | num_refs = btrfs_extent_refs(leaf, ei); | |
166 | extent_flags = btrfs_extent_flags(leaf, ei); | |
d8ba2a91 JB |
167 | owner = btrfs_get_extent_owner_root(fs_info, leaf, |
168 | path->slots[0]); | |
a22285a6 | 169 | } else { |
182741d2 QW |
170 | ret = -EUCLEAN; |
171 | btrfs_err(fs_info, | |
172 | "unexpected extent item size, has %u expect >= %zu", | |
173 | item_size, sizeof(*ei)); | |
ba3c2b19 NB |
174 | if (trans) |
175 | btrfs_abort_transaction(trans, ret); | |
176 | else | |
177 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
178 | ||
179 | goto out_free; | |
a22285a6 | 180 | } |
ba3c2b19 | 181 | |
a22285a6 YZ |
182 | BUG_ON(num_refs == 0); |
183 | } else { | |
184 | num_refs = 0; | |
185 | extent_flags = 0; | |
186 | ret = 0; | |
187 | } | |
188 | ||
189 | if (!trans) | |
190 | goto out; | |
191 | ||
192 | delayed_refs = &trans->transaction->delayed_refs; | |
193 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 194 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
195 | if (head) { |
196 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 197 | refcount_inc(&head->refs); |
a22285a6 YZ |
198 | spin_unlock(&delayed_refs->lock); |
199 | ||
b3b4aa74 | 200 | btrfs_release_path(path); |
a22285a6 | 201 | |
8cc33e5c DS |
202 | /* |
203 | * Mutex was contended, block until it's released and try | |
204 | * again | |
205 | */ | |
a22285a6 YZ |
206 | mutex_lock(&head->mutex); |
207 | mutex_unlock(&head->mutex); | |
d278850e | 208 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 209 | goto search_again; |
a22285a6 | 210 | } |
d7df2c79 | 211 | spin_lock(&head->lock); |
a22285a6 YZ |
212 | if (head->extent_op && head->extent_op->update_flags) |
213 | extent_flags |= head->extent_op->flags_to_set; | |
214 | else | |
215 | BUG_ON(num_refs == 0); | |
216 | ||
d278850e | 217 | num_refs += head->ref_mod; |
d7df2c79 | 218 | spin_unlock(&head->lock); |
a22285a6 YZ |
219 | mutex_unlock(&head->mutex); |
220 | } | |
221 | spin_unlock(&delayed_refs->lock); | |
222 | out: | |
223 | WARN_ON(num_refs == 0); | |
224 | if (refs) | |
225 | *refs = num_refs; | |
226 | if (flags) | |
227 | *flags = extent_flags; | |
d8ba2a91 JB |
228 | if (owning_root) |
229 | *owning_root = owner; | |
a22285a6 YZ |
230 | out_free: |
231 | btrfs_free_path(path); | |
232 | return ret; | |
233 | } | |
234 | ||
d8d5f3e1 CM |
235 | /* |
236 | * Back reference rules. Back refs have three main goals: | |
237 | * | |
238 | * 1) differentiate between all holders of references to an extent so that | |
239 | * when a reference is dropped we can make sure it was a valid reference | |
240 | * before freeing the extent. | |
241 | * | |
242 | * 2) Provide enough information to quickly find the holders of an extent | |
243 | * if we notice a given block is corrupted or bad. | |
244 | * | |
245 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
246 | * maintenance. This is actually the same as #2, but with a slightly | |
247 | * different use case. | |
248 | * | |
5d4f98a2 YZ |
249 | * There are two kinds of back refs. The implicit back refs is optimized |
250 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
251 | * back refs of this kind provide information about the block's owner tree | |
252 | * and the pointer's key. These information allow us to find the block by | |
253 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
254 | * referenced by their owner trees. The location of tree block is recorded | |
255 | * in the back refs. Actually the full back refs is generic, and can be | |
256 | * used in all cases the implicit back refs is used. The major shortcoming | |
257 | * of the full back refs is its overhead. Every time a tree block gets | |
258 | * COWed, we have to update back refs entry for all pointers in it. | |
259 | * | |
260 | * For a newly allocated tree block, we use implicit back refs for | |
261 | * pointers in it. This means most tree related operations only involve | |
262 | * implicit back refs. For a tree block created in old transaction, the | |
263 | * only way to drop a reference to it is COW it. So we can detect the | |
264 | * event that tree block loses its owner tree's reference and do the | |
265 | * back refs conversion. | |
266 | * | |
01327610 | 267 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
268 | * |
269 | * The reference count of the block is one and the tree is the block's | |
270 | * owner tree. Nothing to do in this case. | |
271 | * | |
272 | * The reference count of the block is one and the tree is not the | |
273 | * block's owner tree. In this case, full back refs is used for pointers | |
274 | * in the block. Remove these full back refs, add implicit back refs for | |
275 | * every pointers in the new block. | |
276 | * | |
277 | * The reference count of the block is greater than one and the tree is | |
278 | * the block's owner tree. In this case, implicit back refs is used for | |
279 | * pointers in the block. Add full back refs for every pointers in the | |
280 | * block, increase lower level extents' reference counts. The original | |
281 | * implicit back refs are entailed to the new block. | |
282 | * | |
283 | * The reference count of the block is greater than one and the tree is | |
284 | * not the block's owner tree. Add implicit back refs for every pointer in | |
285 | * the new block, increase lower level extents' reference count. | |
286 | * | |
287 | * Back Reference Key composing: | |
288 | * | |
289 | * The key objectid corresponds to the first byte in the extent, | |
290 | * The key type is used to differentiate between types of back refs. | |
291 | * There are different meanings of the key offset for different types | |
292 | * of back refs. | |
293 | * | |
d8d5f3e1 CM |
294 | * File extents can be referenced by: |
295 | * | |
296 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 297 | * - different files inside a single subvolume |
d8d5f3e1 CM |
298 | * - different offsets inside a file (bookend extents in file.c) |
299 | * | |
5d4f98a2 | 300 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
301 | * |
302 | * - Objectid of the subvolume root | |
d8d5f3e1 | 303 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
304 | * - original offset in the file |
305 | * - how many bookend extents | |
d8d5f3e1 | 306 | * |
5d4f98a2 YZ |
307 | * The key offset for the implicit back refs is hash of the first |
308 | * three fields. | |
d8d5f3e1 | 309 | * |
5d4f98a2 | 310 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 311 | * |
5d4f98a2 | 312 | * - number of pointers in the tree leaf |
d8d5f3e1 | 313 | * |
5d4f98a2 YZ |
314 | * The key offset for the implicit back refs is the first byte of |
315 | * the tree leaf | |
d8d5f3e1 | 316 | * |
5d4f98a2 YZ |
317 | * When a file extent is allocated, The implicit back refs is used. |
318 | * the fields are filled in: | |
d8d5f3e1 | 319 | * |
5d4f98a2 | 320 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 321 | * |
5d4f98a2 YZ |
322 | * When a file extent is removed file truncation, we find the |
323 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 324 | * |
5d4f98a2 | 325 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 326 | * |
5d4f98a2 | 327 | * Btree extents can be referenced by: |
d8d5f3e1 | 328 | * |
5d4f98a2 | 329 | * - Different subvolumes |
d8d5f3e1 | 330 | * |
5d4f98a2 YZ |
331 | * Both the implicit back refs and the full back refs for tree blocks |
332 | * only consist of key. The key offset for the implicit back refs is | |
333 | * objectid of block's owner tree. The key offset for the full back refs | |
334 | * is the first byte of parent block. | |
d8d5f3e1 | 335 | * |
5d4f98a2 YZ |
336 | * When implicit back refs is used, information about the lowest key and |
337 | * level of the tree block are required. These information are stored in | |
338 | * tree block info structure. | |
d8d5f3e1 | 339 | */ |
31840ae1 | 340 | |
167ce953 LB |
341 | /* |
342 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 343 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
344 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
345 | */ | |
346 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
347 | struct btrfs_extent_inline_ref *iref, | |
348 | enum btrfs_inline_ref_type is_data) | |
349 | { | |
d9a620f7 | 350 | struct btrfs_fs_info *fs_info = eb->fs_info; |
167ce953 | 351 | int type = btrfs_extent_inline_ref_type(eb, iref); |
64ecdb64 | 352 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 | 353 | |
d9a620f7 BB |
354 | if (type == BTRFS_EXTENT_OWNER_REF_KEY) { |
355 | ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA)); | |
356 | return type; | |
357 | } | |
358 | ||
167ce953 LB |
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 | 366 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
d9a620f7 | 367 | ASSERT(fs_info); |
64ecdb64 | 368 | /* |
ea57788e QW |
369 | * Every shared one has parent tree block, |
370 | * which must be aligned to sector size. | |
64ecdb64 | 371 | */ |
d9a620f7 | 372 | if (offset && IS_ALIGNED(offset, fs_info->sectorsize)) |
64ecdb64 LB |
373 | return type; |
374 | } | |
167ce953 | 375 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 376 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 377 | return type; |
64ecdb64 | 378 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
d9a620f7 | 379 | ASSERT(fs_info); |
64ecdb64 | 380 | /* |
ea57788e QW |
381 | * Every shared one has parent tree block, |
382 | * which must be aligned to sector size. | |
64ecdb64 LB |
383 | */ |
384 | if (offset && | |
d9a620f7 | 385 | IS_ALIGNED(offset, fs_info->sectorsize)) |
64ecdb64 LB |
386 | return type; |
387 | } | |
167ce953 LB |
388 | } else { |
389 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
390 | return type; | |
391 | } | |
392 | } | |
393 | ||
25761430 | 394 | WARN_ON(1); |
6c75a589 | 395 | btrfs_print_leaf(eb); |
d9a620f7 | 396 | btrfs_err(fs_info, |
ea57788e QW |
397 | "eb %llu iref 0x%lx invalid extent inline ref type %d", |
398 | eb->start, (unsigned long)iref, type); | |
167ce953 LB |
399 | |
400 | return BTRFS_REF_TYPE_INVALID; | |
401 | } | |
402 | ||
0785a9aa | 403 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
404 | { |
405 | u32 high_crc = ~(u32)0; | |
406 | u32 low_crc = ~(u32)0; | |
407 | __le64 lenum; | |
408 | ||
409 | lenum = cpu_to_le64(root_objectid); | |
03e86348 | 410 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 411 | lenum = cpu_to_le64(owner); |
03e86348 | 412 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 413 | lenum = cpu_to_le64(offset); |
03e86348 | 414 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
415 | |
416 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
417 | } | |
418 | ||
419 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
420 | struct btrfs_extent_data_ref *ref) | |
421 | { | |
422 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
423 | btrfs_extent_data_ref_objectid(leaf, ref), | |
424 | btrfs_extent_data_ref_offset(leaf, ref)); | |
425 | } | |
426 | ||
427 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
428 | struct btrfs_extent_data_ref *ref, | |
429 | u64 root_objectid, u64 owner, u64 offset) | |
430 | { | |
431 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
432 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
433 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
434 | return 0; | |
435 | return 1; | |
436 | } | |
437 | ||
438 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
439 | struct btrfs_path *path, |
440 | u64 bytenr, u64 parent, | |
441 | u64 root_objectid, | |
442 | u64 owner, u64 offset) | |
443 | { | |
29cbcf40 | 444 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
5d4f98a2 YZ |
445 | struct btrfs_key key; |
446 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 447 | struct extent_buffer *leaf; |
5d4f98a2 | 448 | u32 nritems; |
5d4f98a2 | 449 | int recow; |
1618aa3c | 450 | int ret; |
74493f7a | 451 | |
31840ae1 | 452 | key.objectid = bytenr; |
5d4f98a2 YZ |
453 | if (parent) { |
454 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
455 | key.offset = parent; | |
456 | } else { | |
457 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
458 | key.offset = hash_extent_data_ref(root_objectid, | |
459 | owner, offset); | |
460 | } | |
461 | again: | |
462 | recow = 0; | |
463 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1618aa3c AJ |
464 | if (ret < 0) |
465 | return ret; | |
31840ae1 | 466 | |
5d4f98a2 | 467 | if (parent) { |
1618aa3c AJ |
468 | if (ret) |
469 | return -ENOENT; | |
470 | return 0; | |
31840ae1 ZY |
471 | } |
472 | ||
1618aa3c | 473 | ret = -ENOENT; |
31840ae1 | 474 | leaf = path->nodes[0]; |
5d4f98a2 YZ |
475 | nritems = btrfs_header_nritems(leaf); |
476 | while (1) { | |
477 | if (path->slots[0] >= nritems) { | |
478 | ret = btrfs_next_leaf(root, path); | |
1618aa3c AJ |
479 | if (ret) { |
480 | if (ret > 1) | |
481 | return -ENOENT; | |
482 | return ret; | |
483 | } | |
5d4f98a2 YZ |
484 | |
485 | leaf = path->nodes[0]; | |
486 | nritems = btrfs_header_nritems(leaf); | |
487 | recow = 1; | |
488 | } | |
489 | ||
490 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
491 | if (key.objectid != bytenr || | |
492 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
493 | goto fail; | |
494 | ||
495 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
496 | struct btrfs_extent_data_ref); | |
497 | ||
498 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
499 | owner, offset)) { | |
500 | if (recow) { | |
b3b4aa74 | 501 | btrfs_release_path(path); |
5d4f98a2 YZ |
502 | goto again; |
503 | } | |
1618aa3c | 504 | ret = 0; |
5d4f98a2 YZ |
505 | break; |
506 | } | |
507 | path->slots[0]++; | |
31840ae1 | 508 | } |
5d4f98a2 | 509 | fail: |
1618aa3c | 510 | return ret; |
31840ae1 ZY |
511 | } |
512 | ||
5d4f98a2 | 513 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 514 | struct btrfs_path *path, |
b4b5934a JB |
515 | struct btrfs_delayed_ref_node *node, |
516 | u64 bytenr) | |
31840ae1 | 517 | { |
29cbcf40 | 518 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
31840ae1 ZY |
519 | struct btrfs_key key; |
520 | struct extent_buffer *leaf; | |
b4b5934a JB |
521 | u64 owner = btrfs_delayed_ref_owner(node); |
522 | u64 offset = btrfs_delayed_ref_offset(node); | |
5d4f98a2 | 523 | u32 size; |
31840ae1 ZY |
524 | u32 num_refs; |
525 | int ret; | |
74493f7a | 526 | |
74493f7a | 527 | key.objectid = bytenr; |
b4b5934a | 528 | if (node->parent) { |
5d4f98a2 | 529 | key.type = BTRFS_SHARED_DATA_REF_KEY; |
b4b5934a | 530 | key.offset = node->parent; |
5d4f98a2 YZ |
531 | size = sizeof(struct btrfs_shared_data_ref); |
532 | } else { | |
533 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
b4b5934a | 534 | key.offset = hash_extent_data_ref(node->ref_root, owner, offset); |
5d4f98a2 YZ |
535 | size = sizeof(struct btrfs_extent_data_ref); |
536 | } | |
74493f7a | 537 | |
5d4f98a2 YZ |
538 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
539 | if (ret && ret != -EEXIST) | |
540 | goto fail; | |
541 | ||
542 | leaf = path->nodes[0]; | |
b4b5934a | 543 | if (node->parent) { |
5d4f98a2 | 544 | struct btrfs_shared_data_ref *ref; |
31840ae1 | 545 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
546 | struct btrfs_shared_data_ref); |
547 | if (ret == 0) { | |
b4b5934a | 548 | btrfs_set_shared_data_ref_count(leaf, ref, node->ref_mod); |
5d4f98a2 YZ |
549 | } else { |
550 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
b4b5934a | 551 | num_refs += node->ref_mod; |
5d4f98a2 | 552 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); |
31840ae1 | 553 | } |
5d4f98a2 YZ |
554 | } else { |
555 | struct btrfs_extent_data_ref *ref; | |
556 | while (ret == -EEXIST) { | |
557 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
558 | struct btrfs_extent_data_ref); | |
b4b5934a | 559 | if (match_extent_data_ref(leaf, ref, node->ref_root, |
5d4f98a2 YZ |
560 | owner, offset)) |
561 | break; | |
b3b4aa74 | 562 | btrfs_release_path(path); |
5d4f98a2 YZ |
563 | key.offset++; |
564 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
565 | size); | |
566 | if (ret && ret != -EEXIST) | |
567 | goto fail; | |
31840ae1 | 568 | |
5d4f98a2 YZ |
569 | leaf = path->nodes[0]; |
570 | } | |
571 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
572 | struct btrfs_extent_data_ref); | |
573 | if (ret == 0) { | |
b4b5934a | 574 | btrfs_set_extent_data_ref_root(leaf, ref, node->ref_root); |
5d4f98a2 YZ |
575 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); |
576 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
b4b5934a | 577 | btrfs_set_extent_data_ref_count(leaf, ref, node->ref_mod); |
5d4f98a2 YZ |
578 | } else { |
579 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
b4b5934a | 580 | num_refs += node->ref_mod; |
5d4f98a2 | 581 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); |
31840ae1 | 582 | } |
31840ae1 | 583 | } |
50564b65 | 584 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
585 | ret = 0; |
586 | fail: | |
b3b4aa74 | 587 | btrfs_release_path(path); |
7bb86316 | 588 | return ret; |
74493f7a CM |
589 | } |
590 | ||
5d4f98a2 | 591 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
76d76e78 | 592 | struct btrfs_root *root, |
5d4f98a2 | 593 | struct btrfs_path *path, |
5b2a54bb | 594 | int refs_to_drop) |
31840ae1 | 595 | { |
5d4f98a2 YZ |
596 | struct btrfs_key key; |
597 | struct btrfs_extent_data_ref *ref1 = NULL; | |
598 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 599 | struct extent_buffer *leaf; |
5d4f98a2 | 600 | u32 num_refs = 0; |
31840ae1 ZY |
601 | int ret = 0; |
602 | ||
603 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
604 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
605 | ||
606 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
607 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
608 | struct btrfs_extent_data_ref); | |
609 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
610 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
611 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
612 | struct btrfs_shared_data_ref); | |
613 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 | 614 | } else { |
182741d2 QW |
615 | btrfs_err(trans->fs_info, |
616 | "unrecognized backref key (%llu %u %llu)", | |
617 | key.objectid, key.type, key.offset); | |
618 | btrfs_abort_transaction(trans, -EUCLEAN); | |
619 | return -EUCLEAN; | |
5d4f98a2 YZ |
620 | } |
621 | ||
56bec294 CM |
622 | BUG_ON(num_refs < refs_to_drop); |
623 | num_refs -= refs_to_drop; | |
5d4f98a2 | 624 | |
31840ae1 | 625 | if (num_refs == 0) { |
76d76e78 | 626 | ret = btrfs_del_item(trans, root, path); |
31840ae1 | 627 | } else { |
5d4f98a2 YZ |
628 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
629 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
630 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
631 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
50564b65 | 632 | btrfs_mark_buffer_dirty(trans, leaf); |
31840ae1 | 633 | } |
31840ae1 ZY |
634 | return ret; |
635 | } | |
636 | ||
9ed0dea0 | 637 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 638 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 639 | { |
5d4f98a2 YZ |
640 | struct btrfs_key key; |
641 | struct extent_buffer *leaf; | |
642 | struct btrfs_extent_data_ref *ref1; | |
643 | struct btrfs_shared_data_ref *ref2; | |
644 | u32 num_refs = 0; | |
3de28d57 | 645 | int type; |
5d4f98a2 YZ |
646 | |
647 | leaf = path->nodes[0]; | |
648 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 | 649 | |
5d4f98a2 | 650 | if (iref) { |
3de28d57 LB |
651 | /* |
652 | * If type is invalid, we should have bailed out earlier than | |
653 | * this call. | |
654 | */ | |
655 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
656 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
657 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
658 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
659 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
660 | } else { | |
661 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
662 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
663 | } | |
664 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
665 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
666 | struct btrfs_extent_data_ref); | |
667 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
668 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
669 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
670 | struct btrfs_shared_data_ref); | |
671 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
672 | } else { |
673 | WARN_ON(1); | |
674 | } | |
675 | return num_refs; | |
676 | } | |
15916de8 | 677 | |
5d4f98a2 | 678 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
679 | struct btrfs_path *path, |
680 | u64 bytenr, u64 parent, | |
681 | u64 root_objectid) | |
1f3c79a2 | 682 | { |
29cbcf40 | 683 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
5d4f98a2 | 684 | struct btrfs_key key; |
1f3c79a2 | 685 | int ret; |
1f3c79a2 | 686 | |
5d4f98a2 YZ |
687 | key.objectid = bytenr; |
688 | if (parent) { | |
689 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
690 | key.offset = parent; | |
691 | } else { | |
692 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
693 | key.offset = root_objectid; | |
1f3c79a2 LH |
694 | } |
695 | ||
5d4f98a2 YZ |
696 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
697 | if (ret > 0) | |
698 | ret = -ENOENT; | |
5d4f98a2 | 699 | return ret; |
1f3c79a2 LH |
700 | } |
701 | ||
5d4f98a2 | 702 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 703 | struct btrfs_path *path, |
b4b5934a JB |
704 | struct btrfs_delayed_ref_node *node, |
705 | u64 bytenr) | |
31840ae1 | 706 | { |
29cbcf40 | 707 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
5d4f98a2 | 708 | struct btrfs_key key; |
31840ae1 | 709 | int ret; |
31840ae1 | 710 | |
5d4f98a2 | 711 | key.objectid = bytenr; |
b4b5934a | 712 | if (node->parent) { |
5d4f98a2 | 713 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; |
b4b5934a | 714 | key.offset = node->parent; |
5d4f98a2 YZ |
715 | } else { |
716 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
b4b5934a | 717 | key.offset = node->ref_root; |
5d4f98a2 YZ |
718 | } |
719 | ||
29cbcf40 | 720 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); |
b3b4aa74 | 721 | btrfs_release_path(path); |
31840ae1 ZY |
722 | return ret; |
723 | } | |
724 | ||
5d4f98a2 | 725 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 726 | { |
5d4f98a2 YZ |
727 | int type; |
728 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
729 | if (parent > 0) | |
730 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
731 | else | |
732 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
733 | } else { | |
734 | if (parent > 0) | |
735 | type = BTRFS_SHARED_DATA_REF_KEY; | |
736 | else | |
737 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
738 | } | |
739 | return type; | |
31840ae1 | 740 | } |
56bec294 | 741 | |
2c47e605 YZ |
742 | static int find_next_key(struct btrfs_path *path, int level, |
743 | struct btrfs_key *key) | |
56bec294 | 744 | |
02217ed2 | 745 | { |
2c47e605 | 746 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
747 | if (!path->nodes[level]) |
748 | break; | |
5d4f98a2 YZ |
749 | if (path->slots[level] + 1 >= |
750 | btrfs_header_nritems(path->nodes[level])) | |
751 | continue; | |
752 | if (level == 0) | |
753 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
754 | path->slots[level] + 1); | |
755 | else | |
756 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
757 | path->slots[level] + 1); | |
758 | return 0; | |
759 | } | |
760 | return 1; | |
761 | } | |
037e6390 | 762 | |
5d4f98a2 YZ |
763 | /* |
764 | * look for inline back ref. if back ref is found, *ref_ret is set | |
765 | * to the address of inline back ref, and 0 is returned. | |
766 | * | |
767 | * if back ref isn't found, *ref_ret is set to the address where it | |
768 | * should be inserted, and -ENOENT is returned. | |
769 | * | |
770 | * if insert is true and there are too many inline back refs, the path | |
771 | * points to the extent item, and -EAGAIN is returned. | |
772 | * | |
773 | * NOTE: inline back refs are ordered in the same way that back ref | |
774 | * items in the tree are ordered. | |
775 | */ | |
776 | static noinline_for_stack | |
777 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
778 | struct btrfs_path *path, |
779 | struct btrfs_extent_inline_ref **ref_ret, | |
780 | u64 bytenr, u64 num_bytes, | |
781 | u64 parent, u64 root_objectid, | |
782 | u64 owner, u64 offset, int insert) | |
783 | { | |
867cc1fb | 784 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 785 | struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr); |
5d4f98a2 YZ |
786 | struct btrfs_key key; |
787 | struct extent_buffer *leaf; | |
788 | struct btrfs_extent_item *ei; | |
789 | struct btrfs_extent_inline_ref *iref; | |
790 | u64 flags; | |
791 | u64 item_size; | |
792 | unsigned long ptr; | |
793 | unsigned long end; | |
794 | int extra_size; | |
795 | int type; | |
796 | int want; | |
797 | int ret; | |
0b246afa | 798 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 799 | int needed; |
26b8003f | 800 | |
db94535d | 801 | key.objectid = bytenr; |
31840ae1 | 802 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 803 | key.offset = num_bytes; |
31840ae1 | 804 | |
5d4f98a2 YZ |
805 | want = extent_ref_type(parent, owner); |
806 | if (insert) { | |
807 | extra_size = btrfs_extent_inline_ref_size(want); | |
9a664971 | 808 | path->search_for_extension = 1; |
85d4198e | 809 | path->keep_locks = 1; |
5d4f98a2 YZ |
810 | } else |
811 | extra_size = -1; | |
3173a18f JB |
812 | |
813 | /* | |
16d1c062 NB |
814 | * Owner is our level, so we can just add one to get the level for the |
815 | * block we are interested in. | |
3173a18f JB |
816 | */ |
817 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
818 | key.type = BTRFS_METADATA_ITEM_KEY; | |
819 | key.offset = owner; | |
820 | } | |
821 | ||
822 | again: | |
5d4f98a2 | 823 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
cc925b96 | 824 | if (ret < 0) |
5d4f98a2 | 825 | goto out; |
3173a18f JB |
826 | |
827 | /* | |
828 | * We may be a newly converted file system which still has the old fat | |
829 | * extent entries for metadata, so try and see if we have one of those. | |
830 | */ | |
831 | if (ret > 0 && skinny_metadata) { | |
832 | skinny_metadata = false; | |
833 | if (path->slots[0]) { | |
834 | path->slots[0]--; | |
835 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
836 | path->slots[0]); | |
837 | if (key.objectid == bytenr && | |
838 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
839 | key.offset == num_bytes) | |
840 | ret = 0; | |
841 | } | |
842 | if (ret) { | |
9ce49a0b | 843 | key.objectid = bytenr; |
3173a18f JB |
844 | key.type = BTRFS_EXTENT_ITEM_KEY; |
845 | key.offset = num_bytes; | |
846 | btrfs_release_path(path); | |
847 | goto again; | |
848 | } | |
849 | } | |
850 | ||
79787eaa | 851 | if (ret && !insert) { |
cc925b96 | 852 | ret = -ENOENT; |
79787eaa | 853 | goto out; |
fae7f21c | 854 | } else if (WARN_ON(ret)) { |
7f72f505 QW |
855 | btrfs_print_leaf(path->nodes[0]); |
856 | btrfs_err(fs_info, | |
857 | "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu", | |
858 | bytenr, num_bytes, parent, root_objectid, owner, | |
859 | offset); | |
eba444f1 | 860 | ret = -EUCLEAN; |
492104c8 | 861 | goto out; |
79787eaa | 862 | } |
5d4f98a2 YZ |
863 | |
864 | leaf = path->nodes[0]; | |
3212fa14 | 865 | item_size = btrfs_item_size(leaf, path->slots[0]); |
6d8ff4e4 | 866 | if (unlikely(item_size < sizeof(*ei))) { |
cc925b96 | 867 | ret = -EUCLEAN; |
182741d2 QW |
868 | btrfs_err(fs_info, |
869 | "unexpected extent item size, has %llu expect >= %zu", | |
870 | item_size, sizeof(*ei)); | |
cc925b96 | 871 | btrfs_abort_transaction(trans, ret); |
ba3c2b19 NB |
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 | ||
cc925b96 | 891 | ret = -ENOENT; |
da8848ac | 892 | while (ptr < end) { |
5d4f98a2 | 893 | iref = (struct btrfs_extent_inline_ref *)ptr; |
3de28d57 | 894 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
d9a620f7 BB |
895 | if (type == BTRFS_EXTENT_OWNER_REF_KEY) { |
896 | ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA)); | |
897 | ptr += btrfs_extent_inline_ref_size(type); | |
898 | continue; | |
899 | } | |
3de28d57 | 900 | if (type == BTRFS_REF_TYPE_INVALID) { |
cc925b96 | 901 | ret = -EUCLEAN; |
3de28d57 LB |
902 | goto out; |
903 | } | |
904 | ||
5d4f98a2 YZ |
905 | if (want < type) |
906 | break; | |
907 | if (want > type) { | |
908 | ptr += btrfs_extent_inline_ref_size(type); | |
909 | continue; | |
910 | } | |
911 | ||
912 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
913 | struct btrfs_extent_data_ref *dref; | |
914 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
915 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
916 | owner, offset)) { | |
cc925b96 | 917 | ret = 0; |
5d4f98a2 YZ |
918 | break; |
919 | } | |
920 | if (hash_extent_data_ref_item(leaf, dref) < | |
921 | hash_extent_data_ref(root_objectid, owner, offset)) | |
922 | break; | |
923 | } else { | |
924 | u64 ref_offset; | |
925 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
926 | if (parent > 0) { | |
927 | if (parent == ref_offset) { | |
cc925b96 | 928 | ret = 0; |
5d4f98a2 YZ |
929 | break; |
930 | } | |
931 | if (ref_offset < parent) | |
932 | break; | |
933 | } else { | |
934 | if (root_objectid == ref_offset) { | |
cc925b96 | 935 | ret = 0; |
5d4f98a2 YZ |
936 | break; |
937 | } | |
938 | if (ref_offset < root_objectid) | |
939 | break; | |
940 | } | |
941 | } | |
942 | ptr += btrfs_extent_inline_ref_size(type); | |
943 | } | |
da8848ac FM |
944 | |
945 | if (unlikely(ptr > end)) { | |
946 | ret = -EUCLEAN; | |
947 | btrfs_print_leaf(path->nodes[0]); | |
948 | btrfs_crit(fs_info, | |
949 | "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu", | |
950 | path->slots[0], root_objectid, owner, offset, parent); | |
951 | goto out; | |
952 | } | |
953 | ||
cc925b96 | 954 | if (ret == -ENOENT && insert) { |
5d4f98a2 YZ |
955 | if (item_size + extra_size >= |
956 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
cc925b96 | 957 | ret = -EAGAIN; |
5d4f98a2 YZ |
958 | goto out; |
959 | } | |
960 | /* | |
961 | * To add new inline back ref, we have to make sure | |
962 | * there is no corresponding back ref item. | |
963 | * For simplicity, we just do not add new inline back | |
964 | * ref if there is any kind of item for this block | |
965 | */ | |
2c47e605 YZ |
966 | if (find_next_key(path, 0, &key) == 0 && |
967 | key.objectid == bytenr && | |
85d4198e | 968 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
