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" | |
995946dd | 21 | #include "tree-log.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" |
6ab0a202 | 29 | #include "sysfs.h" |
fcebe456 | 30 | #include "qgroup.h" |
fd708b81 | 31 | #include "ref-verify.h" |
8719aaae | 32 | #include "space-info.h" |
d12ffdd1 | 33 | #include "block-rsv.h" |
86736342 | 34 | #include "delalloc-space.h" |
b0643e59 | 35 | #include "discard.h" |
c57dd1f2 | 36 | #include "rcu-string.h" |
169e0da9 | 37 | #include "zoned.h" |
6143c23c | 38 | #include "dev-replace.h" |
c7f13d42 | 39 | #include "fs.h" |
07e81dc9 | 40 | #include "accessors.h" |
45c40c8f | 41 | #include "root-tree.h" |
7c8ede16 | 42 | #include "file-item.h" |
aa5d3003 | 43 | #include "orphan.h" |
103c1972 | 44 | #include "tree-checker.h" |
02c372e1 | 45 | #include "raid-stripe-tree.h" |
fec577fb | 46 | |
709c0486 AJ |
47 | #undef SCRAMBLE_DELAYED_REFS |
48 | ||
9f9b8e8d | 49 | |
5d4f98a2 | 50 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
cecbb533 | 51 | struct btrfs_delayed_ref_head *href, |
e72cb923 NB |
52 | struct btrfs_delayed_ref_node *node, u64 parent, |
53 | u64 root_objectid, u64 owner_objectid, | |
1df6b3c0 | 54 | u64 owner_offset, |
e72cb923 | 55 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
56 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
57 | struct extent_buffer *leaf, | |
58 | struct btrfs_extent_item *ei); | |
59 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
60 | u64 parent, u64 root_objectid, |
61 | u64 flags, u64 owner, u64 offset, | |
2672a051 | 62 | struct btrfs_key *ins, int ref_mod, u64 oref_root); |
5d4f98a2 | 63 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 64 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 65 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
66 | static int find_next_key(struct btrfs_path *path, int level, |
67 | struct btrfs_key *key); | |
6a63209f | 68 | |
32da5386 | 69 | static int block_group_bits(struct btrfs_block_group *cache, u64 bits) |
0f9dd46c JB |
70 | { |
71 | return (cache->flags & bits) == bits; | |
72 | } | |
73 | ||
1a4ed8fd | 74 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 75 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 | 76 | { |
29cbcf40 | 77 | struct btrfs_root *root = btrfs_extent_root(fs_info, start); |
e02119d5 CM |
78 | int ret; |
79 | struct btrfs_key key; | |
31840ae1 | 80 | struct btrfs_path *path; |
e02119d5 | 81 | |
31840ae1 | 82 | path = btrfs_alloc_path(); |
d8926bb3 MF |
83 | if (!path) |
84 | return -ENOMEM; | |
85 | ||
e02119d5 CM |
86 | key.objectid = start; |
87 | key.offset = len; | |
3173a18f | 88 | key.type = BTRFS_EXTENT_ITEM_KEY; |
29cbcf40 | 89 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
31840ae1 | 90 | btrfs_free_path(path); |
7bb86316 CM |
91 | return ret; |
92 | } | |
93 | ||
a22285a6 | 94 | /* |
3173a18f | 95 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
96 | * |
97 | * the head node for delayed ref is used to store the sum of all the | |
98 | * reference count modifications queued up in the rbtree. the head | |
99 | * node may also store the extent flags to set. This way you can check | |
100 | * to see what the reference count and extent flags would be if all of | |
101 | * the delayed refs are not processed. | |
102 | */ | |
103 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 104 | struct btrfs_fs_info *fs_info, u64 bytenr, |
d8ba2a91 JB |
105 | u64 offset, int metadata, u64 *refs, u64 *flags, |
106 | u64 *owning_root) | |
a22285a6 | 107 | { |
29cbcf40 | 108 | struct btrfs_root *extent_root; |
a22285a6 YZ |
109 | struct btrfs_delayed_ref_head *head; |
110 | struct btrfs_delayed_ref_root *delayed_refs; | |
111 | struct btrfs_path *path; | |
112 | struct btrfs_extent_item *ei; | |
113 | struct extent_buffer *leaf; | |
114 | struct btrfs_key key; | |
115 | u32 item_size; | |
116 | u64 num_refs; | |
117 | u64 extent_flags; | |
d8ba2a91 | 118 | u64 owner = 0; |
a22285a6 YZ |
119 | int ret; |
120 | ||
3173a18f JB |
121 | /* |
122 | * If we don't have skinny metadata, don't bother doing anything | |
123 | * different | |
124 | */ | |
0b246afa JM |
125 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
126 | offset = fs_info->nodesize; | |
3173a18f JB |
127 | metadata = 0; |
128 | } | |
129 | ||
a22285a6 YZ |
130 | path = btrfs_alloc_path(); |
131 | if (!path) | |
132 | return -ENOMEM; | |
133 | ||
a22285a6 YZ |
134 | if (!trans) { |
135 | path->skip_locking = 1; | |
136 | path->search_commit_root = 1; | |
137 | } | |
639eefc8 FDBM |
138 | |
139 | search_again: | |
140 | key.objectid = bytenr; | |
141 | key.offset = offset; | |
142 | if (metadata) | |
143 | key.type = BTRFS_METADATA_ITEM_KEY; | |
144 | else | |
145 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
146 | ||
29cbcf40 JB |
147 | extent_root = btrfs_extent_root(fs_info, bytenr); |
148 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
a22285a6 YZ |
149 | if (ret < 0) |
150 | goto out_free; | |
151 | ||
3173a18f | 152 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
153 | if (path->slots[0]) { |
154 | path->slots[0]--; | |
155 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
156 | path->slots[0]); | |
157 | if (key.objectid == bytenr && | |
158 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 159 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
160 | ret = 0; |
161 | } | |
3173a18f JB |
162 | } |
163 | ||
a22285a6 YZ |
164 | if (ret == 0) { |
165 | leaf = path->nodes[0]; | |
3212fa14 | 166 | item_size = btrfs_item_size(leaf, path->slots[0]); |
a22285a6 YZ |
167 | if (item_size >= sizeof(*ei)) { |
168 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
169 | struct btrfs_extent_item); | |
170 | num_refs = btrfs_extent_refs(leaf, ei); | |
171 | extent_flags = btrfs_extent_flags(leaf, ei); | |
d8ba2a91 JB |
172 | owner = btrfs_get_extent_owner_root(fs_info, leaf, |
173 | path->slots[0]); | |
a22285a6 | 174 | } else { |
182741d2 QW |
175 | ret = -EUCLEAN; |
176 | btrfs_err(fs_info, | |
177 | "unexpected extent item size, has %u expect >= %zu", | |
178 | item_size, sizeof(*ei)); | |
ba3c2b19 NB |
179 | if (trans) |
180 | btrfs_abort_transaction(trans, ret); | |
181 | else | |
182 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
183 | ||
184 | goto out_free; | |
a22285a6 | 185 | } |
ba3c2b19 | 186 | |
a22285a6 YZ |
187 | BUG_ON(num_refs == 0); |
188 | } else { | |
189 | num_refs = 0; | |
190 | extent_flags = 0; | |
191 | ret = 0; | |
192 | } | |
193 | ||
194 | if (!trans) | |
195 | goto out; | |
196 | ||
197 | delayed_refs = &trans->transaction->delayed_refs; | |
198 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 199 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
200 | if (head) { |
201 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 202 | refcount_inc(&head->refs); |
a22285a6 YZ |
203 | spin_unlock(&delayed_refs->lock); |
204 | ||
b3b4aa74 | 205 | btrfs_release_path(path); |
a22285a6 | 206 | |
8cc33e5c DS |
207 | /* |
208 | * Mutex was contended, block until it's released and try | |
209 | * again | |
210 | */ | |
a22285a6 YZ |
211 | mutex_lock(&head->mutex); |
212 | mutex_unlock(&head->mutex); | |
d278850e | 213 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 214 | goto search_again; |
a22285a6 | 215 | } |
d7df2c79 | 216 | spin_lock(&head->lock); |
a22285a6 YZ |
217 | if (head->extent_op && head->extent_op->update_flags) |
218 | extent_flags |= head->extent_op->flags_to_set; | |
219 | else | |
220 | BUG_ON(num_refs == 0); | |
221 | ||
d278850e | 222 | num_refs += head->ref_mod; |
d7df2c79 | 223 | spin_unlock(&head->lock); |
a22285a6 YZ |
224 | mutex_unlock(&head->mutex); |
225 | } | |
226 | spin_unlock(&delayed_refs->lock); | |
227 | out: | |
228 | WARN_ON(num_refs == 0); | |
229 | if (refs) | |
230 | *refs = num_refs; | |
231 | if (flags) | |
232 | *flags = extent_flags; | |
d8ba2a91 JB |
233 | if (owning_root) |
234 | *owning_root = owner; | |
a22285a6 YZ |
235 | out_free: |
236 | btrfs_free_path(path); | |
237 | return ret; | |
238 | } | |
239 | ||
d8d5f3e1 CM |
240 | /* |
241 | * Back reference rules. Back refs have three main goals: | |
242 | * | |
243 | * 1) differentiate between all holders of references to an extent so that | |
244 | * when a reference is dropped we can make sure it was a valid reference | |
245 | * before freeing the extent. | |
246 | * | |
247 | * 2) Provide enough information to quickly find the holders of an extent | |
248 | * if we notice a given block is corrupted or bad. | |
249 | * | |
250 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
251 | * maintenance. This is actually the same as #2, but with a slightly | |
252 | * different use case. | |
253 | * | |
5d4f98a2 YZ |
254 | * There are two kinds of back refs. The implicit back refs is optimized |
255 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
256 | * back refs of this kind provide information about the block's owner tree | |
257 | * and the pointer's key. These information allow us to find the block by | |
258 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
259 | * referenced by their owner trees. The location of tree block is recorded | |
260 | * in the back refs. Actually the full back refs is generic, and can be | |
261 | * used in all cases the implicit back refs is used. The major shortcoming | |
262 | * of the full back refs is its overhead. Every time a tree block gets | |
263 | * COWed, we have to update back refs entry for all pointers in it. | |
264 | * | |
265 | * For a newly allocated tree block, we use implicit back refs for | |
266 | * pointers in it. This means most tree related operations only involve | |
267 | * implicit back refs. For a tree block created in old transaction, the | |
268 | * only way to drop a reference to it is COW it. So we can detect the | |
269 | * event that tree block loses its owner tree's reference and do the | |
270 | * back refs conversion. | |
271 | * | |
01327610 | 272 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
273 | * |
274 | * The reference count of the block is one and the tree is the block's | |
275 | * owner tree. Nothing to do in this case. | |
276 | * | |
277 | * The reference count of the block is one and the tree is not the | |
278 | * block's owner tree. In this case, full back refs is used for pointers | |
279 | * in the block. Remove these full back refs, add implicit back refs for | |
280 | * every pointers in the new block. | |
281 | * | |
282 | * The reference count of the block is greater than one and the tree is | |
283 | * the block's owner tree. In this case, implicit back refs is used for | |
284 | * pointers in the block. Add full back refs for every pointers in the | |
285 | * block, increase lower level extents' reference counts. The original | |
286 | * implicit back refs are entailed to the new block. | |
287 | * | |
288 | * The reference count of the block is greater than one and the tree is | |
289 | * not the block's owner tree. Add implicit back refs for every pointer in | |
290 | * the new block, increase lower level extents' reference count. | |
291 | * | |
292 | * Back Reference Key composing: | |
293 | * | |
294 | * The key objectid corresponds to the first byte in the extent, | |
295 | * The key type is used to differentiate between types of back refs. | |
296 | * There are different meanings of the key offset for different types | |
297 | * of back refs. | |
298 | * | |
d8d5f3e1 CM |
299 | * File extents can be referenced by: |
300 | * | |
301 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 302 | * - different files inside a single subvolume |
d8d5f3e1 CM |
303 | * - different offsets inside a file (bookend extents in file.c) |
304 | * | |
5d4f98a2 | 305 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
306 | * |
307 | * - Objectid of the subvolume root | |
d8d5f3e1 | 308 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
309 | * - original offset in the file |
310 | * - how many bookend extents | |
d8d5f3e1 | 311 | * |
5d4f98a2 YZ |
312 | * The key offset for the implicit back refs is hash of the first |
313 | * three fields. | |
d8d5f3e1 | 314 | * |
5d4f98a2 | 315 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 316 | * |
5d4f98a2 | 317 | * - number of pointers in the tree leaf |
d8d5f3e1 | 318 | * |
5d4f98a2 YZ |
319 | * The key offset for the implicit back refs is the first byte of |
320 | * the tree leaf | |
d8d5f3e1 | 321 | * |
5d4f98a2 YZ |
322 | * When a file extent is allocated, The implicit back refs is used. |
323 | * the fields are filled in: | |
d8d5f3e1 | 324 | * |
5d4f98a2 | 325 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 326 | * |
5d4f98a2 YZ |
327 | * When a file extent is removed file truncation, we find the |
328 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 329 | * |
5d4f98a2 | 330 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 331 | * |
5d4f98a2 | 332 | * Btree extents can be referenced by: |
d8d5f3e1 | 333 | * |
5d4f98a2 | 334 | * - Different subvolumes |
d8d5f3e1 | 335 | * |
5d4f98a2 YZ |
336 | * Both the implicit back refs and the full back refs for tree blocks |
337 | * only consist of key. The key offset for the implicit back refs is | |
338 | * objectid of block's owner tree. The key offset for the full back refs | |
339 | * is the first byte of parent block. | |
d8d5f3e1 | 340 | * |
5d4f98a2 YZ |
341 | * When implicit back refs is used, information about the lowest key and |
342 | * level of the tree block are required. These information are stored in | |
343 | * tree block info structure. | |
d8d5f3e1 | 344 | */ |
31840ae1 | 345 | |
167ce953 LB |
346 | /* |
347 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 348 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
349 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
350 | */ | |
351 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
352 | struct btrfs_extent_inline_ref *iref, | |
353 | enum btrfs_inline_ref_type is_data) | |
354 | { | |
d9a620f7 | 355 | struct btrfs_fs_info *fs_info = eb->fs_info; |
167ce953 | 356 | int type = btrfs_extent_inline_ref_type(eb, iref); |
64ecdb64 | 357 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 | 358 | |
d9a620f7 BB |
359 | if (type == BTRFS_EXTENT_OWNER_REF_KEY) { |
360 | ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA)); | |
361 | return type; | |
362 | } | |
363 | ||
167ce953 LB |
364 | if (type == BTRFS_TREE_BLOCK_REF_KEY || |
365 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
366 | type == BTRFS_SHARED_DATA_REF_KEY || | |
367 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
368 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 369 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 370 | return type; |
64ecdb64 | 371 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
d9a620f7 | 372 | ASSERT(fs_info); |
64ecdb64 | 373 | /* |
ea57788e QW |
374 | * Every shared one has parent tree block, |
375 | * which must be aligned to sector size. | |
64ecdb64 | 376 | */ |
d9a620f7 | 377 | if (offset && IS_ALIGNED(offset, fs_info->sectorsize)) |
64ecdb64 LB |
378 | return type; |
379 | } | |
167ce953 | 380 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 381 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 382 | return type; |
64ecdb64 | 383 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
d9a620f7 | 384 | ASSERT(fs_info); |
64ecdb64 | 385 | /* |
ea57788e QW |
386 | * Every shared one has parent tree block, |
387 | * which must be aligned to sector size. | |
64ecdb64 LB |
388 | */ |
389 | if (offset && | |
d9a620f7 | 390 | IS_ALIGNED(offset, fs_info->sectorsize)) |
64ecdb64 LB |
391 | return type; |
392 | } | |
167ce953 LB |
393 | } else { |
394 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
395 | return type; | |
396 | } | |
397 | } | |
398 | ||
25761430 | 399 | WARN_ON(1); |
6c75a589 | 400 | btrfs_print_leaf(eb); |
d9a620f7 | 401 | btrfs_err(fs_info, |
ea57788e QW |
402 | "eb %llu iref 0x%lx invalid extent inline ref type %d", |
403 | eb->start, (unsigned long)iref, type); | |
167ce953 LB |
404 | |
405 | return BTRFS_REF_TYPE_INVALID; | |
406 | } | |
407 | ||
0785a9aa | 408 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
409 | { |
410 | u32 high_crc = ~(u32)0; | |
411 | u32 low_crc = ~(u32)0; | |
412 | __le64 lenum; | |
413 | ||
414 | lenum = cpu_to_le64(root_objectid); | |
03e86348 | 415 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 416 | lenum = cpu_to_le64(owner); |
03e86348 | 417 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 418 | lenum = cpu_to_le64(offset); |
03e86348 | 419 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
420 | |
421 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
422 | } | |
423 | ||
424 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
425 | struct btrfs_extent_data_ref *ref) | |
426 | { | |
427 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
428 | btrfs_extent_data_ref_objectid(leaf, ref), | |
429 | btrfs_extent_data_ref_offset(leaf, ref)); | |
430 | } | |
431 | ||
432 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
433 | struct btrfs_extent_data_ref *ref, | |
434 | u64 root_objectid, u64 owner, u64 offset) | |
435 | { | |
436 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
437 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
438 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
439 | return 0; | |
440 | return 1; | |
441 | } | |
442 | ||
443 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
444 | struct btrfs_path *path, |
445 | u64 bytenr, u64 parent, | |
446 | u64 root_objectid, | |
447 | u64 owner, u64 offset) | |
448 | { | |
29cbcf40 | 449 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
5d4f98a2 YZ |
450 | struct btrfs_key key; |
451 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 452 | struct extent_buffer *leaf; |
5d4f98a2 | 453 | u32 nritems; |
74493f7a | 454 | int ret; |
5d4f98a2 YZ |
455 | int recow; |
456 | int err = -ENOENT; | |
74493f7a | 457 | |
31840ae1 | 458 | key.objectid = bytenr; |
5d4f98a2 YZ |
459 | if (parent) { |
460 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
461 | key.offset = parent; | |
462 | } else { | |
463 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
464 | key.offset = hash_extent_data_ref(root_objectid, | |
465 | owner, offset); | |
466 | } | |
467 | again: | |
468 | recow = 0; | |
469 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
470 | if (ret < 0) { | |
471 | err = ret; | |
472 | goto fail; | |
473 | } | |
31840ae1 | 474 | |
5d4f98a2 YZ |
475 | if (parent) { |
476 | if (!ret) | |
477 | return 0; | |
5d4f98a2 | 478 | goto fail; |
31840ae1 ZY |
479 | } |
480 | ||
481 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
482 | nritems = btrfs_header_nritems(leaf); |
483 | while (1) { | |
484 | if (path->slots[0] >= nritems) { | |
485 | ret = btrfs_next_leaf(root, path); | |
486 | if (ret < 0) | |
487 | err = ret; | |
488 | if (ret) | |
489 | goto fail; | |
490 | ||
491 | leaf = path->nodes[0]; | |
492 | nritems = btrfs_header_nritems(leaf); | |
493 | recow = 1; | |
494 | } | |
495 | ||
496 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
497 | if (key.objectid != bytenr || | |
498 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
499 | goto fail; | |
500 | ||
501 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
502 | struct btrfs_extent_data_ref); | |
503 | ||
504 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
505 | owner, offset)) { | |
506 | if (recow) { | |
b3b4aa74 | 507 | btrfs_release_path(path); |
5d4f98a2 YZ |
508 | goto again; |
509 | } | |
510 | err = 0; | |
511 | break; | |
512 | } | |
513 | path->slots[0]++; | |
31840ae1 | 514 | } |
5d4f98a2 YZ |
515 | fail: |
516 | return err; | |
31840ae1 ZY |
517 | } |
518 | ||
5d4f98a2 | 519 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
520 | struct btrfs_path *path, |
521 | u64 bytenr, u64 parent, | |
522 | u64 root_objectid, u64 owner, | |
523 | u64 offset, int refs_to_add) | |
31840ae1 | 524 | { |
29cbcf40 | 525 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
31840ae1 ZY |
526 | struct btrfs_key key; |
527 | struct extent_buffer *leaf; | |
5d4f98a2 | 528 | u32 size; |
31840ae1 ZY |
529 | u32 num_refs; |
530 | int ret; | |
74493f7a | 531 | |
74493f7a | 532 | key.objectid = bytenr; |
5d4f98a2 YZ |
533 | if (parent) { |
534 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
535 | key.offset = parent; | |
536 | size = sizeof(struct btrfs_shared_data_ref); | |
537 | } else { | |
538 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
539 | key.offset = hash_extent_data_ref(root_objectid, | |
540 | owner, offset); | |
541 | size = sizeof(struct btrfs_extent_data_ref); | |
542 | } | |
74493f7a | 543 | |
5d4f98a2 YZ |
544 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
545 | if (ret && ret != -EEXIST) | |
546 | goto fail; | |
547 | ||
548 | leaf = path->nodes[0]; | |
549 | if (parent) { | |
550 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 551 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
552 | struct btrfs_shared_data_ref); |
553 | if (ret == 0) { | |
554 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
555 | } else { | |
556 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
557 | num_refs += refs_to_add; | |
558 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 559 | } |
5d4f98a2 YZ |
560 | } else { |
561 | struct btrfs_extent_data_ref *ref; | |
562 | while (ret == -EEXIST) { | |
563 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
564 | struct btrfs_extent_data_ref); | |
565 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
566 | owner, offset)) | |
567 | break; | |
b3b4aa74 | 568 | btrfs_release_path(path); |
5d4f98a2 YZ |
569 | key.offset++; |
570 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
571 | size); | |
572 | if (ret && ret != -EEXIST) | |
573 | goto fail; | |
31840ae1 | 574 | |
5d4f98a2 YZ |
575 | leaf = path->nodes[0]; |
576 | } | |
577 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
578 | struct btrfs_extent_data_ref); | |