cc925b96 | 969 | ret = -EAGAIN; |
5d4f98a2 YZ |
970 | goto out; |
971 | } | |
972 | } | |
973 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
974 | out: | |
85d4198e | 975 | if (insert) { |
5d4f98a2 | 976 | path->keep_locks = 0; |
9a664971 | 977 | path->search_for_extension = 0; |
5d4f98a2 YZ |
978 | btrfs_unlock_up_safe(path, 1); |
979 | } | |
cc925b96 | 980 | return ret; |
5d4f98a2 YZ |
981 | } |
982 | ||
983 | /* | |
984 | * helper to add new inline back ref | |
985 | */ | |
986 | static noinline_for_stack | |
50564b65 | 987 | void setup_inline_extent_backref(struct btrfs_trans_handle *trans, |
143bede5 JM |
988 | struct btrfs_path *path, |
989 | struct btrfs_extent_inline_ref *iref, | |
990 | u64 parent, u64 root_objectid, | |
991 | u64 owner, u64 offset, int refs_to_add, | |
992 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
993 | { |
994 | struct extent_buffer *leaf; | |
995 | struct btrfs_extent_item *ei; | |
996 | unsigned long ptr; | |
997 | unsigned long end; | |
998 | unsigned long item_offset; | |
999 | u64 refs; | |
1000 | int size; | |
1001 | int type; | |
5d4f98a2 YZ |
1002 | |
1003 | leaf = path->nodes[0]; | |
1004 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1005 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1006 | ||
1007 | type = extent_ref_type(parent, owner); | |
1008 | size = btrfs_extent_inline_ref_size(type); | |
1009 | ||
50564b65 | 1010 | btrfs_extend_item(trans, path, size); |
5d4f98a2 YZ |
1011 | |
1012 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1013 | refs = btrfs_extent_refs(leaf, ei); | |
1014 | refs += refs_to_add; | |
1015 | btrfs_set_extent_refs(leaf, ei, refs); | |
1016 | if (extent_op) | |
1017 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1018 | ||
1019 | ptr = (unsigned long)ei + item_offset; | |
3212fa14 | 1020 | end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]); |
5d4f98a2 YZ |
1021 | if (ptr < end - size) |
1022 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1023 | end - size - ptr); | |
1024 | ||
1025 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1026 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1027 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1028 | struct btrfs_extent_data_ref *dref; | |
1029 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1030 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1031 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1032 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1033 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1034 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1035 | struct btrfs_shared_data_ref *sref; | |
1036 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1037 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1038 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1039 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1040 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1041 | } else { | |
1042 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1043 | } | |
50564b65 | 1044 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
1045 | } |
1046 | ||
1047 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1048 | struct btrfs_path *path, |
1049 | struct btrfs_extent_inline_ref **ref_ret, | |
1050 | u64 bytenr, u64 num_bytes, u64 parent, | |
1051 | u64 root_objectid, u64 owner, u64 offset) | |
1052 | { | |
1053 | int ret; | |
1054 | ||
867cc1fb NB |
1055 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1056 | num_bytes, parent, root_objectid, | |
1057 | owner, offset, 0); | |
5d4f98a2 | 1058 | if (ret != -ENOENT) |
54aa1f4d | 1059 | return ret; |
5d4f98a2 | 1060 | |
b3b4aa74 | 1061 | btrfs_release_path(path); |
5d4f98a2 YZ |
1062 | *ref_ret = NULL; |
1063 | ||
1064 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1065 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1066 | root_objectid); | |
5d4f98a2 | 1067 | } else { |
bd1d53ef NB |
1068 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1069 | root_objectid, owner, offset); | |
b9473439 | 1070 | } |
5d4f98a2 YZ |
1071 | return ret; |
1072 | } | |
31840ae1 | 1073 | |
5d4f98a2 YZ |
1074 | /* |
1075 | * helper to update/remove inline back ref | |
1076 | */ | |
50564b65 FM |
1077 | static noinline_for_stack int update_inline_extent_backref( |
1078 | struct btrfs_trans_handle *trans, | |
1079 | struct btrfs_path *path, | |
143bede5 JM |
1080 | struct btrfs_extent_inline_ref *iref, |
1081 | int refs_to_mod, | |
5b2a54bb | 1082 | struct btrfs_delayed_extent_op *extent_op) |
5d4f98a2 | 1083 | { |
61a18f1c | 1084 | struct extent_buffer *leaf = path->nodes[0]; |
25761430 | 1085 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5d4f98a2 YZ |
1086 | struct btrfs_extent_item *ei; |
1087 | struct btrfs_extent_data_ref *dref = NULL; | |
1088 | struct btrfs_shared_data_ref *sref = NULL; | |
1089 | unsigned long ptr; | |
1090 | unsigned long end; | |
1091 | u32 item_size; | |
1092 | int size; | |
1093 | int type; | |
5d4f98a2 YZ |
1094 | u64 refs; |
1095 | ||
5d4f98a2 YZ |
1096 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1097 | refs = btrfs_extent_refs(leaf, ei); | |
25761430 QW |
1098 | if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) { |
1099 | struct btrfs_key key; | |
1100 | u32 extent_size; | |
1101 | ||
1102 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1103 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
1104 | extent_size = fs_info->nodesize; | |
1105 | else | |
1106 | extent_size = key.offset; | |
1107 | btrfs_print_leaf(leaf); | |
1108 | btrfs_err(fs_info, | |
1109 | "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu", | |
1110 | key.objectid, extent_size, refs_to_mod, refs); | |
1111 | return -EUCLEAN; | |
1112 | } | |
5d4f98a2 YZ |
1113 | refs += refs_to_mod; |
1114 | btrfs_set_extent_refs(leaf, ei, refs); | |
1115 | if (extent_op) | |
1116 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1117 | ||
25761430 | 1118 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); |
3de28d57 | 1119 | /* |
25761430 QW |
1120 | * Function btrfs_get_extent_inline_ref_type() has already printed |
1121 | * error messages. | |
3de28d57 | 1122 | */ |
25761430 QW |
1123 | if (unlikely(type == BTRFS_REF_TYPE_INVALID)) |
1124 | return -EUCLEAN; | |
5d4f98a2 YZ |
1125 | |
1126 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1127 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1128 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1129 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1130 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1131 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1132 | } else { | |
1133 | refs = 1; | |
25761430 QW |
1134 | /* |
1135 | * For tree blocks we can only drop one ref for it, and tree | |
1136 | * blocks should not have refs > 1. | |
1137 | * | |
1138 | * Furthermore if we're inserting a new inline backref, we | |
1139 | * won't reach this path either. That would be | |
1140 | * setup_inline_extent_backref(). | |
1141 | */ | |
1142 | if (unlikely(refs_to_mod != -1)) { | |
1143 | struct btrfs_key key; | |
1144 | ||
1145 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1146 | ||
1147 | btrfs_print_leaf(leaf); | |
1148 | btrfs_err(fs_info, | |
1149 | "invalid refs_to_mod for tree block %llu, has %d expect -1", | |
1150 | key.objectid, refs_to_mod); | |
1151 | return -EUCLEAN; | |
1152 | } | |
56bec294 | 1153 | } |
31840ae1 | 1154 | |
25761430 QW |
1155 | if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) { |
1156 | struct btrfs_key key; | |
1157 | u32 extent_size; | |
1158 | ||
1159 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1160 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
1161 | extent_size = fs_info->nodesize; | |
1162 | else | |
1163 | extent_size = key.offset; | |
1164 | btrfs_print_leaf(leaf); | |
1165 | btrfs_err(fs_info, | |
1166 | "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu", | |
1167 | (unsigned long)iref, key.objectid, extent_size, | |
1168 | refs_to_mod, refs); | |
1169 | return -EUCLEAN; | |
1170 | } | |
5d4f98a2 YZ |
1171 | refs += refs_to_mod; |
1172 | ||
1173 | if (refs > 0) { | |
1174 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1175 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1176 | else | |
1177 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1178 | } else { | |
1179 | size = btrfs_extent_inline_ref_size(type); | |
3212fa14 | 1180 | item_size = btrfs_item_size(leaf, path->slots[0]); |
5d4f98a2 YZ |
1181 | ptr = (unsigned long)iref; |
1182 | end = (unsigned long)ei + item_size; | |
1183 | if (ptr + size < end) | |
1184 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1185 | end - ptr - size); | |
1186 | item_size -= size; | |
50564b65 | 1187 | btrfs_truncate_item(trans, path, item_size, 1); |
5d4f98a2 | 1188 | } |
50564b65 | 1189 | btrfs_mark_buffer_dirty(trans, leaf); |
25761430 | 1190 | return 0; |
5d4f98a2 YZ |
1191 | } |
1192 | ||
1193 | static noinline_for_stack | |
1194 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1195 | struct btrfs_path *path, |
1196 | u64 bytenr, u64 num_bytes, u64 parent, | |
1197 | u64 root_objectid, u64 owner, | |
1198 | u64 offset, int refs_to_add, | |
1199 | struct btrfs_delayed_extent_op *extent_op) | |
1200 | { | |
1201 | struct btrfs_extent_inline_ref *iref; | |
1202 | int ret; | |
1203 | ||
867cc1fb NB |
1204 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1205 | num_bytes, parent, root_objectid, | |
1206 | owner, offset, 1); | |
5d4f98a2 | 1207 | if (ret == 0) { |
07cce5cf QW |
1208 | /* |
1209 | * We're adding refs to a tree block we already own, this | |
1210 | * should not happen at all. | |
1211 | */ | |
1212 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
eee3b811 | 1213 | btrfs_print_leaf(path->nodes[0]); |
07cce5cf | 1214 | btrfs_crit(trans->fs_info, |
eee3b811 QW |
1215 | "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u", |
1216 | bytenr, num_bytes, root_objectid, path->slots[0]); | |
07cce5cf QW |
1217 | return -EUCLEAN; |
1218 | } | |
50564b65 FM |
1219 | ret = update_inline_extent_backref(trans, path, iref, |
1220 | refs_to_add, extent_op); | |
5d4f98a2 | 1221 | } else if (ret == -ENOENT) { |
50564b65 | 1222 | setup_inline_extent_backref(trans, path, iref, parent, |
143bede5 JM |
1223 | root_objectid, owner, offset, |
1224 | refs_to_add, extent_op); | |
1225 | ret = 0; | |
771ed689 | 1226 | } |
5d4f98a2 YZ |
1227 | return ret; |
1228 | } | |
31840ae1 | 1229 | |
5d4f98a2 | 1230 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
76d76e78 | 1231 | struct btrfs_root *root, |
5d4f98a2 YZ |
1232 | struct btrfs_path *path, |
1233 | struct btrfs_extent_inline_ref *iref, | |
5b2a54bb | 1234 | int refs_to_drop, int is_data) |
5d4f98a2 | 1235 | { |
143bede5 | 1236 | int ret = 0; |
b9473439 | 1237 | |
5d4f98a2 | 1238 | BUG_ON(!is_data && refs_to_drop != 1); |
5b2a54bb | 1239 | if (iref) |
50564b65 FM |
1240 | ret = update_inline_extent_backref(trans, path, iref, |
1241 | -refs_to_drop, NULL); | |
5b2a54bb JB |
1242 | else if (is_data) |
1243 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1244 | else | |
76d76e78 | 1245 | ret = btrfs_del_item(trans, root, path); |
5d4f98a2 YZ |
1246 | return ret; |
1247 | } | |
1248 | ||
d04c6b88 JM |
1249 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1250 | u64 *discarded_bytes) | |
5d4f98a2 | 1251 | { |
86557861 JM |
1252 | int j, ret = 0; |
1253 | u64 bytes_left, end; | |
adbe7e38 | 1254 | u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT); |
d04c6b88 | 1255 | |
a208b3f1 DS |
1256 | /* Adjust the range to be aligned to 512B sectors if necessary. */ |
1257 | if (start != aligned_start) { | |
4d89d377 | 1258 | len -= aligned_start - start; |
adbe7e38 | 1259 | len = round_down(len, 1 << SECTOR_SHIFT); |
4d89d377 JM |
1260 | start = aligned_start; |
1261 | } | |
d04c6b88 | 1262 | |
4d89d377 | 1263 | *discarded_bytes = 0; |
86557861 JM |
1264 | |
1265 | if (!len) | |
1266 | return 0; | |
1267 | ||
1268 | end = start + len; | |
1269 | bytes_left = len; | |
1270 | ||
1271 | /* Skip any superblocks on this device. */ | |
1272 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1273 | u64 sb_start = btrfs_sb_offset(j); | |
1274 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1275 | u64 size = sb_start - start; | |
1276 | ||
1277 | if (!in_range(sb_start, start, bytes_left) && | |
1278 | !in_range(sb_end, start, bytes_left) && | |
1279 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1280 | continue; | |
1281 | ||
1282 | /* | |
1283 | * Superblock spans beginning of range. Adjust start and | |
1284 | * try again. | |
1285 | */ | |
1286 | if (sb_start <= start) { | |
1287 | start += sb_end - start; | |
1288 | if (start > end) { | |
1289 | bytes_left = 0; | |
1290 | break; | |
1291 | } | |
1292 | bytes_left = end - start; | |
1293 | continue; | |
1294 | } | |
1295 | ||
1296 | if (size) { | |
29e70be2 AJ |
1297 | ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT, |
1298 | size >> SECTOR_SHIFT, | |
44abff2c | 1299 | GFP_NOFS); |
86557861 JM |
1300 | if (!ret) |
1301 | *discarded_bytes += size; | |
1302 | else if (ret != -EOPNOTSUPP) | |
1303 | return ret; | |
1304 | } | |
1305 | ||
1306 | start = sb_end; | |
1307 | if (start > end) { | |
1308 | bytes_left = 0; | |
1309 | break; | |
1310 | } | |
1311 | bytes_left = end - start; | |
1312 | } | |
1313 | ||
1314 | if (bytes_left) { | |
29e70be2 AJ |
1315 | ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT, |
1316 | bytes_left >> SECTOR_SHIFT, | |
44abff2c | 1317 | GFP_NOFS); |
4d89d377 | 1318 | if (!ret) |
86557861 | 1319 | *discarded_bytes += bytes_left; |
4d89d377 | 1320 | } |
d04c6b88 | 1321 | return ret; |
5d4f98a2 | 1322 | } |
5d4f98a2 | 1323 | |
a4012f06 | 1324 | static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes) |
6143c23c NA |
1325 | { |
1326 | struct btrfs_device *dev = stripe->dev; | |
1327 | struct btrfs_fs_info *fs_info = dev->fs_info; | |
1328 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1329 | u64 phys = stripe->physical; | |
1330 | u64 len = stripe->length; | |
1331 | u64 discarded = 0; | |
1332 | int ret = 0; | |
1333 | ||
1334 | /* Zone reset on a zoned filesystem */ | |
1335 | if (btrfs_can_zone_reset(dev, phys, len)) { | |
1336 | u64 src_disc; | |
1337 | ||
1338 | ret = btrfs_reset_device_zone(dev, phys, len, &discarded); | |
1339 | if (ret) | |
1340 | goto out; | |
1341 | ||
1342 | if (!btrfs_dev_replace_is_ongoing(dev_replace) || | |
1343 | dev != dev_replace->srcdev) | |
1344 | goto out; | |
1345 | ||
1346 | src_disc = discarded; | |
1347 | ||
1348 | /* Send to replace target as well */ | |
1349 | ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len, | |
1350 | &discarded); | |
1351 | discarded += src_disc; | |
70200574 | 1352 | } else if (bdev_max_discard_sectors(stripe->dev->bdev)) { |
6143c23c NA |
1353 | ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded); |
1354 | } else { | |
1355 | ret = 0; | |
1356 | *bytes = 0; | |
1357 | } | |
1358 | ||
1359 | out: | |
1360 | *bytes = discarded; | |
1361 | return ret; | |
1362 | } | |
1363 | ||
2ff7e61e | 1364 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1365 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1366 | { |
6b7faadd | 1367 | int ret = 0; |
5378e607 | 1368 | u64 discarded_bytes = 0; |
6b7faadd QW |
1369 | u64 end = bytenr + num_bytes; |
1370 | u64 cur = bytenr; | |
e244a0ae | 1371 | |
2999241d | 1372 | /* |
a4012f06 CH |
1373 | * Avoid races with device replace and make sure the devices in the |
1374 | * stripes don't go away while we are discarding. | |
2999241d | 1375 | */ |
0b246afa | 1376 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd | 1377 | while (cur < end) { |
a4012f06 CH |
1378 | struct btrfs_discard_stripe *stripes; |
1379 | unsigned int num_stripes; | |
5d4f98a2 YZ |
1380 | int i; |
1381 | ||
6b7faadd | 1382 | num_bytes = end - cur; |
a4012f06 CH |
1383 | stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes); |
1384 | if (IS_ERR(stripes)) { | |
1385 | ret = PTR_ERR(stripes); | |
1386 | if (ret == -EOPNOTSUPP) | |
1387 | ret = 0; | |
1388 | break; | |
1389 | } | |
5d4f98a2 | 1390 | |
a4012f06 CH |
1391 | for (i = 0; i < num_stripes; i++) { |
1392 | struct btrfs_discard_stripe *stripe = stripes + i; | |
d04c6b88 | 1393 | u64 bytes; |
38b5f68e | 1394 | |
a4012f06 | 1395 | if (!stripe->dev->bdev) { |
627e0873 FM |
1396 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); |
1397 | continue; | |
1398 | } | |
dcba6e48 | 1399 | |
a4012f06 CH |
1400 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
1401 | &stripe->dev->dev_state)) | |
5e753a81 AJ |
1402 | continue; |
1403 | ||
6143c23c | 1404 | ret = do_discard_extent(stripe, &bytes); |
a4012f06 | 1405 | if (ret) { |
6b7faadd | 1406 | /* |
a4012f06 CH |
1407 | * Keep going if discard is not supported by the |
1408 | * device. | |
6b7faadd | 1409 | */ |
a4012f06 CH |
1410 | if (ret != -EOPNOTSUPP) |
1411 | break; | |
1412 | ret = 0; | |
1413 | } else { | |
1414 | discarded_bytes += bytes; | |
6b7faadd | 1415 | } |
5d4f98a2 | 1416 | } |
a4012f06 CH |
1417 | kfree(stripes); |
1418 | if (ret) | |
1419 | break; | |
6b7faadd | 1420 | cur += num_bytes; |
5d4f98a2 | 1421 | } |
0b246afa | 1422 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1423 | if (actual_bytes) |
1424 | *actual_bytes = discarded_bytes; | |
5d4f98a2 | 1425 | return ret; |
5d4f98a2 YZ |
1426 | } |
1427 | ||
79787eaa | 1428 | /* Can return -ENOMEM */ |
5d4f98a2 | 1429 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1430 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1431 | { |
82fa113f | 1432 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 1433 | int ret; |
66d7e7f0 | 1434 | |
82fa113f QW |
1435 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1436 | generic_ref->action); | |
1437 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
f2e69a77 | 1438 | generic_ref->ref_root == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 1439 | |
82fa113f | 1440 | if (generic_ref->type == BTRFS_REF_METADATA) |
2187374f | 1441 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL); |
82fa113f | 1442 | else |
2187374f | 1443 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0); |
d7eae340 | 1444 | |
82fa113f | 1445 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1446 | |
5d4f98a2 YZ |
1447 | return ret; |
1448 | } | |
1449 | ||
bd3c685e | 1450 | /* |
9580503b | 1451 | * Insert backreference for a given extent. |
bd3c685e | 1452 | * |
07cce5cf QW |
1453 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1454 | * how it works. | |
1455 | * | |
bd3c685e NB |
1456 | * @trans: Handle of transaction |
1457 | * | |
1458 | * @node: The delayed ref node used to get the bytenr/length for | |
1459 | * extent whose references are incremented. | |
1460 | * | |
bd3c685e NB |
1461 | * @extent_op Pointer to a structure, holding information necessary when |
1462 | * updating a tree block's flags | |
1463 | * | |
1464 | */ | |
5d4f98a2 | 1465 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1466 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1467 | struct btrfs_delayed_extent_op *extent_op) |
1468 | { | |
1469 | struct btrfs_path *path; | |
1470 | struct extent_buffer *leaf; | |
1471 | struct btrfs_extent_item *item; | |
fcebe456 | 1472 | struct btrfs_key key; |
c682f9b3 QW |
1473 | u64 bytenr = node->bytenr; |
1474 | u64 num_bytes = node->num_bytes; | |
a502f112 JB |
1475 | u64 owner = btrfs_delayed_ref_owner(node); |
1476 | u64 offset = btrfs_delayed_ref_offset(node); | |
5d4f98a2 | 1477 | u64 refs; |
88b2d088 | 1478 | int refs_to_add = node->ref_mod; |
5d4f98a2 | 1479 | int ret; |
5d4f98a2 YZ |
1480 | |
1481 | path = btrfs_alloc_path(); | |
1482 | if (!path) | |
1483 | return -ENOMEM; | |
1484 | ||
5d4f98a2 | 1485 | /* this will setup the path even if it fails to insert the back ref */ |
a639cdeb | 1486 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
a502f112 | 1487 | node->parent, node->ref_root, owner, |
a639cdeb | 1488 | offset, refs_to_add, extent_op); |
0ed4792a | 1489 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1490 | goto out; |
fcebe456 JB |
1491 | |
1492 | /* | |
1493 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1494 | * inline extent ref, so just update the reference count and add a | |
1495 | * normal backref. | |
1496 | */ | |
5d4f98a2 | 1497 | leaf = path->nodes[0]; |
fcebe456 | 1498 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1499 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1500 | refs = btrfs_extent_refs(leaf, item); | |
1501 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1502 | if (extent_op) | |
1503 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1504 | |
50564b65 | 1505 | btrfs_mark_buffer_dirty(trans, leaf); |
b3b4aa74 | 1506 | btrfs_release_path(path); |
56bec294 | 1507 | |
56bec294 | 1508 | /* now insert the actual backref */ |
d2f79e63 | 1509 | if (owner < BTRFS_FIRST_FREE_OBJECTID) |
b4b5934a | 1510 | ret = insert_tree_block_ref(trans, path, node, bytenr); |
d2f79e63 | 1511 | else |
b4b5934a | 1512 | ret = insert_extent_data_ref(trans, path, node, bytenr); |
d2f79e63 | 1513 | |
79787eaa | 1514 | if (ret) |
66642832 | 1515 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1516 | out: |
56bec294 | 1517 | btrfs_free_path(path); |
30d133fc | 1518 | return ret; |
56bec294 CM |
1519 | } |
1520 | ||
e85a0ada BB |
1521 | static void free_head_ref_squota_rsv(struct btrfs_fs_info *fs_info, |
1522 | struct btrfs_delayed_ref_head *href) | |
1523 | { | |
1524 | u64 root = href->owning_root; | |
1525 | ||
1526 | /* | |
1527 | * Don't check must_insert_reserved, as this is called from contexts | |
1528 | * where it has already been unset. | |
1529 | */ | |
1530 | if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE || | |
1531 | !href->is_data || !is_fstree(root)) | |
1532 | return; | |
1533 | ||
1534 | btrfs_qgroup_free_refroot(fs_info, root, href->reserved_bytes, | |
1535 | BTRFS_QGROUP_RSV_DATA); | |
1536 | } | |
1537 | ||
5d4f98a2 | 1538 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
cecbb533 | 1539 | struct btrfs_delayed_ref_head *href, |
5d4f98a2 YZ |
1540 | struct btrfs_delayed_ref_node *node, |
1541 | struct btrfs_delayed_extent_op *extent_op, | |
61c681fe | 1542 | bool insert_reserved) |
56bec294 | 1543 | { |
5d4f98a2 | 1544 | int ret = 0; |
5d4f98a2 | 1545 | u64 parent = 0; |
5d4f98a2 YZ |
1546 | u64 flags = 0; |
1547 | ||
1bff6d4f | 1548 | trace_run_delayed_data_ref(trans->fs_info, node); |
599c75ec | 1549 | |
5d4f98a2 | 1550 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
cf4f0432 | 1551 | parent = node->parent; |
5d4f98a2 YZ |
1552 | |
1553 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
7cce0d69 | 1554 | struct btrfs_key key; |
cecbb533 BB |
1555 | struct btrfs_squota_delta delta = { |
1556 | .root = href->owning_root, | |
1557 | .num_bytes = node->num_bytes, | |
cecbb533 BB |
1558 | .is_data = true, |
1559 | .is_inc = true, | |
bd7c1ea3 | 1560 | .generation = trans->transid, |
cecbb533 | 1561 | }; |
44cc2e38 JB |
1562 | u64 owner = btrfs_delayed_ref_owner(node); |
1563 | u64 offset = btrfs_delayed_ref_offset(node); | |
7cce0d69 | 1564 | |
3173a18f | 1565 | if (extent_op) |
5d4f98a2 | 1566 | flags |= extent_op->flags_to_set; |
7cce0d69 FM |
1567 | |
1568 | key.objectid = node->bytenr; | |
1569 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1570 | key.offset = node->num_bytes; | |
1571 | ||
cf4f0432 | 1572 | ret = alloc_reserved_file_extent(trans, parent, node->ref_root, |
44cc2e38 JB |
1573 | flags, owner, offset, &key, |
1574 | node->ref_mod, | |
1575 | href->owning_root); | |
e85a0ada | 1576 | free_head_ref_squota_rsv(trans->fs_info, href); |
cecbb533 BB |
1577 | if (!ret) |
1578 | ret = btrfs_record_squota_delta(trans->fs_info, &delta); | |
5d4f98a2 | 1579 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
a502f112 | 1580 | ret = __btrfs_inc_extent_ref(trans, node, extent_op); |
5d4f98a2 | 1581 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
85bb9f54 | 1582 | ret = __btrfs_free_extent(trans, href, node, extent_op); |
5d4f98a2 YZ |
1583 | } else { |
1584 | BUG(); | |
1585 | } | |
1586 | return ret; | |
1587 | } | |
1588 | ||
1589 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1590 | struct extent_buffer *leaf, | |
1591 | struct btrfs_extent_item *ei) | |
1592 | { | |
1593 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1594 | if (extent_op->update_flags) { | |
1595 | flags |= extent_op->flags_to_set; | |
1596 | btrfs_set_extent_flags(leaf, ei, flags); | |
1597 | } | |
1598 | ||
1599 | if (extent_op->update_key) { | |
1600 | struct btrfs_tree_block_info *bi; | |
1601 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1602 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1603 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1604 | } | |
1605 | } | |
1606 | ||
1607 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1608 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1609 | struct btrfs_delayed_extent_op *extent_op) |
1610 | { | |
20b9a2d6 | 1611 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 1612 | struct btrfs_root *root; |
5d4f98a2 YZ |
1613 | struct btrfs_key key; |
1614 | struct btrfs_path *path; | |
1615 | struct btrfs_extent_item *ei; | |
1616 | struct extent_buffer *leaf; | |
1617 | u32 item_size; | |
56bec294 | 1618 | int ret; |
0e3696f8 | 1619 | int metadata = 1; |
5d4f98a2 | 1620 | |
bf31f87f | 1621 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1622 | return 0; |
1623 | ||
0e3696f8 | 1624 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1625 | metadata = 0; |
1626 | ||
5d4f98a2 YZ |
1627 | path = btrfs_alloc_path(); |
1628 | if (!path) | |
1629 | return -ENOMEM; | |
1630 | ||
d278850e | 1631 | key.objectid = head->bytenr; |
5d4f98a2 | 1632 | |
3173a18f | 1633 | if (metadata) { |
3173a18f | 1634 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1635 | key.offset = extent_op->level; |
3173a18f JB |
1636 | } else { |
1637 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1638 | key.offset = head->num_bytes; |
3173a18f JB |
1639 | } |
1640 | ||
29cbcf40 | 1641 | root = btrfs_extent_root(fs_info, key.objectid); |
3173a18f | 1642 | again: |
29cbcf40 | 1643 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
5d4f98a2 | 1644 | if (ret < 0) { |
5d4f98a2 | 1645 | goto out; |
20fb05a6 | 1646 | } else if (ret > 0) { |
3173a18f | 1647 | if (metadata) { |
55994887 FDBM |
1648 | if (path->slots[0] > 0) { |
1649 | path->slots[0]--; | |
1650 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1651 | path->slots[0]); | |
d278850e | 1652 | if (key.objectid == head->bytenr && |
55994887 | 1653 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1654 | key.offset == head->num_bytes) |
55994887 FDBM |
1655 | ret = 0; |
1656 | } | |
1657 | if (ret > 0) { | |
1658 | btrfs_release_path(path); | |
1659 | metadata = 0; | |
3173a18f | 1660 | |
d278850e JB |
1661 | key.objectid = head->bytenr; |
1662 | key.offset = head->num_bytes; | |
55994887 FDBM |
1663 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1664 | goto again; | |
1665 | } | |
1666 | } else { | |
20fb05a6 | 1667 | ret = -EUCLEAN; |
8ec0a4a5 FM |
1668 | btrfs_err(fs_info, |
1669 | "missing extent item for extent %llu num_bytes %llu level %d", | |
1670 | head->bytenr, head->num_bytes, extent_op->level); | |
55994887 | 1671 | goto out; |
3173a18f | 1672 | } |
5d4f98a2 YZ |
1673 | } |
1674 | ||
1675 | leaf = path->nodes[0]; | |
3212fa14 | 1676 | item_size = btrfs_item_size(leaf, path->slots[0]); |
ba3c2b19 | 1677 | |
6d8ff4e4 | 1678 | if (unlikely(item_size < sizeof(*ei))) { |
20fb05a6 | 1679 | ret = -EUCLEAN; |
182741d2 QW |
1680 | btrfs_err(fs_info, |
1681 | "unexpected extent item size, has %u expect >= %zu", | |
1682 | item_size, sizeof(*ei)); | |
20fb05a6 | 1683 | btrfs_abort_transaction(trans, ret); |
ba3c2b19 NB |
1684 | goto out; |
1685 | } | |
1686 | ||
5d4f98a2 YZ |
1687 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1688 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1689 | |
50564b65 | 1690 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
1691 | out: |
1692 | btrfs_free_path(path); | |
20fb05a6 | 1693 | return ret; |
56bec294 CM |
1694 | } |
1695 | ||
5d4f98a2 | 1696 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
cecbb533 | 1697 | struct btrfs_delayed_ref_head *href, |
5d4f98a2 YZ |
1698 | struct btrfs_delayed_ref_node *node, |
1699 | struct btrfs_delayed_extent_op *extent_op, | |
61c681fe | 1700 | bool insert_reserved) |
56bec294 CM |
1701 | { |
1702 | int ret = 0; | |
cecbb533 | 1703 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1704 | u64 parent = 0; |
1705 | u64 ref_root = 0; | |
56bec294 | 1706 | |
1bff6d4f | 1707 | trace_run_delayed_tree_ref(trans->fs_info, node); |
599c75ec | 1708 | |
5d4f98a2 | 1709 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
cf4f0432 JB |
1710 | parent = node->parent; |
1711 | ref_root = node->ref_root; | |
5d4f98a2 | 1712 | |
1bf76df3 | 1713 | if (unlikely(node->ref_mod != 1)) { |
f97806f2 | 1714 | btrfs_err(trans->fs_info, |
1bf76df3 | 1715 | "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu", |
02794222 LB |
1716 | node->bytenr, node->ref_mod, node->action, ref_root, |
1717 | parent); | |
1bf76df3 | 1718 | return -EUCLEAN; |
02794222 | 1719 | } |
5d4f98a2 | 1720 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
cecbb533 BB |
1721 | struct btrfs_squota_delta delta = { |
1722 | .root = href->owning_root, | |
1723 | .num_bytes = fs_info->nodesize, | |
cecbb533 BB |
1724 | .is_data = false, |
1725 | .is_inc = true, | |
bd7c1ea3 | 1726 | .generation = trans->transid, |
cecbb533 BB |
1727 | }; |
1728 | ||
3173a18f | 1729 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1730 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
cecbb533 BB |
1731 | if (!ret) |
1732 | btrfs_record_squota_delta(fs_info, &delta); | |
5d4f98a2 | 1733 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
a502f112 | 1734 | ret = __btrfs_inc_extent_ref(trans, node, extent_op); |
5d4f98a2 | 1735 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
85bb9f54 | 1736 | ret = __btrfs_free_extent(trans, href, node, extent_op); |
5d4f98a2 YZ |
1737 | } else { |
1738 | BUG(); | |
1739 | } | |
56bec294 CM |
1740 | return ret; |
1741 | } | |
1742 | ||
1743 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1744 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
cecbb533 | 1745 | struct btrfs_delayed_ref_head *href, |
5d4f98a2 YZ |
1746 | struct btrfs_delayed_ref_node *node, |
1747 | struct btrfs_delayed_extent_op *extent_op, | |
61c681fe | 1748 | bool insert_reserved) |
56bec294 | 1749 | { |
79787eaa JM |
1750 | int ret = 0; |
1751 | ||
bf31f87f | 1752 | if (TRANS_ABORTED(trans)) { |
e85a0ada | 1753 | if (insert_reserved) { |
b25c36f8 | 1754 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
e85a0ada BB |
1755 | free_head_ref_squota_rsv(trans->fs_info, href); |
1756 | } | |
79787eaa | 1757 | return 0; |
857cc2fc | 1758 | } |
79787eaa | 1759 | |
5d4f98a2 YZ |
1760 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1761 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
cecbb533 | 1762 | ret = run_delayed_tree_ref(trans, href, node, extent_op, |
5d4f98a2 YZ |
1763 | insert_reserved); |
1764 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1765 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
cecbb533 | 1766 | ret = run_delayed_data_ref(trans, href, node, extent_op, |
5d4f98a2 | 1767 | insert_reserved); |
d9a620f7 BB |
1768 | else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY) |
1769 | ret = 0; | |
5d4f98a2 YZ |
1770 | else |
1771 | BUG(); | |
80ee54bf | 1772 | if (ret && insert_reserved) |
b25c36f8 | 1773 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
39f501d6 QW |
1774 | if (ret < 0) |
1775 | btrfs_err(trans->fs_info, | |
1776 | "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d", | |
1777 | node->bytenr, node->num_bytes, node->type, | |
1778 | node->action, node->ref_mod, ret); | |
5d4f98a2 | 1779 | return ret; |
56bec294 CM |
1780 | } |
1781 | ||
c6fc2454 | 1782 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1783 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1784 | { | |
cffc3374 FM |
1785 | struct btrfs_delayed_ref_node *ref; |
1786 | ||
e3d03965 | 1787 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1788 | return NULL; |
d7df2c79 | 1789 | |
cffc3374 FM |
1790 | /* |
1791 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1792 | * This is to prevent a ref count from going down to zero, which deletes | |
1793 | * the extent item from the extent tree, when there still are references | |
1794 | * to add, which would fail because they would not find the extent item. | |
1795 | */ | |
1d57ee94 WX |
1796 | if (!list_empty(&head->ref_add_list)) |
1797 | return list_first_entry(&head->ref_add_list, | |
1798 | struct btrfs_delayed_ref_node, add_list); | |
1799 | ||
e3d03965 | 1800 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1801 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1802 | ASSERT(list_empty(&ref->add_list)); |
1803 | return ref; | |
56bec294 CM |
1804 | } |
1805 | ||
2eadaa22 JB |
1806 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1807 | struct btrfs_delayed_ref_head *head) | |
1808 | { | |
1809 | spin_lock(&delayed_refs->lock); | |
61c681fe | 1810 | head->processing = false; |
2eadaa22 JB |
1811 | delayed_refs->num_heads_ready++; |
1812 | spin_unlock(&delayed_refs->lock); | |
1813 | btrfs_delayed_ref_unlock(head); | |
1814 | } | |
1815 | ||
bedc6617 JB |
1816 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1817 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1818 | { |
1819 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1820 | |
1821 | if (!extent_op) | |
bedc6617 JB |
1822 | return NULL; |
1823 | ||
b00e6250 | 1824 | if (head->must_insert_reserved) { |
bedc6617 | 1825 | head->extent_op = NULL; |
b00e6250 | 1826 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1827 | return NULL; |
b00e6250 | 1828 | } |
bedc6617 JB |
1829 | return extent_op; |
1830 | } | |
1831 | ||
1832 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1833 | struct btrfs_delayed_ref_head *head) | |
1834 | { | |
1835 | struct btrfs_delayed_extent_op *extent_op; | |
1836 | int ret; | |
1837 | ||
1838 | extent_op = cleanup_extent_op(head); | |
1839 | if (!extent_op) | |
1840 | return 0; | |
1841 | head->extent_op = NULL; | |
b00e6250 | 1842 | spin_unlock(&head->lock); |
20b9a2d6 | 1843 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1844 | btrfs_free_delayed_extent_op(extent_op); |
1845 | return ret ? ret : 1; | |
1846 | } | |
1847 | ||
8a526c44 | 1848 | u64 btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
31890da0 JB |
1849 | struct btrfs_delayed_ref_root *delayed_refs, |
1850 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1851 | { |
e85a0ada BB |
1852 | u64 ret = 0; |
1853 | ||
81e75ac7 JB |
1854 | /* |
1855 | * We had csum deletions accounted for in our delayed refs rsv, we need | |
1856 | * to drop the csum leaves for this update from our delayed_refs_rsv. | |
1857 | */ | |
1858 | if (head->total_ref_mod < 0 && head->is_data) { | |
adb86dbe | 1859 | int nr_csums; |
3ee56a58 | 1860 | |
81e75ac7 JB |
1861 | spin_lock(&delayed_refs->lock); |
1862 | delayed_refs->pending_csums -= head->num_bytes; | |
1863 | spin_unlock(&delayed_refs->lock); | |
adb86dbe | 1864 | nr_csums = btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes); |
3ee56a58 | 1865 | |
adb86dbe | 1866 | btrfs_delayed_refs_rsv_release(fs_info, 0, nr_csums); |
81e75ac7 | 1867 | |
e85a0ada | 1868 | ret = btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums); |
3ee56a58 | 1869 | } |
e85a0ada BB |
1870 | /* must_insert_reserved can be set only if we didn't run the head ref. */ |
1871 | if (head->must_insert_reserved) | |
1872 | free_head_ref_squota_rsv(fs_info, head); | |
8a526c44 | 1873 | |
e85a0ada | 1874 | return ret; |
07c47775 JB |
1875 | } |
1876 | ||
194ab0bc | 1877 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
8a526c44 FM |
1878 | struct btrfs_delayed_ref_head *head, |
1879 | u64 *bytes_released) | |
194ab0bc | 1880 | { |
f9871edd NB |
1881 | |
1882 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1883 | struct btrfs_delayed_ref_root *delayed_refs; |
1884 | int ret; | |
1885 | ||
1886 | delayed_refs = &trans->transaction->delayed_refs; | |
1887 | ||
bedc6617 | 1888 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1889 | if (ret < 0) { |
1890 | unselect_delayed_ref_head(delayed_refs, head); | |
1891 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1892 | return ret; | |
1893 | } else if (ret) { | |
1894 | return ret; | |
1895 | } | |
1896 | ||
1897 | /* | |
1898 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1899 | * and then re-check to make sure nobody got added. | |
1900 | */ | |
1901 | spin_unlock(&head->lock); | |
1902 | spin_lock(&delayed_refs->lock); | |
1903 | spin_lock(&head->lock); | |
e3d03965 | 1904 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1905 | spin_unlock(&head->lock); |
1906 | spin_unlock(&delayed_refs->lock); | |
1907 | return 1; | |
1908 | } | |
d7baffda | 1909 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1910 | spin_unlock(&head->lock); |
1e7a1421 | 1911 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1912 | |
c1103f7a | 1913 | if (head->must_insert_reserved) { |
b25c36f8 | 1914 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1915 | if (head->is_data) { |
fc28b25e JB |
1916 | struct btrfs_root *csum_root; |
1917 | ||
1918 | csum_root = btrfs_csum_root(fs_info, head->bytenr); | |
1919 | ret = btrfs_del_csums(trans, csum_root, head->bytenr, | |
1920 | head->num_bytes); | |
c1103f7a JB |
1921 | } |
1922 | } | |
1923 | ||
3ee56a58 | 1924 | *bytes_released += btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1925 | |
1926 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1927 | btrfs_delayed_ref_unlock(head); |
d278850e | 1928 | btrfs_put_delayed_ref_head(head); |
856bd270 | 1929 | return ret; |
194ab0bc JB |
1930 | } |
1931 | ||
b1cdbcb5 NB |
1932 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1933 | struct btrfs_trans_handle *trans) | |
1934 | { | |
1935 | struct btrfs_delayed_ref_root *delayed_refs = | |
1936 | &trans->transaction->delayed_refs; | |
1937 | struct btrfs_delayed_ref_head *head = NULL; | |
1938 | int ret; | |
1939 | ||
1940 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1941 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1942 | if (!head) { |
1943 | spin_unlock(&delayed_refs->lock); | |
1944 | return head; | |
1945 | } | |
1946 | ||
1947 | /* | |
1948 | * Grab the lock that says we are going to process all the refs for | |
1949 | * this head | |
1950 | */ | |
9e920a6f | 1951 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1952 | spin_unlock(&delayed_refs->lock); |
1953 | ||
1954 | /* | |
1955 | * We may have dropped the spin lock to get the head mutex lock, and | |
1956 | * that might have given someone else time to free the head. If that's | |
1957 | * true, it has been removed from our list and we can move on. | |
1958 | */ | |
1959 | if (ret == -EAGAIN) | |
1960 | head = ERR_PTR(-EAGAIN); | |
1961 | ||
1962 | return head; | |
1963 | } | |
1964 | ||
e7261386 | 1965 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
3ee56a58 FM |
1966 | struct btrfs_delayed_ref_head *locked_ref, |
1967 | u64 *bytes_released) | |
e7261386 NB |
1968 | { |
1969 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1970 | struct btrfs_delayed_ref_root *delayed_refs; | |
1971 | struct btrfs_delayed_extent_op *extent_op; | |
1972 | struct btrfs_delayed_ref_node *ref; | |
61c681fe | 1973 | bool must_insert_reserved; |
e7261386 NB |
1974 | int ret; |
1975 | ||
1976 | delayed_refs = &trans->transaction->delayed_refs; | |
1977 | ||
0110a4c4 NB |
1978 | lockdep_assert_held(&locked_ref->mutex); |
1979 | lockdep_assert_held(&locked_ref->lock); | |
1980 | ||
e7261386 NB |
1981 | while ((ref = select_delayed_ref(locked_ref))) { |
1982 | if (ref->seq && | |
1983 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1984 | spin_unlock(&locked_ref->lock); | |
1985 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1986 | return -EAGAIN; | |
1987 | } | |
1988 | ||
e7261386 NB |
1989 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); |
1990 | RB_CLEAR_NODE(&ref->ref_node); | |
1991 | if (!list_empty(&ref->add_list)) | |
1992 | list_del(&ref->add_list); | |
1993 | /* | |
1994 | * When we play the delayed ref, also correct the ref_mod on | |
1995 | * head | |
1996 | */ | |
1997 | switch (ref->action) { | |
1998 | case BTRFS_ADD_DELAYED_REF: | |
1999 | case BTRFS_ADD_DELAYED_EXTENT: | |
2000 | locked_ref->ref_mod -= ref->ref_mod; | |
2001 | break; | |
2002 | case BTRFS_DROP_DELAYED_REF: | |
2003 | locked_ref->ref_mod += ref->ref_mod; | |
2004 | break; | |
2005 | default: | |
2006 | WARN_ON(1); | |
2007 | } | |
2008 | atomic_dec(&delayed_refs->num_entries); | |
2009 | ||
2010 | /* | |
2011 | * Record the must_insert_reserved flag before we drop the | |
2012 | * spin lock. | |
2013 | */ | |
2014 | must_insert_reserved = locked_ref->must_insert_reserved; | |
e85a0ada BB |
2015 | /* |
2016 | * Unsetting this on the head ref relinquishes ownership of | |
2017 | * the rsv_bytes, so it is critical that every possible code | |
2018 | * path from here forward frees all reserves including qgroup | |
2019 | * reserve. | |
2020 | */ | |
61c681fe | 2021 | locked_ref->must_insert_reserved = false; |
e7261386 NB |
2022 | |
2023 | extent_op = locked_ref->extent_op; | |
2024 | locked_ref->extent_op = NULL; | |
2025 | spin_unlock(&locked_ref->lock); | |
2026 | ||
cecbb533 | 2027 | ret = run_one_delayed_ref(trans, locked_ref, ref, extent_op, |
e7261386 | 2028 | must_insert_reserved); |
adb86dbe | 2029 | btrfs_delayed_refs_rsv_release(fs_info, 1, 0); |
3ee56a58 | 2030 | *bytes_released += btrfs_calc_delayed_ref_bytes(fs_info, 1); |
cecbb533 | 2031 | |
e7261386 NB |
2032 | btrfs_free_delayed_extent_op(extent_op); |
2033 | if (ret) { | |
2034 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2035 | btrfs_put_delayed_ref(ref); | |
e7261386 NB |
2036 | return ret; |
2037 | } | |
2038 | ||
2039 | btrfs_put_delayed_ref(ref); | |
2040 | cond_resched(); | |
2041 | ||
2042 | spin_lock(&locked_ref->lock); | |
0c555c97 | 2043 | btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref); |
e7261386 NB |
2044 | } |
2045 | ||
2046 | return 0; | |
2047 | } | |
2048 | ||
79787eaa JM |
2049 | /* |
2050 | * Returns 0 on success or if called with an already aborted transaction. | |
2051 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2052 | */ | |
d7df2c79 | 2053 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
8a526c44 | 2054 | u64 min_bytes) |
56bec294 | 2055 | { |
0a1e458a | 2056 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2057 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 CM |
2058 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
2059 | int ret; | |
d7df2c79 | 2060 | unsigned long count = 0; |
8a526c44 FM |
2061 | unsigned long max_count = 0; |
2062 | u64 bytes_processed = 0; | |
56bec294 CM |
2063 | |
2064 | delayed_refs = &trans->transaction->delayed_refs; | |
8a526c44 FM |
2065 | if (min_bytes == 0) { |
2066 | max_count = delayed_refs->num_heads_ready; | |
2067 | min_bytes = U64_MAX; | |
2068 | } | |
2069 | ||
0110a4c4 | 2070 | do { |
56bec294 | 2071 | if (!locked_ref) { |
b1cdbcb5 | 2072 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2073 | if (IS_ERR_OR_NULL(locked_ref)) { |
2074 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2075 | continue; | |
2076 | } else { | |
2077 | break; | |
2078 | } | |
56bec294 | 2079 | } |
0110a4c4 | 2080 | count++; |
56bec294 | 2081 | } |
2c3cf7d5 FM |
2082 | /* |
2083 | * We need to try and merge add/drops of the same ref since we | |
2084 | * can run into issues with relocate dropping the implicit ref | |
2085 | * and then it being added back again before the drop can | |
2086 | * finish. If we merged anything we need to re-loop so we can | |
2087 | * get a good ref. | |
2088 | * Or we can get node references of the same type that weren't | |
2089 | * merged when created due to bumps in the tree mod seq, and | |
2090 | * we need to merge them to prevent adding an inline extent | |
2091 | * backref before dropping it (triggering a BUG_ON at | |
2092 | * insert_inline_extent_backref()). | |
2093 | */ | |
d7df2c79 | 2094 | spin_lock(&locked_ref->lock); |
0c555c97 | 2095 | btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref); |
ae1e206b | 2096 | |
3ee56a58 | 2097 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, &bytes_processed); |
0110a4c4 NB |
2098 | if (ret < 0 && ret != -EAGAIN) { |
2099 | /* | |
2100 | * Error, btrfs_run_delayed_refs_for_head already | |
2101 | * unlocked everything so just bail out | |
2102 | */ | |
2103 | return ret; | |
2104 | } else if (!ret) { | |
2105 | /* | |
2106 | * Success, perform the usual cleanup of a processed | |
2107 | * head | |
2108 | */ | |
3ee56a58 | 2109 | ret = cleanup_ref_head(trans, locked_ref, &bytes_processed); |
194ab0bc | 2110 | if (ret > 0 ) { |
b00e6250 JB |
2111 | /* We dropped our lock, we need to loop. */ |
2112 | ret = 0; | |
d7df2c79 | 2113 | continue; |
194ab0bc JB |
2114 | } else if (ret) { |
2115 | return ret; | |
5d4f98a2 | 2116 | } |
22cd2e7d | 2117 | } |
1ce7a5ec | 2118 | |
b00e6250 | 2119 | /* |
0110a4c4 NB |
2120 | * Either success case or btrfs_run_delayed_refs_for_head |
2121 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2122 | */ |
b00e6250 | 2123 | |
0110a4c4 | 2124 | locked_ref = NULL; |
c3e69d58 | 2125 | cond_resched(); |
8a526c44 FM |
2126 | } while ((min_bytes != U64_MAX && bytes_processed < min_bytes) || |
2127 | (max_count > 0 && count < max_count) || | |
2128 | locked_ref); | |
0a2b2a84 | 2129 | |
d7df2c79 | 2130 | return 0; |
c3e69d58 CM |
2131 | } |
2132 | ||
709c0486 AJ |
2133 | #ifdef SCRAMBLE_DELAYED_REFS |
2134 | /* | |
2135 | * Normally delayed refs get processed in ascending bytenr order. This | |
2136 | * correlates in most cases to the order added. To expose dependencies on this | |
2137 | * order, we start to process the tree in the middle instead of the beginning | |
2138 | */ | |
2139 | static u64 find_middle(struct rb_root *root) | |
2140 | { | |
2141 | struct rb_node *n = root->rb_node; | |
2142 | struct btrfs_delayed_ref_node *entry; | |
2143 | int alt = 1; | |
2144 | u64 middle; | |
2145 | u64 first = 0, last = 0; | |
2146 | ||
2147 | n = rb_first(root); | |
2148 | if (n) { | |
2149 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2150 | first = entry->bytenr; | |
2151 | } | |
2152 | n = rb_last(root); | |
2153 | if (n) { | |
2154 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2155 | last = entry->bytenr; | |
2156 | } | |
2157 | n = root->rb_node; | |
2158 | ||
2159 | while (n) { | |
2160 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2161 | WARN_ON(!entry->in_tree); | |
2162 | ||
2163 | middle = entry->bytenr; | |
2164 | ||
2165 | if (alt) | |
2166 | n = n->rb_left; | |
2167 | else | |
2168 | n = n->rb_right; | |
2169 | ||
2170 | alt = 1 - alt; | |
2171 | } | |
2172 | return middle; | |
2173 | } | |
2174 | #endif | |
2175 | ||
c3e69d58 | 2176 | /* |
8a526c44 FM |
2177 | * Start processing the delayed reference count updates and extent insertions |
2178 | * we have queued up so far. | |
2179 | * | |
2180 | * @trans: Transaction handle. | |
2181 | * @min_bytes: How many bytes of delayed references to process. After this | |
2182 | * many bytes we stop processing delayed references if there are | |
2183 | * any more. If 0 it means to run all existing delayed references, | |
2184 | * but not new ones added after running all existing ones. | |
2185 | * Use (u64)-1 (U64_MAX) to run all existing delayed references | |
2186 | * plus any new ones that are added. | |
79787eaa JM |
2187 | * |
2188 | * Returns 0 on success or if called with an aborted transaction | |
2189 | * Returns <0 on error and aborts the transaction | |
c3e69d58 | 2190 | */ |
8a526c44 | 2191 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, u64 min_bytes) |
c3e69d58 | 2192 | { |
c79a70b1 | 2193 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 | 2194 | struct btrfs_delayed_ref_root *delayed_refs; |
c3e69d58 | 2195 | int ret; |
c3e69d58 | 2196 | |
79787eaa | 2197 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2198 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2199 | return 0; |
2200 | ||
0b246afa | 2201 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2202 | return 0; |
2203 | ||
c3e69d58 | 2204 | delayed_refs = &trans->transaction->delayed_refs; |
c3e69d58 | 2205 | again: |
709c0486 AJ |
2206 | #ifdef SCRAMBLE_DELAYED_REFS |
2207 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2208 | #endif | |
8a526c44 | 2209 | ret = __btrfs_run_delayed_refs(trans, min_bytes); |
d7df2c79 | 2210 | if (ret < 0) { |
66642832 | 2211 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2212 | return ret; |
eb099670 | 2213 | } |
c3e69d58 | 2214 | |
8a526c44 | 2215 | if (min_bytes == U64_MAX) { |
119e80df | 2216 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2217 | |
d7df2c79 | 2218 | spin_lock(&delayed_refs->lock); |
3cbb9f51 | 2219 | if (RB_EMPTY_ROOT(&delayed_refs->href_root.rb_root)) { |
d7df2c79 | 2220 | spin_unlock(&delayed_refs->lock); |
3cbb9f51 | 2221 | return 0; |
d7df2c79 | 2222 | } |
d278850e | 2223 | spin_unlock(&delayed_refs->lock); |
e9d0b13b | 2224 | |
d7df2c79 | 2225 | cond_resched(); |
56bec294 | 2226 | goto again; |
5f39d397 | 2227 | } |
3cbb9f51 | 2228 | |
a28ec197 CM |
2229 | return 0; |
2230 | } | |
2231 | ||
5d4f98a2 | 2232 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
4aec05fa | 2233 | struct extent_buffer *eb, u64 flags) |
5d4f98a2 YZ |
2234 | { |
2235 | struct btrfs_delayed_extent_op *extent_op; | |
4aec05fa | 2236 | int level = btrfs_header_level(eb); |
5d4f98a2 YZ |
2237 | int ret; |
2238 | ||
78a6184a | 2239 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2240 | if (!extent_op) |
2241 | return -ENOMEM; | |
2242 | ||
2243 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2244 | extent_op->update_flags = true; |
2245 | extent_op->update_key = false; | |
b1c79e09 | 2246 | extent_op->level = level; |
5d4f98a2 | 2247 | |
42c9d0b5 | 2248 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2249 | if (ret) |
78a6184a | 2250 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2251 | return ret; |
2252 | } | |
2253 | ||
e4c3b2dc | 2254 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2255 | struct btrfs_path *path, |
2256 | u64 objectid, u64 offset, u64 bytenr) | |
2257 | { | |
2258 | struct btrfs_delayed_ref_head *head; | |
2259 | struct btrfs_delayed_ref_node *ref; | |
5d4f98a2 | 2260 | struct btrfs_delayed_ref_root *delayed_refs; |
e4c3b2dc | 2261 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2262 | struct rb_node *node; |
5d4f98a2 YZ |
2263 | int ret = 0; |
2264 | ||
998ac6d2 | 2265 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2266 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2267 | if (cur_trans) |
2268 | refcount_inc(&cur_trans->use_count); | |
2269 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2270 | if (!cur_trans) |
2271 | return 0; | |
2272 | ||
2273 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2274 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2275 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2276 | if (!head) { |
2277 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2278 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2279 | return 0; |
2280 | } | |
5d4f98a2 YZ |
2281 | |
2282 | if (!mutex_trylock(&head->mutex)) { | |
26ce9114 JB |
2283 | if (path->nowait) { |
2284 | spin_unlock(&delayed_refs->lock); | |
2285 | btrfs_put_transaction(cur_trans); | |
2286 | return -EAGAIN; | |
2287 | } | |
2288 | ||
d278850e | 2289 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2290 | spin_unlock(&delayed_refs->lock); |
2291 | ||
b3b4aa74 | 2292 | btrfs_release_path(path); |
5d4f98a2 | 2293 | |
8cc33e5c DS |
2294 | /* |
2295 | * Mutex was contended, block until it's released and let | |
2296 | * caller try again | |
2297 | */ | |
5d4f98a2 YZ |
2298 | mutex_lock(&head->mutex); |
2299 | mutex_unlock(&head->mutex); | |
d278850e | 2300 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2301 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2302 | return -EAGAIN; |
2303 | } | |
d7df2c79 | 2304 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2305 | |
d7df2c79 | 2306 | spin_lock(&head->lock); |
0e0adbcf JB |
2307 | /* |
2308 | * XXX: We should replace this with a proper search function in the | |
2309 | * future. | |
2310 | */ | |
e3d03965 LB |
2311 | for (node = rb_first_cached(&head->ref_tree); node; |
2312 | node = rb_next(node)) { | |
44cc2e38 JB |
2313 | u64 ref_owner; |
2314 | u64 ref_offset; | |
2315 | ||
0e0adbcf | 2316 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2317 | /* If it's a shared ref we know a cross reference exists */ |
2318 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2319 | ret = 1; | |
2320 | break; | |
2321 | } | |
5d4f98a2 | 2322 | |
44cc2e38 JB |
2323 | ref_owner = btrfs_delayed_ref_owner(ref); |
2324 | ref_offset = btrfs_delayed_ref_offset(ref); | |
5d4f98a2 | 2325 | |
d7df2c79 JB |
2326 | /* |
2327 | * If our ref doesn't match the one we're currently looking at | |
2328 | * then we have a cross reference. | |
2329 | */ | |
e094f480 | 2330 | if (ref->ref_root != btrfs_root_id(root) || |
44cc2e38 | 2331 | ref_owner != objectid || ref_offset != offset) { |
d7df2c79 JB |
2332 | ret = 1; |
2333 | break; | |
2334 | } | |
5d4f98a2 | 2335 | } |
d7df2c79 | 2336 | spin_unlock(&head->lock); |
5d4f98a2 | 2337 | mutex_unlock(&head->mutex); |
998ac6d2 | 2338 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2339 | return ret; |
2340 | } | |
2341 | ||
e4c3b2dc | 2342 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2343 | struct btrfs_path *path, |
a84d5d42 BB |
2344 | u64 objectid, u64 offset, u64 bytenr, |
2345 | bool strict) | |
be20aa9d | 2346 | { |
0b246afa | 2347 | struct btrfs_fs_info *fs_info = root->fs_info; |
29cbcf40 | 2348 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr); |
f321e491 | 2349 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2350 | struct btrfs_extent_data_ref *ref; |
2351 | struct btrfs_extent_inline_ref *iref; | |
2352 | struct btrfs_extent_item *ei; | |
f321e491 | 2353 | struct btrfs_key key; |
5d4f98a2 | 2354 | u32 item_size; |
d9a620f7 | 2355 | u32 expected_size; |
3de28d57 | 2356 | int type; |
be20aa9d | 2357 | int ret; |
925baedd | 2358 | |
be20aa9d | 2359 | key.objectid = bytenr; |
31840ae1 | 2360 | key.offset = (u64)-1; |
f321e491 | 2361 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2362 | |
be20aa9d CM |
2363 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2364 | if (ret < 0) | |
2365 | goto out; | |
a4259b6c DS |
2366 | if (ret == 0) { |
2367 | /* | |
2368 | * Key with offset -1 found, there would have to exist an extent | |
2369 | * item with such offset, but this is out of the valid range. | |
2370 | */ | |
2371 | ret = -EUCLEAN; | |
2372 | goto out; | |
2373 | } | |
80ff3856 YZ |
2374 | |
2375 | ret = -ENOENT; | |
2376 | if (path->slots[0] == 0) | |
31840ae1 | 2377 | goto out; |
be20aa9d | 2378 | |
31840ae1 | 2379 | path->slots[0]--; |
f321e491 | 2380 | leaf = path->nodes[0]; |
5d4f98a2 | 2381 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2382 | |
5d4f98a2 | 2383 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2384 | goto out; |
f321e491 | 2385 | |
5d4f98a2 | 2386 | ret = 1; |
3212fa14 | 2387 | item_size = btrfs_item_size(leaf, path->slots[0]); |
5d4f98a2 | 2388 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
d9a620f7 BB |
2389 | expected_size = sizeof(*ei) + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY); |
2390 | ||
2391 | /* No inline refs; we need to bail before checking for owner ref. */ | |
2392 | if (item_size == sizeof(*ei)) | |
2393 | goto out; | |
2394 | ||
2395 | /* Check for an owner ref; skip over it to the real inline refs. */ | |
2396 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2397 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
2398 | if (btrfs_fs_incompat(fs_info, SIMPLE_QUOTA) && type == BTRFS_EXTENT_OWNER_REF_KEY) { | |
2399 | expected_size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY); | |
2400 | iref = (struct btrfs_extent_inline_ref *)(iref + 1); | |
2401 | } | |
bd09835d | 2402 | |
a6bd9cd1 | 2403 | /* If extent item has more than 1 inline ref then it's shared */ |
d9a620f7 | 2404 | if (item_size != expected_size) |
5d4f98a2 | 2405 | goto out; |
be20aa9d | 2406 | |
a84d5d42 BB |
2407 | /* |
2408 | * If extent created before last snapshot => it's shared unless the | |
2409 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2410 | */ | |
2411 | if (!strict && | |
2412 | (btrfs_extent_generation(leaf, ei) <= | |
2413 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2414 | goto out; |
2415 | ||
a6bd9cd1 | 2416 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2417 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2418 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2419 | goto out; |
2420 | ||
2421 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2422 | if (btrfs_extent_refs(leaf, ei) != | |
2423 | btrfs_extent_data_ref_count(leaf, ref) || | |
e094f480 | 2424 | btrfs_extent_data_ref_root(leaf, ref) != btrfs_root_id(root) || |
5d4f98a2 YZ |
2425 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || |
2426 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2427 | goto out; | |
2428 | ||
2429 | ret = 0; | |
2430 | out: | |
2431 | return ret; | |
2432 | } | |
2433 | ||
e4c3b2dc | 2434 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
1a89f173 | 2435 | u64 bytenr, bool strict, struct btrfs_path *path) |
5d4f98a2 | 2436 | { |
5d4f98a2 | 2437 | int ret; |
5d4f98a2 | 2438 | |
5d4f98a2 | 2439 | do { |
e4c3b2dc | 2440 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2441 | offset, bytenr, strict); |
5d4f98a2 | 2442 | if (ret && ret != -ENOENT) |
f321e491 | 2443 | goto out; |
80ff3856 | 2444 | |
380fd066 MT |
2445 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2446 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2447 | |
be20aa9d | 2448 | out: |
1a89f173 | 2449 | btrfs_release_path(path); |
37f00a6d | 2450 | if (btrfs_is_data_reloc_root(root)) |
f0486c68 | 2451 | WARN_ON(ret > 0); |
f321e491 | 2452 | return ret; |
be20aa9d | 2453 | } |
c5739bba | 2454 | |
5d4f98a2 | 2455 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2456 | struct btrfs_root *root, |
5d4f98a2 | 2457 | struct extent_buffer *buf, |
e339a6b0 | 2458 | int full_backref, int inc) |
31840ae1 | 2459 | { |
0b246afa | 2460 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 2461 | u64 parent; |
31840ae1 | 2462 | u64 ref_root; |
31840ae1 | 2463 | u32 nritems; |
31840ae1 ZY |
2464 | struct btrfs_key key; |
2465 | struct btrfs_file_extent_item *fi; | |
82fa113f | 2466 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); |
31840ae1 | 2467 | int i; |
82fa113f | 2468 | int action; |
31840ae1 ZY |
2469 | int level; |
2470 | int ret = 0; | |
fccb84c9 | 2471 | |
0b246afa | 2472 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2473 | return 0; |
fccb84c9 | 2474 | |
31840ae1 | 2475 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2476 | nritems = btrfs_header_nritems(buf); |
2477 | level = btrfs_header_level(buf); | |
2478 | ||
92a7cc42 | 2479 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2480 | return 0; |
31840ae1 | 2481 | |
5d4f98a2 YZ |
2482 | if (full_backref) |
2483 | parent = buf->start; | |
2484 | else | |
2485 | parent = 0; | |
82fa113f QW |
2486 | if (inc) |
2487 | action = BTRFS_ADD_DELAYED_REF; | |
2488 | else | |
2489 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2490 | |
2491 | for (i = 0; i < nritems; i++) { | |
4d09b4e9 JB |
2492 | struct btrfs_ref ref = { |
2493 | .action = action, | |
2494 | .parent = parent, | |
f2e69a77 | 2495 | .ref_root = ref_root, |
4d09b4e9 JB |
2496 | }; |
2497 | ||
31840ae1 | 2498 | if (level == 0) { |
5d4f98a2 | 2499 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2500 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2501 | continue; |
5d4f98a2 | 2502 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2503 | struct btrfs_file_extent_item); |
2504 | if (btrfs_file_extent_type(buf, fi) == | |
2505 | BTRFS_FILE_EXTENT_INLINE) | |
2506 | continue; | |
4d09b4e9 JB |
2507 | ref.bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
2508 | if (ref.bytenr == 0) | |
31840ae1 | 2509 | continue; |
5d4f98a2 | 2510 | |
12390e42 | 2511 | ref.num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); |
4d09b4e9 JB |
2512 | ref.owning_root = ref_root; |
2513 | ||
5d4f98a2 | 2514 | key.offset -= btrfs_file_extent_offset(buf, fi); |
f2e69a77 | 2515 | btrfs_init_data_ref(&ref, key.objectid, key.offset, |
e094f480 | 2516 | btrfs_root_id(root), for_reloc); |
dd28b6a5 | 2517 | if (inc) |
4d09b4e9 | 2518 | ret = btrfs_inc_extent_ref(trans, &ref); |
dd28b6a5 | 2519 | else |
4d09b4e9 | 2520 | ret = btrfs_free_extent(trans, &ref); |
31840ae1 ZY |
2521 | if (ret) |
2522 | goto fail; | |
2523 | } else { | |
4d09b4e9 JB |
2524 | /* We don't know the owning_root, leave as 0. */ |