579 | if (ret == 0) { | |
580 | btrfs_set_extent_data_ref_root(leaf, ref, | |
581 | root_objectid); | |
582 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
583 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
584 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
585 | } else { | |
586 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
587 | num_refs += refs_to_add; | |
588 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 589 | } |
31840ae1 | 590 | } |
50564b65 | 591 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
592 | ret = 0; |
593 | fail: | |
b3b4aa74 | 594 | btrfs_release_path(path); |
7bb86316 | 595 | return ret; |
74493f7a CM |
596 | } |
597 | ||
5d4f98a2 | 598 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
76d76e78 | 599 | struct btrfs_root *root, |
5d4f98a2 | 600 | struct btrfs_path *path, |
5b2a54bb | 601 | int refs_to_drop) |
31840ae1 | 602 | { |
5d4f98a2 YZ |
603 | struct btrfs_key key; |
604 | struct btrfs_extent_data_ref *ref1 = NULL; | |
605 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 606 | struct extent_buffer *leaf; |
5d4f98a2 | 607 | u32 num_refs = 0; |
31840ae1 ZY |
608 | int ret = 0; |
609 | ||
610 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
611 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
612 | ||
613 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
614 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
615 | struct btrfs_extent_data_ref); | |
616 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
617 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
618 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
619 | struct btrfs_shared_data_ref); | |
620 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 | 621 | } else { |
182741d2 QW |
622 | btrfs_err(trans->fs_info, |
623 | "unrecognized backref key (%llu %u %llu)", | |
624 | key.objectid, key.type, key.offset); | |
625 | btrfs_abort_transaction(trans, -EUCLEAN); | |
626 | return -EUCLEAN; | |
5d4f98a2 YZ |
627 | } |
628 | ||
56bec294 CM |
629 | BUG_ON(num_refs < refs_to_drop); |
630 | num_refs -= refs_to_drop; | |
5d4f98a2 | 631 | |
31840ae1 | 632 | if (num_refs == 0) { |
76d76e78 | 633 | ret = btrfs_del_item(trans, root, path); |
31840ae1 | 634 | } else { |
5d4f98a2 YZ |
635 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
636 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
637 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
638 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
50564b65 | 639 | btrfs_mark_buffer_dirty(trans, leaf); |
31840ae1 | 640 | } |
31840ae1 ZY |
641 | return ret; |
642 | } | |
643 | ||
9ed0dea0 | 644 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 645 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 646 | { |
5d4f98a2 YZ |
647 | struct btrfs_key key; |
648 | struct extent_buffer *leaf; | |
649 | struct btrfs_extent_data_ref *ref1; | |
650 | struct btrfs_shared_data_ref *ref2; | |
651 | u32 num_refs = 0; | |
3de28d57 | 652 | int type; |
5d4f98a2 YZ |
653 | |
654 | leaf = path->nodes[0]; | |
655 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 | 656 | |
5d4f98a2 | 657 | if (iref) { |
3de28d57 LB |
658 | /* |
659 | * If type is invalid, we should have bailed out earlier than | |
660 | * this call. | |
661 | */ | |
662 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
663 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
664 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
665 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
666 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
667 | } else { | |
668 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
669 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
670 | } | |
671 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
672 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
673 | struct btrfs_extent_data_ref); | |
674 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
675 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
676 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
677 | struct btrfs_shared_data_ref); | |
678 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
679 | } else { |
680 | WARN_ON(1); | |
681 | } | |
682 | return num_refs; | |
683 | } | |
15916de8 | 684 | |
5d4f98a2 | 685 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
686 | struct btrfs_path *path, |
687 | u64 bytenr, u64 parent, | |
688 | u64 root_objectid) | |
1f3c79a2 | 689 | { |
29cbcf40 | 690 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
5d4f98a2 | 691 | struct btrfs_key key; |
1f3c79a2 | 692 | int ret; |
1f3c79a2 | 693 | |
5d4f98a2 YZ |
694 | key.objectid = bytenr; |
695 | if (parent) { | |
696 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
697 | key.offset = parent; | |
698 | } else { | |
699 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
700 | key.offset = root_objectid; | |
1f3c79a2 LH |
701 | } |
702 | ||
5d4f98a2 YZ |
703 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
704 | if (ret > 0) | |
705 | ret = -ENOENT; | |
5d4f98a2 | 706 | return ret; |
1f3c79a2 LH |
707 | } |
708 | ||
5d4f98a2 | 709 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
710 | struct btrfs_path *path, |
711 | u64 bytenr, u64 parent, | |
712 | u64 root_objectid) | |
31840ae1 | 713 | { |
29cbcf40 | 714 | struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr); |
5d4f98a2 | 715 | struct btrfs_key key; |
31840ae1 | 716 | int ret; |
31840ae1 | 717 | |
5d4f98a2 YZ |
718 | key.objectid = bytenr; |
719 | if (parent) { | |
720 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
721 | key.offset = parent; | |
722 | } else { | |
723 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
724 | key.offset = root_objectid; | |
725 | } | |
726 | ||
29cbcf40 | 727 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); |
b3b4aa74 | 728 | btrfs_release_path(path); |
31840ae1 ZY |
729 | return ret; |
730 | } | |
731 | ||
5d4f98a2 | 732 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 733 | { |
5d4f98a2 YZ |
734 | int type; |
735 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
736 | if (parent > 0) | |
737 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
738 | else | |
739 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
740 | } else { | |
741 | if (parent > 0) | |
742 | type = BTRFS_SHARED_DATA_REF_KEY; | |
743 | else | |
744 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
745 | } | |
746 | return type; | |
31840ae1 | 747 | } |
56bec294 | 748 | |
2c47e605 YZ |
749 | static int find_next_key(struct btrfs_path *path, int level, |
750 | struct btrfs_key *key) | |
56bec294 | 751 | |
02217ed2 | 752 | { |
2c47e605 | 753 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
754 | if (!path->nodes[level]) |
755 | break; | |
5d4f98a2 YZ |
756 | if (path->slots[level] + 1 >= |
757 | btrfs_header_nritems(path->nodes[level])) | |
758 | continue; | |
759 | if (level == 0) | |
760 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
761 | path->slots[level] + 1); | |
762 | else | |
763 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
764 | path->slots[level] + 1); | |
765 | return 0; | |
766 | } | |
767 | return 1; | |
768 | } | |
037e6390 | 769 | |
5d4f98a2 YZ |
770 | /* |
771 | * look for inline back ref. if back ref is found, *ref_ret is set | |
772 | * to the address of inline back ref, and 0 is returned. | |
773 | * | |
774 | * if back ref isn't found, *ref_ret is set to the address where it | |
775 | * should be inserted, and -ENOENT is returned. | |
776 | * | |
777 | * if insert is true and there are too many inline back refs, the path | |
778 | * points to the extent item, and -EAGAIN is returned. | |
779 | * | |
780 | * NOTE: inline back refs are ordered in the same way that back ref | |
781 | * items in the tree are ordered. | |
782 | */ | |
783 | static noinline_for_stack | |
784 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
785 | struct btrfs_path *path, |
786 | struct btrfs_extent_inline_ref **ref_ret, | |
787 | u64 bytenr, u64 num_bytes, | |
788 | u64 parent, u64 root_objectid, | |
789 | u64 owner, u64 offset, int insert) | |
790 | { | |
867cc1fb | 791 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 792 | struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr); |
5d4f98a2 YZ |
793 | struct btrfs_key key; |
794 | struct extent_buffer *leaf; | |
795 | struct btrfs_extent_item *ei; | |
796 | struct btrfs_extent_inline_ref *iref; | |
797 | u64 flags; | |
798 | u64 item_size; | |
799 | unsigned long ptr; | |
800 | unsigned long end; | |
801 | int extra_size; | |
802 | int type; | |
803 | int want; | |
804 | int ret; | |
0b246afa | 805 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 806 | int needed; |
26b8003f | 807 | |
db94535d | 808 | key.objectid = bytenr; |
31840ae1 | 809 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 810 | key.offset = num_bytes; |
31840ae1 | 811 | |
5d4f98a2 YZ |
812 | want = extent_ref_type(parent, owner); |
813 | if (insert) { | |
814 | extra_size = btrfs_extent_inline_ref_size(want); | |
9a664971 | 815 | path->search_for_extension = 1; |
85d4198e | 816 | path->keep_locks = 1; |
5d4f98a2 YZ |
817 | } else |
818 | extra_size = -1; | |
3173a18f JB |
819 | |
820 | /* | |
16d1c062 NB |
821 | * Owner is our level, so we can just add one to get the level for the |
822 | * block we are interested in. | |
3173a18f JB |
823 | */ |
824 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
825 | key.type = BTRFS_METADATA_ITEM_KEY; | |
826 | key.offset = owner; | |
827 | } | |
828 | ||
829 | again: | |
5d4f98a2 | 830 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
cc925b96 | 831 | if (ret < 0) |
5d4f98a2 | 832 | goto out; |
3173a18f JB |
833 | |
834 | /* | |
835 | * We may be a newly converted file system which still has the old fat | |
836 | * extent entries for metadata, so try and see if we have one of those. | |
837 | */ | |
838 | if (ret > 0 && skinny_metadata) { | |
839 | skinny_metadata = false; | |
840 | if (path->slots[0]) { | |
841 | path->slots[0]--; | |
842 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
843 | path->slots[0]); | |
844 | if (key.objectid == bytenr && | |
845 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
846 | key.offset == num_bytes) | |
847 | ret = 0; | |
848 | } | |
849 | if (ret) { | |
9ce49a0b | 850 | key.objectid = bytenr; |
3173a18f JB |
851 | key.type = BTRFS_EXTENT_ITEM_KEY; |
852 | key.offset = num_bytes; | |
853 | btrfs_release_path(path); | |
854 | goto again; | |
855 | } | |
856 | } | |
857 | ||
79787eaa | 858 | if (ret && !insert) { |
cc925b96 | 859 | ret = -ENOENT; |
79787eaa | 860 | goto out; |
fae7f21c | 861 | } else if (WARN_ON(ret)) { |
7f72f505 QW |
862 | btrfs_print_leaf(path->nodes[0]); |
863 | btrfs_err(fs_info, | |
864 | "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu", | |
865 | bytenr, num_bytes, parent, root_objectid, owner, | |
866 | offset); | |
eba444f1 | 867 | ret = -EUCLEAN; |
492104c8 | 868 | goto out; |
79787eaa | 869 | } |
5d4f98a2 YZ |
870 | |
871 | leaf = path->nodes[0]; | |
3212fa14 | 872 | item_size = btrfs_item_size(leaf, path->slots[0]); |
6d8ff4e4 | 873 | if (unlikely(item_size < sizeof(*ei))) { |
cc925b96 | 874 | ret = -EUCLEAN; |
182741d2 QW |
875 | btrfs_err(fs_info, |
876 | "unexpected extent item size, has %llu expect >= %zu", | |
877 | item_size, sizeof(*ei)); | |
cc925b96 | 878 | btrfs_abort_transaction(trans, ret); |
ba3c2b19 NB |
879 | goto out; |
880 | } | |
5d4f98a2 | 881 | |
5d4f98a2 YZ |
882 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
883 | flags = btrfs_extent_flags(leaf, ei); | |
884 | ||
885 | ptr = (unsigned long)(ei + 1); | |
886 | end = (unsigned long)ei + item_size; | |
887 | ||
3173a18f | 888 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
889 | ptr += sizeof(struct btrfs_tree_block_info); |
890 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
891 | } |
892 | ||
3de28d57 LB |
893 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
894 | needed = BTRFS_REF_TYPE_DATA; | |
895 | else | |
896 | needed = BTRFS_REF_TYPE_BLOCK; | |
897 | ||
cc925b96 | 898 | ret = -ENOENT; |
da8848ac | 899 | while (ptr < end) { |
5d4f98a2 | 900 | iref = (struct btrfs_extent_inline_ref *)ptr; |
3de28d57 | 901 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
d9a620f7 BB |
902 | if (type == BTRFS_EXTENT_OWNER_REF_KEY) { |
903 | ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA)); | |
904 | ptr += btrfs_extent_inline_ref_size(type); | |
905 | continue; | |
906 | } | |
3de28d57 | 907 | if (type == BTRFS_REF_TYPE_INVALID) { |
cc925b96 | 908 | ret = -EUCLEAN; |
3de28d57 LB |
909 | goto out; |
910 | } | |
911 | ||
5d4f98a2 YZ |
912 | if (want < type) |
913 | break; | |
914 | if (want > type) { | |
915 | ptr += btrfs_extent_inline_ref_size(type); | |
916 | continue; | |
917 | } | |
918 | ||
919 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
920 | struct btrfs_extent_data_ref *dref; | |
921 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
922 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
923 | owner, offset)) { | |
cc925b96 | 924 | ret = 0; |
5d4f98a2 YZ |
925 | break; |
926 | } | |
927 | if (hash_extent_data_ref_item(leaf, dref) < | |
928 | hash_extent_data_ref(root_objectid, owner, offset)) | |
929 | break; | |
930 | } else { | |
931 | u64 ref_offset; | |
932 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
933 | if (parent > 0) { | |
934 | if (parent == ref_offset) { | |
cc925b96 | 935 | ret = 0; |
5d4f98a2 YZ |
936 | break; |
937 | } | |
938 | if (ref_offset < parent) | |
939 | break; | |
940 | } else { | |
941 | if (root_objectid == ref_offset) { | |
cc925b96 | 942 | ret = 0; |
5d4f98a2 YZ |
943 | break; |
944 | } | |
945 | if (ref_offset < root_objectid) | |
946 | break; | |
947 | } | |
948 | } | |
949 | ptr += btrfs_extent_inline_ref_size(type); | |
950 | } | |
da8848ac FM |
951 | |
952 | if (unlikely(ptr > end)) { | |
953 | ret = -EUCLEAN; | |
954 | btrfs_print_leaf(path->nodes[0]); | |
955 | btrfs_crit(fs_info, | |
956 | "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu", | |
957 | path->slots[0], root_objectid, owner, offset, parent); | |
958 | goto out; | |
959 | } | |
960 | ||
cc925b96 | 961 | if (ret == -ENOENT && insert) { |
5d4f98a2 YZ |
962 | if (item_size + extra_size >= |
963 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
cc925b96 | 964 | ret = -EAGAIN; |
5d4f98a2 YZ |
965 | goto out; |
966 | } | |
967 | /* | |
968 | * To add new inline back ref, we have to make sure | |
969 | * there is no corresponding back ref item. | |
970 | * For simplicity, we just do not add new inline back | |
971 | * ref if there is any kind of item for this block | |
972 | */ | |
2c47e605 YZ |
973 | if (find_next_key(path, 0, &key) == 0 && |
974 | key.objectid == bytenr && | |
85d4198e | 975 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
cc925b96 | 976 | ret = -EAGAIN; |
5d4f98a2 YZ |
977 | goto out; |
978 | } | |
979 | } | |
980 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
981 | out: | |
85d4198e | 982 | if (insert) { |
5d4f98a2 | 983 | path->keep_locks = 0; |
9a664971 | 984 | path->search_for_extension = 0; |
5d4f98a2 YZ |
985 | btrfs_unlock_up_safe(path, 1); |
986 | } | |
cc925b96 | 987 | return ret; |
5d4f98a2 YZ |
988 | } |
989 | ||
990 | /* | |
991 | * helper to add new inline back ref | |
992 | */ | |
993 | static noinline_for_stack | |
50564b65 | 994 | void setup_inline_extent_backref(struct btrfs_trans_handle *trans, |
143bede5 JM |
995 | struct btrfs_path *path, |
996 | struct btrfs_extent_inline_ref *iref, | |
997 | u64 parent, u64 root_objectid, | |
998 | u64 owner, u64 offset, int refs_to_add, | |
999 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1000 | { |
1001 | struct extent_buffer *leaf; | |
1002 | struct btrfs_extent_item *ei; | |
1003 | unsigned long ptr; | |
1004 | unsigned long end; | |
1005 | unsigned long item_offset; | |
1006 | u64 refs; | |
1007 | int size; | |
1008 | int type; | |
5d4f98a2 YZ |
1009 | |
1010 | leaf = path->nodes[0]; | |
1011 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1012 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1013 | ||
1014 | type = extent_ref_type(parent, owner); | |
1015 | size = btrfs_extent_inline_ref_size(type); | |
1016 | ||
50564b65 | 1017 | btrfs_extend_item(trans, path, size); |
5d4f98a2 YZ |
1018 | |
1019 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1020 | refs = btrfs_extent_refs(leaf, ei); | |
1021 | refs += refs_to_add; | |
1022 | btrfs_set_extent_refs(leaf, ei, refs); | |
1023 | if (extent_op) | |
1024 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1025 | ||
1026 | ptr = (unsigned long)ei + item_offset; | |
3212fa14 | 1027 | end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]); |
5d4f98a2 YZ |
1028 | if (ptr < end - size) |
1029 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1030 | end - size - ptr); | |
1031 | ||
1032 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1033 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1034 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1035 | struct btrfs_extent_data_ref *dref; | |
1036 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1037 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1038 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1039 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1040 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1041 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1042 | struct btrfs_shared_data_ref *sref; | |
1043 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1044 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1045 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1046 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1047 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1048 | } else { | |
1049 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1050 | } | |
50564b65 | 1051 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
1052 | } |
1053 | ||
1054 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1055 | struct btrfs_path *path, |
1056 | struct btrfs_extent_inline_ref **ref_ret, | |
1057 | u64 bytenr, u64 num_bytes, u64 parent, | |
1058 | u64 root_objectid, u64 owner, u64 offset) | |
1059 | { | |
1060 | int ret; | |
1061 | ||
867cc1fb NB |
1062 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1063 | num_bytes, parent, root_objectid, | |
1064 | owner, offset, 0); | |
5d4f98a2 | 1065 | if (ret != -ENOENT) |
54aa1f4d | 1066 | return ret; |
5d4f98a2 | 1067 | |
b3b4aa74 | 1068 | btrfs_release_path(path); |
5d4f98a2 YZ |
1069 | *ref_ret = NULL; |
1070 | ||
1071 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1072 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1073 | root_objectid); | |
5d4f98a2 | 1074 | } else { |
bd1d53ef NB |
1075 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1076 | root_objectid, owner, offset); | |
b9473439 | 1077 | } |
5d4f98a2 YZ |
1078 | return ret; |
1079 | } | |
31840ae1 | 1080 | |
5d4f98a2 YZ |
1081 | /* |
1082 | * helper to update/remove inline back ref | |
1083 | */ | |
50564b65 FM |
1084 | static noinline_for_stack int update_inline_extent_backref( |
1085 | struct btrfs_trans_handle *trans, | |
1086 | struct btrfs_path *path, | |
143bede5 JM |
1087 | struct btrfs_extent_inline_ref *iref, |
1088 | int refs_to_mod, | |
5b2a54bb | 1089 | struct btrfs_delayed_extent_op *extent_op) |
5d4f98a2 | 1090 | { |
61a18f1c | 1091 | struct extent_buffer *leaf = path->nodes[0]; |
25761430 | 1092 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5d4f98a2 YZ |
1093 | struct btrfs_extent_item *ei; |
1094 | struct btrfs_extent_data_ref *dref = NULL; | |
1095 | struct btrfs_shared_data_ref *sref = NULL; | |
1096 | unsigned long ptr; | |
1097 | unsigned long end; | |
1098 | u32 item_size; | |
1099 | int size; | |
1100 | int type; | |
5d4f98a2 YZ |
1101 | u64 refs; |
1102 | ||
5d4f98a2 YZ |
1103 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1104 | refs = btrfs_extent_refs(leaf, ei); | |
25761430 QW |
1105 | if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) { |
1106 | struct btrfs_key key; | |
1107 | u32 extent_size; | |
1108 | ||
1109 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1110 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
1111 | extent_size = fs_info->nodesize; | |
1112 | else | |
1113 | extent_size = key.offset; | |
1114 | btrfs_print_leaf(leaf); | |
1115 | btrfs_err(fs_info, | |
1116 | "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu", | |
1117 | key.objectid, extent_size, refs_to_mod, refs); | |
1118 | return -EUCLEAN; | |
1119 | } | |
5d4f98a2 YZ |
1120 | refs += refs_to_mod; |
1121 | btrfs_set_extent_refs(leaf, ei, refs); | |
1122 | if (extent_op) | |
1123 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1124 | ||
25761430 | 1125 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); |
3de28d57 | 1126 | /* |
25761430 QW |
1127 | * Function btrfs_get_extent_inline_ref_type() has already printed |
1128 | * error messages. | |
3de28d57 | 1129 | */ |
25761430 QW |
1130 | if (unlikely(type == BTRFS_REF_TYPE_INVALID)) |
1131 | return -EUCLEAN; | |
5d4f98a2 YZ |
1132 | |
1133 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1134 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1135 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1136 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1137 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1138 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1139 | } else { | |
1140 | refs = 1; | |
25761430 QW |
1141 | /* |
1142 | * For tree blocks we can only drop one ref for it, and tree | |
1143 | * blocks should not have refs > 1. | |
1144 | * | |
1145 | * Furthermore if we're inserting a new inline backref, we | |
1146 | * won't reach this path either. That would be | |
1147 | * setup_inline_extent_backref(). | |
1148 | */ | |
1149 | if (unlikely(refs_to_mod != -1)) { | |
1150 | struct btrfs_key key; | |
1151 | ||
1152 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1153 | ||