2525 | ref.bytenr = btrfs_node_blockptr(buf, i); | |
12390e42 | 2526 | ref.num_bytes = fs_info->nodesize; |
4d09b4e9 | 2527 | |
f2e69a77 | 2528 | btrfs_init_tree_ref(&ref, level - 1, |
e094f480 | 2529 | btrfs_root_id(root), for_reloc); |
dd28b6a5 | 2530 | if (inc) |
4d09b4e9 | 2531 | ret = btrfs_inc_extent_ref(trans, &ref); |
dd28b6a5 | 2532 | else |
4d09b4e9 | 2533 | ret = btrfs_free_extent(trans, &ref); |
31840ae1 ZY |
2534 | if (ret) |
2535 | goto fail; | |
2536 | } | |
2537 | } | |
2538 | return 0; | |
2539 | fail: | |
5d4f98a2 YZ |
2540 | return ret; |
2541 | } | |
2542 | ||
2543 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2544 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2545 | { |
e339a6b0 | 2546 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2547 | } |
2548 | ||
2549 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2550 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2551 | { |
e339a6b0 | 2552 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2553 | } |
2554 | ||
1b86826d | 2555 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2556 | { |
0b246afa | 2557 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2558 | u64 flags; |
53b381b3 | 2559 | u64 ret; |
9ed74f2d | 2560 | |
b742bb82 YZ |
2561 | if (data) |
2562 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2563 | else if (root == fs_info->chunk_root) |
b742bb82 | 2564 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2565 | else |
b742bb82 | 2566 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2567 | |
878d7b67 | 2568 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2569 | return ret; |
6a63209f | 2570 | } |
9ed74f2d | 2571 | |
0eb997bf | 2572 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info) |
a061fc8d | 2573 | { |
08dddb29 FM |
2574 | struct rb_node *leftmost; |
2575 | u64 bytenr = 0; | |
a1897fdd | 2576 | |
16b0c258 | 2577 | read_lock(&fs_info->block_group_cache_lock); |
0eb997bf | 2578 | /* Get the block group with the lowest logical start address. */ |
08dddb29 FM |
2579 | leftmost = rb_first_cached(&fs_info->block_group_cache_tree); |
2580 | if (leftmost) { | |
2581 | struct btrfs_block_group *bg; | |
0f9dd46c | 2582 | |
08dddb29 FM |
2583 | bg = rb_entry(leftmost, struct btrfs_block_group, cache_node); |
2584 | bytenr = bg->start; | |
2585 | } | |
16b0c258 | 2586 | read_unlock(&fs_info->block_group_cache_lock); |
d2fb3437 YZ |
2587 | |
2588 | return bytenr; | |
a061fc8d CM |
2589 | } |
2590 | ||
6690d071 NB |
2591 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2592 | struct btrfs_block_group *cache, | |
f0486c68 | 2593 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2594 | { |
fdf08605 DS |
2595 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2596 | ||
11833d66 YZ |
2597 | spin_lock(&cache->space_info->lock); |
2598 | spin_lock(&cache->lock); | |
2599 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2600 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2601 | num_bytes); | |
11833d66 YZ |
2602 | if (reserved) { |
2603 | cache->reserved -= num_bytes; | |
2604 | cache->space_info->bytes_reserved -= num_bytes; | |
2605 | } | |
2606 | spin_unlock(&cache->lock); | |
2607 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2608 | |
fe1a598c | 2609 | set_extent_bit(&trans->transaction->pinned_extents, bytenr, |
1d126800 | 2610 | bytenr + num_bytes - 1, EXTENT_DIRTY, NULL); |
f0486c68 YZ |
2611 | return 0; |
2612 | } | |
68b38550 | 2613 | |
b25c36f8 | 2614 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2615 | u64 bytenr, u64 num_bytes, int reserved) |
2616 | { | |
32da5386 | 2617 | struct btrfs_block_group *cache; |
68b38550 | 2618 | |
b25c36f8 | 2619 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2620 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2621 | |
6690d071 | 2622 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2623 | |
2624 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2625 | return 0; |
2626 | } | |
2627 | ||
9fce5704 | 2628 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
007dec8c | 2629 | const struct extent_buffer *eb) |
e688b725 | 2630 | { |
32da5386 | 2631 | struct btrfs_block_group *cache; |
b50c6e25 | 2632 | int ret; |
e688b725 | 2633 | |
007dec8c | 2634 | cache = btrfs_lookup_block_group(trans->fs_info, eb->start); |
b50c6e25 JB |
2635 | if (!cache) |
2636 | return -EINVAL; | |
e688b725 CM |
2637 | |
2638 | /* | |
ced8ecf0 OS |
2639 | * Fully cache the free space first so that our pin removes the free space |
2640 | * from the cache. | |
e688b725 | 2641 | */ |
ced8ecf0 | 2642 | ret = btrfs_cache_block_group(cache, true); |
9ad6d91f FM |
2643 | if (ret) |
2644 | goto out; | |
e688b725 | 2645 | |
007dec8c | 2646 | pin_down_extent(trans, cache, eb->start, eb->len, 0); |
e688b725 CM |
2647 | |
2648 | /* remove us from the free space cache (if we're there at all) */ | |
007dec8c | 2649 | ret = btrfs_remove_free_space(cache, eb->start, eb->len); |
9ad6d91f | 2650 | out: |
e688b725 | 2651 | btrfs_put_block_group(cache); |
b50c6e25 | 2652 | return ret; |
e688b725 CM |
2653 | } |
2654 | ||
2ff7e61e JM |
2655 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2656 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2657 | { |
2658 | int ret; | |
32da5386 | 2659 | struct btrfs_block_group *block_group; |
8c2a1a30 | 2660 | |
0b246afa | 2661 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2662 | if (!block_group) |
2663 | return -EINVAL; | |
2664 | ||
ced8ecf0 | 2665 | ret = btrfs_cache_block_group(block_group, true); |
9ad6d91f FM |
2666 | if (ret) |
2667 | goto out; | |
8c2a1a30 | 2668 | |
9ad6d91f FM |
2669 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2670 | out: | |
8c2a1a30 JB |
2671 | btrfs_put_block_group(block_group); |
2672 | return ret; | |
2673 | } | |
2674 | ||
bcdc428c | 2675 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2676 | { |
bcdc428c | 2677 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2678 | struct btrfs_file_extent_item *item; |
2679 | struct btrfs_key key; | |
2680 | int found_type; | |
2681 | int i; | |
b89311ef | 2682 | int ret = 0; |
8c2a1a30 | 2683 | |
2ff7e61e | 2684 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2685 | return 0; |
2686 | ||
2687 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2688 | btrfs_item_key_to_cpu(eb, &key, i); | |
2689 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2690 | continue; | |
2691 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2692 | found_type = btrfs_file_extent_type(eb, item); | |
2693 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2694 | continue; | |
2695 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2696 | continue; | |
2697 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2698 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2699 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2700 | if (ret) | |
2701 | break; | |
8c2a1a30 JB |
2702 | } |
2703 | ||
b89311ef | 2704 | return ret; |
8c2a1a30 JB |
2705 | } |
2706 | ||
9cfa3e34 | 2707 | static void |
32da5386 | 2708 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2709 | { |
2710 | atomic_inc(&bg->reservations); | |
2711 | } | |
2712 | ||
c759c4e1 JB |
2713 | /* |
2714 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2715 | * what it should be based on the mount options. | |
2716 | */ | |
2717 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2718 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2719 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2720 | { |
2721 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2722 | |
2723 | *empty_cluster = 0; | |
2724 | if (btrfs_mixed_space_info(space_info)) | |
2725 | return ret; | |
2726 | ||
c759c4e1 | 2727 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2728 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2729 | if (btrfs_test_opt(fs_info, SSD)) |
2730 | *empty_cluster = SZ_2M; | |
2731 | else | |
ee22184b | 2732 | *empty_cluster = SZ_64K; |
583b7231 HK |
2733 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2734 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2735 | *empty_cluster = SZ_2M; | |
0b246afa | 2736 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2737 | } |
2738 | ||
2739 | return ret; | |
2740 | } | |
2741 | ||
2ff7e61e JM |
2742 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2743 | u64 start, u64 end, | |
678886bd | 2744 | const bool return_free_space) |
ccd467d6 | 2745 | { |
32da5386 | 2746 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2747 | struct btrfs_space_info *space_info; |
2748 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2749 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2750 | u64 len; |
c759c4e1 JB |
2751 | u64 total_unpinned = 0; |
2752 | u64 empty_cluster = 0; | |
7b398f8e | 2753 | bool readonly; |
44a6c343 | 2754 | int ret = 0; |
ccd467d6 | 2755 | |
11833d66 | 2756 | while (start <= end) { |
7b398f8e | 2757 | readonly = false; |
11833d66 | 2758 | if (!cache || |
b3470b5d | 2759 | start >= cache->start + cache->length) { |
11833d66 YZ |
2760 | if (cache) |
2761 | btrfs_put_block_group(cache); | |
c759c4e1 | 2762 | total_unpinned = 0; |
11833d66 | 2763 | cache = btrfs_lookup_block_group(fs_info, start); |
44a6c343 DS |
2764 | if (cache == NULL) { |
2765 | /* Logic error, something removed the block group. */ | |
2766 | ret = -EUCLEAN; | |
2767 | goto out; | |
2768 | } | |
c759c4e1 | 2769 | |
2ff7e61e | 2770 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2771 | cache->space_info, |
2772 | &empty_cluster); | |
2773 | empty_cluster <<= 1; | |
11833d66 YZ |
2774 | } |
2775 | ||
b3470b5d | 2776 | len = cache->start + cache->length - start; |
11833d66 YZ |
2777 | len = min(len, end + 1 - start); |
2778 | ||
48ff7083 OS |
2779 | if (return_free_space) |
2780 | btrfs_add_free_space(cache, start, len); | |
11833d66 | 2781 | |
f0486c68 | 2782 | start += len; |
c759c4e1 | 2783 | total_unpinned += len; |
7b398f8e | 2784 | space_info = cache->space_info; |
f0486c68 | 2785 | |
c759c4e1 JB |
2786 | /* |
2787 | * If this space cluster has been marked as fragmented and we've | |
2788 | * unpinned enough in this block group to potentially allow a | |
2789 | * cluster to be created inside of it go ahead and clear the | |
2790 | * fragmented check. | |
2791 | */ | |
2792 | if (cluster && cluster->fragmented && | |
2793 | total_unpinned > empty_cluster) { | |
2794 | spin_lock(&cluster->lock); | |
2795 | cluster->fragmented = 0; | |
2796 | spin_unlock(&cluster->lock); | |
2797 | } | |
2798 | ||
7b398f8e | 2799 | spin_lock(&space_info->lock); |
11833d66 YZ |
2800 | spin_lock(&cache->lock); |
2801 | cache->pinned -= len; | |
bb96c4e5 | 2802 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2803 | space_info->max_extent_size = 0; |
7b398f8e JB |
2804 | if (cache->ro) { |
2805 | space_info->bytes_readonly += len; | |
2806 | readonly = true; | |
169e0da9 NA |
2807 | } else if (btrfs_is_zoned(fs_info)) { |
2808 | /* Need reset before reusing in a zoned block group */ | |
2809 | space_info->bytes_zone_unusable += len; | |
2810 | readonly = true; | |
7b398f8e | 2811 | } |
11833d66 | 2812 | spin_unlock(&cache->lock); |
957780eb JB |
2813 | if (!readonly && return_free_space && |
2814 | global_rsv->space_info == space_info) { | |
7b398f8e JB |
2815 | spin_lock(&global_rsv->lock); |
2816 | if (!global_rsv->full) { | |
c4bf1909 JC |
2817 | u64 to_add = min(len, global_rsv->size - |
2818 | global_rsv->reserved); | |
2819 | ||
957780eb | 2820 | global_rsv->reserved += to_add; |
bb96c4e5 JB |
2821 | btrfs_space_info_update_bytes_may_use(fs_info, |
2822 | space_info, to_add); | |
7b398f8e JB |
2823 | if (global_rsv->reserved >= global_rsv->size) |
2824 | global_rsv->full = 1; | |
957780eb | 2825 | len -= to_add; |
7b398f8e JB |
2826 | } |
2827 | spin_unlock(&global_rsv->lock); | |
2828 | } | |
2732798c JB |
2829 | /* Add to any tickets we may have */ |
2830 | if (!readonly && return_free_space && len) | |
2831 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2832 | spin_unlock(&space_info->lock); |
ccd467d6 | 2833 | } |
11833d66 YZ |
2834 | |
2835 | if (cache) | |
2836 | btrfs_put_block_group(cache); | |
44a6c343 DS |
2837 | out: |
2838 | return ret; | |
ccd467d6 CM |
2839 | } |
2840 | ||
5ead2dd0 | 2841 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2842 | { |
5ead2dd0 | 2843 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2844 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2845 | struct list_head *deleted_bgs; |
11833d66 | 2846 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2847 | u64 start; |
2848 | u64 end; | |
a28ec197 | 2849 | int ret; |
a28ec197 | 2850 | |
fe119a6e | 2851 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2852 | |
bf31f87f | 2853 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2854 | struct extent_state *cached_state = NULL; |
2855 | ||
d4b450cd | 2856 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
e5860f82 FM |
2857 | if (!find_first_extent_bit(unpin, 0, &start, &end, |
2858 | EXTENT_DIRTY, &cached_state)) { | |
d4b450cd | 2859 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
a28ec197 | 2860 | break; |
d4b450cd | 2861 | } |
1f3c79a2 | 2862 | |
46b27f50 | 2863 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2864 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2865 | end + 1 - start, NULL); |
1f3c79a2 | 2866 | |
0e6ec385 | 2867 | clear_extent_dirty(unpin, start, end, &cached_state); |
44a6c343 DS |
2868 | ret = unpin_extent_range(fs_info, start, end, true); |
2869 | BUG_ON(ret); | |
d4b450cd | 2870 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2871 | free_extent_state(cached_state); |
b9473439 | 2872 | cond_resched(); |
a28ec197 | 2873 | } |
817d52f8 | 2874 | |
a2309300 DZ |
2875 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2876 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2877 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2878 | } |
b0643e59 | 2879 | |
e33e17ee JM |
2880 | /* |
2881 | * Transaction is finished. We don't need the lock anymore. We | |
2882 | * do need to clean up the block groups in case of a transaction | |
2883 | * abort. | |
2884 | */ | |
2885 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2886 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2887 | u64 trimmed = 0; | |
2888 | ||
2889 | ret = -EROFS; | |
bf31f87f | 2890 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2891 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2892 | block_group->start, |
2893 | block_group->length, | |
e33e17ee JM |
2894 | &trimmed); |
2895 | ||
2896 | list_del_init(&block_group->bg_list); | |
6b7304af | 2897 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2898 | btrfs_put_block_group(block_group); |
2899 | ||
2900 | if (ret) { | |
2901 | const char *errstr = btrfs_decode_error(ret); | |
2902 | btrfs_warn(fs_info, | |
913e1535 | 2903 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2904 | ret, errstr); |
2905 | } | |
2906 | } | |
2907 | ||
e20d96d6 CM |
2908 | return 0; |
2909 | } | |
2910 | ||
8d299091 BB |
2911 | /* |
2912 | * Parse an extent item's inline extents looking for a simple quotas owner ref. | |
2913 | * | |
2914 | * @fs_info: the btrfs_fs_info for this mount | |
2915 | * @leaf: a leaf in the extent tree containing the extent item | |
2916 | * @slot: the slot in the leaf where the extent item is found | |
2917 | * | |
2918 | * Returns the objectid of the root that originally allocated the extent item | |
2919 | * if the inline owner ref is expected and present, otherwise 0. | |
2920 | * | |
2921 | * If an extent item has an owner ref item, it will be the first inline ref | |
2922 | * item. Therefore the logic is to check whether there are any inline ref | |
2923 | * items, then check the type of the first one. | |
2924 | */ | |
2925 | u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info, | |
2926 | struct extent_buffer *leaf, int slot) | |
2927 | { | |
2928 | struct btrfs_extent_item *ei; | |
2929 | struct btrfs_extent_inline_ref *iref; | |
2930 | struct btrfs_extent_owner_ref *oref; | |
2931 | unsigned long ptr; | |
2932 | unsigned long end; | |
2933 | int type; | |
2934 | ||
2935 | if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA)) | |
2936 | return 0; | |
2937 | ||
2938 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
2939 | ptr = (unsigned long)(ei + 1); | |
2940 | end = (unsigned long)ei + btrfs_item_size(leaf, slot); | |
2941 | ||
2942 | /* No inline ref items of any kind, can't check type. */ | |
2943 | if (ptr == end) | |
2944 | return 0; | |
2945 | ||
2946 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
2947 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
2948 | ||
2949 | /* We found an owner ref, get the root out of it. */ | |
2950 | if (type == BTRFS_EXTENT_OWNER_REF_KEY) { | |
2951 | oref = (struct btrfs_extent_owner_ref *)(&iref->offset); | |
2952 | return btrfs_extent_owner_ref_root_id(leaf, oref); | |
2953 | } | |
2954 | ||
2955 | /* We have inline refs, but not an owner ref. */ | |
2956 | return 0; | |
2957 | } | |
2958 | ||
8f8aa4c7 | 2959 | static int do_free_extent_accounting(struct btrfs_trans_handle *trans, |
cecbb533 | 2960 | u64 bytenr, struct btrfs_squota_delta *delta) |
8f8aa4c7 JB |
2961 | { |
2962 | int ret; | |
cecbb533 | 2963 | u64 num_bytes = delta->num_bytes; |
8f8aa4c7 | 2964 | |
cecbb533 | 2965 | if (delta->is_data) { |
8f8aa4c7 JB |
2966 | struct btrfs_root *csum_root; |
2967 | ||
2968 | csum_root = btrfs_csum_root(trans->fs_info, bytenr); | |
2969 | ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes); | |
2970 | if (ret) { | |
2971 | btrfs_abort_transaction(trans, ret); | |
2972 | return ret; | |
2973 | } | |
ca41504e JT |
2974 | |
2975 | ret = btrfs_delete_raid_extent(trans, bytenr, num_bytes); | |
2976 | if (ret) { | |
2977 | btrfs_abort_transaction(trans, ret); | |
2978 | return ret; | |
2979 | } | |
8f8aa4c7 JB |
2980 | } |
2981 | ||
cecbb533 BB |
2982 | ret = btrfs_record_squota_delta(trans->fs_info, delta); |
2983 | if (ret) { | |
2984 | btrfs_abort_transaction(trans, ret); | |
2985 | return ret; | |
2986 | } | |
2987 | ||
8f8aa4c7 JB |
2988 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
2989 | if (ret) { | |
2990 | btrfs_abort_transaction(trans, ret); | |
2991 | return ret; | |
2992 | } | |
2993 | ||
2994 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, false); | |
2995 | if (ret) | |
2996 | btrfs_abort_transaction(trans, ret); | |
2997 | ||
2998 | return ret; | |
2999 | } | |
3000 | ||
eee3b811 QW |
3001 | #define abort_and_dump(trans, path, fmt, args...) \ |
3002 | ({ \ | |
3003 | btrfs_abort_transaction(trans, -EUCLEAN); \ | |
3004 | btrfs_print_leaf(path->nodes[0]); \ | |
3005 | btrfs_crit(trans->fs_info, fmt, ##args); \ | |
3006 | }) | |
3007 | ||
1c2a07f5 QW |
3008 | /* |
3009 | * Drop one or more refs of @node. | |
3010 | * | |
3011 | * 1. Locate the extent refs. | |
3012 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
3013 | * Locate it, then reduce the refs number or remove the ref line completely. | |
3014 | * | |
3015 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
3016 | * | |
3017 | * Inline backref case: | |
3018 | * | |
3019 | * in extent tree we have: | |
3020 | * | |
3021 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
3022 | * refs 2 gen 6 flags DATA | |
3023 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
3024 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
3025 | * | |
3026 | * This function gets called with: | |
3027 | * | |
3028 | * node->bytenr = 13631488 | |
3029 | * node->num_bytes = 1048576 | |
3030 | * root_objectid = FS_TREE | |
3031 | * owner_objectid = 257 | |
3032 | * owner_offset = 0 | |
3033 | * refs_to_drop = 1 | |
3034 | * | |
3035 | * Then we should get some like: | |
3036 | * | |
3037 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
3038 | * refs 1 gen 6 flags DATA | |
3039 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
3040 | * | |
3041 | * Keyed backref case: | |
3042 | * | |
3043 | * in extent tree we have: | |
3044 | * | |
3045 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
3046 | * refs 754 gen 6 flags DATA | |
3047 | * [...] | |
3048 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
3049 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
3050 | * | |
3051 | * This function get called with: | |
3052 | * | |
3053 | * node->bytenr = 13631488 | |
3054 | * node->num_bytes = 1048576 | |
3055 | * root_objectid = FS_TREE | |
3056 | * owner_objectid = 866 | |
3057 | * owner_offset = 0 | |
3058 | * refs_to_drop = 1 | |
3059 | * | |
3060 | * Then we should get some like: | |
3061 | * | |
3062 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
3063 | * refs 753 gen 6 flags DATA | |
3064 | * | |
3065 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
3066 | */ | |
5d4f98a2 | 3067 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
cecbb533 | 3068 | struct btrfs_delayed_ref_head *href, |
85bb9f54 | 3069 | struct btrfs_delayed_ref_node *node, |
e72cb923 | 3070 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 3071 | { |
e72cb923 | 3072 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 3073 | struct btrfs_key key; |
5d4f98a2 | 3074 | struct btrfs_path *path; |
29cbcf40 | 3075 | struct btrfs_root *extent_root; |
5f39d397 | 3076 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3077 | struct btrfs_extent_item *ei; |
3078 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 3079 | int ret; |
5d4f98a2 | 3080 | int is_data; |
952fccac CM |
3081 | int extent_slot = 0; |
3082 | int found_extent = 0; | |
3083 | int num_to_del = 1; | |
1df6b3c0 | 3084 | int refs_to_drop = node->ref_mod; |
5d4f98a2 YZ |
3085 | u32 item_size; |
3086 | u64 refs; | |
c682f9b3 QW |
3087 | u64 bytenr = node->bytenr; |
3088 | u64 num_bytes = node->num_bytes; | |
85bb9f54 JB |
3089 | u64 owner_objectid = btrfs_delayed_ref_owner(node); |
3090 | u64 owner_offset = btrfs_delayed_ref_offset(node); | |
0b246afa | 3091 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
cecbb533 | 3092 | u64 delayed_ref_root = href->owning_root; |
037e6390 | 3093 | |
29cbcf40 | 3094 | extent_root = btrfs_extent_root(info, bytenr); |
abed4aaa | 3095 | ASSERT(extent_root); |
29cbcf40 | 3096 | |
5caf2a00 | 3097 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3098 | if (!path) |
3099 | return -ENOMEM; | |
5f26f772 | 3100 | |
5d4f98a2 | 3101 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
1c2a07f5 QW |
3102 | |
3103 | if (!is_data && refs_to_drop != 1) { | |
3104 | btrfs_crit(info, | |
3105 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
3106 | node->bytenr, refs_to_drop); | |
3107 | ret = -EINVAL; | |
3108 | btrfs_abort_transaction(trans, ret); | |
3109 | goto out; | |
3110 | } | |
5d4f98a2 | 3111 | |
3173a18f | 3112 | if (is_data) |
897ca819 | 3113 | skinny_metadata = false; |
3173a18f | 3114 | |
fbe4801b | 3115 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
85bb9f54 | 3116 | node->parent, node->ref_root, owner_objectid, |
5d4f98a2 | 3117 | owner_offset); |
7bb86316 | 3118 | if (ret == 0) { |
1c2a07f5 QW |
3119 | /* |
3120 | * Either the inline backref or the SHARED_DATA_REF/ | |
3121 | * SHARED_BLOCK_REF is found | |
3122 | * | |
3123 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
3124 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
3125 | */ | |
952fccac | 3126 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3127 | while (extent_slot >= 0) { |
3128 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3129 | extent_slot); |
5d4f98a2 | 3130 | if (key.objectid != bytenr) |
952fccac | 3131 | break; |
5d4f98a2 YZ |
3132 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3133 | key.offset == num_bytes) { | |
952fccac CM |
3134 | found_extent = 1; |
3135 | break; | |
3136 | } | |
3173a18f JB |
3137 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3138 | key.offset == owner_objectid) { | |
3139 | found_extent = 1; | |
3140 | break; | |
3141 | } | |
1c2a07f5 QW |
3142 | |
3143 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
3144 | if (path->slots[0] - extent_slot > 5) |
3145 | break; | |
5d4f98a2 | 3146 | extent_slot--; |
952fccac | 3147 | } |
a79865c6 | 3148 | |
31840ae1 | 3149 | if (!found_extent) { |
1c2a07f5 | 3150 | if (iref) { |
eee3b811 QW |
3151 | abort_and_dump(trans, path, |
3152 | "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref", | |
3153 | path->slots[0]); | |
3154 | ret = -EUCLEAN; | |
3155 | goto out; | |
1c2a07f5 QW |
3156 | } |
3157 | /* Must be SHARED_* item, remove the backref first */ | |
76d76e78 | 3158 | ret = remove_extent_backref(trans, extent_root, path, |
5b2a54bb | 3159 | NULL, refs_to_drop, is_data); |
005d6427 | 3160 | if (ret) { |
66642832 | 3161 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3162 | goto out; |
3163 | } | |
b3b4aa74 | 3164 | btrfs_release_path(path); |
5d4f98a2 | 3165 | |
1c2a07f5 | 3166 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3167 | key.objectid = bytenr; |
3168 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3169 | key.offset = num_bytes; | |
3170 | ||
3173a18f JB |
3171 | if (!is_data && skinny_metadata) { |
3172 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3173 | key.offset = owner_objectid; | |
3174 | } | |
3175 | ||
31840ae1 ZY |
3176 | ret = btrfs_search_slot(trans, extent_root, |
3177 | &key, path, -1, 1); | |
3173a18f JB |
3178 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3179 | /* | |
3180 | * Couldn't find our skinny metadata item, | |
3181 | * see if we have ye olde extent item. | |
3182 | */ | |
3183 | path->slots[0]--; | |
3184 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3185 | path->slots[0]); | |
3186 | if (key.objectid == bytenr && | |
3187 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3188 | key.offset == num_bytes) | |
3189 | ret = 0; | |
3190 | } | |
3191 | ||
3192 | if (ret > 0 && skinny_metadata) { | |
3193 | skinny_metadata = false; | |
9ce49a0b | 3194 | key.objectid = bytenr; |
3173a18f JB |
3195 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3196 | key.offset = num_bytes; | |
3197 | btrfs_release_path(path); | |
3198 | ret = btrfs_search_slot(trans, extent_root, | |
3199 | &key, path, -1, 1); | |
3200 | } | |
3201 | ||
f3465ca4 | 3202 | if (ret) { |
b783e62d | 3203 | if (ret > 0) |
a4f78750 | 3204 | btrfs_print_leaf(path->nodes[0]); |
eee3b811 QW |
3205 | btrfs_err(info, |
3206 | "umm, got %d back from search, was looking for %llu, slot %d", | |
3207 | ret, bytenr, path->slots[0]); | |
f3465ca4 | 3208 | } |
005d6427 | 3209 | if (ret < 0) { |
66642832 | 3210 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3211 | goto out; |
3212 | } | |
31840ae1 ZY |
3213 | extent_slot = path->slots[0]; |
3214 | } | |
fae7f21c | 3215 | } else if (WARN_ON(ret == -ENOENT)) { |
eee3b811 QW |
3216 | abort_and_dump(trans, path, |
3217 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d", | |
85bb9f54 | 3218 | bytenr, node->parent, node->ref_root, owner_objectid, |
eee3b811 | 3219 | owner_offset, path->slots[0]); |
c4a050bb | 3220 | goto out; |
79787eaa | 3221 | } else { |
66642832 | 3222 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3223 | goto out; |
7bb86316 | 3224 | } |
5f39d397 CM |
3225 | |
3226 | leaf = path->nodes[0]; | |
3212fa14 | 3227 | item_size = btrfs_item_size(leaf, extent_slot); |
6d8ff4e4 | 3228 | if (unlikely(item_size < sizeof(*ei))) { |
182741d2 QW |
3229 | ret = -EUCLEAN; |
3230 | btrfs_err(trans->fs_info, | |
3231 | "unexpected extent item size, has %u expect >= %zu", | |
3232 | item_size, sizeof(*ei)); | |
ba3c2b19 NB |
3233 | btrfs_abort_transaction(trans, ret); |
3234 | goto out; | |
3235 | } | |
952fccac | 3236 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3237 | struct btrfs_extent_item); |
3173a18f JB |
3238 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3239 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3240 | struct btrfs_tree_block_info *bi; |
eee3b811 | 3241 | |
1c2a07f5 | 3242 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
eee3b811 QW |
3243 | abort_and_dump(trans, path, |
3244 | "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu", | |
3245 | key.objectid, key.type, key.offset, | |
3246 | path->slots[0], owner_objectid, item_size, | |
3247 | sizeof(*ei) + sizeof(*bi)); | |
3248 | ret = -EUCLEAN; | |
3249 | goto out; | |
1c2a07f5 | 3250 | } |
5d4f98a2 YZ |
3251 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3252 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3253 | } | |
56bec294 | 3254 | |
5d4f98a2 | 3255 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3256 | if (refs < refs_to_drop) { |
eee3b811 QW |
3257 | abort_and_dump(trans, path, |
3258 | "trying to drop %d refs but we only have %llu for bytenr %llu slot %u", | |
3259 | refs_to_drop, refs, bytenr, path->slots[0]); | |
3260 | ret = -EUCLEAN; | |
3261 | goto out; | |
32b02538 | 3262 | } |
56bec294 | 3263 | refs -= refs_to_drop; |
5f39d397 | 3264 | |
5d4f98a2 YZ |
3265 | if (refs > 0) { |
3266 | if (extent_op) | |
3267 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3268 | /* | |
3269 | * In the case of inline back ref, reference count will | |
3270 | * be updated by remove_extent_backref | |
952fccac | 3271 | */ |
5d4f98a2 | 3272 | if (iref) { |
1c2a07f5 | 3273 | if (!found_extent) { |
eee3b811 QW |
3274 | abort_and_dump(trans, path, |
3275 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u", | |
3276 | path->slots[0]); | |
3277 | ret = -EUCLEAN; | |
3278 | goto out; | |
1c2a07f5 | 3279 | } |
5d4f98a2 YZ |
3280 | } else { |
3281 | btrfs_set_extent_refs(leaf, ei, refs); | |
50564b65 | 3282 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
3283 | } |
3284 | if (found_extent) { | |
76d76e78 | 3285 | ret = remove_extent_backref(trans, extent_root, path, |
5b2a54bb | 3286 | iref, refs_to_drop, is_data); |
005d6427 | 3287 | if (ret) { |
66642832 | 3288 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3289 | goto out; |
3290 | } | |
952fccac | 3291 | } |
5d4f98a2 | 3292 | } else { |
cecbb533 BB |
3293 | struct btrfs_squota_delta delta = { |
3294 | .root = delayed_ref_root, | |
3295 | .num_bytes = num_bytes, | |
cecbb533 BB |
3296 | .is_data = is_data, |
3297 | .is_inc = false, | |
bd7c1ea3 | 3298 | .generation = btrfs_extent_generation(leaf, ei), |
cecbb533 BB |
3299 | }; |
3300 | ||
1c2a07f5 | 3301 | /* In this branch refs == 1 */ |
5d4f98a2 | 3302 | if (found_extent) { |
1c2a07f5 QW |
3303 | if (is_data && refs_to_drop != |
3304 | extent_data_ref_count(path, iref)) { | |
eee3b811 QW |
3305 | abort_and_dump(trans, path, |
3306 | "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u", | |
3307 | extent_data_ref_count(path, iref), | |
3308 | refs_to_drop, path->slots[0]); | |