1154 | btrfs_print_leaf(leaf); | |
1155 | btrfs_err(fs_info, | |
1156 | "invalid refs_to_mod for tree block %llu, has %d expect -1", | |
1157 | key.objectid, refs_to_mod); | |
1158 | return -EUCLEAN; | |
1159 | } | |
56bec294 | 1160 | } |
31840ae1 | 1161 | |
25761430 QW |
1162 | if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) { |
1163 | struct btrfs_key key; | |
1164 | u32 extent_size; | |
1165 | ||
1166 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1167 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
1168 | extent_size = fs_info->nodesize; | |
1169 | else | |
1170 | extent_size = key.offset; | |
1171 | btrfs_print_leaf(leaf); | |
1172 | btrfs_err(fs_info, | |
1173 | "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu", | |
1174 | (unsigned long)iref, key.objectid, extent_size, | |
1175 | refs_to_mod, refs); | |
1176 | return -EUCLEAN; | |
1177 | } | |
5d4f98a2 YZ |
1178 | refs += refs_to_mod; |
1179 | ||
1180 | if (refs > 0) { | |
1181 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1182 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1183 | else | |
1184 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1185 | } else { | |
1186 | size = btrfs_extent_inline_ref_size(type); | |
3212fa14 | 1187 | item_size = btrfs_item_size(leaf, path->slots[0]); |
5d4f98a2 YZ |
1188 | ptr = (unsigned long)iref; |
1189 | end = (unsigned long)ei + item_size; | |
1190 | if (ptr + size < end) | |
1191 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1192 | end - ptr - size); | |
1193 | item_size -= size; | |
50564b65 | 1194 | btrfs_truncate_item(trans, path, item_size, 1); |
5d4f98a2 | 1195 | } |
50564b65 | 1196 | btrfs_mark_buffer_dirty(trans, leaf); |
25761430 | 1197 | return 0; |
5d4f98a2 YZ |
1198 | } |
1199 | ||
1200 | static noinline_for_stack | |
1201 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1202 | struct btrfs_path *path, |
1203 | u64 bytenr, u64 num_bytes, u64 parent, | |
1204 | u64 root_objectid, u64 owner, | |
1205 | u64 offset, int refs_to_add, | |
1206 | struct btrfs_delayed_extent_op *extent_op) | |
1207 | { | |
1208 | struct btrfs_extent_inline_ref *iref; | |
1209 | int ret; | |
1210 | ||
867cc1fb NB |
1211 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1212 | num_bytes, parent, root_objectid, | |
1213 | owner, offset, 1); | |
5d4f98a2 | 1214 | if (ret == 0) { |
07cce5cf QW |
1215 | /* |
1216 | * We're adding refs to a tree block we already own, this | |
1217 | * should not happen at all. | |
1218 | */ | |
1219 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
eee3b811 | 1220 | btrfs_print_leaf(path->nodes[0]); |
07cce5cf | 1221 | btrfs_crit(trans->fs_info, |
eee3b811 QW |
1222 | "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u", |
1223 | bytenr, num_bytes, root_objectid, path->slots[0]); | |
07cce5cf QW |
1224 | return -EUCLEAN; |
1225 | } | |
50564b65 FM |
1226 | ret = update_inline_extent_backref(trans, path, iref, |
1227 | refs_to_add, extent_op); | |
5d4f98a2 | 1228 | } else if (ret == -ENOENT) { |
50564b65 | 1229 | setup_inline_extent_backref(trans, path, iref, parent, |
143bede5 JM |
1230 | root_objectid, owner, offset, |
1231 | refs_to_add, extent_op); | |
1232 | ret = 0; | |
771ed689 | 1233 | } |
5d4f98a2 YZ |
1234 | return ret; |
1235 | } | |
31840ae1 | 1236 | |
5d4f98a2 | 1237 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
76d76e78 | 1238 | struct btrfs_root *root, |
5d4f98a2 YZ |
1239 | struct btrfs_path *path, |
1240 | struct btrfs_extent_inline_ref *iref, | |
5b2a54bb | 1241 | int refs_to_drop, int is_data) |
5d4f98a2 | 1242 | { |
143bede5 | 1243 | int ret = 0; |
b9473439 | 1244 | |
5d4f98a2 | 1245 | BUG_ON(!is_data && refs_to_drop != 1); |
5b2a54bb | 1246 | if (iref) |
50564b65 FM |
1247 | ret = update_inline_extent_backref(trans, path, iref, |
1248 | -refs_to_drop, NULL); | |
5b2a54bb JB |
1249 | else if (is_data) |
1250 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1251 | else | |
76d76e78 | 1252 | ret = btrfs_del_item(trans, root, path); |
5d4f98a2 YZ |
1253 | return ret; |
1254 | } | |
1255 | ||
d04c6b88 JM |
1256 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1257 | u64 *discarded_bytes) | |
5d4f98a2 | 1258 | { |
86557861 JM |
1259 | int j, ret = 0; |
1260 | u64 bytes_left, end; | |
adbe7e38 | 1261 | u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT); |
d04c6b88 | 1262 | |
4d89d377 JM |
1263 | if (WARN_ON(start != aligned_start)) { |
1264 | len -= aligned_start - start; | |
adbe7e38 | 1265 | len = round_down(len, 1 << SECTOR_SHIFT); |
4d89d377 JM |
1266 | start = aligned_start; |
1267 | } | |
d04c6b88 | 1268 | |
4d89d377 | 1269 | *discarded_bytes = 0; |
86557861 JM |
1270 | |
1271 | if (!len) | |
1272 | return 0; | |
1273 | ||
1274 | end = start + len; | |
1275 | bytes_left = len; | |
1276 | ||
1277 | /* Skip any superblocks on this device. */ | |
1278 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1279 | u64 sb_start = btrfs_sb_offset(j); | |
1280 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1281 | u64 size = sb_start - start; | |
1282 | ||
1283 | if (!in_range(sb_start, start, bytes_left) && | |
1284 | !in_range(sb_end, start, bytes_left) && | |
1285 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1286 | continue; | |
1287 | ||
1288 | /* | |
1289 | * Superblock spans beginning of range. Adjust start and | |
1290 | * try again. | |
1291 | */ | |
1292 | if (sb_start <= start) { | |
1293 | start += sb_end - start; | |
1294 | if (start > end) { | |
1295 | bytes_left = 0; | |
1296 | break; | |
1297 | } | |
1298 | bytes_left = end - start; | |
1299 | continue; | |
1300 | } | |
1301 | ||
1302 | if (size) { | |
29e70be2 AJ |
1303 | ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT, |
1304 | size >> SECTOR_SHIFT, | |
44abff2c | 1305 | GFP_NOFS); |
86557861 JM |
1306 | if (!ret) |
1307 | *discarded_bytes += size; | |
1308 | else if (ret != -EOPNOTSUPP) | |
1309 | return ret; | |
1310 | } | |
1311 | ||
1312 | start = sb_end; | |
1313 | if (start > end) { | |
1314 | bytes_left = 0; | |
1315 | break; | |
1316 | } | |
1317 | bytes_left = end - start; | |
1318 | } | |
1319 | ||
1320 | if (bytes_left) { | |
29e70be2 AJ |
1321 | ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT, |
1322 | bytes_left >> SECTOR_SHIFT, | |
44abff2c | 1323 | GFP_NOFS); |
4d89d377 | 1324 | if (!ret) |
86557861 | 1325 | *discarded_bytes += bytes_left; |
4d89d377 | 1326 | } |
d04c6b88 | 1327 | return ret; |
5d4f98a2 | 1328 | } |
5d4f98a2 | 1329 | |
a4012f06 | 1330 | static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes) |
6143c23c NA |
1331 | { |
1332 | struct btrfs_device *dev = stripe->dev; | |
1333 | struct btrfs_fs_info *fs_info = dev->fs_info; | |
1334 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1335 | u64 phys = stripe->physical; | |
1336 | u64 len = stripe->length; | |
1337 | u64 discarded = 0; | |
1338 | int ret = 0; | |
1339 | ||
1340 | /* Zone reset on a zoned filesystem */ | |
1341 | if (btrfs_can_zone_reset(dev, phys, len)) { | |
1342 | u64 src_disc; | |
1343 | ||
1344 | ret = btrfs_reset_device_zone(dev, phys, len, &discarded); | |
1345 | if (ret) | |
1346 | goto out; | |
1347 | ||
1348 | if (!btrfs_dev_replace_is_ongoing(dev_replace) || | |
1349 | dev != dev_replace->srcdev) | |
1350 | goto out; | |
1351 | ||
1352 | src_disc = discarded; | |
1353 | ||
1354 | /* Send to replace target as well */ | |
1355 | ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len, | |
1356 | &discarded); | |
1357 | discarded += src_disc; | |
70200574 | 1358 | } else if (bdev_max_discard_sectors(stripe->dev->bdev)) { |
6143c23c NA |
1359 | ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded); |
1360 | } else { | |
1361 | ret = 0; | |
1362 | *bytes = 0; | |
1363 | } | |
1364 | ||
1365 | out: | |
1366 | *bytes = discarded; | |
1367 | return ret; | |
1368 | } | |
1369 | ||
2ff7e61e | 1370 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1371 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1372 | { |
6b7faadd | 1373 | int ret = 0; |
5378e607 | 1374 | u64 discarded_bytes = 0; |
6b7faadd QW |
1375 | u64 end = bytenr + num_bytes; |
1376 | u64 cur = bytenr; | |
e244a0ae | 1377 | |
2999241d | 1378 | /* |
a4012f06 CH |
1379 | * Avoid races with device replace and make sure the devices in the |
1380 | * stripes don't go away while we are discarding. | |
2999241d | 1381 | */ |
0b246afa | 1382 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd | 1383 | while (cur < end) { |
a4012f06 CH |
1384 | struct btrfs_discard_stripe *stripes; |
1385 | unsigned int num_stripes; | |
5d4f98a2 YZ |
1386 | int i; |
1387 | ||
6b7faadd | 1388 | num_bytes = end - cur; |
a4012f06 CH |
1389 | stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes); |
1390 | if (IS_ERR(stripes)) { | |
1391 | ret = PTR_ERR(stripes); | |
1392 | if (ret == -EOPNOTSUPP) | |
1393 | ret = 0; | |
1394 | break; | |
1395 | } | |
5d4f98a2 | 1396 | |
a4012f06 CH |
1397 | for (i = 0; i < num_stripes; i++) { |
1398 | struct btrfs_discard_stripe *stripe = stripes + i; | |
d04c6b88 | 1399 | u64 bytes; |
38b5f68e | 1400 | |
a4012f06 | 1401 | if (!stripe->dev->bdev) { |
627e0873 FM |
1402 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); |
1403 | continue; | |
1404 | } | |
dcba6e48 | 1405 | |
a4012f06 CH |
1406 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
1407 | &stripe->dev->dev_state)) | |
5e753a81 AJ |
1408 | continue; |
1409 | ||
6143c23c | 1410 | ret = do_discard_extent(stripe, &bytes); |
a4012f06 | 1411 | if (ret) { |
6b7faadd | 1412 | /* |
a4012f06 CH |
1413 | * Keep going if discard is not supported by the |
1414 | * device. | |
6b7faadd | 1415 | */ |
a4012f06 CH |
1416 | if (ret != -EOPNOTSUPP) |
1417 | break; | |
1418 | ret = 0; | |
1419 | } else { | |
1420 | discarded_bytes += bytes; | |
6b7faadd | 1421 | } |
5d4f98a2 | 1422 | } |
a4012f06 CH |
1423 | kfree(stripes); |
1424 | if (ret) | |
1425 | break; | |
6b7faadd | 1426 | cur += num_bytes; |
5d4f98a2 | 1427 | } |
0b246afa | 1428 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1429 | if (actual_bytes) |
1430 | *actual_bytes = discarded_bytes; | |
5d4f98a2 | 1431 | return ret; |
5d4f98a2 YZ |
1432 | } |
1433 | ||
79787eaa | 1434 | /* Can return -ENOMEM */ |
5d4f98a2 | 1435 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1436 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1437 | { |
82fa113f | 1438 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 1439 | int ret; |
66d7e7f0 | 1440 | |
82fa113f QW |
1441 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1442 | generic_ref->action); | |
1443 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
610647d7 | 1444 | generic_ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 1445 | |
82fa113f | 1446 | if (generic_ref->type == BTRFS_REF_METADATA) |
2187374f | 1447 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL); |
82fa113f | 1448 | else |
2187374f | 1449 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0); |
d7eae340 | 1450 | |
82fa113f | 1451 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1452 | |
5d4f98a2 YZ |
1453 | return ret; |
1454 | } | |
1455 | ||
bd3c685e | 1456 | /* |
9580503b | 1457 | * Insert backreference for a given extent. |
bd3c685e | 1458 | * |
07cce5cf QW |
1459 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1460 | * how it works. | |
1461 | * | |
bd3c685e NB |
1462 | * @trans: Handle of transaction |
1463 | * | |
1464 | * @node: The delayed ref node used to get the bytenr/length for | |
1465 | * extent whose references are incremented. | |
1466 | * | |
1467 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1468 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1469 | * bytenr of the parent block. Since new extents are always | |
1470 | * created with indirect references, this will only be the case | |
1471 | * when relocating a shared extent. In that case, root_objectid | |
1a9fd417 | 1472 | * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must |
bd3c685e NB |
1473 | * be 0 |
1474 | * | |
1475 | * @root_objectid: The id of the root where this modification has originated, | |
1476 | * this can be either one of the well-known metadata trees or | |
1477 | * the subvolume id which references this extent. | |
1478 | * | |
1479 | * @owner: For data extents it is the inode number of the owning file. | |
1480 | * For metadata extents this parameter holds the level in the | |
1481 | * tree of the extent. | |
1482 | * | |
1483 | * @offset: For metadata extents the offset is ignored and is currently | |
1484 | * always passed as 0. For data extents it is the fileoffset | |
1485 | * this extent belongs to. | |
1486 | * | |
bd3c685e NB |
1487 | * @extent_op Pointer to a structure, holding information necessary when |
1488 | * updating a tree block's flags | |
1489 | * | |
1490 | */ | |
5d4f98a2 | 1491 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1492 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 | 1493 | u64 parent, u64 root_objectid, |
88b2d088 | 1494 | u64 owner, u64 offset, |
5d4f98a2 YZ |
1495 | struct btrfs_delayed_extent_op *extent_op) |
1496 | { | |
1497 | struct btrfs_path *path; | |
1498 | struct extent_buffer *leaf; | |
1499 | struct btrfs_extent_item *item; | |
fcebe456 | 1500 | struct btrfs_key key; |
c682f9b3 QW |
1501 | u64 bytenr = node->bytenr; |
1502 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 | 1503 | u64 refs; |
88b2d088 | 1504 | int refs_to_add = node->ref_mod; |
5d4f98a2 | 1505 | int ret; |
5d4f98a2 YZ |
1506 | |
1507 | path = btrfs_alloc_path(); | |
1508 | if (!path) | |
1509 | return -ENOMEM; | |
1510 | ||
5d4f98a2 | 1511 | /* this will setup the path even if it fails to insert the back ref */ |
a639cdeb NB |
1512 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1513 | parent, root_objectid, owner, | |
1514 | offset, refs_to_add, extent_op); | |
0ed4792a | 1515 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1516 | goto out; |
fcebe456 JB |
1517 | |
1518 | /* | |
1519 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1520 | * inline extent ref, so just update the reference count and add a | |
1521 | * normal backref. | |
1522 | */ | |
5d4f98a2 | 1523 | leaf = path->nodes[0]; |
fcebe456 | 1524 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1525 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1526 | refs = btrfs_extent_refs(leaf, item); | |
1527 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1528 | if (extent_op) | |
1529 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1530 | |
50564b65 | 1531 | btrfs_mark_buffer_dirty(trans, leaf); |
b3b4aa74 | 1532 | btrfs_release_path(path); |
56bec294 | 1533 | |
56bec294 | 1534 | /* now insert the actual backref */ |
d2f79e63 | 1535 | if (owner < BTRFS_FIRST_FREE_OBJECTID) |
65cd6d9e NB |
1536 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1537 | root_objectid); | |
d2f79e63 | 1538 | else |
65cd6d9e NB |
1539 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1540 | root_objectid, owner, offset, | |
1541 | refs_to_add); | |
d2f79e63 | 1542 | |
79787eaa | 1543 | if (ret) |
66642832 | 1544 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1545 | out: |
56bec294 | 1546 | btrfs_free_path(path); |
30d133fc | 1547 | return ret; |
56bec294 CM |
1548 | } |
1549 | ||
5d4f98a2 | 1550 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
cecbb533 | 1551 | struct btrfs_delayed_ref_head *href, |
5d4f98a2 YZ |
1552 | struct btrfs_delayed_ref_node *node, |
1553 | struct btrfs_delayed_extent_op *extent_op, | |
61c681fe | 1554 | bool insert_reserved) |
56bec294 | 1555 | { |
5d4f98a2 YZ |
1556 | int ret = 0; |
1557 | struct btrfs_delayed_data_ref *ref; | |
5d4f98a2 | 1558 | u64 parent = 0; |
5d4f98a2 YZ |
1559 | u64 flags = 0; |
1560 | ||
5d4f98a2 | 1561 | ref = btrfs_delayed_node_to_data_ref(node); |
2bf98ef3 | 1562 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1563 | |
5d4f98a2 YZ |
1564 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1565 | parent = ref->parent; | |
5d4f98a2 YZ |
1566 | |
1567 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
7cce0d69 | 1568 | struct btrfs_key key; |
cecbb533 BB |
1569 | struct btrfs_squota_delta delta = { |
1570 | .root = href->owning_root, | |
1571 | .num_bytes = node->num_bytes, | |
1572 | .rsv_bytes = href->reserved_bytes, | |
1573 | .is_data = true, | |
1574 | .is_inc = true, | |
bd7c1ea3 | 1575 | .generation = trans->transid, |
cecbb533 | 1576 | }; |
7cce0d69 | 1577 | |
3173a18f | 1578 | if (extent_op) |
5d4f98a2 | 1579 | flags |= extent_op->flags_to_set; |
7cce0d69 FM |
1580 | |
1581 | key.objectid = node->bytenr; | |
1582 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1583 | key.offset = node->num_bytes; | |
1584 | ||
e721043a | 1585 | ret = alloc_reserved_file_extent(trans, parent, ref->root, |
ef89b824 | 1586 | flags, ref->objectid, |
7cce0d69 | 1587 | ref->offset, &key, |
2672a051 | 1588 | node->ref_mod, href->owning_root); |
cecbb533 BB |
1589 | if (!ret) |
1590 | ret = btrfs_record_squota_delta(trans->fs_info, &delta); | |
1591 | else | |
1592 | btrfs_qgroup_free_refroot(trans->fs_info, delta.root, | |
1593 | delta.rsv_bytes, BTRFS_QGROUP_RSV_DATA); | |
5d4f98a2 | 1594 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
e721043a | 1595 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref->root, |
2590d0f1 | 1596 | ref->objectid, ref->offset, |
88b2d088 | 1597 | extent_op); |
5d4f98a2 | 1598 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
cecbb533 | 1599 | ret = __btrfs_free_extent(trans, href, node, parent, |
e721043a | 1600 | ref->root, ref->objectid, |
1df6b3c0 | 1601 | ref->offset, extent_op); |
5d4f98a2 YZ |
1602 | } else { |
1603 | BUG(); | |
1604 | } | |
1605 | return ret; | |
1606 | } | |
1607 | ||
1608 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1609 | struct extent_buffer *leaf, | |
1610 | struct btrfs_extent_item *ei) | |
1611 | { | |
1612 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1613 | if (extent_op->update_flags) { | |
1614 | flags |= extent_op->flags_to_set; | |
1615 | btrfs_set_extent_flags(leaf, ei, flags); | |
1616 | } | |
1617 | ||
1618 | if (extent_op->update_key) { | |
1619 | struct btrfs_tree_block_info *bi; | |
1620 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1621 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1622 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1623 | } | |
1624 | } | |
1625 | ||
1626 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1627 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1628 | struct btrfs_delayed_extent_op *extent_op) |
1629 | { | |
20b9a2d6 | 1630 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 1631 | struct btrfs_root *root; |
5d4f98a2 YZ |
1632 | struct btrfs_key key; |
1633 | struct btrfs_path *path; | |
1634 | struct btrfs_extent_item *ei; | |
1635 | struct extent_buffer *leaf; | |
1636 | u32 item_size; | |
56bec294 | 1637 | int ret; |
0e3696f8 | 1638 | int metadata = 1; |
5d4f98a2 | 1639 | |
bf31f87f | 1640 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1641 | return 0; |
1642 | ||
0e3696f8 | 1643 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1644 | metadata = 0; |
1645 | ||
5d4f98a2 YZ |
1646 | path = btrfs_alloc_path(); |
1647 | if (!path) | |
1648 | return -ENOMEM; | |
1649 | ||
d278850e | 1650 | key.objectid = head->bytenr; |
5d4f98a2 | 1651 | |
3173a18f | 1652 | if (metadata) { |
3173a18f | 1653 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1654 | key.offset = extent_op->level; |
3173a18f JB |
1655 | } else { |
1656 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1657 | key.offset = head->num_bytes; |
3173a18f JB |
1658 | } |
1659 | ||
29cbcf40 | 1660 | root = btrfs_extent_root(fs_info, key.objectid); |
3173a18f | 1661 | again: |
29cbcf40 | 1662 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
5d4f98a2 | 1663 | if (ret < 0) { |
5d4f98a2 | 1664 | goto out; |
20fb05a6 | 1665 | } else if (ret > 0) { |
3173a18f | 1666 | if (metadata) { |
55994887 FDBM |
1667 | if (path->slots[0] > 0) { |
1668 | path->slots[0]--; | |
1669 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1670 | path->slots[0]); | |
d278850e | 1671 | if (key.objectid == head->bytenr && |
55994887 | 1672 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1673 | key.offset == head->num_bytes) |
55994887 FDBM |
1674 | ret = 0; |
1675 | } | |
1676 | if (ret > 0) { | |
1677 | btrfs_release_path(path); | |
1678 | metadata = 0; | |
3173a18f | 1679 | |
d278850e JB |
1680 | key.objectid = head->bytenr; |
1681 | key.offset = head->num_bytes; | |
55994887 FDBM |
1682 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1683 | goto again; | |
1684 | } | |
1685 | } else { | |
20fb05a6 | 1686 | ret = -EUCLEAN; |
8ec0a4a5 FM |
1687 | btrfs_err(fs_info, |
1688 | "missing extent item for extent %llu num_bytes %llu level %d", | |
1689 | head->bytenr, head->num_bytes, extent_op->level); | |
55994887 | 1690 | goto out; |
3173a18f | 1691 | } |
5d4f98a2 YZ |
1692 | } |
1693 | ||
1694 | leaf = path->nodes[0]; | |
3212fa14 | 1695 | item_size = btrfs_item_size(leaf, path->slots[0]); |
ba3c2b19 | 1696 | |
6d8ff4e4 | 1697 | if (unlikely(item_size < sizeof(*ei))) { |
20fb05a6 | 1698 | ret = -EUCLEAN; |
182741d2 QW |
1699 | btrfs_err(fs_info, |
1700 | "unexpected extent item size, has %u expect >= %zu", | |
1701 | item_size, sizeof(*ei)); | |
20fb05a6 | 1702 | btrfs_abort_transaction(trans, ret); |
ba3c2b19 NB |
1703 | goto out; |
1704 | } | |
1705 | ||
5d4f98a2 YZ |
1706 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1707 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1708 | |
50564b65 | 1709 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
1710 | out: |
1711 | btrfs_free_path(path); | |
20fb05a6 | 1712 | return ret; |
56bec294 CM |
1713 | } |
1714 | ||
5d4f98a2 | 1715 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
cecbb533 | 1716 | struct btrfs_delayed_ref_head *href, |
5d4f98a2 YZ |
1717 | struct btrfs_delayed_ref_node *node, |
1718 | struct btrfs_delayed_extent_op *extent_op, | |