3309 | ret = -EUCLEAN; | |
3310 | goto out; | |
1c2a07f5 | 3311 | } |
5d4f98a2 | 3312 | if (iref) { |
1c2a07f5 | 3313 | if (path->slots[0] != extent_slot) { |
eee3b811 QW |
3314 | abort_and_dump(trans, path, |
3315 | "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref", | |
3316 | key.objectid, key.type, | |
3317 | key.offset, path->slots[0]); | |
3318 | ret = -EUCLEAN; | |
3319 | goto out; | |
1c2a07f5 | 3320 | } |
5d4f98a2 | 3321 | } else { |
1c2a07f5 QW |
3322 | /* |
3323 | * No inline ref, we must be at SHARED_* item, | |
3324 | * And it's single ref, it must be: | |
3325 | * | extent_slot ||extent_slot + 1| | |
3326 | * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ] | |
3327 | */ | |
3328 | if (path->slots[0] != extent_slot + 1) { | |
eee3b811 QW |
3329 | abort_and_dump(trans, path, |
3330 | "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM", | |
3331 | path->slots[0]); | |
3332 | ret = -EUCLEAN; | |
3333 | goto out; | |
1c2a07f5 | 3334 | } |
5d4f98a2 YZ |
3335 | path->slots[0] = extent_slot; |
3336 | num_to_del = 2; | |
3337 | } | |
78fae27e | 3338 | } |
cecbb533 BB |
3339 | /* |
3340 | * We can't infer the data owner from the delayed ref, so we need | |
3341 | * to try to get it from the owning ref item. | |
3342 | * | |
3343 | * If it is not present, then that extent was not written under | |
3344 | * simple quotas mode, so we don't need to account for its deletion. | |
3345 | */ | |
3346 | if (is_data) | |
3347 | delta.root = btrfs_get_extent_owner_root(trans->fs_info, | |
3348 | leaf, extent_slot); | |
b9473439 | 3349 | |
952fccac CM |
3350 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3351 | num_to_del); | |
005d6427 | 3352 | if (ret) { |
66642832 | 3353 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3354 | goto out; |
3355 | } | |
b3b4aa74 | 3356 | btrfs_release_path(path); |
21af804c | 3357 | |
cecbb533 | 3358 | ret = do_free_extent_accounting(trans, bytenr, &delta); |
a28ec197 | 3359 | } |
fcebe456 JB |
3360 | btrfs_release_path(path); |
3361 | ||
79787eaa | 3362 | out: |
5caf2a00 | 3363 | btrfs_free_path(path); |
a28ec197 CM |
3364 | return ret; |
3365 | } | |
3366 | ||
1887be66 | 3367 | /* |
f0486c68 | 3368 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3369 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3370 | * a given extent, and if there are no other delayed refs to be processed, it | |
3371 | * removes it from the tree. | |
3372 | */ | |
3373 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3374 | u64 bytenr) |
1887be66 CM |
3375 | { |
3376 | struct btrfs_delayed_ref_head *head; | |
3377 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3378 | int ret = 0; |
1887be66 CM |
3379 | |
3380 | delayed_refs = &trans->transaction->delayed_refs; | |
3381 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3382 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3383 | if (!head) |
cf93da7b | 3384 | goto out_delayed_unlock; |
1887be66 | 3385 | |
d7df2c79 | 3386 | spin_lock(&head->lock); |
e3d03965 | 3387 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3388 | goto out; |
3389 | ||
bedc6617 JB |
3390 | if (cleanup_extent_op(head) != NULL) |
3391 | goto out; | |
5d4f98a2 | 3392 | |
1887be66 CM |
3393 | /* |
3394 | * waiting for the lock here would deadlock. If someone else has it | |
3395 | * locked they are already in the process of dropping it anyway | |
3396 | */ | |
3397 | if (!mutex_trylock(&head->mutex)) | |
3398 | goto out; | |
3399 | ||
d7baffda | 3400 | btrfs_delete_ref_head(delayed_refs, head); |
61c681fe | 3401 | head->processing = false; |
d7baffda | 3402 | |
d7df2c79 | 3403 | spin_unlock(&head->lock); |
1887be66 CM |
3404 | spin_unlock(&delayed_refs->lock); |
3405 | ||
f0486c68 YZ |
3406 | BUG_ON(head->extent_op); |
3407 | if (head->must_insert_reserved) | |
3408 | ret = 1; | |
3409 | ||
31890da0 | 3410 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3411 | mutex_unlock(&head->mutex); |
d278850e | 3412 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3413 | return ret; |
1887be66 | 3414 | out: |
d7df2c79 | 3415 | spin_unlock(&head->lock); |
cf93da7b CM |
3416 | |
3417 | out_delayed_unlock: | |
1887be66 CM |
3418 | spin_unlock(&delayed_refs->lock); |
3419 | return 0; | |
3420 | } | |
3421 | ||
f0486c68 | 3422 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7a163608 | 3423 | u64 root_id, |
f0486c68 | 3424 | struct extent_buffer *buf, |
5581a51a | 3425 | u64 parent, int last_ref) |
f0486c68 | 3426 | { |
7a163608 | 3427 | struct btrfs_fs_info *fs_info = trans->fs_info; |
3ba2d364 | 3428 | struct btrfs_block_group *bg; |
f0486c68 YZ |
3429 | int ret; |
3430 | ||
7a163608 | 3431 | if (root_id != BTRFS_TREE_LOG_OBJECTID) { |
4d09b4e9 JB |
3432 | struct btrfs_ref generic_ref = { |
3433 | .action = BTRFS_DROP_DELAYED_REF, | |
3434 | .bytenr = buf->start, | |
12390e42 | 3435 | .num_bytes = buf->len, |
4d09b4e9 JB |
3436 | .parent = parent, |
3437 | .owning_root = btrfs_header_owner(buf), | |
f2e69a77 | 3438 | .ref_root = root_id, |
4d09b4e9 | 3439 | }; |
74cd8cac | 3440 | |
073bda7a NA |
3441 | /* |
3442 | * Assert that the extent buffer is not cleared due to | |
3443 | * EXTENT_BUFFER_ZONED_ZEROOUT. Please refer | |
3444 | * btrfs_clear_buffer_dirty() and btree_csum_one_bio() for | |
3445 | * detail. | |
3446 | */ | |
3447 | ASSERT(btrfs_header_bytenr(buf) != 0); | |
3448 | ||
f2e69a77 | 3449 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), 0, false); |
8a5040f7 | 3450 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 3451 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL); |
79787eaa | 3452 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
3453 | } |
3454 | ||
3ba2d364 JT |
3455 | if (!last_ref) |
3456 | return; | |
f0486c68 | 3457 | |
3ba2d364 JT |
3458 | if (btrfs_header_generation(buf) != trans->transid) |
3459 | goto out; | |
6219872d | 3460 | |
3ba2d364 JT |
3461 | if (root_id != BTRFS_TREE_LOG_OBJECTID) { |
3462 | ret = check_ref_cleanup(trans, buf->start); | |
3463 | if (!ret) | |
37be25bc | 3464 | goto out; |
3ba2d364 | 3465 | } |
f0486c68 | 3466 | |
3ba2d364 | 3467 | bg = btrfs_lookup_block_group(fs_info, buf->start); |
485df755 | 3468 | |
3ba2d364 JT |
3469 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
3470 | pin_down_extent(trans, bg, buf->start, buf->len, 1); | |
3471 | btrfs_put_block_group(bg); | |
3472 | goto out; | |
3473 | } | |
d3575156 | 3474 | |
3ba2d364 JT |
3475 | /* |
3476 | * If there are tree mod log users we may have recorded mod log | |
3477 | * operations for this node. If we re-allocate this node we | |
3478 | * could replay operations on this node that happened when it | |
3479 | * existed in a completely different root. For example if it | |
3480 | * was part of root A, then was reallocated to root B, and we | |
3481 | * are doing a btrfs_old_search_slot(root b), we could replay | |
3482 | * operations that happened when the block was part of root A, | |
3483 | * giving us an inconsistent view of the btree. | |
3484 | * | |
3485 | * We are safe from races here because at this point no other | |
3486 | * node or root points to this extent buffer, so if after this | |
3487 | * check a new tree mod log user joins we will not have an | |
3488 | * existing log of operations on this node that we have to | |
3489 | * contend with. | |
3490 | */ | |
f0486c68 | 3491 | |
3ba2d364 JT |
3492 | if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags) |
3493 | || btrfs_is_zoned(fs_info)) { | |
3494 | pin_down_extent(trans, bg, buf->start, buf->len, 1); | |
3495 | btrfs_put_block_group(bg); | |
3496 | goto out; | |
f0486c68 | 3497 | } |
3ba2d364 JT |
3498 | |
3499 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3500 | ||
3501 | btrfs_add_free_space(bg, buf->start, buf->len); | |
3502 | btrfs_free_reserved_bytes(bg, buf->len, 0); | |
3503 | btrfs_put_block_group(bg); | |
3504 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); | |
3505 | ||
f0486c68 | 3506 | out: |
3ba2d364 JT |
3507 | |
3508 | /* | |
3509 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3510 | * matter anymore. | |
3511 | */ | |
3512 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
3513 | } |
3514 | ||
79787eaa | 3515 | /* Can return -ENOMEM */ |
ffd4bb2a | 3516 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3517 | { |
ffd4bb2a | 3518 | struct btrfs_fs_info *fs_info = trans->fs_info; |
925baedd CM |
3519 | int ret; |
3520 | ||
f5ee5c9a | 3521 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3522 | return 0; |
fccb84c9 | 3523 | |
56bec294 CM |
3524 | /* |
3525 | * tree log blocks never actually go into the extent allocation | |
3526 | * tree, just update pinning info and exit early. | |
56bec294 | 3527 | */ |
f2e69a77 | 3528 | if (ref->ref_root == BTRFS_TREE_LOG_OBJECTID) { |
12390e42 | 3529 | btrfs_pin_extent(trans, ref->bytenr, ref->num_bytes, 1); |
56bec294 | 3530 | ret = 0; |
ffd4bb2a | 3531 | } else if (ref->type == BTRFS_REF_METADATA) { |
2187374f | 3532 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL); |
5d4f98a2 | 3533 | } else { |
2187374f | 3534 | ret = btrfs_add_delayed_data_ref(trans, ref, 0); |
56bec294 | 3535 | } |
d7eae340 | 3536 | |
f2e69a77 | 3537 | if (ref->ref_root != BTRFS_TREE_LOG_OBJECTID) |
ffd4bb2a | 3538 | btrfs_ref_tree_mod(fs_info, ref); |
8a5040f7 | 3539 | |
925baedd CM |
3540 | return ret; |
3541 | } | |
3542 | ||
817d52f8 | 3543 | enum btrfs_loop_type { |
b9d97cff JB |
3544 | /* |
3545 | * Start caching block groups but do not wait for progress or for them | |
3546 | * to be done. | |
3547 | */ | |
f262fa8d | 3548 | LOOP_CACHING_NOWAIT, |
b9d97cff JB |
3549 | |
3550 | /* | |
3551 | * Wait for the block group free_space >= the space we're waiting for if | |
3552 | * the block group isn't cached. | |
3553 | */ | |
f262fa8d | 3554 | LOOP_CACHING_WAIT, |
b9d97cff JB |
3555 | |
3556 | /* | |
3557 | * Allow allocations to happen from block groups that do not yet have a | |
3558 | * size classification. | |
3559 | */ | |
52bb7a21 | 3560 | LOOP_UNSET_SIZE_CLASS, |
b9d97cff JB |
3561 | |
3562 | /* | |
3563 | * Allocate a chunk and then retry the allocation. | |
3564 | */ | |
f262fa8d | 3565 | LOOP_ALLOC_CHUNK, |
b9d97cff JB |
3566 | |
3567 | /* | |
3568 | * Ignore the size class restrictions for this allocation. | |
3569 | */ | |
52bb7a21 | 3570 | LOOP_WRONG_SIZE_CLASS, |
b9d97cff JB |
3571 | |
3572 | /* | |
3573 | * Ignore the empty size, only try to allocate the number of bytes | |
3574 | * needed for this allocation. | |
3575 | */ | |
f262fa8d | 3576 | LOOP_NO_EMPTY_SIZE, |
817d52f8 JB |
3577 | }; |
3578 | ||
e570fd27 | 3579 | static inline void |
32da5386 | 3580 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3581 | int delalloc) |
3582 | { | |
3583 | if (delalloc) | |
3584 | down_read(&cache->data_rwsem); | |
3585 | } | |
3586 | ||
32da5386 | 3587 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3588 | int delalloc) |
3589 | { | |
3590 | btrfs_get_block_group(cache); | |
3591 | if (delalloc) | |
3592 | down_read(&cache->data_rwsem); | |
3593 | } | |
3594 | ||
32da5386 DS |
3595 | static struct btrfs_block_group *btrfs_lock_cluster( |
3596 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3597 | struct btrfs_free_cluster *cluster, |
3598 | int delalloc) | |
c142c6a4 | 3599 | __acquires(&cluster->refill_lock) |
e570fd27 | 3600 | { |
32da5386 | 3601 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3602 | |
e570fd27 | 3603 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3604 | while (1) { |
3605 | used_bg = cluster->block_group; | |
3606 | if (!used_bg) | |
3607 | return NULL; | |
3608 | ||
3609 | if (used_bg == block_group) | |
e570fd27 MX |
3610 | return used_bg; |
3611 | ||
6719afdc | 3612 | btrfs_get_block_group(used_bg); |
e570fd27 | 3613 | |
6719afdc GU |
3614 | if (!delalloc) |
3615 | return used_bg; | |
e570fd27 | 3616 | |
6719afdc GU |
3617 | if (down_read_trylock(&used_bg->data_rwsem)) |
3618 | return used_bg; | |
e570fd27 | 3619 | |
6719afdc | 3620 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3621 | |
e321f8a8 LB |
3622 | /* We should only have one-level nested. */ |
3623 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3624 | |
6719afdc GU |
3625 | spin_lock(&cluster->refill_lock); |
3626 | if (used_bg == cluster->block_group) | |
3627 | return used_bg; | |
e570fd27 | 3628 | |
6719afdc GU |
3629 | up_read(&used_bg->data_rwsem); |
3630 | btrfs_put_block_group(used_bg); | |
3631 | } | |
e570fd27 MX |
3632 | } |
3633 | ||
3634 | static inline void | |
32da5386 | 3635 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3636 | int delalloc) |
3637 | { | |
3638 | if (delalloc) | |
3639 | up_read(&cache->data_rwsem); | |
3640 | btrfs_put_block_group(cache); | |
3641 | } | |
3642 | ||
d06e3bb6 QW |
3643 | /* |
3644 | * Helper function for find_free_extent(). | |
3645 | * | |
3646 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
d06e3bb6 QW |
3647 | * Return >0 to inform caller that we find nothing |
3648 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3649 | */ | |
32da5386 | 3650 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3651 | struct find_free_extent_ctl *ffe_ctl, |
3652 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3653 | { |
32da5386 | 3654 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3655 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3656 | u64 aligned_cluster; |
3657 | u64 offset; | |
3658 | int ret; | |
3659 | ||
3660 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3661 | if (!cluster_bg) | |
3662 | goto refill_cluster; | |
3663 | if (cluster_bg != bg && (cluster_bg->ro || | |
3664 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3665 | goto release_cluster; | |
3666 | ||
3667 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3668 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3669 | &ffe_ctl->max_extent_size); |
3670 | if (offset) { | |
3671 | /* We have a block, we're done */ | |
3672 | spin_unlock(&last_ptr->refill_lock); | |
cfc2de0f | 3673 | trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl); |
d06e3bb6 QW |
3674 | *cluster_bg_ret = cluster_bg; |
3675 | ffe_ctl->found_offset = offset; | |
3676 | return 0; | |
3677 | } | |
3678 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3679 | ||
3680 | release_cluster: | |
3681 | /* | |
3682 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3683 | * lets just skip it and let the allocator find whatever block it can | |
3684 | * find. If we reach this point, we will have tried the cluster | |
3685 | * allocator plenty of times and not have found anything, so we are | |
3686 | * likely way too fragmented for the clustering stuff to find anything. | |
3687 | * | |
3688 | * However, if the cluster is taken from the current block group, | |
3689 | * release the cluster first, so that we stand a better chance of | |
3690 | * succeeding in the unclustered allocation. | |
3691 | */ | |
3692 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3693 | spin_unlock(&last_ptr->refill_lock); | |
3694 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3695 | return -ENOENT; | |
3696 | } | |
3697 | ||
3698 | /* This cluster didn't work out, free it and start over */ | |
3699 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3700 | ||
3701 | if (cluster_bg != bg) | |
3702 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3703 | ||
3704 | refill_cluster: | |
3705 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3706 | spin_unlock(&last_ptr->refill_lock); | |
3707 | return -ENOENT; | |
3708 | } | |
3709 | ||
3710 | aligned_cluster = max_t(u64, | |
3711 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3712 | bg->full_stripe_len); | |
2ceeae2e DS |
3713 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3714 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3715 | if (ret == 0) { |
3716 | /* Now pull our allocation out of this cluster */ | |
3717 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3718 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3719 | &ffe_ctl->max_extent_size); | |
3720 | if (offset) { | |
3721 | /* We found one, proceed */ | |
3722 | spin_unlock(&last_ptr->refill_lock); | |
d06e3bb6 | 3723 | ffe_ctl->found_offset = offset; |
cfc2de0f | 3724 | trace_btrfs_reserve_extent_cluster(bg, ffe_ctl); |
d06e3bb6 QW |
3725 | return 0; |
3726 | } | |
d06e3bb6 QW |
3727 | } |
3728 | /* | |
3729 | * At this point we either didn't find a cluster or we weren't able to | |
3730 | * allocate a block from our cluster. Free the cluster we've been | |
3731 | * trying to use, and go to the next block group. | |
3732 | */ | |
3733 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3734 | spin_unlock(&last_ptr->refill_lock); | |
3735 | return 1; | |
3736 | } | |
3737 | ||
e1a41848 QW |
3738 | /* |
3739 | * Return >0 to inform caller that we find nothing | |
3740 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
e1a41848 | 3741 | */ |
32da5386 | 3742 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3743 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3744 | { |
897cae79 | 3745 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3746 | u64 offset; |
3747 | ||
3748 | /* | |
3749 | * We are doing an unclustered allocation, set the fragmented flag so | |
3750 | * we don't bother trying to setup a cluster again until we get more | |
3751 | * space. | |
3752 | */ | |
3753 | if (unlikely(last_ptr)) { | |
3754 | spin_lock(&last_ptr->lock); | |
3755 | last_ptr->fragmented = 1; | |
3756 | spin_unlock(&last_ptr->lock); | |
3757 | } | |
3758 | if (ffe_ctl->cached) { | |
3759 | struct btrfs_free_space_ctl *free_space_ctl; | |
3760 | ||
3761 | free_space_ctl = bg->free_space_ctl; | |
3762 | spin_lock(&free_space_ctl->tree_lock); | |
3763 | if (free_space_ctl->free_space < | |
3764 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3765 | ffe_ctl->empty_size) { | |
3766 | ffe_ctl->total_free_space = max_t(u64, | |
3767 | ffe_ctl->total_free_space, | |
3768 | free_space_ctl->free_space); | |
3769 | spin_unlock(&free_space_ctl->tree_lock); | |
3770 | return 1; | |
3771 | } | |
3772 | spin_unlock(&free_space_ctl->tree_lock); | |
3773 | } | |
3774 | ||
3775 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3776 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3777 | &ffe_ctl->max_extent_size); | |
cd361199 | 3778 | if (!offset) |
e1a41848 | 3779 | return 1; |
e1a41848 QW |
3780 | ffe_ctl->found_offset = offset; |
3781 | return 0; | |
3782 | } | |
3783 | ||
c668690d NA |
3784 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3785 | struct find_free_extent_ctl *ffe_ctl, | |
3786 | struct btrfs_block_group **bg_ret) | |
3787 | { | |
3788 | int ret; | |
3789 | ||
3790 | /* We want to try and use the cluster allocator, so lets look there */ | |
3791 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3792 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
cd361199 | 3793 | if (ret >= 0) |
c668690d NA |
3794 | return ret; |
3795 | /* ret == -ENOENT case falls through */ | |
3796 | } | |
3797 | ||
897cae79 | 3798 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3799 | } |
3800 | ||
40ab3be1 NA |
3801 | /* |
3802 | * Tree-log block group locking | |
3803 | * ============================ | |
3804 | * | |
3805 | * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which | |
3806 | * indicates the starting address of a block group, which is reserved only | |
3807 | * for tree-log metadata. | |
3808 | * | |
3809 | * Lock nesting | |
3810 | * ============ | |
3811 | * | |
3812 | * space_info::lock | |
3813 | * block_group::lock | |
3814 | * fs_info::treelog_bg_lock | |
3815 | */ | |
3816 | ||
2eda5708 NA |
3817 | /* |
3818 | * Simple allocator for sequential-only block group. It only allows sequential | |
3819 | * allocation. No need to play with trees. This function also reserves the | |
3820 | * bytes as in btrfs_add_reserved_bytes. | |
3821 | */ | |
3822 | static int do_allocation_zoned(struct btrfs_block_group *block_group, | |
3823 | struct find_free_extent_ctl *ffe_ctl, | |
3824 | struct btrfs_block_group **bg_ret) | |
3825 | { | |
40ab3be1 | 3826 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
2eda5708 NA |
3827 | struct btrfs_space_info *space_info = block_group->space_info; |
3828 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3829 | u64 start = block_group->start; | |
3830 | u64 num_bytes = ffe_ctl->num_bytes; | |
3831 | u64 avail; | |
40ab3be1 NA |
3832 | u64 bytenr = block_group->start; |
3833 | u64 log_bytenr; | |
c2707a25 | 3834 | u64 data_reloc_bytenr; |
2eda5708 | 3835 | int ret = 0; |
2d81eb1c | 3836 | bool skip = false; |
2eda5708 NA |
3837 | |
3838 | ASSERT(btrfs_is_zoned(block_group->fs_info)); | |
3839 | ||
40ab3be1 NA |
3840 | /* |
3841 | * Do not allow non-tree-log blocks in the dedicated tree-log block | |
3842 | * group, and vice versa. | |
3843 | */ | |
3844 | spin_lock(&fs_info->treelog_bg_lock); | |
3845 | log_bytenr = fs_info->treelog_bg; | |
2d81eb1c JT |
3846 | if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) || |
3847 | (!ffe_ctl->for_treelog && bytenr == log_bytenr))) | |
3848 | skip = true; | |
40ab3be1 NA |
3849 | spin_unlock(&fs_info->treelog_bg_lock); |
3850 | if (skip) | |
3851 | return 1; | |
3852 | ||
c2707a25 JT |
3853 | /* |
3854 | * Do not allow non-relocation blocks in the dedicated relocation block | |
3855 | * group, and vice versa. | |
3856 | */ | |
3857 | spin_lock(&fs_info->relocation_bg_lock); | |
3858 | data_reloc_bytenr = fs_info->data_reloc_bg; | |
3859 | if (data_reloc_bytenr && | |
3860 | ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) || | |
3861 | (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr))) | |
3862 | skip = true; | |
3863 | spin_unlock(&fs_info->relocation_bg_lock); | |
3864 | if (skip) | |
3865 | return 1; | |
1ada69f6 | 3866 | |
2e654e4b NA |
3867 | /* Check RO and no space case before trying to activate it */ |
3868 | spin_lock(&block_group->lock); | |
1bfd4767 | 3869 | if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) { |
1ada69f6 NA |
3870 | ret = 1; |
3871 | /* | |
3872 | * May need to clear fs_info->{treelog,data_reloc}_bg. | |
3873 | * Return the error after taking the locks. | |
3874 | */ | |
2e654e4b NA |
3875 | } |
3876 | spin_unlock(&block_group->lock); | |
3877 | ||
5a7d107e NA |
3878 | /* Metadata block group is activated at write time. */ |
3879 | if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) && | |
3880 | !btrfs_zone_activate(block_group)) { | |
1ada69f6 NA |
3881 | ret = 1; |
3882 | /* | |
3883 | * May need to clear fs_info->{treelog,data_reloc}_bg. | |
3884 | * Return the error after taking the locks. | |
3885 | */ | |
3886 | } | |
2e654e4b | 3887 | |
2eda5708 NA |
3888 | spin_lock(&space_info->lock); |
3889 | spin_lock(&block_group->lock); | |
40ab3be1 | 3890 | spin_lock(&fs_info->treelog_bg_lock); |
c2707a25 | 3891 | spin_lock(&fs_info->relocation_bg_lock); |
40ab3be1 | 3892 | |
1ada69f6 NA |
3893 | if (ret) |
3894 | goto out; | |
3895 | ||
40ab3be1 NA |
3896 | ASSERT(!ffe_ctl->for_treelog || |
3897 | block_group->start == fs_info->treelog_bg || | |
3898 | fs_info->treelog_bg == 0); | |
c2707a25 JT |
3899 | ASSERT(!ffe_ctl->for_data_reloc || |
3900 | block_group->start == fs_info->data_reloc_bg || | |
3901 | fs_info->data_reloc_bg == 0); | |
2eda5708 | 3902 | |
3349b57f | 3903 | if (block_group->ro || |
332581bd NA |
3904 | (!ffe_ctl->for_data_reloc && |
3905 | test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) { | |
2eda5708 NA |
3906 | ret = 1; |
3907 | goto out; | |
3908 | } | |
3909 | ||
40ab3be1 NA |
3910 | /* |
3911 | * Do not allow currently using block group to be tree-log dedicated | |
3912 | * block group. | |
3913 | */ | |
3914 | if (ffe_ctl->for_treelog && !fs_info->treelog_bg && | |
3915 | (block_group->used || block_group->reserved)) { | |
3916 | ret = 1; | |
3917 | goto out; | |
3918 | } | |
3919 | ||
c2707a25 JT |
3920 | /* |
3921 | * Do not allow currently used block group to be the data relocation | |
3922 | * dedicated block group. | |
3923 | */ | |
3924 | if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg && | |
3925 | (block_group->used || block_group->reserved)) { | |
3926 | ret = 1; | |
3927 | goto out; | |
3928 | } | |
3929 | ||
98173255 NA |
3930 | WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity); |
3931 | avail = block_group->zone_capacity - block_group->alloc_offset; | |
2eda5708 NA |
3932 | if (avail < num_bytes) { |
3933 | if (ffe_ctl->max_extent_size < avail) { | |
3934 | /* | |
3935 | * With sequential allocator, free space is always | |
3936 | * contiguous | |
3937 | */ | |
3938 | ffe_ctl->max_extent_size = avail; | |
3939 | ffe_ctl->total_free_space = avail; | |
3940 | } | |
3941 | ret = 1; | |
3942 | goto out; | |
3943 | } | |
3944 | ||
40ab3be1 NA |
3945 | if (ffe_ctl->for_treelog && !fs_info->treelog_bg) |
3946 | fs_info->treelog_bg = block_group->start; | |
3947 | ||
332581bd NA |
3948 | if (ffe_ctl->for_data_reloc) { |
3949 | if (!fs_info->data_reloc_bg) | |
3950 | fs_info->data_reloc_bg = block_group->start; | |
3951 | /* | |
3952 | * Do not allow allocations from this block group, unless it is | |
3953 | * for data relocation. Compared to increasing the ->ro, setting | |
3954 | * the ->zoned_data_reloc_ongoing flag still allows nocow | |
3955 | * writers to come in. See btrfs_inc_nocow_writers(). | |
3956 | * | |
3957 | * We need to disable an allocation to avoid an allocation of | |
3958 | * regular (non-relocation data) extent. With mix of relocation | |
3959 | * extents and regular extents, we can dispatch WRITE commands | |
3960 | * (for relocation extents) and ZONE APPEND commands (for | |
3961 | * regular extents) at the same time to the same zone, which | |
3962 | * easily break the write pointer. | |
3963 | * | |
3964 | * Also, this flag avoids this block group to be zone finished. | |
3965 | */ | |
3966 | set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags); | |
3967 | } | |
c2707a25 | 3968 | |
2eda5708 NA |
3969 | ffe_ctl->found_offset = start + block_group->alloc_offset; |
3970 | block_group->alloc_offset += num_bytes; | |
3971 | spin_lock(&ctl->tree_lock); | |
3972 | ctl->free_space -= num_bytes; | |
3973 | spin_unlock(&ctl->tree_lock); | |
3974 | ||
3975 | /* | |
3976 | * We do not check if found_offset is aligned to stripesize. The | |
3977 | * address is anyway rewritten when using zone append writing. | |
3978 | */ | |
3979 | ||
3980 | ffe_ctl->search_start = ffe_ctl->found_offset; | |
3981 | ||
3982 | out: | |
40ab3be1 NA |
3983 | if (ret && ffe_ctl->for_treelog) |
3984 | fs_info->treelog_bg = 0; | |
332581bd | 3985 | if (ret && ffe_ctl->for_data_reloc) |
c2707a25 JT |
3986 | fs_info->data_reloc_bg = 0; |
3987 | spin_unlock(&fs_info->relocation_bg_lock); | |
40ab3be1 | 3988 | spin_unlock(&fs_info->treelog_bg_lock); |
2eda5708 NA |
3989 | spin_unlock(&block_group->lock); |
3990 | spin_unlock(&space_info->lock); | |
3991 | return ret; | |
3992 | } | |
3993 | ||
c668690d NA |
3994 | static int do_allocation(struct btrfs_block_group *block_group, |
3995 | struct find_free_extent_ctl *ffe_ctl, | |
3996 | struct btrfs_block_group **bg_ret) | |
3997 | { | |
3998 | switch (ffe_ctl->policy) { | |
3999 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4000 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
2eda5708 NA |
4001 | case BTRFS_EXTENT_ALLOC_ZONED: |
4002 | return do_allocation_zoned(block_group, ffe_ctl, bg_ret); | |
c668690d NA |
4003 | default: |
4004 | BUG(); | |
4005 | } | |
4006 | } | |
4007 | ||
baba5062 NA |
4008 | static void release_block_group(struct btrfs_block_group *block_group, |
4009 | struct find_free_extent_ctl *ffe_ctl, | |
4010 | int delalloc) | |
4011 | { | |
4012 | switch (ffe_ctl->policy) { | |
4013 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
cd361199 | 4014 | ffe_ctl->retry_uncached = false; |
baba5062 | 4015 | break; |
2eda5708 NA |
4016 | case BTRFS_EXTENT_ALLOC_ZONED: |
4017 | /* Nothing to do */ | |
4018 | break; | |
baba5062 NA |
4019 | default: |
4020 | BUG(); | |
4021 | } | |
4022 | ||
4023 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
4024 | ffe_ctl->index); | |
4025 | btrfs_release_block_group(block_group, delalloc); | |
4026 | } | |
4027 | ||
0ab9724b NA |
4028 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
4029 | struct btrfs_key *ins) | |
4030 | { | |
4031 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
4032 | ||
4033 | if (!ffe_ctl->use_cluster && last_ptr) { | |
4034 | spin_lock(&last_ptr->lock); | |
4035 | last_ptr->window_start = ins->objectid; | |
4036 | spin_unlock(&last_ptr->lock); | |
4037 | } | |
4038 | } | |
4039 | ||
4040 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
4041 | struct btrfs_key *ins) | |
4042 | { | |
4043 | switch (ffe_ctl->policy) { | |
4044 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4045 | found_extent_clustered(ffe_ctl, ins); | |
4046 | break; | |
2eda5708 NA |
4047 | case BTRFS_EXTENT_ALLOC_ZONED: |
4048 | /* Nothing to do */ | |
4049 | break; | |
0ab9724b NA |
4050 | default: |
4051 | BUG(); | |
4052 | } | |
4053 | } | |
4054 | ||
393f646e NA |
4055 | static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info, |
4056 | struct find_free_extent_ctl *ffe_ctl) | |
4057 | { | |
5a7d107e NA |
4058 | /* Block group's activeness is not a requirement for METADATA block groups. */ |
4059 | if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)) | |
4060 | return 0; | |
4061 | ||
393f646e NA |
4062 | /* If we can activate new zone, just allocate a chunk and use it */ |
4063 | if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags)) | |
4064 | return 0; | |
4065 | ||
4066 | /* | |
4067 | * We already reached the max active zones. Try to finish one block | |
4068 | * group to make a room for a new block group. This is only possible | |
4069 | * for a data block group because btrfs_zone_finish() may need to wait | |
4070 | * for a running transaction which can cause a deadlock for metadata | |
4071 | * allocation. | |
4072 | */ | |
4073 | if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) { | |
4074 | int ret = btrfs_zone_finish_one_bg(fs_info); | |
4075 | ||
4076 | if (ret == 1) | |
4077 | return 0; | |
4078 | else if (ret < 0) | |
4079 | return ret; | |
4080 | } | |
4081 | ||
4082 | /* | |
4083 | * If we have enough free space left in an already active block group | |
4084 | * and we can't activate any other zone now, do not allow allocating a | |
4085 | * new chunk and let find_free_extent() retry with a smaller size. | |
4086 | */ | |
4087 | if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size) | |
4088 | return -ENOSPC; | |
4089 | ||
898793d9 NA |
4090 | /* |
4091 | * Even min_alloc_size is not left in any block groups. Since we cannot | |