61c681fe | 1719 | bool insert_reserved) |
56bec294 CM |
1720 | { |
1721 | int ret = 0; | |
cecbb533 | 1722 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 1723 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1724 | u64 parent = 0; |
1725 | u64 ref_root = 0; | |
56bec294 | 1726 | |
5d4f98a2 | 1727 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1728 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1729 | |
5d4f98a2 YZ |
1730 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1731 | parent = ref->parent; | |
fcebe456 | 1732 | ref_root = ref->root; |
5d4f98a2 | 1733 | |
1bf76df3 | 1734 | if (unlikely(node->ref_mod != 1)) { |
f97806f2 | 1735 | btrfs_err(trans->fs_info, |
1bf76df3 | 1736 | "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu", |
02794222 LB |
1737 | node->bytenr, node->ref_mod, node->action, ref_root, |
1738 | parent); | |
1bf76df3 | 1739 | return -EUCLEAN; |
02794222 | 1740 | } |
5d4f98a2 | 1741 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
cecbb533 BB |
1742 | struct btrfs_squota_delta delta = { |
1743 | .root = href->owning_root, | |
1744 | .num_bytes = fs_info->nodesize, | |
1745 | .rsv_bytes = 0, | |
1746 | .is_data = false, | |
1747 | .is_inc = true, | |
bd7c1ea3 | 1748 | .generation = trans->transid, |
cecbb533 BB |
1749 | }; |
1750 | ||
3173a18f | 1751 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1752 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
cecbb533 BB |
1753 | if (!ret) |
1754 | btrfs_record_squota_delta(fs_info, &delta); | |
5d4f98a2 | 1755 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 | 1756 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
88b2d088 | 1757 | ref->level, 0, extent_op); |
5d4f98a2 | 1758 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
cecbb533 | 1759 | ret = __btrfs_free_extent(trans, href, node, parent, ref_root, |
1df6b3c0 | 1760 | ref->level, 0, extent_op); |
5d4f98a2 YZ |
1761 | } else { |
1762 | BUG(); | |
1763 | } | |
56bec294 CM |
1764 | return ret; |
1765 | } | |
1766 | ||
1767 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1768 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
cecbb533 | 1769 | struct btrfs_delayed_ref_head *href, |
5d4f98a2 YZ |
1770 | struct btrfs_delayed_ref_node *node, |
1771 | struct btrfs_delayed_extent_op *extent_op, | |
61c681fe | 1772 | bool insert_reserved) |
56bec294 | 1773 | { |
79787eaa JM |
1774 | int ret = 0; |
1775 | ||
bf31f87f | 1776 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1777 | if (insert_reserved) |
b25c36f8 | 1778 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1779 | return 0; |
857cc2fc | 1780 | } |
79787eaa | 1781 | |
5d4f98a2 YZ |
1782 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1783 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
cecbb533 | 1784 | ret = run_delayed_tree_ref(trans, href, node, extent_op, |
5d4f98a2 YZ |
1785 | insert_reserved); |
1786 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1787 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
cecbb533 | 1788 | ret = run_delayed_data_ref(trans, href, node, extent_op, |
5d4f98a2 | 1789 | insert_reserved); |
d9a620f7 BB |
1790 | else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY) |
1791 | ret = 0; | |
5d4f98a2 YZ |
1792 | else |
1793 | BUG(); | |
80ee54bf | 1794 | if (ret && insert_reserved) |
b25c36f8 | 1795 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
39f501d6 QW |
1796 | if (ret < 0) |
1797 | btrfs_err(trans->fs_info, | |
1798 | "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d", | |
1799 | node->bytenr, node->num_bytes, node->type, | |
1800 | node->action, node->ref_mod, ret); | |
5d4f98a2 | 1801 | return ret; |
56bec294 CM |
1802 | } |
1803 | ||
c6fc2454 | 1804 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1805 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1806 | { | |
cffc3374 FM |
1807 | struct btrfs_delayed_ref_node *ref; |
1808 | ||
e3d03965 | 1809 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1810 | return NULL; |
d7df2c79 | 1811 | |
cffc3374 FM |
1812 | /* |
1813 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1814 | * This is to prevent a ref count from going down to zero, which deletes | |
1815 | * the extent item from the extent tree, when there still are references | |
1816 | * to add, which would fail because they would not find the extent item. | |
1817 | */ | |
1d57ee94 WX |
1818 | if (!list_empty(&head->ref_add_list)) |
1819 | return list_first_entry(&head->ref_add_list, | |
1820 | struct btrfs_delayed_ref_node, add_list); | |
1821 | ||
e3d03965 | 1822 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1823 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1824 | ASSERT(list_empty(&ref->add_list)); |
1825 | return ref; | |
56bec294 CM |
1826 | } |
1827 | ||
2eadaa22 JB |
1828 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1829 | struct btrfs_delayed_ref_head *head) | |
1830 | { | |
1831 | spin_lock(&delayed_refs->lock); | |
61c681fe | 1832 | head->processing = false; |
2eadaa22 JB |
1833 | delayed_refs->num_heads_ready++; |
1834 | spin_unlock(&delayed_refs->lock); | |
1835 | btrfs_delayed_ref_unlock(head); | |
1836 | } | |
1837 | ||
bedc6617 JB |
1838 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1839 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1840 | { |
1841 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1842 | |
1843 | if (!extent_op) | |
bedc6617 JB |
1844 | return NULL; |
1845 | ||
b00e6250 | 1846 | if (head->must_insert_reserved) { |
bedc6617 | 1847 | head->extent_op = NULL; |
b00e6250 | 1848 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1849 | return NULL; |
b00e6250 | 1850 | } |
bedc6617 JB |
1851 | return extent_op; |
1852 | } | |
1853 | ||
1854 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1855 | struct btrfs_delayed_ref_head *head) | |
1856 | { | |
1857 | struct btrfs_delayed_extent_op *extent_op; | |
1858 | int ret; | |
1859 | ||
1860 | extent_op = cleanup_extent_op(head); | |
1861 | if (!extent_op) | |
1862 | return 0; | |
1863 | head->extent_op = NULL; | |
b00e6250 | 1864 | spin_unlock(&head->lock); |
20b9a2d6 | 1865 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1866 | btrfs_free_delayed_extent_op(extent_op); |
1867 | return ret ? ret : 1; | |
1868 | } | |
1869 | ||
8a526c44 | 1870 | u64 btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
31890da0 JB |
1871 | struct btrfs_delayed_ref_root *delayed_refs, |
1872 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1873 | { |
81e75ac7 JB |
1874 | /* |
1875 | * We had csum deletions accounted for in our delayed refs rsv, we need | |
1876 | * to drop the csum leaves for this update from our delayed_refs_rsv. | |
1877 | */ | |
1878 | if (head->total_ref_mod < 0 && head->is_data) { | |
adb86dbe | 1879 | int nr_csums; |
3ee56a58 | 1880 | |
81e75ac7 JB |
1881 | spin_lock(&delayed_refs->lock); |
1882 | delayed_refs->pending_csums -= head->num_bytes; | |
1883 | spin_unlock(&delayed_refs->lock); | |
adb86dbe | 1884 | nr_csums = btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes); |
3ee56a58 | 1885 | |
adb86dbe | 1886 | btrfs_delayed_refs_rsv_release(fs_info, 0, nr_csums); |
81e75ac7 | 1887 | |
adb86dbe | 1888 | return btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums); |
3ee56a58 | 1889 | } |
cecbb533 BB |
1890 | if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE && |
1891 | head->must_insert_reserved && head->is_data) | |
1892 | btrfs_qgroup_free_refroot(fs_info, head->owning_root, | |
1893 | head->reserved_bytes, BTRFS_QGROUP_RSV_DATA); | |
8a526c44 | 1894 | |
3ee56a58 | 1895 | return 0; |
07c47775 JB |
1896 | } |
1897 | ||
194ab0bc | 1898 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
8a526c44 FM |
1899 | struct btrfs_delayed_ref_head *head, |
1900 | u64 *bytes_released) | |
194ab0bc | 1901 | { |
f9871edd NB |
1902 | |
1903 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1904 | struct btrfs_delayed_ref_root *delayed_refs; |
1905 | int ret; | |
1906 | ||
1907 | delayed_refs = &trans->transaction->delayed_refs; | |
1908 | ||
bedc6617 | 1909 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1910 | if (ret < 0) { |
1911 | unselect_delayed_ref_head(delayed_refs, head); | |
1912 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1913 | return ret; | |
1914 | } else if (ret) { | |
1915 | return ret; | |
1916 | } | |
1917 | ||
1918 | /* | |
1919 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1920 | * and then re-check to make sure nobody got added. | |
1921 | */ | |
1922 | spin_unlock(&head->lock); | |
1923 | spin_lock(&delayed_refs->lock); | |
1924 | spin_lock(&head->lock); | |
e3d03965 | 1925 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1926 | spin_unlock(&head->lock); |
1927 | spin_unlock(&delayed_refs->lock); | |
1928 | return 1; | |
1929 | } | |
d7baffda | 1930 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1931 | spin_unlock(&head->lock); |
1e7a1421 | 1932 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1933 | |
c1103f7a | 1934 | if (head->must_insert_reserved) { |
b25c36f8 | 1935 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1936 | if (head->is_data) { |
fc28b25e JB |
1937 | struct btrfs_root *csum_root; |
1938 | ||
1939 | csum_root = btrfs_csum_root(fs_info, head->bytenr); | |
1940 | ret = btrfs_del_csums(trans, csum_root, head->bytenr, | |
1941 | head->num_bytes); | |
c1103f7a JB |
1942 | } |
1943 | } | |
1944 | ||
3ee56a58 | 1945 | *bytes_released += btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1946 | |
1947 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1948 | btrfs_delayed_ref_unlock(head); |
d278850e | 1949 | btrfs_put_delayed_ref_head(head); |
856bd270 | 1950 | return ret; |
194ab0bc JB |
1951 | } |
1952 | ||
b1cdbcb5 NB |
1953 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1954 | struct btrfs_trans_handle *trans) | |
1955 | { | |
1956 | struct btrfs_delayed_ref_root *delayed_refs = | |
1957 | &trans->transaction->delayed_refs; | |
1958 | struct btrfs_delayed_ref_head *head = NULL; | |
1959 | int ret; | |
1960 | ||
1961 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1962 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1963 | if (!head) { |
1964 | spin_unlock(&delayed_refs->lock); | |
1965 | return head; | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * Grab the lock that says we are going to process all the refs for | |
1970 | * this head | |
1971 | */ | |
9e920a6f | 1972 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1973 | spin_unlock(&delayed_refs->lock); |
1974 | ||
1975 | /* | |
1976 | * We may have dropped the spin lock to get the head mutex lock, and | |
1977 | * that might have given someone else time to free the head. If that's | |
1978 | * true, it has been removed from our list and we can move on. | |
1979 | */ | |
1980 | if (ret == -EAGAIN) | |
1981 | head = ERR_PTR(-EAGAIN); | |
1982 | ||
1983 | return head; | |
1984 | } | |
1985 | ||
e7261386 | 1986 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
3ee56a58 FM |
1987 | struct btrfs_delayed_ref_head *locked_ref, |
1988 | u64 *bytes_released) | |
e7261386 NB |
1989 | { |
1990 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1991 | struct btrfs_delayed_ref_root *delayed_refs; | |
1992 | struct btrfs_delayed_extent_op *extent_op; | |
1993 | struct btrfs_delayed_ref_node *ref; | |
61c681fe | 1994 | bool must_insert_reserved; |
e7261386 NB |
1995 | int ret; |
1996 | ||
1997 | delayed_refs = &trans->transaction->delayed_refs; | |
1998 | ||
0110a4c4 NB |
1999 | lockdep_assert_held(&locked_ref->mutex); |
2000 | lockdep_assert_held(&locked_ref->lock); | |
2001 | ||
e7261386 NB |
2002 | while ((ref = select_delayed_ref(locked_ref))) { |
2003 | if (ref->seq && | |
2004 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
2005 | spin_unlock(&locked_ref->lock); | |
2006 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2007 | return -EAGAIN; | |
2008 | } | |
2009 | ||
e7261386 NB |
2010 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); |
2011 | RB_CLEAR_NODE(&ref->ref_node); | |
2012 | if (!list_empty(&ref->add_list)) | |
2013 | list_del(&ref->add_list); | |
2014 | /* | |
2015 | * When we play the delayed ref, also correct the ref_mod on | |
2016 | * head | |
2017 | */ | |
2018 | switch (ref->action) { | |
2019 | case BTRFS_ADD_DELAYED_REF: | |
2020 | case BTRFS_ADD_DELAYED_EXTENT: | |
2021 | locked_ref->ref_mod -= ref->ref_mod; | |
2022 | break; | |
2023 | case BTRFS_DROP_DELAYED_REF: | |
2024 | locked_ref->ref_mod += ref->ref_mod; | |
2025 | break; | |
2026 | default: | |
2027 | WARN_ON(1); | |
2028 | } | |
2029 | atomic_dec(&delayed_refs->num_entries); | |
2030 | ||
2031 | /* | |
2032 | * Record the must_insert_reserved flag before we drop the | |
2033 | * spin lock. | |
2034 | */ | |
2035 | must_insert_reserved = locked_ref->must_insert_reserved; | |
61c681fe | 2036 | locked_ref->must_insert_reserved = false; |
e7261386 NB |
2037 | |
2038 | extent_op = locked_ref->extent_op; | |
2039 | locked_ref->extent_op = NULL; | |
2040 | spin_unlock(&locked_ref->lock); | |
2041 | ||
cecbb533 | 2042 | ret = run_one_delayed_ref(trans, locked_ref, ref, extent_op, |
e7261386 | 2043 | must_insert_reserved); |
adb86dbe | 2044 | btrfs_delayed_refs_rsv_release(fs_info, 1, 0); |
3ee56a58 | 2045 | *bytes_released += btrfs_calc_delayed_ref_bytes(fs_info, 1); |
cecbb533 | 2046 | |
e7261386 NB |
2047 | btrfs_free_delayed_extent_op(extent_op); |
2048 | if (ret) { | |
2049 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2050 | btrfs_put_delayed_ref(ref); | |
e7261386 NB |
2051 | return ret; |
2052 | } | |
2053 | ||
2054 | btrfs_put_delayed_ref(ref); | |
2055 | cond_resched(); | |
2056 | ||
2057 | spin_lock(&locked_ref->lock); | |
0c555c97 | 2058 | btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref); |
e7261386 NB |
2059 | } |
2060 | ||
2061 | return 0; | |
2062 | } | |
2063 | ||
79787eaa JM |
2064 | /* |
2065 | * Returns 0 on success or if called with an already aborted transaction. | |
2066 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2067 | */ | |
d7df2c79 | 2068 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
8a526c44 | 2069 | u64 min_bytes) |
56bec294 | 2070 | { |
0a1e458a | 2071 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2072 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 CM |
2073 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
2074 | int ret; | |
d7df2c79 | 2075 | unsigned long count = 0; |
8a526c44 FM |
2076 | unsigned long max_count = 0; |
2077 | u64 bytes_processed = 0; | |
56bec294 CM |
2078 | |
2079 | delayed_refs = &trans->transaction->delayed_refs; | |
8a526c44 FM |
2080 | if (min_bytes == 0) { |
2081 | max_count = delayed_refs->num_heads_ready; | |
2082 | min_bytes = U64_MAX; | |
2083 | } | |
2084 | ||
0110a4c4 | 2085 | do { |
56bec294 | 2086 | if (!locked_ref) { |
b1cdbcb5 | 2087 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2088 | if (IS_ERR_OR_NULL(locked_ref)) { |
2089 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2090 | continue; | |
2091 | } else { | |
2092 | break; | |
2093 | } | |
56bec294 | 2094 | } |
0110a4c4 | 2095 | count++; |
56bec294 | 2096 | } |
2c3cf7d5 FM |
2097 | /* |
2098 | * We need to try and merge add/drops of the same ref since we | |
2099 | * can run into issues with relocate dropping the implicit ref | |
2100 | * and then it being added back again before the drop can | |
2101 | * finish. If we merged anything we need to re-loop so we can | |
2102 | * get a good ref. | |
2103 | * Or we can get node references of the same type that weren't | |
2104 | * merged when created due to bumps in the tree mod seq, and | |
2105 | * we need to merge them to prevent adding an inline extent | |
2106 | * backref before dropping it (triggering a BUG_ON at | |
2107 | * insert_inline_extent_backref()). | |
2108 | */ | |
d7df2c79 | 2109 | spin_lock(&locked_ref->lock); |
0c555c97 | 2110 | btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref); |
ae1e206b | 2111 | |
3ee56a58 | 2112 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, &bytes_processed); |
0110a4c4 NB |
2113 | if (ret < 0 && ret != -EAGAIN) { |
2114 | /* | |
2115 | * Error, btrfs_run_delayed_refs_for_head already | |
2116 | * unlocked everything so just bail out | |
2117 | */ | |
2118 | return ret; | |
2119 | } else if (!ret) { | |
2120 | /* | |
2121 | * Success, perform the usual cleanup of a processed | |
2122 | * head | |
2123 | */ | |
3ee56a58 | 2124 | ret = cleanup_ref_head(trans, locked_ref, &bytes_processed); |
194ab0bc | 2125 | if (ret > 0 ) { |
b00e6250 JB |
2126 | /* We dropped our lock, we need to loop. */ |
2127 | ret = 0; | |
d7df2c79 | 2128 | continue; |
194ab0bc JB |
2129 | } else if (ret) { |
2130 | return ret; | |
5d4f98a2 | 2131 | } |
22cd2e7d | 2132 | } |
1ce7a5ec | 2133 | |
b00e6250 | 2134 | /* |
0110a4c4 NB |
2135 | * Either success case or btrfs_run_delayed_refs_for_head |
2136 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2137 | */ |
b00e6250 | 2138 | |
0110a4c4 | 2139 | locked_ref = NULL; |
c3e69d58 | 2140 | cond_resched(); |
8a526c44 FM |
2141 | } while ((min_bytes != U64_MAX && bytes_processed < min_bytes) || |
2142 | (max_count > 0 && count < max_count) || | |
2143 | locked_ref); | |
0a2b2a84 | 2144 | |
d7df2c79 | 2145 | return 0; |
c3e69d58 CM |
2146 | } |
2147 | ||
709c0486 AJ |
2148 | #ifdef SCRAMBLE_DELAYED_REFS |
2149 | /* | |
2150 | * Normally delayed refs get processed in ascending bytenr order. This | |
2151 | * correlates in most cases to the order added. To expose dependencies on this | |
2152 | * order, we start to process the tree in the middle instead of the beginning | |
2153 | */ | |
2154 | static u64 find_middle(struct rb_root *root) | |
2155 | { | |
2156 | struct rb_node *n = root->rb_node; | |
2157 | struct btrfs_delayed_ref_node *entry; | |
2158 | int alt = 1; | |
2159 | u64 middle; | |
2160 | u64 first = 0, last = 0; | |
2161 | ||
2162 | n = rb_first(root); | |
2163 | if (n) { | |
2164 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2165 | first = entry->bytenr; | |
2166 | } | |
2167 | n = rb_last(root); | |
2168 | if (n) { | |
2169 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2170 | last = entry->bytenr; | |
2171 | } | |
2172 | n = root->rb_node; | |
2173 | ||
2174 | while (n) { | |
2175 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2176 | WARN_ON(!entry->in_tree); | |
2177 | ||
2178 | middle = entry->bytenr; | |
2179 | ||
2180 | if (alt) | |
2181 | n = n->rb_left; | |
2182 | else | |
2183 | n = n->rb_right; | |
2184 | ||
2185 | alt = 1 - alt; | |
2186 | } | |
2187 | return middle; | |
2188 | } | |
2189 | #endif | |
2190 | ||
c3e69d58 | 2191 | /* |
8a526c44 FM |
2192 | * Start processing the delayed reference count updates and extent insertions |
2193 | * we have queued up so far. | |
2194 | * | |
2195 | * @trans: Transaction handle. | |
2196 | * @min_bytes: How many bytes of delayed references to process. After this | |
2197 | * many bytes we stop processing delayed references if there are | |
2198 | * any more. If 0 it means to run all existing delayed references, | |
2199 | * but not new ones added after running all existing ones. | |
2200 | * Use (u64)-1 (U64_MAX) to run all existing delayed references | |
2201 | * plus any new ones that are added. | |
79787eaa JM |
2202 | * |
2203 | * Returns 0 on success or if called with an aborted transaction | |
2204 | * Returns <0 on error and aborts the transaction | |
c3e69d58 | 2205 | */ |
8a526c44 | 2206 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, u64 min_bytes) |
c3e69d58 | 2207 | { |
c79a70b1 | 2208 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 | 2209 | struct btrfs_delayed_ref_root *delayed_refs; |
c3e69d58 | 2210 | int ret; |
c3e69d58 | 2211 | |
79787eaa | 2212 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2213 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2214 | return 0; |
2215 | ||
0b246afa | 2216 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2217 | return 0; |
2218 | ||
c3e69d58 | 2219 | delayed_refs = &trans->transaction->delayed_refs; |
c3e69d58 | 2220 | again: |
709c0486 AJ |
2221 | #ifdef SCRAMBLE_DELAYED_REFS |
2222 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2223 | #endif | |
8a526c44 | 2224 | ret = __btrfs_run_delayed_refs(trans, min_bytes); |
d7df2c79 | 2225 | if (ret < 0) { |
66642832 | 2226 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2227 | return ret; |
eb099670 | 2228 | } |
c3e69d58 | 2229 | |
8a526c44 | 2230 | if (min_bytes == U64_MAX) { |
119e80df | 2231 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2232 | |
d7df2c79 | 2233 | spin_lock(&delayed_refs->lock); |
3cbb9f51 | 2234 | if (RB_EMPTY_ROOT(&delayed_refs->href_root.rb_root)) { |
d7df2c79 | 2235 | spin_unlock(&delayed_refs->lock); |
3cbb9f51 | 2236 | return 0; |
d7df2c79 | 2237 | } |
d278850e | 2238 | spin_unlock(&delayed_refs->lock); |
e9d0b13b | 2239 | |
d7df2c79 | 2240 | cond_resched(); |
56bec294 | 2241 | goto again; |
5f39d397 | 2242 | } |
3cbb9f51 | 2243 | |
a28ec197 CM |
2244 | return 0; |
2245 | } | |
2246 | ||
5d4f98a2 | 2247 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
4aec05fa | 2248 | struct extent_buffer *eb, u64 flags) |
5d4f98a2 YZ |
2249 | { |
2250 | struct btrfs_delayed_extent_op *extent_op; | |
4aec05fa | 2251 | int level = btrfs_header_level(eb); |
5d4f98a2 YZ |
2252 | int ret; |
2253 | ||
78a6184a | 2254 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2255 | if (!extent_op) |
2256 | return -ENOMEM; | |
2257 | ||
2258 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2259 | extent_op->update_flags = true; |
2260 | extent_op->update_key = false; | |
b1c79e09 | 2261 | extent_op->level = level; |
5d4f98a2 | 2262 | |
42c9d0b5 | 2263 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2264 | if (ret) |
78a6184a | 2265 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2266 | return ret; |
2267 | } | |
2268 | ||
e4c3b2dc | 2269 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2270 | struct btrfs_path *path, |
2271 | u64 objectid, u64 offset, u64 bytenr) | |
2272 | { | |
2273 | struct btrfs_delayed_ref_head *head; | |
2274 | struct btrfs_delayed_ref_node *ref; | |
2275 | struct btrfs_delayed_data_ref *data_ref; | |
2276 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2277 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2278 | struct rb_node *node; |
5d4f98a2 YZ |
2279 | int ret = 0; |
2280 | ||
998ac6d2 | 2281 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2282 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2283 | if (cur_trans) |
2284 | refcount_inc(&cur_trans->use_count); | |
2285 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2286 | if (!cur_trans) |
2287 | return 0; | |
2288 | ||
2289 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2290 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2291 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2292 | if (!head) { |
2293 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2294 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2295 | return 0; |
2296 | } | |
5d4f98a2 YZ |
2297 | |
2298 | if (!mutex_trylock(&head->mutex)) { | |
26ce9114 JB |
2299 | if (path->nowait) { |
2300 | spin_unlock(&delayed_refs->lock); | |
2301 | btrfs_put_transaction(cur_trans); | |
2302 | return -EAGAIN; | |
2303 | } | |
2304 | ||
d278850e | 2305 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2306 | spin_unlock(&delayed_refs->lock); |
2307 | ||
b3b4aa74 | 2308 | btrfs_release_path(path); |
5d4f98a2 | 2309 | |
8cc33e5c DS |
2310 | /* |
2311 | * Mutex was contended, block until it's released and let | |
2312 | * caller try again | |
2313 | */ | |
5d4f98a2 YZ |
2314 | mutex_lock(&head->mutex); |
2315 | mutex_unlock(&head->mutex); | |
d278850e | 2316 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2317 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2318 | return -EAGAIN; |
2319 | } | |
d7df2c79 | 2320 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2321 | |
d7df2c79 | 2322 | spin_lock(&head->lock); |
0e0adbcf JB |
2323 | /* |
2324 | * XXX: We should replace this with a proper search function in the | |
2325 | * future. | |
2326 | */ | |
e3d03965 LB |
2327 | for (node = rb_first_cached(&head->ref_tree); node; |
2328 | node = rb_next(node)) { | |
0e0adbcf | 2329 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2330 | /* If it's a shared ref we know a cross reference exists */ |
2331 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2332 | ret = 1; | |
2333 | break; | |
2334 | } | |
5d4f98a2 | 2335 | |
d7df2c79 | 2336 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2337 | |
d7df2c79 JB |
2338 | /* |
2339 | * If our ref doesn't match the one we're currently looking at | |
2340 | * then we have a cross reference. | |
2341 | */ | |
2342 | if (data_ref->root != root->root_key.objectid || | |
2343 | data_ref->objectid != objectid || | |
2344 | data_ref->offset != offset) { | |
2345 | ret = 1; | |
2346 | break; | |
2347 | } | |
5d4f98a2 | 2348 | } |
d7df2c79 | 2349 | spin_unlock(&head->lock); |
5d4f98a2 | 2350 | mutex_unlock(&head->mutex); |
998ac6d2 | 2351 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2352 | return ret; |
2353 | } | |
2354 | ||
e4c3b2dc | 2355 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2356 | struct btrfs_path *path, |
a84d5d42 BB |
2357 | u64 objectid, u64 offset, u64 bytenr, |
2358 | bool strict) | |
be20aa9d | 2359 | { |
0b246afa | 2360 | struct btrfs_fs_info *fs_info = root->fs_info; |
29cbcf40 | 2361 | struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr); |
f321e491 | 2362 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2363 | struct btrfs_extent_data_ref *ref; |
2364 | struct btrfs_extent_inline_ref *iref; | |
2365 | struct btrfs_extent_item *ei; | |
f321e491 | 2366 | struct btrfs_key key; |
5d4f98a2 | 2367 | u32 item_size; |
d9a620f7 | 2368 | u32 expected_size; |
3de28d57 | 2369 | int type; |
be20aa9d | 2370 | int ret; |
925baedd | 2371 | |
be20aa9d | 2372 | key.objectid = bytenr; |
31840ae1 | 2373 | key.offset = (u64)-1; |
f321e491 | 2374 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2375 | |
be20aa9d CM |
2376 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2377 | if (ret < 0) | |
2378 | goto out; | |
79787eaa | 2379 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2380 | |
2381 | ret = -ENOENT; | |
2382 | if (path->slots[0] == 0) | |
31840ae1 | 2383 | goto out; |
be20aa9d | 2384 | |
31840ae1 | 2385 | path->slots[0]--; |
f321e491 | 2386 | leaf = path->nodes[0]; |
5d4f98a2 | 2387 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2388 | |
5d4f98a2 | 2389 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2390 | goto out; |
f321e491 | 2391 | |
5d4f98a2 | 2392 | ret = 1; |
3212fa14 | 2393 | item_size = btrfs_item_size(leaf, path->slots[0]); |
5d4f98a2 | 2394 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
d9a620f7 BB |
2395 | expected_size = sizeof(*ei) + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY); |
2396 | ||
2397 | /* No inline refs; we need to bail before checking for owner ref. */ | |
2398 | if (item_size == sizeof(*ei)) | |
2399 | goto out; | |
2400 | ||
2401 | /* Check for an owner ref; skip over it to the real inline refs. */ | |
2402 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2403 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
2404 | if (btrfs_fs_incompat(fs_info, SIMPLE_QUOTA) && type == BTRFS_EXTENT_OWNER_REF_KEY) { | |
2405 | expected_size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY); | |
2406 | iref = (struct btrfs_extent_inline_ref *)(iref + 1); | |
2407 | } | |
bd09835d | 2408 | |
a6bd9cd1 | 2409 | /* If extent item has more than 1 inline ref then it's shared */ |
d9a620f7 | 2410 | if (item_size != expected_size) |
5d4f98a2 | 2411 | goto out; |
be20aa9d | 2412 | |
a84d5d42 BB |
2413 | /* |
2414 | * If extent created before last snapshot => it's shared unless the | |
2415 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2416 | */ | |
2417 | if (!strict && | |
2418 | (btrfs_extent_generation(leaf, ei) <= | |
2419 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2420 | goto out; |
2421 | ||
a6bd9cd1 | 2422 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2423 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2424 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2425 | goto out; |
2426 | ||
2427 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2428 | if (btrfs_extent_refs(leaf, ei) != | |
2429 | btrfs_extent_data_ref_count(leaf, ref) || | |
2430 | btrfs_extent_data_ref_root(leaf, ref) != | |
2431 | root->root_key.objectid || | |
2432 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2433 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2434 | goto out; | |
2435 | ||
2436 | ret = 0; | |
2437 | out: | |
2438 | return ret; | |
2439 | } | |
2440 | ||
e4c3b2dc | 2441 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
1a89f173 | 2442 | u64 bytenr, bool strict, struct btrfs_path *path) |
5d4f98a2 | 2443 | { |
5d4f98a2 | 2444 | int ret; |
5d4f98a2 | 2445 | |
5d4f98a2 | 2446 | do { |
e4c3b2dc | 2447 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2448 | offset, bytenr, strict); |
5d4f98a2 | 2449 | if (ret && ret != -ENOENT) |
f321e491 | 2450 | goto out; |
80ff3856 | 2451 | |
380fd066 MT |
2452 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2453 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2454 | |
be20aa9d | 2455 | out: |
1a89f173 | 2456 | btrfs_release_path(path); |
37f00a6d | 2457 | if (btrfs_is_data_reloc_root(root)) |
f0486c68 | 2458 | WARN_ON(ret > 0); |
f321e491 | 2459 | return ret; |
be20aa9d | 2460 | } |
c5739bba | 2461 | |
5d4f98a2 | 2462 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2463 | struct btrfs_root *root, |
5d4f98a2 | 2464 | struct extent_buffer *buf, |
e339a6b0 | 2465 | int full_backref, int inc) |
31840ae1 | 2466 | { |
0b246afa | 2467 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2468 | u64 bytenr; |
5d4f98a2 YZ |
2469 | u64 num_bytes; |
2470 | u64 parent; | |
31840ae1 | 2471 | u64 ref_root; |
31840ae1 | 2472 | u32 nritems; |
31840ae1 ZY |
2473 | struct btrfs_key key; |
2474 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2475 | struct btrfs_ref generic_ref = { 0 }; |
2476 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2477 | int i; |
82fa113f | 2478 | int action; |
31840ae1 ZY |
2479 | int level; |
2480 | int ret = 0; | |
fccb84c9 | 2481 | |
0b246afa | 2482 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2483 | return 0; |
fccb84c9 | 2484 | |
31840ae1 | 2485 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2486 | nritems = btrfs_header_nritems(buf); |
2487 | level = btrfs_header_level(buf); | |
2488 | ||
92a7cc42 | 2489 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2490 | return 0; |
31840ae1 | 2491 | |
5d4f98a2 YZ |
2492 | if (full_backref) |
2493 | parent = buf->start; | |
2494 | else | |
2495 | parent = 0; | |
82fa113f QW |
2496 | if (inc) |
2497 | action = BTRFS_ADD_DELAYED_REF; | |
2498 | else | |
2499 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2500 | |
2501 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2502 | if (level == 0) { |
5d4f98a2 | 2503 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2504 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2505 | continue; |
5d4f98a2 | 2506 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2507 | struct btrfs_file_extent_item); |
2508 | if (btrfs_file_extent_type(buf, fi) == | |
2509 | BTRFS_FILE_EXTENT_INLINE) | |
2510 | continue; | |
2511 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2512 | if (bytenr == 0) | |
2513 | continue; | |
5d4f98a2 YZ |
2514 | |
2515 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2516 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f | 2517 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
457cb1dd | 2518 | num_bytes, parent, ref_root); |
82fa113f | 2519 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, |
f42c5da6 NB |
2520 | key.offset, root->root_key.objectid, |
2521 | for_reloc); | |
dd28b6a5 | 2522 | if (inc) |
82fa113f | 2523 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2524 | else |
ffd4bb2a | 2525 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2526 | if (ret) |
2527 | goto fail; | |
2528 | } else { | |
5d4f98a2 | 2529 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2530 | num_bytes = fs_info->nodesize; |
457cb1dd | 2531 | /* We don't know the owning_root, use 0. */ |
82fa113f | 2532 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
457cb1dd | 2533 | num_bytes, parent, 0); |
f42c5da6 NB |
2534 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root, |
2535 | root->root_key.objectid, for_reloc); | |
dd28b6a5 | 2536 | if (inc) |
82fa113f | 2537 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2538 | else |
ffd4bb2a | 2539 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2540 | if (ret) |
2541 | goto fail; | |
2542 | } | |
2543 | } | |
2544 | return 0; | |
2545 | fail: | |
5d4f98a2 YZ |
2546 | return ret; |
2547 | } | |
2548 | ||
2549 | int btrfs_inc_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, 1); |
5d4f98a2 YZ |
2553 | } |
2554 | ||
2555 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2556 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2557 | { |
e339a6b0 | 2558 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2559 | } |
2560 | ||
1b86826d | 2561 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2562 | { |
0b246afa | 2563 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2564 | u64 flags; |
53b381b3 | 2565 | u64 ret; |
9ed74f2d | 2566 | |
b742bb82 YZ |
2567 | if (data) |
2568 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2569 | else if (root == fs_info->chunk_root) |
b742bb82 | 2570 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2571 | else |
b742bb82 | 2572 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2573 | |
878d7b67 | 2574 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2575 | return ret; |
6a63209f | 2576 | } |
9ed74f2d | 2577 | |
0eb997bf | 2578 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info) |
a061fc8d | 2579 | { |
08dddb29 FM |
2580 | struct rb_node *leftmost; |
2581 | u64 bytenr = 0; | |
a1897fdd | 2582 | |
16b0c258 | 2583 | read_lock(&fs_info->block_group_cache_lock); |
0eb997bf | 2584 | /* Get the block group with the lowest logical start address. */ |
08dddb29 FM |
2585 | leftmost = rb_first_cached(&fs_info->block_group_cache_tree); |
2586 | if (leftmost) { | |
2587 | struct btrfs_block_group *bg; | |
0f9dd46c | 2588 | |
08dddb29 FM |
2589 | bg = rb_entry(leftmost, struct btrfs_block_group, cache_node); |
2590 | bytenr = bg->start; | |
2591 | } | |
16b0c258 | 2592 | read_unlock(&fs_info->block_group_cache_lock); |
d2fb3437 YZ |
2593 | |
2594 | return bytenr; | |
a061fc8d CM |
2595 | } |
2596 | ||
6690d071 NB |
2597 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2598 | struct btrfs_block_group *cache, | |
f0486c68 | 2599 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2600 | { |
fdf08605 DS |
2601 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2602 | ||
11833d66 YZ |
2603 | spin_lock(&cache->space_info->lock); |
2604 | spin_lock(&cache->lock); | |
2605 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2606 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2607 | num_bytes); | |
11833d66 YZ |
2608 | if (reserved) { |
2609 | cache->reserved -= num_bytes; | |
2610 | cache->space_info->bytes_reserved -= num_bytes; | |
2611 | } | |
2612 | spin_unlock(&cache->lock); | |
2613 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2614 | |
fe1a598c | 2615 | set_extent_bit(&trans->transaction->pinned_extents, bytenr, |
1d126800 | 2616 | bytenr + num_bytes - 1, EXTENT_DIRTY, NULL); |
f0486c68 YZ |
2617 | return 0; |
2618 | } | |
68b38550 | 2619 | |
b25c36f8 | 2620 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2621 | u64 bytenr, u64 num_bytes, int reserved) |
2622 | { | |
32da5386 | 2623 | struct btrfs_block_group *cache; |
68b38550 | 2624 | |
b25c36f8 | 2625 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2626 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2627 | |
6690d071 | 2628 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2629 | |
2630 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2631 | return 0; |
2632 | } | |
2633 | ||
9fce5704 | 2634 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
007dec8c | 2635 | const struct extent_buffer *eb) |
e688b725 | 2636 | { |
32da5386 | 2637 | struct btrfs_block_group *cache; |
b50c6e25 | 2638 | int ret; |
e688b725 | 2639 | |
007dec8c | 2640 | cache = btrfs_lookup_block_group(trans->fs_info, eb->start); |
b50c6e25 JB |
2641 | if (!cache) |
2642 | return -EINVAL; | |
e688b725 CM |
2643 | |
2644 | /* | |
ced8ecf0 OS |
2645 | * Fully cache the free space first so that our pin removes the free space |
2646 | * from the cache. | |
e688b725 | 2647 | */ |
ced8ecf0 | 2648 | ret = btrfs_cache_block_group(cache, true); |
9ad6d91f FM |
2649 | if (ret) |
2650 | goto out; | |
e688b725 | 2651 | |
007dec8c | 2652 | pin_down_extent(trans, cache, eb->start, eb->len, 0); |
e688b725 CM |
2653 | |
2654 | /* remove us from the free space cache (if we're there at all) */ | |
007dec8c | 2655 | ret = btrfs_remove_free_space(cache, eb->start, eb->len); |
9ad6d91f | 2656 | out: |
e688b725 | 2657 | btrfs_put_block_group(cache); |
b50c6e25 | 2658 | return ret; |
e688b725 CM |
2659 | } |
2660 | ||
2ff7e61e JM |
2661 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2662 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2663 | { |
2664 | int ret; | |
32da5386 | 2665 | struct btrfs_block_group *block_group; |
8c2a1a30 | 2666 | |
0b246afa | 2667 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2668 | if (!block_group) |
2669 | return -EINVAL; | |
2670 | ||
ced8ecf0 | 2671 | ret = btrfs_cache_block_group(block_group, true); |
9ad6d91f FM |
2672 | if (ret) |
2673 | goto out; | |
8c2a1a30 | 2674 | |
9ad6d91f FM |
2675 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2676 | out: | |
8c2a1a30 JB |
2677 | btrfs_put_block_group(block_group); |
2678 | return ret; | |
2679 | } | |
2680 | ||
bcdc428c | 2681 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2682 | { |
bcdc428c | 2683 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2684 | struct btrfs_file_extent_item *item; |
2685 | struct btrfs_key key; | |
2686 | int found_type; | |
2687 | int i; | |
b89311ef | 2688 | int ret = 0; |
8c2a1a30 | 2689 | |
2ff7e61e | 2690 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2691 | return 0; |
2692 | ||
2693 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2694 | btrfs_item_key_to_cpu(eb, &key, i); | |
2695 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2696 | continue; | |
2697 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2698 | found_type = btrfs_file_extent_type(eb, item); | |
2699 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2700 | continue; | |
2701 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2702 | continue; | |
2703 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2704 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2705 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2706 | if (ret) | |
2707 | break; | |
8c2a1a30 JB |
2708 | } |
2709 | ||
b89311ef | 2710 | return ret; |
8c2a1a30 JB |
2711 | } |
2712 | ||
9cfa3e34 | 2713 | static void |
32da5386 | 2714 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2715 | { |
2716 | atomic_inc(&bg->reservations); | |
2717 | } | |
2718 | ||
c759c4e1 JB |
2719 | /* |
2720 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2721 | * what it should be based on the mount options. | |
2722 | */ | |
2723 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2724 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2725 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2726 | { |
2727 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2728 | |
2729 | *empty_cluster = 0; | |
2730 | if (btrfs_mixed_space_info(space_info)) | |
2731 | return ret; | |
2732 | ||
c759c4e1 | 2733 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2734 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2735 | if (btrfs_test_opt(fs_info, SSD)) |
2736 | *empty_cluster = SZ_2M; | |
2737 | else | |
ee22184b | 2738 | *empty_cluster = SZ_64K; |
583b7231 HK |
2739 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2740 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2741 | *empty_cluster = SZ_2M; | |
0b246afa | 2742 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2743 | } |
2744 | ||
2745 | return ret; | |
2746 | } | |
2747 | ||
2ff7e61e JM |
2748 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2749 | u64 start, u64 end, | |
678886bd | 2750 | const bool return_free_space) |
ccd467d6 | 2751 | { |
32da5386 | 2752 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2753 | struct btrfs_space_info *space_info; |
2754 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2755 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2756 | u64 len; |
c759c4e1 JB |
2757 | u64 total_unpinned = 0; |
2758 | u64 empty_cluster = 0; | |
7b398f8e | 2759 | bool readonly; |
ccd467d6 | 2760 | |
11833d66 | 2761 | while (start <= end) { |
7b398f8e | 2762 | readonly = false; |
11833d66 | 2763 | if (!cache || |
b3470b5d | 2764 | start >= cache->start + cache->length) { |
11833d66 YZ |
2765 | if (cache) |
2766 | btrfs_put_block_group(cache); | |
c759c4e1 | 2767 | total_unpinned = 0; |
11833d66 | 2768 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2769 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2770 | |
2ff7e61e | 2771 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2772 | cache->space_info, |
2773 | &empty_cluster); | |
2774 | empty_cluster <<= 1; | |
11833d66 YZ |
2775 | } |
2776 | ||
b3470b5d | 2777 | len = cache->start + cache->length - start; |
11833d66 YZ |
2778 | len = min(len, end + 1 - start); |
2779 | ||
48ff7083 OS |
2780 | if (return_free_space) |
2781 | btrfs_add_free_space(cache, start, len); | |
11833d66 | 2782 | |
f0486c68 | 2783 | start += len; |
c759c4e1 | 2784 | total_unpinned += len; |
7b398f8e | 2785 | space_info = cache->space_info; |
f0486c68 | 2786 | |
c759c4e1 JB |
2787 | /* |
2788 | * If this space cluster has been marked as fragmented and we've | |
2789 | * unpinned enough in this block group to potentially allow a | |
2790 | * cluster to be created inside of it go ahead and clear the | |
2791 | * fragmented check. | |
2792 | */ | |
2793 | if (cluster && cluster->fragmented && | |
2794 | total_unpinned > empty_cluster) { | |
2795 | spin_lock(&cluster->lock); | |
2796 | cluster->fragmented = 0; | |
2797 | spin_unlock(&cluster->lock); | |
2798 | } | |
2799 | ||
7b398f8e | 2800 | spin_lock(&space_info->lock); |
11833d66 YZ |
2801 | spin_lock(&cache->lock); |
2802 | cache->pinned -= len; | |
bb96c4e5 | 2803 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2804 | space_info->max_extent_size = 0; |
7b398f8e JB |
2805 | if (cache->ro) { |
2806 | space_info->bytes_readonly += len; | |
2807 | readonly = true; | |
169e0da9 NA |
2808 | } else if (btrfs_is_zoned(fs_info)) { |
2809 | /* Need reset before reusing in a zoned block group */ | |
2810 | space_info->bytes_zone_unusable += len; | |
2811 | readonly = true; | |
7b398f8e | 2812 | } |
11833d66 | 2813 | spin_unlock(&cache->lock); |
957780eb JB |
2814 | if (!readonly && return_free_space && |
2815 | global_rsv->space_info == space_info) { | |
7b398f8e JB |
2816 | spin_lock(&global_rsv->lock); |
2817 | if (!global_rsv->full) { | |
c4bf1909 JC |
2818 | u64 to_add = min(len, global_rsv->size - |
2819 | global_rsv->reserved); | |
2820 | ||
957780eb | 2821 | global_rsv->reserved += to_add; |
bb96c4e5 JB |
2822 | btrfs_space_info_update_bytes_may_use(fs_info, |
2823 | space_info, to_add); | |
7b398f8e JB |
2824 | if (global_rsv->reserved >= global_rsv->size) |
2825 | global_rsv->full = 1; | |
957780eb | 2826 | len -= to_add; |
7b398f8e JB |
2827 | } |
2828 | spin_unlock(&global_rsv->lock); | |
2829 | } | |
2732798c JB |