4092 | * activate a new block group, allocating it may not help. Let's tell a | |
4093 | * caller to try again and hope it progress something by writing some | |
4094 | * parts of the region. That is only possible for data block groups, | |
4095 | * where a part of the region can be written. | |
4096 | */ | |
4097 | if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) | |
4098 | return -EAGAIN; | |
4099 | ||
393f646e NA |
4100 | /* |
4101 | * We cannot activate a new block group and no enough space left in any | |
4102 | * block groups. So, allocating a new block group may not help. But, | |
4103 | * there is nothing to do anyway, so let's go with it. | |
4104 | */ | |
4105 | return 0; | |
4106 | } | |
4107 | ||
bb9950d3 NA |
4108 | static int can_allocate_chunk(struct btrfs_fs_info *fs_info, |
4109 | struct find_free_extent_ctl *ffe_ctl) | |
50475cd5 NA |
4110 | { |
4111 | switch (ffe_ctl->policy) { | |
4112 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
bb9950d3 | 4113 | return 0; |
50475cd5 | 4114 | case BTRFS_EXTENT_ALLOC_ZONED: |
393f646e | 4115 | return can_allocate_chunk_zoned(fs_info, ffe_ctl); |
50475cd5 NA |
4116 | default: |
4117 | BUG(); | |
4118 | } | |
4119 | } | |
4120 | ||
e72d79d6 QW |
4121 | /* |
4122 | * Return >0 means caller needs to re-search for free extent | |
4123 | * Return 0 means we have the needed free extent. | |
4124 | * Return <0 means we failed to locate any free extent. | |
4125 | */ | |
4126 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
4127 | struct btrfs_key *ins, |
4128 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 4129 | bool full_search) |
e72d79d6 | 4130 | { |
8e1d0290 | 4131 | struct btrfs_root *root = fs_info->chunk_root; |
e72d79d6 QW |
4132 | int ret; |
4133 | ||
4134 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
4135 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
4136 | ffe_ctl->orig_have_caching_bg = true; | |
4137 | ||
e72d79d6 | 4138 | if (ins->objectid) { |
0ab9724b | 4139 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
4140 | return 0; |
4141 | } | |
4142 | ||
a85f05e5 NA |
4143 | if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg) |
4144 | return 1; | |
4145 | ||
4146 | ffe_ctl->index++; | |
4147 | if (ffe_ctl->index < BTRFS_NR_RAID_TYPES) | |
4148 | return 1; | |
4149 | ||
b9d97cff | 4150 | /* See the comments for btrfs_loop_type for an explanation of the phases. */ |
e72d79d6 QW |
4151 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { |
4152 | ffe_ctl->index = 0; | |
52bb7a21 BB |
4153 | /* |
4154 | * We want to skip the LOOP_CACHING_WAIT step if we don't have | |
4155 | * any uncached bgs and we've already done a full search | |
4156 | * through. | |
4157 | */ | |
4158 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT && | |
4159 | (!ffe_ctl->orig_have_caching_bg && full_search)) | |
e72d79d6 | 4160 | ffe_ctl->loop++; |
52bb7a21 | 4161 | ffe_ctl->loop++; |
e72d79d6 QW |
4162 | |
4163 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
4164 | struct btrfs_trans_handle *trans; | |
4165 | int exist = 0; | |
4166 | ||
52bb7a21 | 4167 | /* Check if allocation policy allows to create a new chunk */ |
bb9950d3 NA |
4168 | ret = can_allocate_chunk(fs_info, ffe_ctl); |
4169 | if (ret) | |
4170 | return ret; | |
50475cd5 | 4171 | |
e72d79d6 QW |
4172 | trans = current->journal_info; |
4173 | if (trans) | |
4174 | exist = 1; | |
4175 | else | |
4176 | trans = btrfs_join_transaction(root); | |
4177 | ||
4178 | if (IS_ERR(trans)) { | |
4179 | ret = PTR_ERR(trans); | |
4180 | return ret; | |
4181 | } | |
4182 | ||
fc471cb0 | 4183 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
760e69c4 | 4184 | CHUNK_ALLOC_FORCE_FOR_EXTENT); |
e72d79d6 | 4185 | |
e72d79d6 | 4186 | /* Do not bail out on ENOSPC since we can do more. */ |
52bb7a21 BB |
4187 | if (ret == -ENOSPC) { |
4188 | ret = 0; | |
4189 | ffe_ctl->loop++; | |
4190 | } | |
c70e2139 | 4191 | else if (ret < 0) |
e72d79d6 QW |
4192 | btrfs_abort_transaction(trans, ret); |
4193 | else | |
4194 | ret = 0; | |
4195 | if (!exist) | |
4196 | btrfs_end_transaction(trans); | |
4197 | if (ret) | |
4198 | return ret; | |
4199 | } | |
4200 | ||
4201 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
4202 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
4203 | return -ENOSPC; | |
4204 | ||
e72d79d6 QW |
4205 | /* |
4206 | * Don't loop again if we already have no empty_size and | |
4207 | * no empty_cluster. | |
4208 | */ | |
4209 | if (ffe_ctl->empty_size == 0 && | |
4210 | ffe_ctl->empty_cluster == 0) | |
4211 | return -ENOSPC; | |
4212 | ffe_ctl->empty_size = 0; | |
4213 | ffe_ctl->empty_cluster = 0; | |
4214 | } | |
4215 | return 1; | |
4216 | } | |
4217 | return -ENOSPC; | |
4218 | } | |
4219 | ||
52bb7a21 BB |
4220 | static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl, |
4221 | struct btrfs_block_group *bg) | |
4222 | { | |
4223 | if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED) | |
4224 | return true; | |
cb0922f2 | 4225 | if (!btrfs_block_group_should_use_size_class(bg)) |
52bb7a21 BB |
4226 | return true; |
4227 | if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS) | |
4228 | return true; | |
4229 | if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS && | |
4230 | bg->size_class == BTRFS_BG_SZ_NONE) | |
4231 | return true; | |
4232 | return ffe_ctl->size_class == bg->size_class; | |
4233 | } | |
4234 | ||
7e895409 NA |
4235 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
4236 | struct find_free_extent_ctl *ffe_ctl, | |
4237 | struct btrfs_space_info *space_info, | |
4238 | struct btrfs_key *ins) | |
4239 | { | |
4240 | /* | |
4241 | * If our free space is heavily fragmented we may not be able to make | |
4242 | * big contiguous allocations, so instead of doing the expensive search | |
4243 | * for free space, simply return ENOSPC with our max_extent_size so we | |
4244 | * can go ahead and search for a more manageable chunk. | |
4245 | * | |
4246 | * If our max_extent_size is large enough for our allocation simply | |
4247 | * disable clustering since we will likely not be able to find enough | |
4248 | * space to create a cluster and induce latency trying. | |
4249 | */ | |
4250 | if (space_info->max_extent_size) { | |
4251 | spin_lock(&space_info->lock); | |
4252 | if (space_info->max_extent_size && | |
4253 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
4254 | ins->offset = space_info->max_extent_size; | |
4255 | spin_unlock(&space_info->lock); | |
4256 | return -ENOSPC; | |
4257 | } else if (space_info->max_extent_size) { | |
4258 | ffe_ctl->use_cluster = false; | |
4259 | } | |
4260 | spin_unlock(&space_info->lock); | |
4261 | } | |
4262 | ||
4263 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
4264 | &ffe_ctl->empty_cluster); | |
4265 | if (ffe_ctl->last_ptr) { | |
4266 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
4267 | ||
4268 | spin_lock(&last_ptr->lock); | |
4269 | if (last_ptr->block_group) | |
4270 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4271 | if (last_ptr->fragmented) { | |
4272 | /* | |
4273 | * We still set window_start so we can keep track of the | |
4274 | * last place we found an allocation to try and save | |
4275 | * some time. | |
4276 | */ | |
4277 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4278 | ffe_ctl->use_cluster = false; | |
4279 | } | |
4280 | spin_unlock(&last_ptr->lock); | |
4281 | } | |
4282 | ||
4283 | return 0; | |
4284 | } | |
4285 | ||
b271fee9 NA |
4286 | static int prepare_allocation_zoned(struct btrfs_fs_info *fs_info, |
4287 | struct find_free_extent_ctl *ffe_ctl) | |
4288 | { | |
4289 | if (ffe_ctl->for_treelog) { | |
4290 | spin_lock(&fs_info->treelog_bg_lock); | |
4291 | if (fs_info->treelog_bg) | |
4292 | ffe_ctl->hint_byte = fs_info->treelog_bg; | |
4293 | spin_unlock(&fs_info->treelog_bg_lock); | |
4294 | } else if (ffe_ctl->for_data_reloc) { | |
4295 | spin_lock(&fs_info->relocation_bg_lock); | |
4296 | if (fs_info->data_reloc_bg) | |
4297 | ffe_ctl->hint_byte = fs_info->data_reloc_bg; | |
4298 | spin_unlock(&fs_info->relocation_bg_lock); | |
02444f2a NA |
4299 | } else if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) { |
4300 | struct btrfs_block_group *block_group; | |
4301 | ||
4302 | spin_lock(&fs_info->zone_active_bgs_lock); | |
4303 | list_for_each_entry(block_group, &fs_info->zone_active_bgs, active_bg_list) { | |
4304 | /* | |
4305 | * No lock is OK here because avail is monotinically | |
4306 | * decreasing, and this is just a hint. | |
4307 | */ | |
4308 | u64 avail = block_group->zone_capacity - block_group->alloc_offset; | |
4309 | ||
4310 | if (block_group_bits(block_group, ffe_ctl->flags) && | |
4311 | avail >= ffe_ctl->num_bytes) { | |
4312 | ffe_ctl->hint_byte = block_group->start; | |
4313 | break; | |
4314 | } | |
4315 | } | |
4316 | spin_unlock(&fs_info->zone_active_bgs_lock); | |
b271fee9 NA |
4317 | } |
4318 | ||
4319 | return 0; | |
4320 | } | |
4321 | ||
7e895409 NA |
4322 | static int prepare_allocation(struct btrfs_fs_info *fs_info, |
4323 | struct find_free_extent_ctl *ffe_ctl, | |
4324 | struct btrfs_space_info *space_info, | |
4325 | struct btrfs_key *ins) | |
4326 | { | |
4327 | switch (ffe_ctl->policy) { | |
4328 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4329 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
4330 | space_info, ins); | |
2eda5708 | 4331 | case BTRFS_EXTENT_ALLOC_ZONED: |
b271fee9 | 4332 | return prepare_allocation_zoned(fs_info, ffe_ctl); |
7e895409 NA |
4333 | default: |
4334 | BUG(); | |
4335 | } | |
4336 | } | |
4337 | ||
fec577fb CM |
4338 | /* |
4339 | * walks the btree of allocated extents and find a hole of a given size. | |
4340 | * The key ins is changed to record the hole: | |
a4820398 | 4341 | * ins->objectid == start position |
62e2749e | 4342 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4343 | * ins->offset == the size of the hole. |
fec577fb | 4344 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4345 | * |
4346 | * If there is no suitable free space, we will record the max size of | |
4347 | * the free space extent currently. | |
e72d79d6 QW |
4348 | * |
4349 | * The overall logic and call chain: | |
4350 | * | |
4351 | * find_free_extent() | |
4352 | * |- Iterate through all block groups | |
4353 | * | |- Get a valid block group | |
4354 | * | |- Try to do clustered allocation in that block group | |
4355 | * | |- Try to do unclustered allocation in that block group | |
4356 | * | |- Check if the result is valid | |
4357 | * | | |- If valid, then exit | |
4358 | * | |- Jump to next block group | |
4359 | * | | |
4360 | * |- Push harder to find free extents | |
4361 | * |- If not found, re-iterate all block groups | |
fec577fb | 4362 | */ |
437490fe | 4363 | static noinline int find_free_extent(struct btrfs_root *root, |
a12b0dc0 NA |
4364 | struct btrfs_key *ins, |
4365 | struct find_free_extent_ctl *ffe_ctl) | |
fec577fb | 4366 | { |
437490fe | 4367 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4368 | int ret = 0; |
db8fe64f | 4369 | int cache_block_group_error = 0; |
32da5386 | 4370 | struct btrfs_block_group *block_group = NULL; |
80eb234a | 4371 | struct btrfs_space_info *space_info; |
a5e681d9 | 4372 | bool full_search = false; |
fec577fb | 4373 | |
a12b0dc0 | 4374 | WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize); |
b4bd745d | 4375 | |
a12b0dc0 NA |
4376 | ffe_ctl->search_start = 0; |
4377 | /* For clustered allocation */ | |
4378 | ffe_ctl->empty_cluster = 0; | |
4379 | ffe_ctl->last_ptr = NULL; | |
4380 | ffe_ctl->use_cluster = true; | |
4381 | ffe_ctl->have_caching_bg = false; | |
4382 | ffe_ctl->orig_have_caching_bg = false; | |
4383 | ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags); | |
4384 | ffe_ctl->loop = 0; | |
cd361199 | 4385 | ffe_ctl->retry_uncached = false; |
a12b0dc0 NA |
4386 | ffe_ctl->cached = 0; |
4387 | ffe_ctl->max_extent_size = 0; | |
4388 | ffe_ctl->total_free_space = 0; | |
4389 | ffe_ctl->found_offset = 0; | |
4390 | ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED; | |
52bb7a21 | 4391 | ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes); |
c10859be | 4392 | |
2eda5708 | 4393 | if (btrfs_is_zoned(fs_info)) |
a12b0dc0 | 4394 | ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED; |
2eda5708 | 4395 | |
962a298f | 4396 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4397 | ins->objectid = 0; |
4398 | ins->offset = 0; | |
b1a4d965 | 4399 | |
cfc2de0f | 4400 | trace_find_free_extent(root, ffe_ctl); |
3f7de037 | 4401 | |
a12b0dc0 | 4402 | space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags); |
1b1d1f66 | 4403 | if (!space_info) { |
a12b0dc0 | 4404 | btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags); |
1b1d1f66 JB |
4405 | return -ENOSPC; |
4406 | } | |
2552d17e | 4407 | |
a12b0dc0 | 4408 | ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins); |
7e895409 NA |
4409 | if (ret < 0) |
4410 | return ret; | |
fa9c0d79 | 4411 | |
a12b0dc0 | 4412 | ffe_ctl->search_start = max(ffe_ctl->search_start, |
0eb997bf | 4413 | first_logical_byte(fs_info)); |
a12b0dc0 NA |
4414 | ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte); |
4415 | if (ffe_ctl->search_start == ffe_ctl->hint_byte) { | |
b4bd745d | 4416 | block_group = btrfs_lookup_block_group(fs_info, |
a12b0dc0 | 4417 | ffe_ctl->search_start); |
817d52f8 JB |
4418 | /* |
4419 | * we don't want to use the block group if it doesn't match our | |
4420 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4421 | * |
4422 | * However if we are re-searching with an ideal block group | |
4423 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4424 | */ |
a12b0dc0 | 4425 | if (block_group && block_group_bits(block_group, ffe_ctl->flags) && |
285ff5af | 4426 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4427 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4428 | if (list_empty(&block_group->list) || |
4429 | block_group->ro) { | |
4430 | /* | |
4431 | * someone is removing this block group, | |
4432 | * we can't jump into the have_block_group | |
4433 | * target because our list pointers are not | |
4434 | * valid | |
4435 | */ | |
4436 | btrfs_put_block_group(block_group); | |
4437 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4438 | } else { |
a12b0dc0 NA |
4439 | ffe_ctl->index = btrfs_bg_flags_to_raid_index( |
4440 | block_group->flags); | |
4441 | btrfs_lock_block_group(block_group, | |
4442 | ffe_ctl->delalloc); | |
854c2f36 | 4443 | ffe_ctl->hinted = true; |
44fb5511 | 4444 | goto have_block_group; |
ccf0e725 | 4445 | } |
2552d17e | 4446 | } else if (block_group) { |
fa9c0d79 | 4447 | btrfs_put_block_group(block_group); |
2552d17e | 4448 | } |
42e70e7a | 4449 | } |
2552d17e | 4450 | search: |
854c2f36 | 4451 | trace_find_free_extent_search_loop(root, ffe_ctl); |
a12b0dc0 NA |
4452 | ffe_ctl->have_caching_bg = false; |
4453 | if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) || | |
4454 | ffe_ctl->index == 0) | |
a5e681d9 | 4455 | full_search = true; |
80eb234a | 4456 | down_read(&space_info->groups_sem); |
b4bd745d | 4457 | list_for_each_entry(block_group, |
a12b0dc0 | 4458 | &space_info->block_groups[ffe_ctl->index], list) { |
c668690d NA |
4459 | struct btrfs_block_group *bg_ret; |
4460 | ||
854c2f36 | 4461 | ffe_ctl->hinted = false; |
14443937 | 4462 | /* If the block group is read-only, we can skip it entirely. */ |
40ab3be1 | 4463 | if (unlikely(block_group->ro)) { |
a12b0dc0 | 4464 | if (ffe_ctl->for_treelog) |
40ab3be1 | 4465 | btrfs_clear_treelog_bg(block_group); |
c2707a25 JT |
4466 | if (ffe_ctl->for_data_reloc) |
4467 | btrfs_clear_data_reloc_bg(block_group); | |
14443937 | 4468 | continue; |
40ab3be1 | 4469 | } |
14443937 | 4470 | |
a12b0dc0 NA |
4471 | btrfs_grab_block_group(block_group, ffe_ctl->delalloc); |
4472 | ffe_ctl->search_start = block_group->start; | |
42e70e7a | 4473 | |
83a50de9 CM |
4474 | /* |
4475 | * this can happen if we end up cycling through all the | |
4476 | * raid types, but we want to make sure we only allocate | |
4477 | * for the proper type. | |
4478 | */ | |
a12b0dc0 | 4479 | if (!block_group_bits(block_group, ffe_ctl->flags)) { |
bece2e82 | 4480 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4481 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4482 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4483 | BTRFS_BLOCK_GROUP_RAID10; |
4484 | ||
4485 | /* | |
4486 | * if they asked for extra copies and this block group | |
4487 | * doesn't provide them, bail. This does allow us to | |
4488 | * fill raid0 from raid1. | |
4489 | */ | |
a12b0dc0 | 4490 | if ((ffe_ctl->flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4491 | goto loop; |
2a28468e QW |
4492 | |
4493 | /* | |
4494 | * This block group has different flags than we want. | |
4495 | * It's possible that we have MIXED_GROUP flag but no | |
4496 | * block group is mixed. Just skip such block group. | |
4497 | */ | |
a12b0dc0 | 4498 | btrfs_release_block_group(block_group, ffe_ctl->delalloc); |
2a28468e | 4499 | continue; |
83a50de9 CM |
4500 | } |
4501 | ||
2552d17e | 4502 | have_block_group: |
854c2f36 | 4503 | trace_find_free_extent_have_block_group(root, ffe_ctl, block_group); |
a12b0dc0 NA |
4504 | ffe_ctl->cached = btrfs_block_group_done(block_group); |
4505 | if (unlikely(!ffe_ctl->cached)) { | |
4506 | ffe_ctl->have_caching_bg = true; | |
ced8ecf0 | 4507 | ret = btrfs_cache_block_group(block_group, false); |
db8fe64f JB |
4508 | |
4509 | /* | |
4510 | * If we get ENOMEM here or something else we want to | |
4511 | * try other block groups, because it may not be fatal. | |
4512 | * However if we can't find anything else we need to | |
4513 | * save our return here so that we return the actual | |
4514 | * error that caused problems, not ENOSPC. | |
4515 | */ | |
4516 | if (ret < 0) { | |
4517 | if (!cache_block_group_error) | |
4518 | cache_block_group_error = ret; | |
4519 | ret = 0; | |
4520 | goto loop; | |
4521 | } | |
1d4284bd | 4522 | ret = 0; |
817d52f8 JB |
4523 | } |
4524 | ||
92fb94b6 JB |
4525 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) { |
4526 | if (!cache_block_group_error) | |
4527 | cache_block_group_error = -EIO; | |
36cce922 | 4528 | goto loop; |
92fb94b6 | 4529 | } |
0f9dd46c | 4530 | |
52bb7a21 BB |
4531 | if (!find_free_extent_check_size_class(ffe_ctl, block_group)) |
4532 | goto loop; | |
4533 | ||
c668690d | 4534 | bg_ret = NULL; |
a12b0dc0 | 4535 | ret = do_allocation(block_group, ffe_ctl, &bg_ret); |
cd361199 | 4536 | if (ret > 0) |
1cdda9b8 | 4537 | goto loop; |
cd361199 JB |
4538 | |
4539 | if (bg_ret && bg_ret != block_group) { | |
4540 | btrfs_release_block_group(block_group, ffe_ctl->delalloc); | |
4541 | block_group = bg_ret; | |
c668690d NA |
4542 | } |
4543 | ||
4544 | /* Checks */ | |
a12b0dc0 NA |
4545 | ffe_ctl->search_start = round_up(ffe_ctl->found_offset, |
4546 | fs_info->stripesize); | |
25179201 | 4547 | |
2552d17e | 4548 | /* move on to the next group */ |
a12b0dc0 | 4549 | if (ffe_ctl->search_start + ffe_ctl->num_bytes > |
b3470b5d | 4550 | block_group->start + block_group->length) { |
2eda5708 | 4551 | btrfs_add_free_space_unused(block_group, |
a12b0dc0 NA |
4552 | ffe_ctl->found_offset, |
4553 | ffe_ctl->num_bytes); | |
2552d17e | 4554 | goto loop; |
6226cb0a | 4555 | } |
f5a31e16 | 4556 | |
a12b0dc0 | 4557 | if (ffe_ctl->found_offset < ffe_ctl->search_start) |
2eda5708 | 4558 | btrfs_add_free_space_unused(block_group, |
a12b0dc0 NA |
4559 | ffe_ctl->found_offset, |
4560 | ffe_ctl->search_start - ffe_ctl->found_offset); | |
2552d17e | 4561 | |
a12b0dc0 NA |
4562 | ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes, |
4563 | ffe_ctl->num_bytes, | |
52bb7a21 BB |
4564 | ffe_ctl->delalloc, |
4565 | ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS); | |
f0486c68 | 4566 | if (ret == -EAGAIN) { |
2eda5708 | 4567 | btrfs_add_free_space_unused(block_group, |
a12b0dc0 NA |
4568 | ffe_ctl->found_offset, |
4569 | ffe_ctl->num_bytes); | |
2552d17e | 4570 | goto loop; |
0f9dd46c | 4571 | } |
9cfa3e34 | 4572 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4573 | |
f0486c68 | 4574 | /* we are all good, lets return */ |
a12b0dc0 NA |
4575 | ins->objectid = ffe_ctl->search_start; |
4576 | ins->offset = ffe_ctl->num_bytes; | |
d2fb3437 | 4577 | |
cfc2de0f | 4578 | trace_btrfs_reserve_extent(block_group, ffe_ctl); |
a12b0dc0 | 4579 | btrfs_release_block_group(block_group, ffe_ctl->delalloc); |
2552d17e JB |
4580 | break; |
4581 | loop: | |
cd361199 JB |
4582 | if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && |
4583 | !ffe_ctl->retry_uncached) { | |
4584 | ffe_ctl->retry_uncached = true; | |
4585 | btrfs_wait_block_group_cache_progress(block_group, | |
4586 | ffe_ctl->num_bytes + | |
4587 | ffe_ctl->empty_cluster + | |
4588 | ffe_ctl->empty_size); | |
4589 | goto have_block_group; | |
4590 | } | |
a12b0dc0 | 4591 | release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc); |
14443937 | 4592 | cond_resched(); |
2552d17e JB |
4593 | } |
4594 | up_read(&space_info->groups_sem); | |
4595 | ||
a12b0dc0 | 4596 | ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search); |
e72d79d6 | 4597 | if (ret > 0) |
b742bb82 YZ |
4598 | goto search; |
4599 | ||
db8fe64f | 4600 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4601 | /* |
4602 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4603 | * any contiguous hole. | |
4604 | */ | |
a12b0dc0 NA |
4605 | if (!ffe_ctl->max_extent_size) |
4606 | ffe_ctl->max_extent_size = ffe_ctl->total_free_space; | |
4f4db217 | 4607 | spin_lock(&space_info->lock); |
a12b0dc0 | 4608 | space_info->max_extent_size = ffe_ctl->max_extent_size; |
4f4db217 | 4609 | spin_unlock(&space_info->lock); |
a12b0dc0 | 4610 | ins->offset = ffe_ctl->max_extent_size; |
db8fe64f JB |
4611 | } else if (ret == -ENOSPC) { |
4612 | ret = cache_block_group_error; | |
4f4db217 | 4613 | } |
0f70abe2 | 4614 | return ret; |
fec577fb | 4615 | } |
ec44a35c | 4616 | |
6f47c706 | 4617 | /* |
9580503b DS |
4618 | * Entry point to the extent allocator. Tries to find a hole that is at least |
4619 | * as big as @num_bytes. | |
6f47c706 NB |
4620 | * |
4621 | * @root - The root that will contain this extent | |
4622 | * | |
4623 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4624 | * is used for accounting purposes. This value differs | |
4625 | * from @num_bytes only in the case of compressed extents. | |
4626 | * | |
4627 | * @num_bytes - Number of bytes to allocate on-disk. | |
4628 | * | |
4629 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4630 | * allocator should try to satisfy. In some cases | |
4631 | * @num_bytes may be larger than what is required and if | |
4632 | * the filesystem is fragmented then allocation fails. | |
4633 | * However, the presence of @min_alloc_size gives a | |
4634 | * chance to try and satisfy the smaller allocation. | |
4635 | * | |
4636 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4637 | * size in bytes the allocator should try to find free | |
4638 | * next to the block it returns. This is just a hint and | |
4639 | * may be ignored by the allocator. | |
4640 | * | |
4641 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4642 | * address passed. It might be ignored. | |
4643 | * | |
4644 | * @ins - This key is modified to record the found hole. It will | |
4645 | * have the following values: | |
4646 | * ins->objectid == start position | |
4647 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4648 | * ins->offset == the size of the hole. | |
4649 | * | |
4650 | * @is_data - Boolean flag indicating whether an extent is | |
4651 | * allocated for data (true) or metadata (false) | |
4652 | * | |
4653 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4654 | * delalloc or not. If 'true' data_rwsem of block groups | |
4655 | * is going to be acquired. | |
4656 | * | |
4657 | * | |
4658 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4659 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4660 | * largest available hole the allocator managed to find. | |
4661 | */ | |
18513091 | 4662 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4663 | u64 num_bytes, u64 min_alloc_size, |
4664 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4665 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4666 | { |
ab8d0fc4 | 4667 | struct btrfs_fs_info *fs_info = root->fs_info; |
a12b0dc0 | 4668 | struct find_free_extent_ctl ffe_ctl = {}; |
36af4e07 | 4669 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4670 | u64 flags; |
fec577fb | 4671 | int ret; |
e094f480 | 4672 | bool for_treelog = (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID); |
c2707a25 | 4673 | bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data); |
925baedd | 4674 | |
1b86826d | 4675 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4676 | again: |
0b246afa | 4677 | WARN_ON(num_bytes < fs_info->sectorsize); |
a12b0dc0 NA |
4678 | |
4679 | ffe_ctl.ram_bytes = ram_bytes; | |
4680 | ffe_ctl.num_bytes = num_bytes; | |
a85f05e5 | 4681 | ffe_ctl.min_alloc_size = min_alloc_size; |
a12b0dc0 NA |
4682 | ffe_ctl.empty_size = empty_size; |
4683 | ffe_ctl.flags = flags; | |
4684 | ffe_ctl.delalloc = delalloc; | |
4685 | ffe_ctl.hint_byte = hint_byte; | |
4686 | ffe_ctl.for_treelog = for_treelog; | |
c2707a25 | 4687 | ffe_ctl.for_data_reloc = for_data_reloc; |
a12b0dc0 NA |
4688 | |
4689 | ret = find_free_extent(root, ins, &ffe_ctl); | |
9cfa3e34 | 4690 | if (!ret && !is_data) { |
ab8d0fc4 | 4691 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4692 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4693 | if (!final_tried && ins->offset) { |
4694 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4695 | num_bytes = round_down(num_bytes, |
0b246afa | 4696 | fs_info->sectorsize); |
9e622d6b | 4697 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4698 | ram_bytes = num_bytes; |
9e622d6b MX |
4699 | if (num_bytes == min_alloc_size) |
4700 | final_tried = true; | |
4701 | goto again; | |
ab8d0fc4 | 4702 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4703 | struct btrfs_space_info *sinfo; |
4704 | ||
280c2908 | 4705 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4706 | btrfs_err(fs_info, |
c2707a25 JT |
4707 | "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d", |
4708 | flags, num_bytes, for_treelog, for_data_reloc); | |
53804280 | 4709 | if (sinfo) |
5da6afeb JB |
4710 | btrfs_dump_space_info(fs_info, sinfo, |
4711 | num_bytes, 1); | |
9e622d6b | 4712 | } |
925baedd | 4713 | } |
0f9dd46c JB |
4714 | |
4715 | return ret; | |
e6dcd2dc CM |
4716 | } |
4717 | ||
a0fbf736 NB |
4718 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4719 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4720 | { |
32da5386 | 4721 | struct btrfs_block_group *cache; |
0f9dd46c | 4722 | |
0b246afa | 4723 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4724 | if (!cache) { |
0b246afa JM |
4725 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4726 | start); | |
0f9dd46c JB |
4727 | return -ENOSPC; |
4728 | } | |
1f3c79a2 | 4729 | |
a0fbf736 NB |
4730 | btrfs_add_free_space(cache, start, len); |
4731 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4732 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4733 | ||
fa9c0d79 | 4734 | btrfs_put_block_group(cache); |
a0fbf736 | 4735 | return 0; |
e6dcd2dc CM |
4736 | } |
4737 | ||
f863c502 DS |
4738 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, |
4739 | const struct extent_buffer *eb) | |
e688b725 | 4740 | { |
7ef54d54 | 4741 | struct btrfs_block_group *cache; |
a0fbf736 | 4742 | int ret = 0; |
7ef54d54 | 4743 | |
f863c502 | 4744 | cache = btrfs_lookup_block_group(trans->fs_info, eb->start); |
7ef54d54 | 4745 | if (!cache) { |
7bfc1007 | 4746 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
f863c502 | 4747 | eb->start); |
7ef54d54 NB |
4748 | return -ENOSPC; |
4749 | } | |
4750 | ||
f863c502 | 4751 | ret = pin_down_extent(trans, cache, eb->start, eb->len, 1); |
7ef54d54 | 4752 | btrfs_put_block_group(cache); |
a0fbf736 | 4753 | return ret; |
e688b725 CM |
4754 | } |
4755 | ||
34666705 JB |
4756 | static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr, |
4757 | u64 num_bytes) | |
4758 | { | |
4759 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
4760 | int ret; | |
4761 | ||
4762 | ret = remove_from_free_space_tree(trans, bytenr, num_bytes); | |
4763 | if (ret) | |
4764 | return ret; | |
4765 | ||
4766 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, true); | |
4767 | if (ret) { | |
4768 | ASSERT(!ret); | |
4769 | btrfs_err(fs_info, "update block group failed for %llu %llu", | |
4770 | bytenr, num_bytes); | |
4771 | return ret; | |
4772 | } | |
4773 | ||
4774 | trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes); | |
4775 | return 0; | |
4776 | } | |
4777 | ||
5d4f98a2 | 4778 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4779 | u64 parent, u64 root_objectid, |
4780 | u64 flags, u64 owner, u64 offset, | |
2672a051 | 4781 | struct btrfs_key *ins, int ref_mod, u64 oref_root) |
e6dcd2dc | 4782 | { |
ef89b824 | 4783 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 4784 | struct btrfs_root *extent_root; |
e6dcd2dc | 4785 | int ret; |
e6dcd2dc | 4786 | struct btrfs_extent_item *extent_item; |
d9a620f7 | 4787 | struct btrfs_extent_owner_ref *oref; |
5d4f98a2 | 4788 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4789 | struct btrfs_path *path; |
5d4f98a2 YZ |
4790 | struct extent_buffer *leaf; |
4791 | int type; | |
4792 | u32 size; | |
d9a620f7 | 4793 | const bool simple_quota = (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE); |
26b8003f | 4794 | |
5d4f98a2 YZ |
4795 | if (parent > 0) |
4796 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4797 | else | |
4798 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4799 | |
d9a620f7 BB |
4800 | size = sizeof(*extent_item); |
4801 | if (simple_quota) | |
4802 | size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY); | |
4803 | size += btrfs_extent_inline_ref_size(type); | |
7bb86316 CM |
4804 | |
4805 | path = btrfs_alloc_path(); | |
db5b493a TI |
4806 | if (!path) |
4807 | return -ENOMEM; | |
47e4bb98 | 4808 | |
29cbcf40 JB |
4809 | extent_root = btrfs_extent_root(fs_info, ins->objectid); |
4810 | ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size); | |
79787eaa JM |
4811 | if (ret) { |
4812 | btrfs_free_path(path); | |
4813 | return ret; | |
4814 | } | |
0f9dd46c | 4815 | |
5d4f98a2 YZ |
4816 | leaf = path->nodes[0]; |
4817 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4818 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4819 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4820 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4821 | btrfs_set_extent_flags(leaf, extent_item, | |