2830 | /* Add to any tickets we may have */ |
2831 | if (!readonly && return_free_space && len) | |
2832 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2833 | spin_unlock(&space_info->lock); |
ccd467d6 | 2834 | } |
11833d66 YZ |
2835 | |
2836 | if (cache) | |
2837 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2838 | return 0; |
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); |
2ff7e61e | 2868 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2869 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2870 | free_extent_state(cached_state); |
b9473439 | 2871 | cond_resched(); |
a28ec197 | 2872 | } |
817d52f8 | 2873 | |
a2309300 DZ |
2874 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2875 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2876 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2877 | } |
b0643e59 | 2878 | |
e33e17ee JM |
2879 | /* |
2880 | * Transaction is finished. We don't need the lock anymore. We | |
2881 | * do need to clean up the block groups in case of a transaction | |
2882 | * abort. | |
2883 | */ | |
2884 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2885 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2886 | u64 trimmed = 0; | |
2887 | ||
2888 | ret = -EROFS; | |
bf31f87f | 2889 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2890 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2891 | block_group->start, |
2892 | block_group->length, | |
e33e17ee JM |
2893 | &trimmed); |
2894 | ||
2895 | list_del_init(&block_group->bg_list); | |
6b7304af | 2896 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2897 | btrfs_put_block_group(block_group); |
2898 | ||
2899 | if (ret) { | |
2900 | const char *errstr = btrfs_decode_error(ret); | |
2901 | btrfs_warn(fs_info, | |
913e1535 | 2902 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2903 | ret, errstr); |
2904 | } | |
2905 | } | |
2906 | ||
e20d96d6 CM |
2907 | return 0; |
2908 | } | |
2909 | ||
8d299091 BB |
2910 | /* |
2911 | * Parse an extent item's inline extents looking for a simple quotas owner ref. | |
2912 | * | |
2913 | * @fs_info: the btrfs_fs_info for this mount | |
2914 | * @leaf: a leaf in the extent tree containing the extent item | |
2915 | * @slot: the slot in the leaf where the extent item is found | |
2916 | * | |
2917 | * Returns the objectid of the root that originally allocated the extent item | |
2918 | * if the inline owner ref is expected and present, otherwise 0. | |
2919 | * | |
2920 | * If an extent item has an owner ref item, it will be the first inline ref | |
2921 | * item. Therefore the logic is to check whether there are any inline ref | |
2922 | * items, then check the type of the first one. | |
2923 | */ | |
2924 | u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info, | |
2925 | struct extent_buffer *leaf, int slot) | |
2926 | { | |
2927 | struct btrfs_extent_item *ei; | |
2928 | struct btrfs_extent_inline_ref *iref; | |
2929 | struct btrfs_extent_owner_ref *oref; | |
2930 | unsigned long ptr; | |
2931 | unsigned long end; | |
2932 | int type; | |
2933 | ||
2934 | if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA)) | |
2935 | return 0; | |
2936 | ||
2937 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); | |
2938 | ptr = (unsigned long)(ei + 1); | |
2939 | end = (unsigned long)ei + btrfs_item_size(leaf, slot); | |
2940 | ||
2941 | /* No inline ref items of any kind, can't check type. */ | |
2942 | if (ptr == end) | |
2943 | return 0; | |
2944 | ||
2945 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
2946 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
2947 | ||
2948 | /* We found an owner ref, get the root out of it. */ | |
2949 | if (type == BTRFS_EXTENT_OWNER_REF_KEY) { | |
2950 | oref = (struct btrfs_extent_owner_ref *)(&iref->offset); | |
2951 | return btrfs_extent_owner_ref_root_id(leaf, oref); | |
2952 | } | |
2953 | ||
2954 | /* We have inline refs, but not an owner ref. */ | |
2955 | return 0; | |
2956 | } | |
2957 | ||
8f8aa4c7 | 2958 | static int do_free_extent_accounting(struct btrfs_trans_handle *trans, |
cecbb533 | 2959 | u64 bytenr, struct btrfs_squota_delta *delta) |
8f8aa4c7 JB |
2960 | { |
2961 | int ret; | |
cecbb533 | 2962 | u64 num_bytes = delta->num_bytes; |
8f8aa4c7 | 2963 | |
cecbb533 | 2964 | if (delta->is_data) { |
8f8aa4c7 JB |
2965 | struct btrfs_root *csum_root; |
2966 | ||
2967 | csum_root = btrfs_csum_root(trans->fs_info, bytenr); | |
2968 | ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes); | |
2969 | if (ret) { | |
2970 | btrfs_abort_transaction(trans, ret); | |
2971 | return ret; | |
2972 | } | |
ca41504e JT |
2973 | |
2974 | ret = btrfs_delete_raid_extent(trans, bytenr, num_bytes); | |
2975 | if (ret) { | |
2976 | btrfs_abort_transaction(trans, ret); | |
2977 | return ret; | |
2978 | } | |
8f8aa4c7 JB |
2979 | } |
2980 | ||
cecbb533 BB |
2981 | ret = btrfs_record_squota_delta(trans->fs_info, delta); |
2982 | if (ret) { | |
2983 | btrfs_abort_transaction(trans, ret); | |
2984 | return ret; | |
2985 | } | |
2986 | ||
8f8aa4c7 JB |
2987 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
2988 | if (ret) { | |
2989 | btrfs_abort_transaction(trans, ret); | |
2990 | return ret; | |
2991 | } | |
2992 | ||
2993 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, false); | |
2994 | if (ret) | |
2995 | btrfs_abort_transaction(trans, ret); | |
2996 | ||
2997 | return ret; | |
2998 | } | |
2999 | ||
eee3b811 QW |
3000 | #define abort_and_dump(trans, path, fmt, args...) \ |
3001 | ({ \ | |
3002 | btrfs_abort_transaction(trans, -EUCLEAN); \ | |
3003 | btrfs_print_leaf(path->nodes[0]); \ | |
3004 | btrfs_crit(trans->fs_info, fmt, ##args); \ | |
3005 | }) | |
3006 | ||
1c2a07f5 QW |
3007 | /* |
3008 | * Drop one or more refs of @node. | |
3009 | * | |
3010 | * 1. Locate the extent refs. | |
3011 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
3012 | * Locate it, then reduce the refs number or remove the ref line completely. | |
3013 | * | |
3014 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
3015 | * | |
3016 | * Inline backref case: | |
3017 | * | |
3018 | * in extent tree we have: | |
3019 | * | |
3020 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
3021 | * refs 2 gen 6 flags DATA | |
3022 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
3023 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
3024 | * | |
3025 | * This function gets called with: | |
3026 | * | |
3027 | * node->bytenr = 13631488 | |
3028 | * node->num_bytes = 1048576 | |
3029 | * root_objectid = FS_TREE | |
3030 | * owner_objectid = 257 | |
3031 | * owner_offset = 0 | |
3032 | * refs_to_drop = 1 | |
3033 | * | |
3034 | * Then we should get some like: | |
3035 | * | |
3036 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
3037 | * refs 1 gen 6 flags DATA | |
3038 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
3039 | * | |
3040 | * Keyed backref case: | |
3041 | * | |
3042 | * in extent tree we have: | |
3043 | * | |
3044 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
3045 | * refs 754 gen 6 flags DATA | |
3046 | * [...] | |
3047 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
3048 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
3049 | * | |
3050 | * This function get called with: | |
3051 | * | |
3052 | * node->bytenr = 13631488 | |
3053 | * node->num_bytes = 1048576 | |
3054 | * root_objectid = FS_TREE | |
3055 | * owner_objectid = 866 | |
3056 | * owner_offset = 0 | |
3057 | * refs_to_drop = 1 | |
3058 | * | |
3059 | * Then we should get some like: | |
3060 | * | |
3061 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
3062 | * refs 753 gen 6 flags DATA | |
3063 | * | |
3064 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
3065 | */ | |
5d4f98a2 | 3066 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
cecbb533 | 3067 | struct btrfs_delayed_ref_head *href, |
e72cb923 NB |
3068 | struct btrfs_delayed_ref_node *node, u64 parent, |
3069 | u64 root_objectid, u64 owner_objectid, | |
1df6b3c0 | 3070 | u64 owner_offset, |
e72cb923 | 3071 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 3072 | { |
e72cb923 | 3073 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 3074 | struct btrfs_key key; |
5d4f98a2 | 3075 | struct btrfs_path *path; |
29cbcf40 | 3076 | struct btrfs_root *extent_root; |
5f39d397 | 3077 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3078 | struct btrfs_extent_item *ei; |
3079 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 3080 | int ret; |
5d4f98a2 | 3081 | int is_data; |
952fccac CM |
3082 | int extent_slot = 0; |
3083 | int found_extent = 0; | |
3084 | int num_to_del = 1; | |
1df6b3c0 | 3085 | int refs_to_drop = node->ref_mod; |
5d4f98a2 YZ |
3086 | u32 item_size; |
3087 | u64 refs; | |
c682f9b3 QW |
3088 | u64 bytenr = node->bytenr; |
3089 | u64 num_bytes = node->num_bytes; | |
0b246afa | 3090 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
cecbb533 | 3091 | u64 delayed_ref_root = href->owning_root; |
037e6390 | 3092 | |
29cbcf40 | 3093 | extent_root = btrfs_extent_root(info, bytenr); |
abed4aaa | 3094 | ASSERT(extent_root); |
29cbcf40 | 3095 | |
5caf2a00 | 3096 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3097 | if (!path) |
3098 | return -ENOMEM; | |
5f26f772 | 3099 | |
5d4f98a2 | 3100 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
1c2a07f5 QW |
3101 | |
3102 | if (!is_data && refs_to_drop != 1) { | |
3103 | btrfs_crit(info, | |
3104 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
3105 | node->bytenr, refs_to_drop); | |
3106 | ret = -EINVAL; | |
3107 | btrfs_abort_transaction(trans, ret); | |
3108 | goto out; | |
3109 | } | |
5d4f98a2 | 3110 | |
3173a18f | 3111 | if (is_data) |
897ca819 | 3112 | skinny_metadata = false; |
3173a18f | 3113 | |
fbe4801b NB |
3114 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
3115 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 3116 | owner_offset); |
7bb86316 | 3117 | if (ret == 0) { |
1c2a07f5 QW |
3118 | /* |
3119 | * Either the inline backref or the SHARED_DATA_REF/ | |
3120 | * SHARED_BLOCK_REF is found | |
3121 | * | |
3122 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
3123 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
3124 | */ | |
952fccac | 3125 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3126 | while (extent_slot >= 0) { |
3127 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3128 | extent_slot); |
5d4f98a2 | 3129 | if (key.objectid != bytenr) |
952fccac | 3130 | break; |
5d4f98a2 YZ |
3131 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3132 | key.offset == num_bytes) { | |
952fccac CM |
3133 | found_extent = 1; |
3134 | break; | |
3135 | } | |
3173a18f JB |
3136 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3137 | key.offset == owner_objectid) { | |
3138 | found_extent = 1; | |
3139 | break; | |
3140 | } | |
1c2a07f5 QW |
3141 | |
3142 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
3143 | if (path->slots[0] - extent_slot > 5) |
3144 | break; | |
5d4f98a2 | 3145 | extent_slot--; |
952fccac | 3146 | } |
a79865c6 | 3147 | |
31840ae1 | 3148 | if (!found_extent) { |
1c2a07f5 | 3149 | if (iref) { |
eee3b811 QW |
3150 | abort_and_dump(trans, path, |
3151 | "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref", | |
3152 | path->slots[0]); | |
3153 | ret = -EUCLEAN; | |
3154 | goto out; | |
1c2a07f5 QW |
3155 | } |
3156 | /* Must be SHARED_* item, remove the backref first */ | |
76d76e78 | 3157 | ret = remove_extent_backref(trans, extent_root, path, |
5b2a54bb | 3158 | NULL, refs_to_drop, is_data); |
005d6427 | 3159 | if (ret) { |
66642832 | 3160 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3161 | goto out; |
3162 | } | |
b3b4aa74 | 3163 | btrfs_release_path(path); |
5d4f98a2 | 3164 | |
1c2a07f5 | 3165 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3166 | key.objectid = bytenr; |
3167 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3168 | key.offset = num_bytes; | |
3169 | ||
3173a18f JB |
3170 | if (!is_data && skinny_metadata) { |
3171 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3172 | key.offset = owner_objectid; | |
3173 | } | |
3174 | ||
31840ae1 ZY |
3175 | ret = btrfs_search_slot(trans, extent_root, |
3176 | &key, path, -1, 1); | |
3173a18f JB |
3177 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3178 | /* | |
3179 | * Couldn't find our skinny metadata item, | |
3180 | * see if we have ye olde extent item. | |
3181 | */ | |
3182 | path->slots[0]--; | |
3183 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3184 | path->slots[0]); | |
3185 | if (key.objectid == bytenr && | |
3186 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3187 | key.offset == num_bytes) | |
3188 | ret = 0; | |
3189 | } | |
3190 | ||
3191 | if (ret > 0 && skinny_metadata) { | |
3192 | skinny_metadata = false; | |
9ce49a0b | 3193 | key.objectid = bytenr; |
3173a18f JB |
3194 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3195 | key.offset = num_bytes; | |
3196 | btrfs_release_path(path); | |
3197 | ret = btrfs_search_slot(trans, extent_root, | |
3198 | &key, path, -1, 1); | |
3199 | } | |
3200 | ||
f3465ca4 | 3201 | if (ret) { |
b783e62d | 3202 | if (ret > 0) |
a4f78750 | 3203 | btrfs_print_leaf(path->nodes[0]); |
eee3b811 QW |
3204 | btrfs_err(info, |
3205 | "umm, got %d back from search, was looking for %llu, slot %d", | |
3206 | ret, bytenr, path->slots[0]); | |
f3465ca4 | 3207 | } |
005d6427 | 3208 | if (ret < 0) { |
66642832 | 3209 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3210 | goto out; |
3211 | } | |
31840ae1 ZY |
3212 | extent_slot = path->slots[0]; |
3213 | } | |
fae7f21c | 3214 | } else if (WARN_ON(ret == -ENOENT)) { |
eee3b811 QW |
3215 | abort_and_dump(trans, path, |
3216 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d", | |
3217 | bytenr, parent, root_objectid, owner_objectid, | |
3218 | owner_offset, path->slots[0]); | |
c4a050bb | 3219 | goto out; |
79787eaa | 3220 | } else { |
66642832 | 3221 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3222 | goto out; |
7bb86316 | 3223 | } |
5f39d397 CM |
3224 | |
3225 | leaf = path->nodes[0]; | |
3212fa14 | 3226 | item_size = btrfs_item_size(leaf, extent_slot); |
6d8ff4e4 | 3227 | if (unlikely(item_size < sizeof(*ei))) { |
182741d2 QW |
3228 | ret = -EUCLEAN; |
3229 | btrfs_err(trans->fs_info, | |
3230 | "unexpected extent item size, has %u expect >= %zu", | |
3231 | item_size, sizeof(*ei)); | |
ba3c2b19 NB |
3232 | btrfs_abort_transaction(trans, ret); |
3233 | goto out; | |
3234 | } | |
952fccac | 3235 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3236 | struct btrfs_extent_item); |
3173a18f JB |
3237 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3238 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3239 | struct btrfs_tree_block_info *bi; |
eee3b811 | 3240 | |
1c2a07f5 | 3241 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
eee3b811 QW |
3242 | abort_and_dump(trans, path, |
3243 | "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu", | |
3244 | key.objectid, key.type, key.offset, | |
3245 | path->slots[0], owner_objectid, item_size, | |
3246 | sizeof(*ei) + sizeof(*bi)); | |
3247 | ret = -EUCLEAN; | |
3248 | goto out; | |
1c2a07f5 | 3249 | } |
5d4f98a2 YZ |
3250 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3251 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3252 | } | |
56bec294 | 3253 | |
5d4f98a2 | 3254 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3255 | if (refs < refs_to_drop) { |
eee3b811 QW |
3256 | abort_and_dump(trans, path, |
3257 | "trying to drop %d refs but we only have %llu for bytenr %llu slot %u", | |
3258 | refs_to_drop, refs, bytenr, path->slots[0]); | |
3259 | ret = -EUCLEAN; | |
3260 | goto out; | |
32b02538 | 3261 | } |
56bec294 | 3262 | refs -= refs_to_drop; |
5f39d397 | 3263 | |
5d4f98a2 YZ |
3264 | if (refs > 0) { |
3265 | if (extent_op) | |
3266 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3267 | /* | |
3268 | * In the case of inline back ref, reference count will | |
3269 | * be updated by remove_extent_backref | |
952fccac | 3270 | */ |
5d4f98a2 | 3271 | if (iref) { |
1c2a07f5 | 3272 | if (!found_extent) { |
eee3b811 QW |
3273 | abort_and_dump(trans, path, |
3274 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u", | |
3275 | path->slots[0]); | |
3276 | ret = -EUCLEAN; | |
3277 | goto out; | |
1c2a07f5 | 3278 | } |
5d4f98a2 YZ |
3279 | } else { |
3280 | btrfs_set_extent_refs(leaf, ei, refs); | |
50564b65 | 3281 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
3282 | } |
3283 | if (found_extent) { | |
76d76e78 | 3284 | ret = remove_extent_backref(trans, extent_root, path, |
5b2a54bb | 3285 | iref, refs_to_drop, is_data); |
005d6427 | 3286 | if (ret) { |
66642832 | 3287 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3288 | goto out; |
3289 | } | |
952fccac | 3290 | } |
5d4f98a2 | 3291 | } else { |
cecbb533 BB |
3292 | struct btrfs_squota_delta delta = { |
3293 | .root = delayed_ref_root, | |
3294 | .num_bytes = num_bytes, | |
3295 | .rsv_bytes = 0, | |
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; |
ed4f255b | 3428 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
3429 | int ret; |
3430 | ||
ed4f255b | 3431 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
457cb1dd | 3432 | buf->start, buf->len, parent, btrfs_header_owner(buf)); |
ed4f255b | 3433 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), |
7a163608 | 3434 | root_id, 0, false); |
ed4f255b | 3435 | |
7a163608 | 3436 | if (root_id != BTRFS_TREE_LOG_OBJECTID) { |
8a5040f7 | 3437 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 3438 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL); |
79787eaa | 3439 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
3440 | } |
3441 | ||
0a16c7d7 | 3442 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3443 | struct btrfs_block_group *cache; |
485df755 | 3444 | bool must_pin = false; |
6219872d | 3445 | |
7a163608 | 3446 | if (root_id != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3447 | ret = check_ref_cleanup(trans, buf->start); |
d3575156 NA |
3448 | if (!ret) { |
3449 | btrfs_redirty_list_add(trans->transaction, buf); | |
37be25bc | 3450 | goto out; |
d3575156 | 3451 | } |
f0486c68 YZ |
3452 | } |
3453 | ||
0b246afa | 3454 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3455 | |
f0486c68 | 3456 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3457 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3458 | btrfs_put_block_group(cache); |
37be25bc | 3459 | goto out; |
f0486c68 YZ |
3460 | } |
3461 | ||
485df755 | 3462 | /* |
968b7158 JB |
3463 | * If there are tree mod log users we may have recorded mod log |
3464 | * operations for this node. If we re-allocate this node we | |
3465 | * could replay operations on this node that happened when it | |
3466 | * existed in a completely different root. For example if it | |
3467 | * was part of root A, then was reallocated to root B, and we | |
3468 | * are doing a btrfs_old_search_slot(root b), we could replay | |
3469 | * operations that happened when the block was part of root A, | |
3470 | * giving us an inconsistent view of the btree. | |
3471 | * | |
485df755 FM |
3472 | * We are safe from races here because at this point no other |
3473 | * node or root points to this extent buffer, so if after this | |
968b7158 JB |
3474 | * check a new tree mod log user joins we will not have an |
3475 | * existing log of operations on this node that we have to | |
3476 | * contend with. | |
485df755 | 3477 | */ |
968b7158 | 3478 | if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags)) |
888dd183 | 3479 | must_pin = true; |
485df755 FM |
3480 | |
3481 | if (must_pin || btrfs_is_zoned(fs_info)) { | |
d3575156 NA |
3482 | btrfs_redirty_list_add(trans->transaction, buf); |
3483 | pin_down_extent(trans, cache, buf->start, buf->len, 1); | |
3484 | btrfs_put_block_group(cache); | |
3485 | goto out; | |
3486 | } | |
3487 | ||
f0486c68 YZ |
3488 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); |
3489 | ||
3490 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3491 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3492 | btrfs_put_block_group(cache); |
71ff6437 | 3493 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3494 | } |
3495 | out: | |
0a16c7d7 OS |
3496 | if (last_ref) { |
3497 | /* | |
3498 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3499 | * matter anymore. | |
3500 | */ | |
3501 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3502 | } | |
f0486c68 YZ |
3503 | } |
3504 | ||
79787eaa | 3505 | /* Can return -ENOMEM */ |
ffd4bb2a | 3506 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3507 | { |
ffd4bb2a | 3508 | struct btrfs_fs_info *fs_info = trans->fs_info; |
925baedd CM |
3509 | int ret; |
3510 | ||
f5ee5c9a | 3511 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3512 | return 0; |
fccb84c9 | 3513 | |
56bec294 CM |
3514 | /* |
3515 | * tree log blocks never actually go into the extent allocation | |
3516 | * tree, just update pinning info and exit early. | |
56bec294 | 3517 | */ |
ffd4bb2a | 3518 | if ((ref->type == BTRFS_REF_METADATA && |
610647d7 | 3519 | ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) || |
ffd4bb2a | 3520 | (ref->type == BTRFS_REF_DATA && |
610647d7 | 3521 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { |
b25c36f8 | 3522 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
56bec294 | 3523 | ret = 0; |
ffd4bb2a | 3524 | } else if (ref->type == BTRFS_REF_METADATA) { |
2187374f | 3525 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL); |
5d4f98a2 | 3526 | } else { |
2187374f | 3527 | ret = btrfs_add_delayed_data_ref(trans, ref, 0); |
56bec294 | 3528 | } |
d7eae340 | 3529 | |
ffd4bb2a | 3530 | if (!((ref->type == BTRFS_REF_METADATA && |
610647d7 | 3531 | ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) || |
ffd4bb2a | 3532 | (ref->type == BTRFS_REF_DATA && |
610647d7 | 3533 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) |
ffd4bb2a | 3534 | btrfs_ref_tree_mod(fs_info, ref); |
8a5040f7 | 3535 | |