4822 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4823 | ||
4824 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
d9a620f7 BB |
4825 | if (simple_quota) { |
4826 | btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY); | |
4827 | oref = (struct btrfs_extent_owner_ref *)(&iref->offset); | |
2672a051 | 4828 | btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root); |
d9a620f7 BB |
4829 | iref = (struct btrfs_extent_inline_ref *)(oref + 1); |
4830 | } | |
5d4f98a2 | 4831 | btrfs_set_extent_inline_ref_type(leaf, iref, type); |
d9a620f7 | 4832 | |
5d4f98a2 YZ |
4833 | if (parent > 0) { |
4834 | struct btrfs_shared_data_ref *ref; | |
4835 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4836 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4837 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4838 | } else { | |
4839 | struct btrfs_extent_data_ref *ref; | |
4840 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4841 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4842 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4843 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4844 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4845 | } | |
47e4bb98 | 4846 | |
50564b65 | 4847 | btrfs_mark_buffer_dirty(trans, path->nodes[0]); |
7bb86316 | 4848 | btrfs_free_path(path); |
f510cfec | 4849 | |
34666705 | 4850 | return alloc_reserved_extent(trans, ins->objectid, ins->offset); |
e6dcd2dc CM |
4851 | } |
4852 | ||
5d4f98a2 | 4853 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4854 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4855 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4856 | { |
9dcdbe01 | 4857 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 4858 | struct btrfs_root *extent_root; |
e6dcd2dc | 4859 | int ret; |
5d4f98a2 | 4860 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4861 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4862 | struct btrfs_tree_block_info *block_info; |
4863 | struct btrfs_extent_inline_ref *iref; | |
4864 | struct btrfs_path *path; | |
4865 | struct extent_buffer *leaf; | |
3173a18f | 4866 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
21ebfbe7 | 4867 | u64 flags = extent_op->flags_to_set; |
efc7d5db JB |
4868 | /* The owner of a tree block is the level. */ |
4869 | int level = btrfs_delayed_ref_owner(node); | |
0b246afa | 4870 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4871 | |
4e6bd4e0 NB |
4872 | extent_key.objectid = node->bytenr; |
4873 | if (skinny_metadata) { | |
efc7d5db JB |
4874 | /* The owner of a tree block is the level. */ |
4875 | extent_key.offset = level; | |
4e6bd4e0 | 4876 | extent_key.type = BTRFS_METADATA_ITEM_KEY; |
4e6bd4e0 NB |
4877 | } else { |
4878 | extent_key.offset = node->num_bytes; | |
4879 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4880 | size += sizeof(*block_info); |
4e6bd4e0 | 4881 | } |
1c2308f8 | 4882 | |
5d4f98a2 | 4883 | path = btrfs_alloc_path(); |
80ee54bf | 4884 | if (!path) |
d8926bb3 | 4885 | return -ENOMEM; |
56bec294 | 4886 | |
29cbcf40 JB |
4887 | extent_root = btrfs_extent_root(fs_info, extent_key.objectid); |
4888 | ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key, | |
4889 | size); | |
79787eaa | 4890 | if (ret) { |
dd825259 | 4891 | btrfs_free_path(path); |
79787eaa JM |
4892 | return ret; |
4893 | } | |
5d4f98a2 YZ |
4894 | |
4895 | leaf = path->nodes[0]; | |
4896 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4897 | struct btrfs_extent_item); | |
4898 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4899 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4900 | btrfs_set_extent_flags(leaf, extent_item, | |
4901 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4902 | |
3173a18f JB |
4903 | if (skinny_metadata) { |
4904 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4905 | } else { | |
4906 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4907 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
efc7d5db | 4908 | btrfs_set_tree_block_level(leaf, block_info, level); |
3173a18f JB |
4909 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4910 | } | |
5d4f98a2 | 4911 | |
d4b20733 | 4912 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4913 | btrfs_set_extent_inline_ref_type(leaf, iref, |
4914 | BTRFS_SHARED_BLOCK_REF_KEY); | |
cf4f0432 | 4915 | btrfs_set_extent_inline_ref_offset(leaf, iref, node->parent); |
5d4f98a2 YZ |
4916 | } else { |
4917 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4918 | BTRFS_TREE_BLOCK_REF_KEY); | |
cf4f0432 | 4919 | btrfs_set_extent_inline_ref_offset(leaf, iref, node->ref_root); |
5d4f98a2 YZ |
4920 | } |
4921 | ||
50564b65 | 4922 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
4923 | btrfs_free_path(path); |
4924 | ||
34666705 | 4925 | return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize); |
5d4f98a2 YZ |
4926 | } |
4927 | ||
4928 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4929 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4930 | u64 offset, u64 ram_bytes, |
4931 | struct btrfs_key *ins) | |
5d4f98a2 | 4932 | { |
4d09b4e9 JB |
4933 | struct btrfs_ref generic_ref = { |
4934 | .action = BTRFS_ADD_DELAYED_EXTENT, | |
4935 | .bytenr = ins->objectid, | |
12390e42 | 4936 | .num_bytes = ins->offset, |
e094f480 JB |
4937 | .owning_root = btrfs_root_id(root), |
4938 | .ref_root = btrfs_root_id(root), | |
4d09b4e9 | 4939 | }; |
5d4f98a2 | 4940 | |
f2e69a77 | 4941 | ASSERT(generic_ref.ref_root != BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4942 | |
2672a051 | 4943 | if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root)) |
4d09b4e9 | 4944 | generic_ref.owning_root = root->relocation_src_root; |
2672a051 | 4945 | |
f2e69a77 | 4946 | btrfs_init_data_ref(&generic_ref, owner, offset, 0, false); |
8a5040f7 | 4947 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
2187374f JB |
4948 | |
4949 | return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes); | |
e6dcd2dc | 4950 | } |
e02119d5 CM |
4951 | |
4952 | /* | |
4953 | * this is used by the tree logging recovery code. It records that | |
4954 | * an extent has been allocated and makes sure to clear the free | |
4955 | * space cache bits as well | |
4956 | */ | |
5d4f98a2 | 4957 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4958 | u64 root_objectid, u64 owner, u64 offset, |
4959 | struct btrfs_key *ins) | |
e02119d5 | 4960 | { |
61da2abf | 4961 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4962 | int ret; |
32da5386 | 4963 | struct btrfs_block_group *block_group; |
ed7a6948 | 4964 | struct btrfs_space_info *space_info; |
cecbb533 BB |
4965 | struct btrfs_squota_delta delta = { |
4966 | .root = root_objectid, | |
4967 | .num_bytes = ins->offset, | |
bd7c1ea3 | 4968 | .generation = trans->transid, |
cecbb533 BB |
4969 | .is_data = true, |
4970 | .is_inc = true, | |
4971 | }; | |
11833d66 | 4972 | |
8c2a1a30 JB |
4973 | /* |
4974 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4975 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4976 | */ |
0b246afa | 4977 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4978 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4979 | ins->offset); | |
b50c6e25 | 4980 | if (ret) |
8c2a1a30 | 4981 | return ret; |
11833d66 YZ |
4982 | } |
4983 | ||
0b246afa | 4984 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4985 | if (!block_group) |
4986 | return -EINVAL; | |
4987 | ||
ed7a6948 WX |
4988 | space_info = block_group->space_info; |
4989 | spin_lock(&space_info->lock); | |
4990 | spin_lock(&block_group->lock); | |
4991 | space_info->bytes_reserved += ins->offset; | |
4992 | block_group->reserved += ins->offset; | |
4993 | spin_unlock(&block_group->lock); | |
4994 | spin_unlock(&space_info->lock); | |
4995 | ||
ef89b824 | 4996 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
2672a051 | 4997 | offset, ins, 1, root_objectid); |
bd727173 | 4998 | if (ret) |
ab9b2c7b | 4999 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
cecbb533 | 5000 | ret = btrfs_record_squota_delta(fs_info, &delta); |
b50c6e25 | 5001 | btrfs_put_block_group(block_group); |
e02119d5 CM |
5002 | return ret; |
5003 | } | |
5004 | ||
150cce2d DS |
5005 | #ifdef CONFIG_BTRFS_DEBUG |
5006 | /* | |
5007 | * Extra safety check in case the extent tree is corrupted and extent allocator | |
5008 | * chooses to use a tree block which is already used and locked. | |
5009 | */ | |
5010 | static bool check_eb_lock_owner(const struct extent_buffer *eb) | |
5011 | { | |
5012 | if (eb->lock_owner == current->pid) { | |
5013 | btrfs_err_rl(eb->fs_info, | |
5014 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
5015 | eb->start, btrfs_header_owner(eb), current->pid); | |
5016 | return true; | |
5017 | } | |
5018 | return false; | |
5019 | } | |
5020 | #else | |
5021 | static bool check_eb_lock_owner(struct extent_buffer *eb) | |
5022 | { | |
5023 | return false; | |
5024 | } | |
5025 | #endif | |
5026 | ||
48a3b636 ES |
5027 | static struct extent_buffer * |
5028 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
9631e4cc JB |
5029 | u64 bytenr, int level, u64 owner, |
5030 | enum btrfs_lock_nesting nest) | |
65b51a00 | 5031 | { |
0b246afa | 5032 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 | 5033 | struct extent_buffer *buf; |
b40130b2 | 5034 | u64 lockdep_owner = owner; |
65b51a00 | 5035 | |
3fbaf258 | 5036 | buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level); |
c871b0f2 LB |
5037 | if (IS_ERR(buf)) |
5038 | return buf; | |
5039 | ||
150cce2d | 5040 | if (check_eb_lock_owner(buf)) { |
b72c3aba QW |
5041 | free_extent_buffer(buf); |
5042 | return ERR_PTR(-EUCLEAN); | |
5043 | } | |
5044 | ||
b40130b2 JB |
5045 | /* |
5046 | * The reloc trees are just snapshots, so we need them to appear to be | |
5047 | * just like any other fs tree WRT lockdep. | |
5048 | * | |
5049 | * The exception however is in replace_path() in relocation, where we | |
5050 | * hold the lock on the original fs root and then search for the reloc | |
5051 | * root. At that point we need to make sure any reloc root buffers are | |
5052 | * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make | |
5053 | * lockdep happy. | |
5054 | */ | |
5055 | if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID && | |
5056 | !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state)) | |
5057 | lockdep_owner = BTRFS_FS_TREE_OBJECTID; | |
5058 | ||
618d1d7d | 5059 | /* btrfs_clear_buffer_dirty() accesses generation field. */ |
cbddcc4f TH |
5060 | btrfs_set_header_generation(buf, trans->transid); |
5061 | ||
e114c545 JB |
5062 | /* |
5063 | * This needs to stay, because we could allocate a freed block from an | |
5064 | * old tree into a new tree, so we need to make sure this new block is | |
5065 | * set to the appropriate level and owner. | |
5066 | */ | |
b40130b2 JB |
5067 | btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level); |
5068 | ||
2066bbfc | 5069 | btrfs_tree_lock_nested(buf, nest); |
190a8339 | 5070 | btrfs_clear_buffer_dirty(trans, buf); |
3083ee2e | 5071 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
cbf44cd9 | 5072 | clear_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &buf->bflags); |
b4ce94de | 5073 | |
4db8c528 | 5074 | set_extent_buffer_uptodate(buf); |
b4ce94de | 5075 | |
bc877d28 NB |
5076 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
5077 | btrfs_set_header_level(buf, level); | |
5078 | btrfs_set_header_bytenr(buf, buf->start); | |
5079 | btrfs_set_header_generation(buf, trans->transid); | |
5080 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
5081 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 5082 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 5083 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
e094f480 | 5084 | if (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 5085 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
5086 | /* |
5087 | * we allow two log transactions at a time, use different | |
52042d8e | 5088 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 5089 | */ |
656f30db | 5090 | if (buf->log_index == 0) |
fe1a598c DS |
5091 | set_extent_bit(&root->dirty_log_pages, buf->start, |
5092 | buf->start + buf->len - 1, | |
1d126800 | 5093 | EXTENT_DIRTY, NULL); |
8cef4e16 | 5094 | else |
eea8686e DS |
5095 | set_extent_bit(&root->dirty_log_pages, buf->start, |
5096 | buf->start + buf->len - 1, | |
1d126800 | 5097 | EXTENT_NEW, NULL); |
d0c803c4 | 5098 | } else { |
656f30db | 5099 | buf->log_index = -1; |
fe1a598c | 5100 | set_extent_bit(&trans->transaction->dirty_pages, buf->start, |
1d126800 | 5101 | buf->start + buf->len - 1, EXTENT_DIRTY, NULL); |
d0c803c4 | 5102 | } |
b4ce94de | 5103 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
5104 | return buf; |
5105 | } | |
5106 | ||
fec577fb | 5107 | /* |
f0486c68 | 5108 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 5109 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 5110 | */ |
4d75f8a9 | 5111 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
5112 | struct btrfs_root *root, |
5113 | u64 parent, u64 root_objectid, | |
5114 | const struct btrfs_disk_key *key, | |
5115 | int level, u64 hint, | |
9631e4cc | 5116 | u64 empty_size, |
60ea105a | 5117 | u64 reloc_src_root, |
9631e4cc | 5118 | enum btrfs_lock_nesting nest) |
fec577fb | 5119 | { |
0b246afa | 5120 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 5121 | struct btrfs_key ins; |
f0486c68 | 5122 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 5123 | struct extent_buffer *buf; |
67b7859e | 5124 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
5125 | u64 flags = 0; |
5126 | int ret; | |
0b246afa JM |
5127 | u32 blocksize = fs_info->nodesize; |
5128 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
60ea105a | 5129 | u64 owning_root; |
fec577fb | 5130 | |
05653ef3 | 5131 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 5132 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 5133 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
9631e4cc | 5134 | level, root_objectid, nest); |
faa2dbf0 JB |
5135 | if (!IS_ERR(buf)) |
5136 | root->alloc_bytenr += blocksize; | |
5137 | return buf; | |
5138 | } | |
05653ef3 | 5139 | #endif |
fccb84c9 | 5140 | |
67f9c220 | 5141 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
5142 | if (IS_ERR(block_rsv)) |
5143 | return ERR_CAST(block_rsv); | |
5144 | ||
18513091 | 5145 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 5146 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
5147 | if (ret) |
5148 | goto out_unuse; | |
55c69072 | 5149 | |
bc877d28 | 5150 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
9631e4cc | 5151 | root_objectid, nest); |
67b7859e OS |
5152 | if (IS_ERR(buf)) { |
5153 | ret = PTR_ERR(buf); | |
5154 | goto out_free_reserved; | |
5155 | } | |
60ea105a | 5156 | owning_root = btrfs_header_owner(buf); |
f0486c68 YZ |
5157 | |
5158 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
5159 | if (parent == 0) | |
5160 | parent = ins.objectid; | |
5161 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
60ea105a | 5162 | owning_root = reloc_src_root; |
f0486c68 YZ |
5163 | } else |
5164 | BUG_ON(parent > 0); | |
5165 | ||
5166 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4d09b4e9 JB |
5167 | struct btrfs_ref generic_ref = { |
5168 | .action = BTRFS_ADD_DELAYED_EXTENT, | |
5169 | .bytenr = ins.objectid, | |
12390e42 | 5170 | .num_bytes = ins.offset, |
4d09b4e9 JB |
5171 | .parent = parent, |
5172 | .owning_root = owning_root, | |
f2e69a77 | 5173 | .ref_root = root_objectid, |
4d09b4e9 | 5174 | }; |
78a6184a | 5175 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
5176 | if (!extent_op) { |
5177 | ret = -ENOMEM; | |
5178 | goto out_free_buf; | |
5179 | } | |
f0486c68 YZ |
5180 | if (key) |
5181 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
5182 | else | |
5183 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
5184 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
5185 | extent_op->update_key = skinny_metadata ? false : true; |
5186 | extent_op->update_flags = true; | |
b1c79e09 | 5187 | extent_op->level = level; |
f0486c68 | 5188 | |
e094f480 | 5189 | btrfs_init_tree_ref(&generic_ref, level, btrfs_root_id(root), false); |
8a5040f7 | 5190 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 5191 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op); |
67b7859e OS |
5192 | if (ret) |
5193 | goto out_free_delayed; | |
f0486c68 | 5194 | } |
fec577fb | 5195 | return buf; |
67b7859e OS |
5196 | |
5197 | out_free_delayed: | |
5198 | btrfs_free_delayed_extent_op(extent_op); | |
5199 | out_free_buf: | |
19ea40dd | 5200 | btrfs_tree_unlock(buf); |
67b7859e OS |
5201 | free_extent_buffer(buf); |
5202 | out_free_reserved: | |
2ff7e61e | 5203 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 5204 | out_unuse: |
67f9c220 | 5205 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 5206 | return ERR_PTR(ret); |
fec577fb | 5207 | } |
a28ec197 | 5208 | |
2c47e605 YZ |
5209 | struct walk_control { |
5210 | u64 refs[BTRFS_MAX_LEVEL]; | |
5211 | u64 flags[BTRFS_MAX_LEVEL]; | |
5212 | struct btrfs_key update_progress; | |
aea6f028 JB |
5213 | struct btrfs_key drop_progress; |
5214 | int drop_level; | |
2c47e605 YZ |
5215 | int stage; |
5216 | int level; | |
5217 | int shared_level; | |
5218 | int update_ref; | |
5219 | int keep_locks; | |
1c4850e2 YZ |
5220 | int reada_slot; |
5221 | int reada_count; | |
78c52d9e | 5222 | int restarted; |
2c47e605 YZ |
5223 | }; |
5224 | ||
5225 | #define DROP_REFERENCE 1 | |
5226 | #define UPDATE_BACKREF 2 | |
5227 | ||
1c4850e2 YZ |
5228 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
5229 | struct btrfs_root *root, | |
5230 | struct walk_control *wc, | |
5231 | struct btrfs_path *path) | |
6407bf6d | 5232 | { |
0b246afa | 5233 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
5234 | u64 bytenr; |
5235 | u64 generation; | |
5236 | u64 refs; | |
94fcca9f | 5237 | u64 flags; |
5d4f98a2 | 5238 | u32 nritems; |
1c4850e2 YZ |
5239 | struct btrfs_key key; |
5240 | struct extent_buffer *eb; | |
6407bf6d | 5241 | int ret; |
1c4850e2 YZ |
5242 | int slot; |
5243 | int nread = 0; | |
6407bf6d | 5244 | |
1c4850e2 YZ |
5245 | if (path->slots[wc->level] < wc->reada_slot) { |
5246 | wc->reada_count = wc->reada_count * 2 / 3; | |
5247 | wc->reada_count = max(wc->reada_count, 2); | |
5248 | } else { | |
5249 | wc->reada_count = wc->reada_count * 3 / 2; | |
5250 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 5251 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 5252 | } |
7bb86316 | 5253 | |
1c4850e2 YZ |
5254 | eb = path->nodes[wc->level]; |
5255 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 5256 | |
1c4850e2 YZ |
5257 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5258 | if (nread >= wc->reada_count) | |
5259 | break; | |
bd56b302 | 5260 | |
2dd3e67b | 5261 | cond_resched(); |
1c4850e2 YZ |
5262 | bytenr = btrfs_node_blockptr(eb, slot); |
5263 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5264 | |
1c4850e2 YZ |
5265 | if (slot == path->slots[wc->level]) |
5266 | goto reada; | |
5d4f98a2 | 5267 | |
1c4850e2 YZ |
5268 | if (wc->stage == UPDATE_BACKREF && |
5269 | generation <= root->root_key.offset) | |
bd56b302 CM |
5270 | continue; |
5271 | ||
94fcca9f | 5272 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 5273 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f | 5274 | wc->level - 1, 1, &refs, |
d8ba2a91 | 5275 | &flags, NULL); |
79787eaa JM |
5276 | /* We don't care about errors in readahead. */ |
5277 | if (ret < 0) | |
5278 | continue; | |
94fcca9f YZ |
5279 | BUG_ON(refs == 0); |
5280 | ||
1c4850e2 | 5281 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5282 | if (refs == 1) |
5283 | goto reada; | |
bd56b302 | 5284 | |
94fcca9f YZ |
5285 | if (wc->level == 1 && |
5286 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5287 | continue; | |
1c4850e2 YZ |
5288 | if (!wc->update_ref || |
5289 | generation <= root->root_key.offset) | |
5290 | continue; | |
5291 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5292 | ret = btrfs_comp_cpu_keys(&key, | |
5293 | &wc->update_progress); | |
5294 | if (ret < 0) | |
5295 | continue; | |
94fcca9f YZ |
5296 | } else { |
5297 | if (wc->level == 1 && | |
5298 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5299 | continue; | |
6407bf6d | 5300 | } |
1c4850e2 | 5301 | reada: |
bfb484d9 | 5302 | btrfs_readahead_node_child(eb, slot); |
1c4850e2 | 5303 | nread++; |
20524f02 | 5304 | } |
1c4850e2 | 5305 | wc->reada_slot = slot; |
20524f02 | 5306 | } |
2c47e605 | 5307 | |
f82d02d9 | 5308 | /* |
2c016dc2 | 5309 | * helper to process tree block while walking down the tree. |
2c47e605 | 5310 | * |
2c47e605 YZ |
5311 | * when wc->stage == UPDATE_BACKREF, this function updates |
5312 | * back refs for pointers in the block. | |
5313 | * | |
5314 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5315 | */ |
2c47e605 | 5316 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5317 | struct btrfs_root *root, |
2c47e605 | 5318 | struct btrfs_path *path, |
94fcca9f | 5319 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5320 | { |
2ff7e61e | 5321 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
5322 | int level = wc->level; |
5323 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5324 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5325 | int ret; |
5326 | ||
e094f480 | 5327 | if (wc->stage == UPDATE_BACKREF && btrfs_header_owner(eb) != btrfs_root_id(root)) |
2c47e605 | 5328 | return 1; |
f82d02d9 | 5329 | |
2c47e605 YZ |
5330 | /* |
5331 | * when reference count of tree block is 1, it won't increase | |
5332 | * again. once full backref flag is set, we never clear it. | |
5333 | */ | |
94fcca9f YZ |
5334 | if (lookup_info && |
5335 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5336 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 5337 | BUG_ON(!path->locks[level]); |
2ff7e61e | 5338 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5339 | eb->start, level, 1, |
2c47e605 | 5340 | &wc->refs[level], |
d8ba2a91 JB |
5341 | &wc->flags[level], |
5342 | NULL); | |
79787eaa JM |
5343 | BUG_ON(ret == -ENOMEM); |
5344 | if (ret) | |
5345 | return ret; | |
2c47e605 YZ |
5346 | BUG_ON(wc->refs[level] == 0); |
5347 | } | |
5d4f98a2 | 5348 | |
2c47e605 YZ |
5349 | if (wc->stage == DROP_REFERENCE) { |
5350 | if (wc->refs[level] > 1) | |
5351 | return 1; | |
f82d02d9 | 5352 | |
2c47e605 | 5353 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 5354 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5355 | path->locks[level] = 0; |
5356 | } | |
5357 | return 0; | |
5358 | } | |
f82d02d9 | 5359 | |
2c47e605 YZ |
5360 | /* wc->stage == UPDATE_BACKREF */ |
5361 | if (!(wc->flags[level] & flag)) { | |
5362 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 5363 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 5364 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 5365 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5366 | BUG_ON(ret); /* -ENOMEM */ |
4aec05fa | 5367 | ret = btrfs_set_disk_extent_flags(trans, eb, flag); |
79787eaa | 5368 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
5369 | wc->flags[level] |= flag; |
5370 | } | |
5371 | ||
5372 | /* | |
5373 | * the block is shared by multiple trees, so it's not good to | |
5374 | * keep the tree lock | |
5375 | */ | |
5376 | if (path->locks[level] && level > 0) { | |
bd681513 | 5377 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5378 | path->locks[level] = 0; |
5379 | } | |
5380 | return 0; | |
5381 | } | |
5382 | ||
78c52d9e JB |
5383 | /* |
5384 | * This is used to verify a ref exists for this root to deal with a bug where we | |
5385 | * would have a drop_progress key that hadn't been updated properly. | |
5386 | */ | |
5387 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
5388 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
5389 | int level) | |
5390 | { | |
5391 | struct btrfs_path *path; | |
5392 | struct btrfs_extent_inline_ref *iref; | |
5393 | int ret; | |
5394 | ||
5395 | path = btrfs_alloc_path(); | |
5396 | if (!path) | |
5397 | return -ENOMEM; | |
5398 | ||
5399 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
5400 | root->fs_info->nodesize, parent, | |
e094f480 | 5401 | btrfs_root_id(root), level, 0); |
78c52d9e JB |
5402 | btrfs_free_path(path); |
5403 | if (ret == -ENOENT) | |
5404 | return 0; | |
5405 | if (ret < 0) | |
5406 | return ret; | |
5407 | return 1; | |
5408 | } | |
5409 | ||
1c4850e2 | 5410 | /* |
2c016dc2 | 5411 | * helper to process tree block pointer. |
1c4850e2 YZ |
5412 | * |
5413 | * when wc->stage == DROP_REFERENCE, this function checks | |
5414 | * reference count of the block pointed to. if the block | |
5415 | * is shared and we need update back refs for the subtree | |
5416 | * rooted at the block, this function changes wc->stage to | |
5417 | * UPDATE_BACKREF. if the block is shared and there is no | |
5418 | * need to update back, this function drops the reference | |
5419 | * to the block. | |
5420 | * | |
5421 | * NOTE: return value 1 means we should stop walking down. | |
5422 | */ | |
5423 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5424 | struct btrfs_root *root, | |
5425 | struct btrfs_path *path, | |
94fcca9f | 5426 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 5427 | { |
0b246afa | 5428 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
5429 | u64 bytenr; |
5430 | u64 generation; | |
d8ba2a91 | 5431 | u64 owner_root = 0; |
789d6a3a | 5432 | struct btrfs_tree_parent_check check = { 0 }; |
1c4850e2 YZ |
5433 | struct btrfs_key key; |
5434 | struct extent_buffer *next; | |
5435 | int level = wc->level; | |
5436 | int reada = 0; | |
5437 | int ret = 0; | |
1152651a | 5438 | bool need_account = false; |
1c4850e2 YZ |
5439 | |
5440 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5441 | path->slots[level]); | |
5442 | /* | |
5443 | * if the lower level block was created before the snapshot | |
5444 | * was created, we know there is no need to update back refs | |
5445 | * for the subtree | |
5446 | */ | |
5447 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5448 | generation <= root->root_key.offset) { |
5449 | *lookup_info = 1; | |
1c4850e2 | 5450 | return 1; |
94fcca9f | 5451 | } |
1c4850e2 YZ |
5452 | |
5453 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
789d6a3a QW |
5454 | |
5455 | check.level = level - 1; | |
5456 | check.transid = generation; | |
e094f480 | 5457 | check.owner_root = btrfs_root_id(root); |
789d6a3a QW |
5458 | check.has_first_key = true; |
5459 | btrfs_node_key_to_cpu(path->nodes[level], &check.first_key, | |
581c1760 | 5460 | path->slots[level]); |
1c4850e2 | 5461 | |
0b246afa | 5462 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5463 | if (!next) { |
3fbaf258 | 5464 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
e094f480 | 5465 | btrfs_root_id(root), level - 1); |
c871b0f2 LB |
5466 | if (IS_ERR(next)) |
5467 | return PTR_ERR(next); | |
1c4850e2 YZ |
5468 | reada = 1; |
5469 | } | |
5470 | btrfs_tree_lock(next); | |
1c4850e2 | 5471 | |
2ff7e61e | 5472 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f | 5473 | &wc->refs[level - 1], |
d8ba2a91 JB |
5474 | &wc->flags[level - 1], |
5475 | &owner_root); | |
4867268c JB |
5476 | if (ret < 0) |
5477 | goto out_unlock; | |
79787eaa | 5478 | |
c2cf52eb | 5479 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5480 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5481 | ret = -EIO; |
5482 | goto out_unlock; | |
c2cf52eb | 5483 | } |
94fcca9f | 5484 | *lookup_info = 0; |
1c4850e2 | 5485 | |
94fcca9f | 5486 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5487 | if (wc->refs[level - 1] > 1) { |
1152651a | 5488 | need_account = true; |
94fcca9f YZ |
5489 | if (level == 1 && |
5490 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5491 | goto skip; | |
5492 | ||
1c4850e2 YZ |
5493 | if (!wc->update_ref || |
5494 | generation <= root->root_key.offset) | |
5495 | goto skip; | |
5496 | ||
5497 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5498 | path->slots[level]); | |
5499 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5500 | if (ret < 0) | |
5501 | goto skip; | |
5502 | ||
5503 | wc->stage = UPDATE_BACKREF; | |
5504 | wc->shared_level = level - 1; | |
5505 | } | |
94fcca9f YZ |
5506 | } else { |
5507 | if (level == 1 && | |
5508 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5509 | goto skip; | |
1c4850e2 YZ |
5510 | } |
5511 | ||
b9fab919 | 5512 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5513 | btrfs_tree_unlock(next); |
5514 | free_extent_buffer(next); | |
5515 | next = NULL; | |
94fcca9f | 5516 | *lookup_info = 1; |
1c4850e2 YZ |
5517 | } |
5518 | ||
5519 | if (!next) { | |
5520 | if (reada && level == 1) | |
5521 | reada_walk_down(trans, root, wc, path); | |
789d6a3a | 5522 | next = read_tree_block(fs_info, bytenr, &check); |
64c043de LB |
5523 | if (IS_ERR(next)) { |
5524 | return PTR_ERR(next); | |
5525 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5526 | free_extent_buffer(next); |
97d9a8a4 | 5527 | return -EIO; |
416bc658 | 5528 | } |
1c4850e2 | 5529 | btrfs_tree_lock(next); |
1c4850e2 YZ |
5530 | } |
5531 | ||
5532 | level--; | |
4867268c JB |
5533 | ASSERT(level == btrfs_header_level(next)); |
5534 | if (level != btrfs_header_level(next)) { | |
5535 | btrfs_err(root->fs_info, "mismatched level"); | |
5536 | ret = -EIO; | |
5537 | goto out_unlock; | |
5538 | } | |
1c4850e2 YZ |
5539 | path->nodes[level] = next; |
5540 | path->slots[level] = 0; | |
ac5887c8 | 5541 | path->locks[level] = BTRFS_WRITE_LOCK; |
1c4850e2 YZ |
5542 | wc->level = level; |
5543 | if (wc->level == 1) | |
5544 | wc->reada_slot = 0; | |
5545 | return 0; | |
5546 | skip: | |
5547 | wc->refs[level - 1] = 0; | |
5548 | wc->flags[level - 1] = 0; | |
94fcca9f | 5549 | if (wc->stage == DROP_REFERENCE) { |
4d09b4e9 JB |
5550 | struct btrfs_ref ref = { |
5551 | .action = BTRFS_DROP_DELAYED_REF, | |
5552 | .bytenr = bytenr, | |
12390e42 | 5553 | .num_bytes = fs_info->nodesize, |
4d09b4e9 | 5554 | .owning_root = owner_root, |
e094f480 | 5555 | .ref_root = btrfs_root_id(root), |
4d09b4e9 | 5556 | }; |
94fcca9f | 5557 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { |
4d09b4e9 | 5558 | ref.parent = path->nodes[level]->start; |
94fcca9f | 5559 | } else { |
e094f480 | 5560 | ASSERT(btrfs_root_id(root) == |
94fcca9f | 5561 | btrfs_header_owner(path->nodes[level])); |
e094f480 | 5562 | if (btrfs_root_id(root) != |
4867268c JB |
5563 | btrfs_header_owner(path->nodes[level])) { |
5564 | btrfs_err(root->fs_info, | |