925baedd CM |
3536 | return ret; |
3537 | } | |
3538 | ||
817d52f8 | 3539 | enum btrfs_loop_type { |
b9d97cff JB |
3540 | /* |
3541 | * Start caching block groups but do not wait for progress or for them | |
3542 | * to be done. | |
3543 | */ | |
f262fa8d | 3544 | LOOP_CACHING_NOWAIT, |
b9d97cff JB |
3545 | |
3546 | /* | |
3547 | * Wait for the block group free_space >= the space we're waiting for if | |
3548 | * the block group isn't cached. | |
3549 | */ | |
f262fa8d | 3550 | LOOP_CACHING_WAIT, |
b9d97cff JB |
3551 | |
3552 | /* | |
3553 | * Allow allocations to happen from block groups that do not yet have a | |
3554 | * size classification. | |
3555 | */ | |
52bb7a21 | 3556 | LOOP_UNSET_SIZE_CLASS, |
b9d97cff JB |
3557 | |
3558 | /* | |
3559 | * Allocate a chunk and then retry the allocation. | |
3560 | */ | |
f262fa8d | 3561 | LOOP_ALLOC_CHUNK, |
b9d97cff JB |
3562 | |
3563 | /* | |
3564 | * Ignore the size class restrictions for this allocation. | |
3565 | */ | |
52bb7a21 | 3566 | LOOP_WRONG_SIZE_CLASS, |
b9d97cff JB |
3567 | |
3568 | /* | |
3569 | * Ignore the empty size, only try to allocate the number of bytes | |
3570 | * needed for this allocation. | |
3571 | */ | |
f262fa8d | 3572 | LOOP_NO_EMPTY_SIZE, |
817d52f8 JB |
3573 | }; |
3574 | ||
e570fd27 | 3575 | static inline void |
32da5386 | 3576 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3577 | int delalloc) |
3578 | { | |
3579 | if (delalloc) | |
3580 | down_read(&cache->data_rwsem); | |
3581 | } | |
3582 | ||
32da5386 | 3583 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3584 | int delalloc) |
3585 | { | |
3586 | btrfs_get_block_group(cache); | |
3587 | if (delalloc) | |
3588 | down_read(&cache->data_rwsem); | |
3589 | } | |
3590 | ||
32da5386 DS |
3591 | static struct btrfs_block_group *btrfs_lock_cluster( |
3592 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3593 | struct btrfs_free_cluster *cluster, |
3594 | int delalloc) | |
c142c6a4 | 3595 | __acquires(&cluster->refill_lock) |
e570fd27 | 3596 | { |
32da5386 | 3597 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3598 | |
e570fd27 | 3599 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3600 | while (1) { |
3601 | used_bg = cluster->block_group; | |
3602 | if (!used_bg) | |
3603 | return NULL; | |
3604 | ||
3605 | if (used_bg == block_group) | |
e570fd27 MX |
3606 | return used_bg; |
3607 | ||
6719afdc | 3608 | btrfs_get_block_group(used_bg); |
e570fd27 | 3609 | |
6719afdc GU |
3610 | if (!delalloc) |
3611 | return used_bg; | |
e570fd27 | 3612 | |
6719afdc GU |
3613 | if (down_read_trylock(&used_bg->data_rwsem)) |
3614 | return used_bg; | |
e570fd27 | 3615 | |
6719afdc | 3616 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3617 | |
e321f8a8 LB |
3618 | /* We should only have one-level nested. */ |
3619 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3620 | |
6719afdc GU |
3621 | spin_lock(&cluster->refill_lock); |
3622 | if (used_bg == cluster->block_group) | |
3623 | return used_bg; | |
e570fd27 | 3624 | |
6719afdc GU |
3625 | up_read(&used_bg->data_rwsem); |
3626 | btrfs_put_block_group(used_bg); | |
3627 | } | |
e570fd27 MX |
3628 | } |
3629 | ||
3630 | static inline void | |
32da5386 | 3631 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3632 | int delalloc) |
3633 | { | |
3634 | if (delalloc) | |
3635 | up_read(&cache->data_rwsem); | |
3636 | btrfs_put_block_group(cache); | |
3637 | } | |
3638 | ||
d06e3bb6 QW |
3639 | /* |
3640 | * Helper function for find_free_extent(). | |
3641 | * | |
3642 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
d06e3bb6 QW |
3643 | * Return >0 to inform caller that we find nothing |
3644 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3645 | */ | |
32da5386 | 3646 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3647 | struct find_free_extent_ctl *ffe_ctl, |
3648 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3649 | { |
32da5386 | 3650 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3651 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3652 | u64 aligned_cluster; |
3653 | u64 offset; | |
3654 | int ret; | |
3655 | ||
3656 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3657 | if (!cluster_bg) | |
3658 | goto refill_cluster; | |
3659 | if (cluster_bg != bg && (cluster_bg->ro || | |
3660 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3661 | goto release_cluster; | |
3662 | ||
3663 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3664 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3665 | &ffe_ctl->max_extent_size); |
3666 | if (offset) { | |
3667 | /* We have a block, we're done */ | |
3668 | spin_unlock(&last_ptr->refill_lock); | |
cfc2de0f | 3669 | trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl); |
d06e3bb6 QW |
3670 | *cluster_bg_ret = cluster_bg; |
3671 | ffe_ctl->found_offset = offset; | |
3672 | return 0; | |
3673 | } | |
3674 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3675 | ||
3676 | release_cluster: | |
3677 | /* | |
3678 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3679 | * lets just skip it and let the allocator find whatever block it can | |
3680 | * find. If we reach this point, we will have tried the cluster | |
3681 | * allocator plenty of times and not have found anything, so we are | |
3682 | * likely way too fragmented for the clustering stuff to find anything. | |
3683 | * | |
3684 | * However, if the cluster is taken from the current block group, | |
3685 | * release the cluster first, so that we stand a better chance of | |
3686 | * succeeding in the unclustered allocation. | |
3687 | */ | |
3688 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3689 | spin_unlock(&last_ptr->refill_lock); | |
3690 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3691 | return -ENOENT; | |
3692 | } | |
3693 | ||
3694 | /* This cluster didn't work out, free it and start over */ | |
3695 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3696 | ||
3697 | if (cluster_bg != bg) | |
3698 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3699 | ||
3700 | refill_cluster: | |
3701 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3702 | spin_unlock(&last_ptr->refill_lock); | |
3703 | return -ENOENT; | |
3704 | } | |
3705 | ||
3706 | aligned_cluster = max_t(u64, | |
3707 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3708 | bg->full_stripe_len); | |
2ceeae2e DS |
3709 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3710 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3711 | if (ret == 0) { |
3712 | /* Now pull our allocation out of this cluster */ | |
3713 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3714 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3715 | &ffe_ctl->max_extent_size); | |
3716 | if (offset) { | |
3717 | /* We found one, proceed */ | |
3718 | spin_unlock(&last_ptr->refill_lock); | |
d06e3bb6 | 3719 | ffe_ctl->found_offset = offset; |
cfc2de0f | 3720 | trace_btrfs_reserve_extent_cluster(bg, ffe_ctl); |
d06e3bb6 QW |
3721 | return 0; |
3722 | } | |
d06e3bb6 QW |
3723 | } |
3724 | /* | |
3725 | * At this point we either didn't find a cluster or we weren't able to | |
3726 | * allocate a block from our cluster. Free the cluster we've been | |
3727 | * trying to use, and go to the next block group. | |
3728 | */ | |
3729 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3730 | spin_unlock(&last_ptr->refill_lock); | |
3731 | return 1; | |
3732 | } | |
3733 | ||
e1a41848 QW |
3734 | /* |
3735 | * Return >0 to inform caller that we find nothing | |
3736 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
e1a41848 | 3737 | */ |
32da5386 | 3738 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3739 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3740 | { |
897cae79 | 3741 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3742 | u64 offset; |
3743 | ||
3744 | /* | |
3745 | * We are doing an unclustered allocation, set the fragmented flag so | |
3746 | * we don't bother trying to setup a cluster again until we get more | |
3747 | * space. | |
3748 | */ | |
3749 | if (unlikely(last_ptr)) { | |
3750 | spin_lock(&last_ptr->lock); | |
3751 | last_ptr->fragmented = 1; | |
3752 | spin_unlock(&last_ptr->lock); | |
3753 | } | |
3754 | if (ffe_ctl->cached) { | |
3755 | struct btrfs_free_space_ctl *free_space_ctl; | |
3756 | ||
3757 | free_space_ctl = bg->free_space_ctl; | |
3758 | spin_lock(&free_space_ctl->tree_lock); | |
3759 | if (free_space_ctl->free_space < | |
3760 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3761 | ffe_ctl->empty_size) { | |
3762 | ffe_ctl->total_free_space = max_t(u64, | |
3763 | ffe_ctl->total_free_space, | |
3764 | free_space_ctl->free_space); | |
3765 | spin_unlock(&free_space_ctl->tree_lock); | |
3766 | return 1; | |
3767 | } | |
3768 | spin_unlock(&free_space_ctl->tree_lock); | |
3769 | } | |
3770 | ||
3771 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3772 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3773 | &ffe_ctl->max_extent_size); | |
cd361199 | 3774 | if (!offset) |
e1a41848 | 3775 | return 1; |
e1a41848 QW |
3776 | ffe_ctl->found_offset = offset; |
3777 | return 0; | |
3778 | } | |
3779 | ||
c668690d NA |
3780 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3781 | struct find_free_extent_ctl *ffe_ctl, | |
3782 | struct btrfs_block_group **bg_ret) | |
3783 | { | |
3784 | int ret; | |
3785 | ||
3786 | /* We want to try and use the cluster allocator, so lets look there */ | |
3787 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3788 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
cd361199 | 3789 | if (ret >= 0) |
c668690d NA |
3790 | return ret; |
3791 | /* ret == -ENOENT case falls through */ | |
3792 | } | |
3793 | ||
897cae79 | 3794 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3795 | } |
3796 | ||
40ab3be1 NA |
3797 | /* |
3798 | * Tree-log block group locking | |
3799 | * ============================ | |
3800 | * | |
3801 | * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which | |
3802 | * indicates the starting address of a block group, which is reserved only | |
3803 | * for tree-log metadata. | |
3804 | * | |
3805 | * Lock nesting | |
3806 | * ============ | |
3807 | * | |
3808 | * space_info::lock | |
3809 | * block_group::lock | |
3810 | * fs_info::treelog_bg_lock | |
3811 | */ | |
3812 | ||
2eda5708 NA |
3813 | /* |
3814 | * Simple allocator for sequential-only block group. It only allows sequential | |
3815 | * allocation. No need to play with trees. This function also reserves the | |
3816 | * bytes as in btrfs_add_reserved_bytes. | |
3817 | */ | |
3818 | static int do_allocation_zoned(struct btrfs_block_group *block_group, | |
3819 | struct find_free_extent_ctl *ffe_ctl, | |
3820 | struct btrfs_block_group **bg_ret) | |
3821 | { | |
40ab3be1 | 3822 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
2eda5708 NA |
3823 | struct btrfs_space_info *space_info = block_group->space_info; |
3824 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3825 | u64 start = block_group->start; | |
3826 | u64 num_bytes = ffe_ctl->num_bytes; | |
3827 | u64 avail; | |
40ab3be1 NA |
3828 | u64 bytenr = block_group->start; |
3829 | u64 log_bytenr; | |
c2707a25 | 3830 | u64 data_reloc_bytenr; |
2eda5708 | 3831 | int ret = 0; |
2d81eb1c | 3832 | bool skip = false; |
2eda5708 NA |
3833 | |
3834 | ASSERT(btrfs_is_zoned(block_group->fs_info)); | |
3835 | ||
40ab3be1 NA |
3836 | /* |
3837 | * Do not allow non-tree-log blocks in the dedicated tree-log block | |
3838 | * group, and vice versa. | |
3839 | */ | |
3840 | spin_lock(&fs_info->treelog_bg_lock); | |
3841 | log_bytenr = fs_info->treelog_bg; | |
2d81eb1c JT |
3842 | if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) || |
3843 | (!ffe_ctl->for_treelog && bytenr == log_bytenr))) | |
3844 | skip = true; | |
40ab3be1 NA |
3845 | spin_unlock(&fs_info->treelog_bg_lock); |
3846 | if (skip) | |
3847 | return 1; | |
3848 | ||
c2707a25 JT |
3849 | /* |
3850 | * Do not allow non-relocation blocks in the dedicated relocation block | |
3851 | * group, and vice versa. | |
3852 | */ | |
3853 | spin_lock(&fs_info->relocation_bg_lock); | |
3854 | data_reloc_bytenr = fs_info->data_reloc_bg; | |
3855 | if (data_reloc_bytenr && | |
3856 | ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) || | |
3857 | (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr))) | |
3858 | skip = true; | |
3859 | spin_unlock(&fs_info->relocation_bg_lock); | |
3860 | if (skip) | |
3861 | return 1; | |
1ada69f6 | 3862 | |
2e654e4b NA |
3863 | /* Check RO and no space case before trying to activate it */ |
3864 | spin_lock(&block_group->lock); | |
1bfd4767 | 3865 | if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) { |
1ada69f6 NA |
3866 | ret = 1; |
3867 | /* | |
3868 | * May need to clear fs_info->{treelog,data_reloc}_bg. | |
3869 | * Return the error after taking the locks. | |
3870 | */ | |
2e654e4b NA |
3871 | } |
3872 | spin_unlock(&block_group->lock); | |
3873 | ||
5a7d107e NA |
3874 | /* Metadata block group is activated at write time. */ |
3875 | if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) && | |
3876 | !btrfs_zone_activate(block_group)) { | |
1ada69f6 NA |
3877 | ret = 1; |
3878 | /* | |
3879 | * May need to clear fs_info->{treelog,data_reloc}_bg. | |
3880 | * Return the error after taking the locks. | |
3881 | */ | |
3882 | } | |
2e654e4b | 3883 | |
2eda5708 NA |
3884 | spin_lock(&space_info->lock); |
3885 | spin_lock(&block_group->lock); | |
40ab3be1 | 3886 | spin_lock(&fs_info->treelog_bg_lock); |
c2707a25 | 3887 | spin_lock(&fs_info->relocation_bg_lock); |
40ab3be1 | 3888 | |
1ada69f6 NA |
3889 | if (ret) |
3890 | goto out; | |
3891 | ||
40ab3be1 NA |
3892 | ASSERT(!ffe_ctl->for_treelog || |
3893 | block_group->start == fs_info->treelog_bg || | |
3894 | fs_info->treelog_bg == 0); | |
c2707a25 JT |
3895 | ASSERT(!ffe_ctl->for_data_reloc || |
3896 | block_group->start == fs_info->data_reloc_bg || | |
3897 | fs_info->data_reloc_bg == 0); | |
2eda5708 | 3898 | |
3349b57f | 3899 | if (block_group->ro || |
332581bd NA |
3900 | (!ffe_ctl->for_data_reloc && |
3901 | test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) { | |
2eda5708 NA |
3902 | ret = 1; |
3903 | goto out; | |
3904 | } | |
3905 | ||
40ab3be1 NA |
3906 | /* |
3907 | * Do not allow currently using block group to be tree-log dedicated | |
3908 | * block group. | |
3909 | */ | |
3910 | if (ffe_ctl->for_treelog && !fs_info->treelog_bg && | |
3911 | (block_group->used || block_group->reserved)) { | |
3912 | ret = 1; | |
3913 | goto out; | |
3914 | } | |
3915 | ||
c2707a25 JT |
3916 | /* |
3917 | * Do not allow currently used block group to be the data relocation | |
3918 | * dedicated block group. | |
3919 | */ | |
3920 | if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg && | |
3921 | (block_group->used || block_group->reserved)) { | |
3922 | ret = 1; | |
3923 | goto out; | |
3924 | } | |
3925 | ||
98173255 NA |
3926 | WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity); |
3927 | avail = block_group->zone_capacity - block_group->alloc_offset; | |
2eda5708 NA |
3928 | if (avail < num_bytes) { |
3929 | if (ffe_ctl->max_extent_size < avail) { | |
3930 | /* | |
3931 | * With sequential allocator, free space is always | |
3932 | * contiguous | |
3933 | */ | |
3934 | ffe_ctl->max_extent_size = avail; | |
3935 | ffe_ctl->total_free_space = avail; | |
3936 | } | |
3937 | ret = 1; | |
3938 | goto out; | |
3939 | } | |
3940 | ||
40ab3be1 NA |
3941 | if (ffe_ctl->for_treelog && !fs_info->treelog_bg) |
3942 | fs_info->treelog_bg = block_group->start; | |
3943 | ||
332581bd NA |
3944 | if (ffe_ctl->for_data_reloc) { |
3945 | if (!fs_info->data_reloc_bg) | |
3946 | fs_info->data_reloc_bg = block_group->start; | |
3947 | /* | |
3948 | * Do not allow allocations from this block group, unless it is | |
3949 | * for data relocation. Compared to increasing the ->ro, setting | |
3950 | * the ->zoned_data_reloc_ongoing flag still allows nocow | |
3951 | * writers to come in. See btrfs_inc_nocow_writers(). | |
3952 | * | |
3953 | * We need to disable an allocation to avoid an allocation of | |
3954 | * regular (non-relocation data) extent. With mix of relocation | |
3955 | * extents and regular extents, we can dispatch WRITE commands | |
3956 | * (for relocation extents) and ZONE APPEND commands (for | |
3957 | * regular extents) at the same time to the same zone, which | |
3958 | * easily break the write pointer. | |
3959 | * | |
3960 | * Also, this flag avoids this block group to be zone finished. | |
3961 | */ | |
3962 | set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags); | |
3963 | } | |
c2707a25 | 3964 | |
2eda5708 NA |
3965 | ffe_ctl->found_offset = start + block_group->alloc_offset; |
3966 | block_group->alloc_offset += num_bytes; | |
3967 | spin_lock(&ctl->tree_lock); | |
3968 | ctl->free_space -= num_bytes; | |
3969 | spin_unlock(&ctl->tree_lock); | |
3970 | ||
3971 | /* | |
3972 | * We do not check if found_offset is aligned to stripesize. The | |
3973 | * address is anyway rewritten when using zone append writing. | |
3974 | */ | |
3975 | ||
3976 | ffe_ctl->search_start = ffe_ctl->found_offset; | |
3977 | ||
3978 | out: | |
40ab3be1 NA |
3979 | if (ret && ffe_ctl->for_treelog) |
3980 | fs_info->treelog_bg = 0; | |
332581bd | 3981 | if (ret && ffe_ctl->for_data_reloc) |
c2707a25 JT |
3982 | fs_info->data_reloc_bg = 0; |
3983 | spin_unlock(&fs_info->relocation_bg_lock); | |
40ab3be1 | 3984 | spin_unlock(&fs_info->treelog_bg_lock); |
2eda5708 NA |
3985 | spin_unlock(&block_group->lock); |
3986 | spin_unlock(&space_info->lock); | |
3987 | return ret; | |
3988 | } | |
3989 | ||
c668690d NA |
3990 | static int do_allocation(struct btrfs_block_group *block_group, |
3991 | struct find_free_extent_ctl *ffe_ctl, | |
3992 | struct btrfs_block_group **bg_ret) | |
3993 | { | |
3994 | switch (ffe_ctl->policy) { | |
3995 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3996 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
2eda5708 NA |
3997 | case BTRFS_EXTENT_ALLOC_ZONED: |
3998 | return do_allocation_zoned(block_group, ffe_ctl, bg_ret); | |
c668690d NA |
3999 | default: |
4000 | BUG(); | |
4001 | } | |
4002 | } | |
4003 | ||
baba5062 NA |
4004 | static void release_block_group(struct btrfs_block_group *block_group, |
4005 | struct find_free_extent_ctl *ffe_ctl, | |
4006 | int delalloc) | |
4007 | { | |
4008 | switch (ffe_ctl->policy) { | |
4009 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
cd361199 | 4010 | ffe_ctl->retry_uncached = false; |
baba5062 | 4011 | break; |
2eda5708 NA |
4012 | case BTRFS_EXTENT_ALLOC_ZONED: |
4013 | /* Nothing to do */ | |
4014 | break; | |
baba5062 NA |
4015 | default: |
4016 | BUG(); | |
4017 | } | |
4018 | ||
4019 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
4020 | ffe_ctl->index); | |
4021 | btrfs_release_block_group(block_group, delalloc); | |
4022 | } | |
4023 | ||
0ab9724b NA |
4024 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
4025 | struct btrfs_key *ins) | |
4026 | { | |
4027 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
4028 | ||
4029 | if (!ffe_ctl->use_cluster && last_ptr) { | |
4030 | spin_lock(&last_ptr->lock); | |
4031 | last_ptr->window_start = ins->objectid; | |
4032 | spin_unlock(&last_ptr->lock); | |
4033 | } | |
4034 | } | |
4035 | ||
4036 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
4037 | struct btrfs_key *ins) | |
4038 | { | |
4039 | switch (ffe_ctl->policy) { | |
4040 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4041 | found_extent_clustered(ffe_ctl, ins); | |
4042 | break; | |
2eda5708 NA |
4043 | case BTRFS_EXTENT_ALLOC_ZONED: |
4044 | /* Nothing to do */ | |
4045 | break; | |
0ab9724b NA |
4046 | default: |
4047 | BUG(); | |
4048 | } | |
4049 | } | |
4050 | ||
393f646e NA |
4051 | static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info, |
4052 | struct find_free_extent_ctl *ffe_ctl) | |
4053 | { | |
5a7d107e NA |
4054 | /* Block group's activeness is not a requirement for METADATA block groups. */ |
4055 | if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)) | |
4056 | return 0; | |
4057 | ||
393f646e NA |
4058 | /* If we can activate new zone, just allocate a chunk and use it */ |
4059 | if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags)) | |
4060 | return 0; | |
4061 | ||
4062 | /* | |
4063 | * We already reached the max active zones. Try to finish one block | |
4064 | * group to make a room for a new block group. This is only possible | |
4065 | * for a data block group because btrfs_zone_finish() may need to wait | |
4066 | * for a running transaction which can cause a deadlock for metadata | |
4067 | * allocation. | |
4068 | */ | |
4069 | if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) { | |
4070 | int ret = btrfs_zone_finish_one_bg(fs_info); | |
4071 | ||
4072 | if (ret == 1) | |
4073 | return 0; | |
4074 | else if (ret < 0) | |
4075 | return ret; | |
4076 | } | |
4077 | ||
4078 | /* | |
4079 | * If we have enough free space left in an already active block group | |
4080 | * and we can't activate any other zone now, do not allow allocating a | |
4081 | * new chunk and let find_free_extent() retry with a smaller size. | |
4082 | */ | |
4083 | if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size) | |
4084 | return -ENOSPC; | |
4085 | ||
898793d9 NA |
4086 | /* |
4087 | * Even min_alloc_size is not left in any block groups. Since we cannot | |
4088 | * activate a new block group, allocating it may not help. Let's tell a | |
4089 | * caller to try again and hope it progress something by writing some | |
4090 | * parts of the region. That is only possible for data block groups, | |
4091 | * where a part of the region can be written. | |
4092 | */ | |
4093 | if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) | |
4094 | return -EAGAIN; | |
4095 | ||
393f646e NA |
4096 | /* |
4097 | * We cannot activate a new block group and no enough space left in any | |
4098 | * block groups. So, allocating a new block group may not help. But, | |