5565 | "mismatched block owner"); | |
5566 | ret = -EIO; | |
5567 | goto out_unlock; | |
5568 | } | |
94fcca9f | 5569 | } |
1c4850e2 | 5570 | |
78c52d9e JB |
5571 | /* |
5572 | * If we had a drop_progress we need to verify the refs are set | |
5573 | * as expected. If we find our ref then we know that from here | |
5574 | * on out everything should be correct, and we can clear the | |
5575 | * ->restarted flag. | |
5576 | */ | |
5577 | if (wc->restarted) { | |
4d09b4e9 | 5578 | ret = check_ref_exists(trans, root, bytenr, ref.parent, |
78c52d9e JB |
5579 | level - 1); |
5580 | if (ret < 0) | |
5581 | goto out_unlock; | |
5582 | if (ret == 0) | |
5583 | goto no_delete; | |
5584 | ret = 0; | |
5585 | wc->restarted = 0; | |
5586 | } | |
5587 | ||
2cd86d30 QW |
5588 | /* |
5589 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5590 | * already accounted them at merge time (replace_path), | |
5591 | * thus we could skip expensive subtree trace here. | |
5592 | */ | |
e094f480 | 5593 | if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID && need_account) { |
deb40627 | 5594 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5595 | generation, level - 1); |
1152651a | 5596 | if (ret) { |
0b246afa | 5597 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5598 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5599 | ret); | |
1152651a MF |
5600 | } |
5601 | } | |
aea6f028 JB |
5602 | |
5603 | /* | |
5604 | * We need to update the next key in our walk control so we can | |
5605 | * update the drop_progress key accordingly. We don't care if | |
5606 | * find_next_key doesn't find a key because that means we're at | |
5607 | * the end and are going to clean up now. | |
5608 | */ | |
5609 | wc->drop_level = level; | |
5610 | find_next_key(path, level, &wc->drop_progress); | |
5611 | ||
f2e69a77 | 5612 | btrfs_init_tree_ref(&ref, level - 1, 0, false); |
ffd4bb2a | 5613 | ret = btrfs_free_extent(trans, &ref); |
4867268c JB |
5614 | if (ret) |
5615 | goto out_unlock; | |
1c4850e2 | 5616 | } |
78c52d9e | 5617 | no_delete: |
4867268c JB |
5618 | *lookup_info = 1; |
5619 | ret = 1; | |
5620 | ||
5621 | out_unlock: | |
1c4850e2 YZ |
5622 | btrfs_tree_unlock(next); |
5623 | free_extent_buffer(next); | |
4867268c JB |
5624 | |
5625 | return ret; | |
1c4850e2 YZ |
5626 | } |
5627 | ||
2c47e605 | 5628 | /* |
2c016dc2 | 5629 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5630 | * |
5631 | * when wc->stage == DROP_REFERENCE, this function drops | |
5632 | * reference count on the block. | |
5633 | * | |
5634 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5635 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5636 | * to UPDATE_BACKREF previously while processing the block. | |
5637 | * | |
5638 | * NOTE: return value 1 means we should stop walking up. | |
5639 | */ | |
5640 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5641 | struct btrfs_root *root, | |
5642 | struct btrfs_path *path, | |
5643 | struct walk_control *wc) | |
5644 | { | |
0b246afa | 5645 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5646 | int ret; |
2c47e605 YZ |
5647 | int level = wc->level; |
5648 | struct extent_buffer *eb = path->nodes[level]; | |
5649 | u64 parent = 0; | |
5650 | ||
5651 | if (wc->stage == UPDATE_BACKREF) { | |
5652 | BUG_ON(wc->shared_level < level); | |
5653 | if (level < wc->shared_level) | |
5654 | goto out; | |
5655 | ||
2c47e605 YZ |
5656 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5657 | if (ret > 0) | |
5658 | wc->update_ref = 0; | |
5659 | ||
5660 | wc->stage = DROP_REFERENCE; | |
5661 | wc->shared_level = -1; | |
5662 | path->slots[level] = 0; | |
5663 | ||
5664 | /* | |
5665 | * check reference count again if the block isn't locked. | |
5666 | * we should start walking down the tree again if reference | |
5667 | * count is one. | |
5668 | */ | |
5669 | if (!path->locks[level]) { | |
5670 | BUG_ON(level == 0); | |
5671 | btrfs_tree_lock(eb); | |
ac5887c8 | 5672 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5673 | |
2ff7e61e | 5674 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5675 | eb->start, level, 1, |
2c47e605 | 5676 | &wc->refs[level], |
d8ba2a91 JB |
5677 | &wc->flags[level], |
5678 | NULL); | |
79787eaa JM |
5679 | if (ret < 0) { |
5680 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5681 | path->locks[level] = 0; |
79787eaa JM |
5682 | return ret; |
5683 | } | |
2c47e605 YZ |
5684 | BUG_ON(wc->refs[level] == 0); |
5685 | if (wc->refs[level] == 1) { | |
bd681513 | 5686 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5687 | path->locks[level] = 0; |
2c47e605 YZ |
5688 | return 1; |
5689 | } | |
f82d02d9 | 5690 | } |
2c47e605 | 5691 | } |
f82d02d9 | 5692 | |
2c47e605 YZ |
5693 | /* wc->stage == DROP_REFERENCE */ |
5694 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5695 | |
2c47e605 YZ |
5696 | if (wc->refs[level] == 1) { |
5697 | if (level == 0) { | |
5698 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5699 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5700 | else |
e339a6b0 | 5701 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5702 | BUG_ON(ret); /* -ENOMEM */ |
e094f480 | 5703 | if (is_fstree(btrfs_root_id(root))) { |
c4140cbf QW |
5704 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); |
5705 | if (ret) { | |
5706 | btrfs_err_rl(fs_info, | |
5707 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5708 | ret); |
c4140cbf | 5709 | } |
1152651a | 5710 | } |
2c47e605 | 5711 | } |
190a8339 | 5712 | /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */ |
d3fb6615 | 5713 | if (!path->locks[level]) { |
2c47e605 | 5714 | btrfs_tree_lock(eb); |
ac5887c8 | 5715 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5716 | } |
190a8339 | 5717 | btrfs_clear_buffer_dirty(trans, eb); |
2c47e605 YZ |
5718 | } |
5719 | ||
5720 | if (eb == root->node) { | |
5721 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5722 | parent = eb->start; | |
e094f480 | 5723 | else if (btrfs_root_id(root) != btrfs_header_owner(eb)) |
65c6e82b | 5724 | goto owner_mismatch; |
2c47e605 YZ |
5725 | } else { |
5726 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5727 | parent = path->nodes[level + 1]->start; | |
e094f480 | 5728 | else if (btrfs_root_id(root) != |
65c6e82b QW |
5729 | btrfs_header_owner(path->nodes[level + 1])) |
5730 | goto owner_mismatch; | |
f82d02d9 | 5731 | } |
f82d02d9 | 5732 | |
7a163608 FM |
5733 | btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent, |
5734 | wc->refs[level] == 1); | |
2c47e605 YZ |
5735 | out: |
5736 | wc->refs[level] = 0; | |
5737 | wc->flags[level] = 0; | |
f0486c68 | 5738 | return 0; |
65c6e82b QW |
5739 | |
5740 | owner_mismatch: | |
5741 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
e094f480 | 5742 | btrfs_header_owner(eb), btrfs_root_id(root)); |
65c6e82b | 5743 | return -EUCLEAN; |
2c47e605 YZ |
5744 | } |
5745 | ||
5746 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5747 | struct btrfs_root *root, | |
5748 | struct btrfs_path *path, | |
5749 | struct walk_control *wc) | |
5750 | { | |
2c47e605 | 5751 | int level = wc->level; |
94fcca9f | 5752 | int lookup_info = 1; |
4e194384 | 5753 | int ret = 0; |
2c47e605 YZ |
5754 | |
5755 | while (level >= 0) { | |
94fcca9f | 5756 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
4e194384 | 5757 | if (ret) |
2c47e605 YZ |
5758 | break; |
5759 | ||
5760 | if (level == 0) | |
5761 | break; | |
5762 | ||
7a7965f8 YZ |
5763 | if (path->slots[level] >= |
5764 | btrfs_header_nritems(path->nodes[level])) | |
5765 | break; | |
5766 | ||
94fcca9f | 5767 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5768 | if (ret > 0) { |
5769 | path->slots[level]++; | |
5770 | continue; | |
90d2c51d | 5771 | } else if (ret < 0) |
4e194384 | 5772 | break; |
1c4850e2 | 5773 | level = wc->level; |
f82d02d9 | 5774 | } |
4e194384 | 5775 | return (ret == 1) ? 0 : ret; |
f82d02d9 YZ |
5776 | } |
5777 | ||
d397712b | 5778 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5779 | struct btrfs_root *root, |
f82d02d9 | 5780 | struct btrfs_path *path, |
2c47e605 | 5781 | struct walk_control *wc, int max_level) |
20524f02 | 5782 | { |
2c47e605 | 5783 | int level = wc->level; |
20524f02 | 5784 | int ret; |
9f3a7427 | 5785 | |
2c47e605 YZ |
5786 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5787 | while (level < max_level && path->nodes[level]) { | |
5788 | wc->level = level; | |
5789 | if (path->slots[level] + 1 < | |
5790 | btrfs_header_nritems(path->nodes[level])) { | |
5791 | path->slots[level]++; | |
20524f02 CM |
5792 | return 0; |
5793 | } else { | |
2c47e605 YZ |
5794 | ret = walk_up_proc(trans, root, path, wc); |
5795 | if (ret > 0) | |
5796 | return 0; | |
65c6e82b QW |
5797 | if (ret < 0) |
5798 | return ret; | |
bd56b302 | 5799 | |
2c47e605 | 5800 | if (path->locks[level]) { |
bd681513 CM |
5801 | btrfs_tree_unlock_rw(path->nodes[level], |
5802 | path->locks[level]); | |
2c47e605 | 5803 | path->locks[level] = 0; |
f82d02d9 | 5804 | } |
2c47e605 YZ |
5805 | free_extent_buffer(path->nodes[level]); |
5806 | path->nodes[level] = NULL; | |
5807 | level++; | |
20524f02 CM |
5808 | } |
5809 | } | |
5810 | return 1; | |
5811 | } | |
5812 | ||
9aca1d51 | 5813 | /* |
2c47e605 YZ |
5814 | * drop a subvolume tree. |
5815 | * | |
5816 | * this function traverses the tree freeing any blocks that only | |
5817 | * referenced by the tree. | |
5818 | * | |
5819 | * when a shared tree block is found. this function decreases its | |
5820 | * reference count by one. if update_ref is true, this function | |
5821 | * also make sure backrefs for the shared block and all lower level | |
5822 | * blocks are properly updated. | |
9d1a2a3a DS |
5823 | * |
5824 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5825 | */ |
0078a9f9 | 5826 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5827 | { |
e094f480 | 5828 | const bool is_reloc_root = (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID); |
ab8d0fc4 | 5829 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5830 | struct btrfs_path *path; |
2c47e605 | 5831 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5832 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5833 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5834 | struct walk_control *wc; |
5835 | struct btrfs_key key; | |
5836 | int err = 0; | |
5837 | int ret; | |
5838 | int level; | |
d29a9f62 | 5839 | bool root_dropped = false; |
b4be6aef | 5840 | bool unfinished_drop = false; |
20524f02 | 5841 | |
e094f480 | 5842 | btrfs_debug(fs_info, "Drop subvolume %llu", btrfs_root_id(root)); |
1152651a | 5843 | |
5caf2a00 | 5844 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5845 | if (!path) { |
5846 | err = -ENOMEM; | |
5847 | goto out; | |
5848 | } | |
20524f02 | 5849 | |
2c47e605 | 5850 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5851 | if (!wc) { |
5852 | btrfs_free_path(path); | |
cb1b69f4 TI |
5853 | err = -ENOMEM; |
5854 | goto out; | |
38a1a919 | 5855 | } |
2c47e605 | 5856 | |
f3e3d9cc QW |
5857 | /* |
5858 | * Use join to avoid potential EINTR from transaction start. See | |
5859 | * wait_reserve_ticket and the whole reservation callchain. | |
5860 | */ | |
5861 | if (for_reloc) | |
5862 | trans = btrfs_join_transaction(tree_root); | |
5863 | else | |
5864 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5865 | if (IS_ERR(trans)) { |
5866 | err = PTR_ERR(trans); | |
5867 | goto out_free; | |
5868 | } | |
98d5dc13 | 5869 | |
0568e82d JB |
5870 | err = btrfs_run_delayed_items(trans); |
5871 | if (err) | |
5872 | goto out_end_trans; | |
5873 | ||
83354f07 JB |
5874 | /* |
5875 | * This will help us catch people modifying the fs tree while we're | |
5876 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5877 | * dropped as we unlock the root node and parent nodes as we walk down | |
5878 | * the tree, assuming nothing will change. If something does change | |
5879 | * then we'll have stale information and drop references to blocks we've | |
5880 | * already dropped. | |
5881 | */ | |
5882 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
b4be6aef JB |
5883 | unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state); |
5884 | ||
9f3a7427 | 5885 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5886 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5887 | path->nodes[level] = btrfs_lock_root_node(root); |
9f3a7427 | 5888 | path->slots[level] = 0; |
ac5887c8 | 5889 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5890 | memset(&wc->update_progress, 0, |
5891 | sizeof(wc->update_progress)); | |
9f3a7427 | 5892 | } else { |
9f3a7427 | 5893 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5894 | memcpy(&wc->update_progress, &key, |
5895 | sizeof(wc->update_progress)); | |
5896 | ||
c8422684 | 5897 | level = btrfs_root_drop_level(root_item); |
2c47e605 | 5898 | BUG_ON(level == 0); |
6702ed49 | 5899 | path->lowest_level = level; |
2c47e605 YZ |
5900 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5901 | path->lowest_level = 0; | |
5902 | if (ret < 0) { | |
5903 | err = ret; | |
79787eaa | 5904 | goto out_end_trans; |
9f3a7427 | 5905 | } |
1c4850e2 | 5906 | WARN_ON(ret > 0); |
2c47e605 | 5907 | |
7d9eb12c CM |
5908 | /* |
5909 | * unlock our path, this is safe because only this | |
5910 | * function is allowed to delete this snapshot | |
5911 | */ | |
5d4f98a2 | 5912 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5913 | |
5914 | level = btrfs_header_level(root->node); | |
5915 | while (1) { | |
5916 | btrfs_tree_lock(path->nodes[level]); | |
ac5887c8 | 5917 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5918 | |
2ff7e61e | 5919 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5920 | path->nodes[level]->start, |
3173a18f | 5921 | level, 1, &wc->refs[level], |
d8ba2a91 | 5922 | &wc->flags[level], NULL); |
79787eaa JM |
5923 | if (ret < 0) { |
5924 | err = ret; | |
5925 | goto out_end_trans; | |
5926 | } | |
2c47e605 YZ |
5927 | BUG_ON(wc->refs[level] == 0); |
5928 | ||
c8422684 | 5929 | if (level == btrfs_root_drop_level(root_item)) |
2c47e605 YZ |
5930 | break; |
5931 | ||
5932 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5933 | path->locks[level] = 0; |
2c47e605 YZ |
5934 | WARN_ON(wc->refs[level] != 1); |
5935 | level--; | |
5936 | } | |
9f3a7427 | 5937 | } |
2c47e605 | 5938 | |
78c52d9e | 5939 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5940 | wc->level = level; |
5941 | wc->shared_level = -1; | |
5942 | wc->stage = DROP_REFERENCE; | |
5943 | wc->update_ref = update_ref; | |
5944 | wc->keep_locks = 0; | |
0b246afa | 5945 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5946 | |
d397712b | 5947 | while (1) { |
9d1a2a3a | 5948 | |
2c47e605 YZ |
5949 | ret = walk_down_tree(trans, root, path, wc); |
5950 | if (ret < 0) { | |
9a93b5a3 | 5951 | btrfs_abort_transaction(trans, ret); |
2c47e605 | 5952 | err = ret; |
20524f02 | 5953 | break; |
2c47e605 | 5954 | } |
9aca1d51 | 5955 | |
2c47e605 YZ |
5956 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5957 | if (ret < 0) { | |
9a93b5a3 | 5958 | btrfs_abort_transaction(trans, ret); |
2c47e605 | 5959 | err = ret; |
20524f02 | 5960 | break; |
2c47e605 YZ |
5961 | } |
5962 | ||
5963 | if (ret > 0) { | |
5964 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5965 | break; |
5966 | } | |
2c47e605 YZ |
5967 | |
5968 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5969 | wc->drop_level = wc->level; |
5970 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5971 | &wc->drop_progress, | |
5972 | path->slots[wc->drop_level]); | |
5973 | } | |
5974 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5975 | &wc->drop_progress); | |
c8422684 | 5976 | btrfs_set_root_drop_level(root_item, wc->drop_level); |
2c47e605 YZ |
5977 | |
5978 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5979 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5980 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5981 | ret = btrfs_update_root(trans, tree_root, |
5982 | &root->root_key, | |
5983 | root_item); | |
79787eaa | 5984 | if (ret) { |
66642832 | 5985 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5986 | err = ret; |
5987 | goto out_end_trans; | |
5988 | } | |
2c47e605 | 5989 | |
12a824dc FM |
5990 | if (!is_reloc_root) |
5991 | btrfs_set_last_root_drop_gen(fs_info, trans->transid); | |
5992 | ||
3a45bb20 | 5993 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5994 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5995 | btrfs_debug(fs_info, |
5996 | "drop snapshot early exit"); | |
3c8f2422 JB |
5997 | err = -EAGAIN; |
5998 | goto out_free; | |
5999 | } | |
6000 | ||
18d3bff4 JB |
6001 | /* |
6002 | * Use join to avoid potential EINTR from transaction | |
6003 | * start. See wait_reserve_ticket and the whole | |
6004 | * reservation callchain. | |
6005 | */ | |
6006 | if (for_reloc) | |
6007 | trans = btrfs_join_transaction(tree_root); | |
6008 | else | |
6009 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
6010 | if (IS_ERR(trans)) { |
6011 | err = PTR_ERR(trans); | |
6012 | goto out_free; | |
6013 | } | |
c3e69d58 | 6014 | } |
20524f02 | 6015 | } |
b3b4aa74 | 6016 | btrfs_release_path(path); |
79787eaa JM |
6017 | if (err) |
6018 | goto out_end_trans; | |
2c47e605 | 6019 | |
ab9ce7d4 | 6020 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 6021 | if (ret) { |
66642832 | 6022 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 6023 | err = ret; |
79787eaa JM |
6024 | goto out_end_trans; |
6025 | } | |
2c47e605 | 6026 | |
12a824dc | 6027 | if (!is_reloc_root) { |
cb517eab MX |
6028 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
6029 | NULL, NULL); | |
79787eaa | 6030 | if (ret < 0) { |
66642832 | 6031 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
6032 | err = ret; |
6033 | goto out_end_trans; | |
6034 | } else if (ret > 0) { | |
84cd948c JB |
6035 | /* if we fail to delete the orphan item this time |
6036 | * around, it'll get picked up the next time. | |
6037 | * | |
6038 | * The most common failure here is just -ENOENT. | |
6039 | */ | |
e094f480 | 6040 | btrfs_del_orphan_item(trans, tree_root, btrfs_root_id(root)); |
76dda93c YZ |
6041 | } |
6042 | } | |
6043 | ||
a3cf0e43 QW |
6044 | /* |
6045 | * This subvolume is going to be completely dropped, and won't be | |
6046 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
6047 | * commit transaction time. So free it here manually. | |
6048 | */ | |
6049 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
6050 | btrfs_qgroup_free_meta_all_pertrans(root); | |
6051 | ||
fc7cbcd4 | 6052 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 6053 | btrfs_add_dropped_root(trans, root); |
8c38938c | 6054 | else |
00246528 | 6055 | btrfs_put_root(root); |
d29a9f62 | 6056 | root_dropped = true; |
79787eaa | 6057 | out_end_trans: |
12a824dc FM |
6058 | if (!is_reloc_root) |
6059 | btrfs_set_last_root_drop_gen(fs_info, trans->transid); | |
6060 | ||
3a45bb20 | 6061 | btrfs_end_transaction_throttle(trans); |
79787eaa | 6062 | out_free: |
2c47e605 | 6063 | kfree(wc); |
5caf2a00 | 6064 | btrfs_free_path(path); |
cb1b69f4 | 6065 | out: |
b4be6aef JB |
6066 | /* |
6067 | * We were an unfinished drop root, check to see if there are any | |
6068 | * pending, and if not clear and wake up any waiters. | |
6069 | */ | |
6070 | if (!err && unfinished_drop) | |
6071 | btrfs_maybe_wake_unfinished_drop(fs_info); | |
6072 | ||
d29a9f62 JB |
6073 | /* |
6074 | * So if we need to stop dropping the snapshot for whatever reason we | |
6075 | * need to make sure to add it back to the dead root list so that we | |
6076 | * keep trying to do the work later. This also cleans up roots if we | |
6077 | * don't have it in the radix (like when we recover after a power fail | |
6078 | * or unmount) so we don't leak memory. | |
6079 | */ | |
897ca819 | 6080 | if (!for_reloc && !root_dropped) |
d29a9f62 | 6081 | btrfs_add_dead_root(root); |
2c536799 | 6082 | return err; |
20524f02 | 6083 | } |
9078a3e1 | 6084 | |
2c47e605 YZ |
6085 | /* |
6086 | * drop subtree rooted at tree block 'node'. | |
6087 | * | |
6088 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 6089 | * only used by relocation code |
2c47e605 | 6090 | */ |
f82d02d9 YZ |
6091 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6092 | struct btrfs_root *root, | |
6093 | struct extent_buffer *node, | |
6094 | struct extent_buffer *parent) | |
6095 | { | |
0b246afa | 6096 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 6097 | struct btrfs_path *path; |
2c47e605 | 6098 | struct walk_control *wc; |
f82d02d9 YZ |
6099 | int level; |
6100 | int parent_level; | |
6101 | int ret = 0; | |
6102 | int wret; | |
6103 | ||
e094f480 | 6104 | BUG_ON(btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID); |
2c47e605 | 6105 | |
f82d02d9 | 6106 | path = btrfs_alloc_path(); |
db5b493a TI |
6107 | if (!path) |
6108 | return -ENOMEM; | |
f82d02d9 | 6109 | |
2c47e605 | 6110 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6111 | if (!wc) { |
6112 | btrfs_free_path(path); | |
6113 | return -ENOMEM; | |
6114 | } | |
2c47e605 | 6115 | |
49d0c642 | 6116 | btrfs_assert_tree_write_locked(parent); |
f82d02d9 | 6117 | parent_level = btrfs_header_level(parent); |
67439dad | 6118 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
6119 | path->nodes[parent_level] = parent; |
6120 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6121 | ||
49d0c642 | 6122 | btrfs_assert_tree_write_locked(node); |
f82d02d9 | 6123 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6124 | path->nodes[level] = node; |
6125 | path->slots[level] = 0; | |
ac5887c8 | 6126 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
6127 | |
6128 | wc->refs[parent_level] = 1; | |
6129 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6130 | wc->level = level; | |
6131 | wc->shared_level = -1; | |
6132 | wc->stage = DROP_REFERENCE; | |
6133 | wc->update_ref = 0; | |
6134 | wc->keep_locks = 1; | |
0b246afa | 6135 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
6136 | |
6137 | while (1) { | |
2c47e605 YZ |
6138 | wret = walk_down_tree(trans, root, path, wc); |
6139 | if (wret < 0) { | |
f82d02d9 | 6140 | ret = wret; |
f82d02d9 | 6141 | break; |
2c47e605 | 6142 | } |
f82d02d9 | 6143 | |
2c47e605 | 6144 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6145 | if (wret < 0) |
6146 | ret = wret; | |
6147 | if (wret != 0) | |
6148 | break; | |
6149 | } | |
6150 | ||
2c47e605 | 6151 | kfree(wc); |
f82d02d9 YZ |
6152 | btrfs_free_path(path); |
6153 | return ret; | |
6154 | } | |
6155 | ||
91701bdf DS |
6156 | /* |
6157 | * Unpin the extent range in an error context and don't add the space back. | |
6158 | * Errors are not propagated further. | |
6159 | */ | |
6160 | void btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, u64 start, u64 end) | |
acce952b | 6161 | { |
91701bdf | 6162 | unpin_extent_range(fs_info, start, end, false); |
acce952b | 6163 | } |
6164 | ||
499f377f JM |
6165 | /* |
6166 | * It used to be that old block groups would be left around forever. | |
6167 | * Iterating over them would be enough to trim unused space. Since we | |
6168 | * now automatically remove them, we also need to iterate over unallocated | |
6169 | * space. | |
6170 | * | |
6171 | * We don't want a transaction for this since the discard may take a | |
6172 | * substantial amount of time. We don't require that a transaction be | |
6173 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
6174 | * to ensure that we're not discarding chunks that were released or |
6175 | * allocated in the current transaction. | |
499f377f JM |
6176 | * |
6177 | * Holding the chunks lock will prevent other threads from allocating | |
6178 | * or releasing chunks, but it won't prevent a running transaction | |
6179 | * from committing and releasing the memory that the pending chunks | |
6180 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
6181 | * transaction and hold the commit root sem. We only need to hold |
6182 | * it while performing the free space search since we have already | |
6183 | * held back allocations. | |
499f377f | 6184 | */ |
8103d10b | 6185 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 6186 | { |
37f85ec3 | 6187 | u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0; |
499f377f JM |
6188 | int ret; |
6189 | ||
6190 | *trimmed = 0; | |
6191 | ||
0be88e36 | 6192 | /* Discard not supported = nothing to do. */ |
70200574 | 6193 | if (!bdev_max_discard_sectors(device->bdev)) |
0be88e36 JM |
6194 | return 0; |
6195 | ||
52042d8e | 6196 | /* Not writable = nothing to do. */ |
ebbede42 | 6197 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
6198 | return 0; |
6199 | ||
6200 | /* No free space = nothing to do. */ | |
6201 | if (device->total_bytes <= device->bytes_used) | |
6202 | return 0; | |
6203 | ||
6204 | ret = 0; | |
6205 | ||
6206 | while (1) { | |
fb456252 | 6207 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
6208 | u64 bytes; |
6209 | ||
6210 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
6211 | if (ret) | |
fee7acc3 | 6212 | break; |
499f377f | 6213 | |
929be17a NB |
6214 | find_first_clear_extent_bit(&device->alloc_state, start, |
6215 | &start, &end, | |
6216 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 6217 | |
c57dd1f2 QW |
6218 | /* Check if there are any CHUNK_* bits left */ |
6219 | if (start > device->total_bytes) { | |
6220 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
6221 | btrfs_warn_in_rcu(fs_info, | |
6222 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
6223 | start, end - start + 1, | |
cb3e217b | 6224 | btrfs_dev_name(device), |
c57dd1f2 QW |
6225 | device->total_bytes); |
6226 | mutex_unlock(&fs_info->chunk_mutex); | |
6227 | ret = 0; | |
6228 | break; | |
6229 | } | |
6230 | ||
37f85ec3 QW |
6231 | /* Ensure we skip the reserved space on each device. */ |
6232 | start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); | |
53460a45 | 6233 | |
929be17a NB |
6234 | /* |
6235 | * If find_first_clear_extent_bit find a range that spans the | |
6236 | * end of the device it will set end to -1, in this case it's up | |
6237 | * to the caller to trim the value to the size of the device. | |
6238 | */ | |
6239 | end = min(end, device->total_bytes - 1); | |
53460a45 | 6240 | |
929be17a | 6241 | len = end - start + 1; |
499f377f | 6242 | |
929be17a NB |
6243 | /* We didn't find any extents */ |
6244 | if (!len) { | |
499f377f | 6245 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 6246 | ret = 0; |
499f377f JM |
6247 | break; |
6248 | } | |
6249 | ||
929be17a NB |
6250 | ret = btrfs_issue_discard(device->bdev, start, len, |
6251 | &bytes); | |
6252 | if (!ret) | |
0acd32c2 | 6253 | set_extent_bit(&device->alloc_state, start, |
1d126800 | 6254 | start + bytes - 1, CHUNK_TRIMMED, NULL); |
499f377f JM |
6255 | mutex_unlock(&fs_info->chunk_mutex); |
6256 | ||
6257 | if (ret) | |
6258 | break; | |
6259 | ||
6260 | start += len; | |
6261 | *trimmed += bytes; | |
6262 | ||
6263 | if (fatal_signal_pending(current)) { | |
6264 | ret = -ERESTARTSYS; | |
6265 | break; | |
6266 | } | |
6267 | ||
6268 | cond_resched(); | |
6269 | } | |
6270 | ||
6271 | return ret; | |
6272 | } | |
6273 | ||
93bba24d QW |
6274 | /* |
6275 | * Trim the whole filesystem by: | |
6276 | * 1) trimming the free space in each block group | |
6277 | * 2) trimming the unallocated space on each device | |
6278 | * | |
6279 | * This will also continue trimming even if a block group or device encounters | |
6280 | * an error. The return value will be the last error, or 0 if nothing bad | |
6281 | * happens. | |
6282 | */ | |
2ff7e61e | 6283 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 6284 | { |
23608d51 | 6285 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
32da5386 | 6286 | struct btrfs_block_group *cache = NULL; |
499f377f | 6287 | struct btrfs_device *device; |
f7039b1d | 6288 | u64 group_trimmed; |
07301df7 | 6289 | u64 range_end = U64_MAX; |
f7039b1d LD |
6290 | u64 start; |
6291 | u64 end; | |
6292 | u64 trimmed = 0; | |
93bba24d QW |
6293 | u64 bg_failed = 0; |
6294 | u64 dev_failed = 0; | |
6295 | int bg_ret = 0; | |
6296 | int dev_ret = 0; | |
f7039b1d LD |
6297 | int ret = 0; |
6298 | ||
f981fec1 JB |
6299 | if (range->start == U64_MAX) |
6300 | return -EINVAL; | |
6301 | ||
07301df7 QW |
6302 | /* |
6303 | * Check range overflow if range->len is set. | |
6304 | * The default range->len is U64_MAX. | |
6305 | */ | |
6306 | if (range->len != U64_MAX && | |
6307 | check_add_overflow(range->start, range->len, &range_end)) | |
6308 | return -EINVAL; | |
6309 | ||
6ba9fc8e | 6310 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 6311 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 6312 | if (cache->start >= range_end) { |
f7039b1d LD |
6313 | btrfs_put_block_group(cache); |
6314 | break; | |
6315 | } | |
6316 | ||
b3470b5d DS |
6317 | start = max(range->start, cache->start); |
6318 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
6319 | |
6320 | if (end - start >= range->minlen) { | |
32da5386 | 6321 | if (!btrfs_block_group_done(cache)) { |
ced8ecf0 | 6322 | ret = btrfs_cache_block_group(cache, true); |
1be41b78 | 6323 | if (ret) { |
93bba24d QW |
6324 | bg_failed++; |
6325 | bg_ret = ret; | |
6326 | continue; | |
1be41b78 | 6327 | } |
f7039b1d LD |
6328 | } |
6329 | ret = btrfs_trim_block_group(cache, | |
6330 | &group_trimmed, | |
6331 | start, | |
6332 | end, | |
6333 | range->minlen); | |
6334 | ||
6335 | trimmed += group_trimmed; | |
6336 | if (ret) { | |
93bba24d QW |
6337 | bg_failed++; |
6338 | bg_ret = ret; | |
6339 | continue; | |
f7039b1d LD |
6340 | } |
6341 | } | |
f7039b1d LD |
6342 | } |
6343 | ||
93bba24d QW |
6344 | if (bg_failed) |
6345 | btrfs_warn(fs_info, | |
6346 | "failed to trim %llu block group(s), last error %d", | |
6347 | bg_failed, bg_ret); | |
23608d51 AJ |
6348 | |
6349 | mutex_lock(&fs_devices->device_list_mutex); | |
6350 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
16a200f6 AJ |
6351 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
6352 | continue; | |
6353 | ||
8103d10b | 6354 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
6355 | if (ret) { |
6356 | dev_failed++; | |
6357 | dev_ret = ret; | |
499f377f | 6358 | break; |
93bba24d | 6359 | } |
499f377f JM |
6360 | |
6361 | trimmed += group_trimmed; | |
6362 | } | |
23608d51 | 6363 | mutex_unlock(&fs_devices->device_list_mutex); |
499f377f | 6364 | |
93bba24d QW |
6365 | if (dev_failed) |
6366 | btrfs_warn(fs_info, | |
6367 | "failed to trim %llu device(s), last error %d", | |
6368 | dev_failed, dev_ret); | |
f7039b1d | 6369 | range->len = trimmed; |
93bba24d QW |
6370 | if (bg_ret) |
6371 | return bg_ret; | |
6372 | return dev_ret; | |
f7039b1d | 6373 | } |