4099 | * there is nothing to do anyway, so let's go with it. | |
4100 | */ | |
4101 | return 0; | |
4102 | } | |
4103 | ||
bb9950d3 NA |
4104 | static int can_allocate_chunk(struct btrfs_fs_info *fs_info, |
4105 | struct find_free_extent_ctl *ffe_ctl) | |
50475cd5 NA |
4106 | { |
4107 | switch (ffe_ctl->policy) { | |
4108 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
bb9950d3 | 4109 | return 0; |
50475cd5 | 4110 | case BTRFS_EXTENT_ALLOC_ZONED: |
393f646e | 4111 | return can_allocate_chunk_zoned(fs_info, ffe_ctl); |
50475cd5 NA |
4112 | default: |
4113 | BUG(); | |
4114 | } | |
4115 | } | |
4116 | ||
e72d79d6 QW |
4117 | /* |
4118 | * Return >0 means caller needs to re-search for free extent | |
4119 | * Return 0 means we have the needed free extent. | |
4120 | * Return <0 means we failed to locate any free extent. | |
4121 | */ | |
4122 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
4123 | struct btrfs_key *ins, |
4124 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 4125 | bool full_search) |
e72d79d6 | 4126 | { |
8e1d0290 | 4127 | struct btrfs_root *root = fs_info->chunk_root; |
e72d79d6 QW |
4128 | int ret; |
4129 | ||
4130 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
4131 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
4132 | ffe_ctl->orig_have_caching_bg = true; | |
4133 | ||
e72d79d6 | 4134 | if (ins->objectid) { |
0ab9724b | 4135 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
4136 | return 0; |
4137 | } | |
4138 | ||
a85f05e5 NA |
4139 | if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg) |
4140 | return 1; | |
4141 | ||
4142 | ffe_ctl->index++; | |
4143 | if (ffe_ctl->index < BTRFS_NR_RAID_TYPES) | |
4144 | return 1; | |
4145 | ||
b9d97cff | 4146 | /* See the comments for btrfs_loop_type for an explanation of the phases. */ |
e72d79d6 QW |
4147 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { |
4148 | ffe_ctl->index = 0; | |
52bb7a21 BB |
4149 | /* |
4150 | * We want to skip the LOOP_CACHING_WAIT step if we don't have | |
4151 | * any uncached bgs and we've already done a full search | |
4152 | * through. | |
4153 | */ | |
4154 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT && | |
4155 | (!ffe_ctl->orig_have_caching_bg && full_search)) | |
e72d79d6 | 4156 | ffe_ctl->loop++; |
52bb7a21 | 4157 | ffe_ctl->loop++; |
e72d79d6 QW |
4158 | |
4159 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
4160 | struct btrfs_trans_handle *trans; | |
4161 | int exist = 0; | |
4162 | ||
52bb7a21 | 4163 | /* Check if allocation policy allows to create a new chunk */ |
bb9950d3 NA |
4164 | ret = can_allocate_chunk(fs_info, ffe_ctl); |
4165 | if (ret) | |
4166 | return ret; | |
50475cd5 | 4167 | |
e72d79d6 QW |
4168 | trans = current->journal_info; |
4169 | if (trans) | |
4170 | exist = 1; | |
4171 | else | |
4172 | trans = btrfs_join_transaction(root); | |
4173 | ||
4174 | if (IS_ERR(trans)) { | |
4175 | ret = PTR_ERR(trans); | |
4176 | return ret; | |
4177 | } | |
4178 | ||
fc471cb0 | 4179 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
760e69c4 | 4180 | CHUNK_ALLOC_FORCE_FOR_EXTENT); |
e72d79d6 | 4181 | |
e72d79d6 | 4182 | /* Do not bail out on ENOSPC since we can do more. */ |
52bb7a21 BB |
4183 | if (ret == -ENOSPC) { |
4184 | ret = 0; | |
4185 | ffe_ctl->loop++; | |
4186 | } | |
c70e2139 | 4187 | else if (ret < 0) |
e72d79d6 QW |
4188 | btrfs_abort_transaction(trans, ret); |
4189 | else | |
4190 | ret = 0; | |
4191 | if (!exist) | |
4192 | btrfs_end_transaction(trans); | |
4193 | if (ret) | |
4194 | return ret; | |
4195 | } | |
4196 | ||
4197 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
4198 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
4199 | return -ENOSPC; | |
4200 | ||
e72d79d6 QW |
4201 | /* |
4202 | * Don't loop again if we already have no empty_size and | |
4203 | * no empty_cluster. | |
4204 | */ | |
4205 | if (ffe_ctl->empty_size == 0 && | |
4206 | ffe_ctl->empty_cluster == 0) | |
4207 | return -ENOSPC; | |
4208 | ffe_ctl->empty_size = 0; | |
4209 | ffe_ctl->empty_cluster = 0; | |
4210 | } | |
4211 | return 1; | |
4212 | } | |
4213 | return -ENOSPC; | |
4214 | } | |
4215 | ||
52bb7a21 BB |
4216 | static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl, |
4217 | struct btrfs_block_group *bg) | |
4218 | { | |
4219 | if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED) | |
4220 | return true; | |
cb0922f2 | 4221 | if (!btrfs_block_group_should_use_size_class(bg)) |
52bb7a21 BB |
4222 | return true; |
4223 | if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS) | |
4224 | return true; | |
4225 | if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS && | |
4226 | bg->size_class == BTRFS_BG_SZ_NONE) | |
4227 | return true; | |
4228 | return ffe_ctl->size_class == bg->size_class; | |
4229 | } | |
4230 | ||
7e895409 NA |
4231 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
4232 | struct find_free_extent_ctl *ffe_ctl, | |
4233 | struct btrfs_space_info *space_info, | |
4234 | struct btrfs_key *ins) | |
4235 | { | |
4236 | /* | |
4237 | * If our free space is heavily fragmented we may not be able to make | |
4238 | * big contiguous allocations, so instead of doing the expensive search | |
4239 | * for free space, simply return ENOSPC with our max_extent_size so we | |
4240 | * can go ahead and search for a more manageable chunk. | |
4241 | * | |
4242 | * If our max_extent_size is large enough for our allocation simply | |
4243 | * disable clustering since we will likely not be able to find enough | |
4244 | * space to create a cluster and induce latency trying. | |
4245 | */ | |
4246 | if (space_info->max_extent_size) { | |
4247 | spin_lock(&space_info->lock); | |
4248 | if (space_info->max_extent_size && | |
4249 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
4250 | ins->offset = space_info->max_extent_size; | |
4251 | spin_unlock(&space_info->lock); | |
4252 | return -ENOSPC; | |
4253 | } else if (space_info->max_extent_size) { | |
4254 | ffe_ctl->use_cluster = false; | |
4255 | } | |
4256 | spin_unlock(&space_info->lock); | |
4257 | } | |
4258 | ||
4259 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
4260 | &ffe_ctl->empty_cluster); | |
4261 | if (ffe_ctl->last_ptr) { | |
4262 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
4263 | ||
4264 | spin_lock(&last_ptr->lock); | |
4265 | if (last_ptr->block_group) | |
4266 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4267 | if (last_ptr->fragmented) { | |
4268 | /* | |
4269 | * We still set window_start so we can keep track of the | |
4270 | * last place we found an allocation to try and save | |
4271 | * some time. | |
4272 | */ | |
4273 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4274 | ffe_ctl->use_cluster = false; | |
4275 | } | |
4276 | spin_unlock(&last_ptr->lock); | |
4277 | } | |
4278 | ||
4279 | return 0; | |
4280 | } | |
4281 | ||
4282 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
4283 | struct find_free_extent_ctl *ffe_ctl, | |
4284 | struct btrfs_space_info *space_info, | |
4285 | struct btrfs_key *ins) | |
4286 | { | |
4287 | switch (ffe_ctl->policy) { | |
4288 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4289 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
4290 | space_info, ins); | |
2eda5708 | 4291 | case BTRFS_EXTENT_ALLOC_ZONED: |
40ab3be1 NA |
4292 | if (ffe_ctl->for_treelog) { |
4293 | spin_lock(&fs_info->treelog_bg_lock); | |
4294 | if (fs_info->treelog_bg) | |
4295 | ffe_ctl->hint_byte = fs_info->treelog_bg; | |
4296 | spin_unlock(&fs_info->treelog_bg_lock); | |
4297 | } | |
c2707a25 JT |
4298 | if (ffe_ctl->for_data_reloc) { |
4299 | spin_lock(&fs_info->relocation_bg_lock); | |
4300 | if (fs_info->data_reloc_bg) | |
4301 | ffe_ctl->hint_byte = fs_info->data_reloc_bg; | |
4302 | spin_unlock(&fs_info->relocation_bg_lock); | |
4303 | } | |
2eda5708 | 4304 | return 0; |
7e895409 NA |
4305 | default: |
4306 | BUG(); | |
4307 | } | |
4308 | } | |
4309 | ||
fec577fb CM |
4310 | /* |
4311 | * walks the btree of allocated extents and find a hole of a given size. | |
4312 | * The key ins is changed to record the hole: | |
a4820398 | 4313 | * ins->objectid == start position |
62e2749e | 4314 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4315 | * ins->offset == the size of the hole. |
fec577fb | 4316 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4317 | * |
4318 | * If there is no suitable free space, we will record the max size of | |
4319 | * the free space extent currently. | |
e72d79d6 QW |
4320 | * |
4321 | * The overall logic and call chain: | |
4322 | * | |
4323 | * find_free_extent() | |
4324 | * |- Iterate through all block groups | |
4325 | * | |- Get a valid block group | |
4326 | * | |- Try to do clustered allocation in that block group | |
4327 | * | |- Try to do unclustered allocation in that block group | |
4328 | * | |- Check if the result is valid | |
4329 | * | | |- If valid, then exit | |
4330 | * | |- Jump to next block group | |
4331 | * | | |
4332 | * |- Push harder to find free extents | |
4333 | * |- If not found, re-iterate all block groups | |
fec577fb | 4334 | */ |
437490fe | 4335 | static noinline int find_free_extent(struct btrfs_root *root, |
a12b0dc0 NA |
4336 | struct btrfs_key *ins, |
4337 | struct find_free_extent_ctl *ffe_ctl) | |
fec577fb | 4338 | { |
437490fe | 4339 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4340 | int ret = 0; |
db8fe64f | 4341 | int cache_block_group_error = 0; |
32da5386 | 4342 | struct btrfs_block_group *block_group = NULL; |
80eb234a | 4343 | struct btrfs_space_info *space_info; |
a5e681d9 | 4344 | bool full_search = false; |
fec577fb | 4345 | |
a12b0dc0 | 4346 | WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize); |
b4bd745d | 4347 | |
a12b0dc0 NA |
4348 | ffe_ctl->search_start = 0; |
4349 | /* For clustered allocation */ | |
4350 | ffe_ctl->empty_cluster = 0; | |
4351 | ffe_ctl->last_ptr = NULL; | |
4352 | ffe_ctl->use_cluster = true; | |
4353 | ffe_ctl->have_caching_bg = false; | |
4354 | ffe_ctl->orig_have_caching_bg = false; | |
4355 | ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags); | |
4356 | ffe_ctl->loop = 0; | |
cd361199 | 4357 | ffe_ctl->retry_uncached = false; |
a12b0dc0 NA |
4358 | ffe_ctl->cached = 0; |
4359 | ffe_ctl->max_extent_size = 0; | |
4360 | ffe_ctl->total_free_space = 0; | |
4361 | ffe_ctl->found_offset = 0; | |
4362 | ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED; | |
52bb7a21 | 4363 | ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes); |
c10859be | 4364 | |
2eda5708 | 4365 | if (btrfs_is_zoned(fs_info)) |
a12b0dc0 | 4366 | ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED; |
2eda5708 | 4367 | |
962a298f | 4368 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4369 | ins->objectid = 0; |
4370 | ins->offset = 0; | |
b1a4d965 | 4371 | |
cfc2de0f | 4372 | trace_find_free_extent(root, ffe_ctl); |
3f7de037 | 4373 | |
a12b0dc0 | 4374 | space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags); |
1b1d1f66 | 4375 | if (!space_info) { |
a12b0dc0 | 4376 | btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags); |
1b1d1f66 JB |
4377 | return -ENOSPC; |
4378 | } | |
2552d17e | 4379 | |
a12b0dc0 | 4380 | ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins); |
7e895409 NA |
4381 | if (ret < 0) |
4382 | return ret; | |
fa9c0d79 | 4383 | |
a12b0dc0 | 4384 | ffe_ctl->search_start = max(ffe_ctl->search_start, |
0eb997bf | 4385 | first_logical_byte(fs_info)); |
a12b0dc0 NA |
4386 | ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte); |
4387 | if (ffe_ctl->search_start == ffe_ctl->hint_byte) { | |
b4bd745d | 4388 | block_group = btrfs_lookup_block_group(fs_info, |
a12b0dc0 | 4389 | ffe_ctl->search_start); |
817d52f8 JB |
4390 | /* |
4391 | * we don't want to use the block group if it doesn't match our | |
4392 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4393 | * |
4394 | * However if we are re-searching with an ideal block group | |
4395 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4396 | */ |
a12b0dc0 | 4397 | if (block_group && block_group_bits(block_group, ffe_ctl->flags) && |
285ff5af | 4398 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4399 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4400 | if (list_empty(&block_group->list) || |
4401 | block_group->ro) { | |
4402 | /* | |
4403 | * someone is removing this block group, | |
4404 | * we can't jump into the have_block_group | |
4405 | * target because our list pointers are not | |
4406 | * valid | |
4407 | */ | |
4408 | btrfs_put_block_group(block_group); | |
4409 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4410 | } else { |
a12b0dc0 NA |
4411 | ffe_ctl->index = btrfs_bg_flags_to_raid_index( |
4412 | block_group->flags); | |
4413 | btrfs_lock_block_group(block_group, | |
4414 | ffe_ctl->delalloc); | |
854c2f36 | 4415 | ffe_ctl->hinted = true; |
44fb5511 | 4416 | goto have_block_group; |
ccf0e725 | 4417 | } |
2552d17e | 4418 | } else if (block_group) { |
fa9c0d79 | 4419 | btrfs_put_block_group(block_group); |
2552d17e | 4420 | } |
42e70e7a | 4421 | } |
2552d17e | 4422 | search: |
854c2f36 | 4423 | trace_find_free_extent_search_loop(root, ffe_ctl); |
a12b0dc0 NA |
4424 | ffe_ctl->have_caching_bg = false; |
4425 | if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) || | |
4426 | ffe_ctl->index == 0) | |
a5e681d9 | 4427 | full_search = true; |
80eb234a | 4428 | down_read(&space_info->groups_sem); |
b4bd745d | 4429 | list_for_each_entry(block_group, |
a12b0dc0 | 4430 | &space_info->block_groups[ffe_ctl->index], list) { |
c668690d NA |
4431 | struct btrfs_block_group *bg_ret; |
4432 | ||
854c2f36 | 4433 | ffe_ctl->hinted = false; |
14443937 | 4434 | /* If the block group is read-only, we can skip it entirely. */ |
40ab3be1 | 4435 | if (unlikely(block_group->ro)) { |
a12b0dc0 | 4436 | if (ffe_ctl->for_treelog) |
40ab3be1 | 4437 | btrfs_clear_treelog_bg(block_group); |
c2707a25 JT |
4438 | if (ffe_ctl->for_data_reloc) |
4439 | btrfs_clear_data_reloc_bg(block_group); | |
14443937 | 4440 | continue; |
40ab3be1 | 4441 | } |
14443937 | 4442 | |
a12b0dc0 NA |
4443 | btrfs_grab_block_group(block_group, ffe_ctl->delalloc); |
4444 | ffe_ctl->search_start = block_group->start; | |
42e70e7a | 4445 | |
83a50de9 CM |
4446 | /* |
4447 | * this can happen if we end up cycling through all the | |
4448 | * raid types, but we want to make sure we only allocate | |
4449 | * for the proper type. | |
4450 | */ | |
a12b0dc0 | 4451 | if (!block_group_bits(block_group, ffe_ctl->flags)) { |
bece2e82 | 4452 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4453 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4454 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4455 | BTRFS_BLOCK_GROUP_RAID10; |
4456 | ||
4457 | /* | |
4458 | * if they asked for extra copies and this block group | |
4459 | * doesn't provide them, bail. This does allow us to | |
4460 | * fill raid0 from raid1. | |
4461 | */ | |
a12b0dc0 | 4462 | if ((ffe_ctl->flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4463 | goto loop; |
2a28468e QW |
4464 | |
4465 | /* | |
4466 | * This block group has different flags than we want. | |
4467 | * It's possible that we have MIXED_GROUP flag but no | |
4468 | * block group is mixed. Just skip such block group. | |
4469 | */ | |
a12b0dc0 | 4470 | btrfs_release_block_group(block_group, ffe_ctl->delalloc); |
2a28468e | 4471 | continue; |
83a50de9 CM |
4472 | } |
4473 | ||
2552d17e | 4474 | have_block_group: |
854c2f36 | 4475 | trace_find_free_extent_have_block_group(root, ffe_ctl, block_group); |
a12b0dc0 NA |
4476 | ffe_ctl->cached = btrfs_block_group_done(block_group); |
4477 | if (unlikely(!ffe_ctl->cached)) { | |
4478 | ffe_ctl->have_caching_bg = true; | |
ced8ecf0 | 4479 | ret = btrfs_cache_block_group(block_group, false); |
db8fe64f JB |
4480 | |
4481 | /* | |
4482 | * If we get ENOMEM here or something else we want to | |
4483 | * try other block groups, because it may not be fatal. | |
4484 | * However if we can't find anything else we need to | |
4485 | * save our return here so that we return the actual | |
4486 | * error that caused problems, not ENOSPC. | |
4487 | */ | |
4488 | if (ret < 0) { | |
4489 | if (!cache_block_group_error) | |
4490 | cache_block_group_error = ret; | |
4491 | ret = 0; | |
4492 | goto loop; | |
4493 | } | |
1d4284bd | 4494 | ret = 0; |
817d52f8 JB |
4495 | } |
4496 | ||
92fb94b6 JB |
4497 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) { |
4498 | if (!cache_block_group_error) | |
4499 | cache_block_group_error = -EIO; | |
36cce922 | 4500 | goto loop; |
92fb94b6 | 4501 | } |
0f9dd46c | 4502 | |
52bb7a21 BB |
4503 | if (!find_free_extent_check_size_class(ffe_ctl, block_group)) |
4504 | goto loop; | |
4505 | ||
c668690d | 4506 | bg_ret = NULL; |
a12b0dc0 | 4507 | ret = do_allocation(block_group, ffe_ctl, &bg_ret); |
cd361199 | 4508 | if (ret > 0) |
1cdda9b8 | 4509 | goto loop; |
cd361199 JB |
4510 | |
4511 | if (bg_ret && bg_ret != block_group) { | |
4512 | btrfs_release_block_group(block_group, ffe_ctl->delalloc); | |
4513 | block_group = bg_ret; | |
c668690d NA |
4514 | } |
4515 | ||
4516 | /* Checks */ | |
a12b0dc0 NA |
4517 | ffe_ctl->search_start = round_up(ffe_ctl->found_offset, |
4518 | fs_info->stripesize); | |
25179201 | 4519 | |
2552d17e | 4520 | /* move on to the next group */ |
a12b0dc0 | 4521 | if (ffe_ctl->search_start + ffe_ctl->num_bytes > |
b3470b5d | 4522 | block_group->start + block_group->length) { |
2eda5708 | 4523 | btrfs_add_free_space_unused(block_group, |
a12b0dc0 NA |
4524 | ffe_ctl->found_offset, |
4525 | ffe_ctl->num_bytes); | |
2552d17e | 4526 | goto loop; |
6226cb0a | 4527 | } |
f5a31e16 | 4528 | |
a12b0dc0 | 4529 | if (ffe_ctl->found_offset < ffe_ctl->search_start) |
2eda5708 | 4530 | btrfs_add_free_space_unused(block_group, |
a12b0dc0 NA |
4531 | ffe_ctl->found_offset, |
4532 | ffe_ctl->search_start - ffe_ctl->found_offset); | |
2552d17e | 4533 | |
a12b0dc0 NA |
4534 | ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes, |
4535 | ffe_ctl->num_bytes, | |
52bb7a21 BB |
4536 | ffe_ctl->delalloc, |
4537 | ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS); | |
f0486c68 | 4538 | if (ret == -EAGAIN) { |
2eda5708 | 4539 | btrfs_add_free_space_unused(block_group, |
a12b0dc0 NA |
4540 | ffe_ctl->found_offset, |
4541 | ffe_ctl->num_bytes); | |
2552d17e | 4542 | goto loop; |
0f9dd46c | 4543 | } |
9cfa3e34 | 4544 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4545 | |
f0486c68 | 4546 | /* we are all good, lets return */ |
a12b0dc0 NA |
4547 | ins->objectid = ffe_ctl->search_start; |
4548 | ins->offset = ffe_ctl->num_bytes; | |
d2fb3437 | 4549 | |
cfc2de0f | 4550 | trace_btrfs_reserve_extent(block_group, ffe_ctl); |
a12b0dc0 | 4551 | btrfs_release_block_group(block_group, ffe_ctl->delalloc); |
2552d17e JB |
4552 | break; |
4553 | loop: | |
cd361199 JB |
4554 | if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && |
4555 | !ffe_ctl->retry_uncached) { | |
4556 | ffe_ctl->retry_uncached = true; | |
4557 | btrfs_wait_block_group_cache_progress(block_group, | |
4558 | ffe_ctl->num_bytes + | |
4559 | ffe_ctl->empty_cluster + | |
4560 | ffe_ctl->empty_size); | |
4561 | goto have_block_group; | |
4562 | } | |
a12b0dc0 | 4563 | release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc); |
14443937 | 4564 | cond_resched(); |
2552d17e JB |
4565 | } |
4566 | up_read(&space_info->groups_sem); | |
4567 | ||
a12b0dc0 | 4568 | ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search); |
e72d79d6 | 4569 | if (ret > 0) |
b742bb82 YZ |
4570 | goto search; |
4571 | ||
db8fe64f | 4572 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4573 | /* |
4574 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4575 | * any contiguous hole. | |
4576 | */ | |
a12b0dc0 NA |
4577 | if (!ffe_ctl->max_extent_size) |
4578 | ffe_ctl->max_extent_size = ffe_ctl->total_free_space; | |
4f4db217 | 4579 | spin_lock(&space_info->lock); |
a12b0dc0 | 4580 | space_info->max_extent_size = ffe_ctl->max_extent_size; |
4f4db217 | 4581 | spin_unlock(&space_info->lock); |
a12b0dc0 | 4582 | ins->offset = ffe_ctl->max_extent_size; |
db8fe64f JB |
4583 | } else if (ret == -ENOSPC) { |
4584 | ret = cache_block_group_error; | |
4f4db217 | 4585 | } |
0f70abe2 | 4586 | return ret; |
fec577fb | 4587 | } |
ec44a35c | 4588 | |
6f47c706 | 4589 | /* |
9580503b DS |
4590 | * Entry point to the extent allocator. Tries to find a hole that is at least |
4591 | * as big as @num_bytes. | |
6f47c706 NB |
4592 | * |
4593 | * @root - The root that will contain this extent | |
4594 | * | |
4595 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4596 | * is used for accounting purposes. This value differs | |
4597 | * from @num_bytes only in the case of compressed extents. | |
4598 | * | |
4599 | * @num_bytes - Number of bytes to allocate on-disk. | |
4600 | * | |
4601 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4602 | * allocator should try to satisfy. In some cases | |
4603 | * @num_bytes may be larger than what is required and if | |
4604 | * the filesystem is fragmented then allocation fails. | |
4605 | * However, the presence of @min_alloc_size gives a | |
4606 | * chance to try and satisfy the smaller allocation. | |
4607 | * | |
4608 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4609 | * size in bytes the allocator should try to find free | |
4610 | * next to the block it returns. This is just a hint and | |
4611 | * may be ignored by the allocator. | |
4612 | * | |
4613 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4614 | * address passed. It might be ignored. | |
4615 | * | |
4616 | * @ins - This key is modified to record the found hole. It will | |
4617 | * have the following values: | |
4618 | * ins->objectid == start position | |
4619 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4620 | * ins->offset == the size of the hole. | |
4621 | * | |
4622 | * @is_data - Boolean flag indicating whether an extent is | |
4623 | * allocated for data (true) or metadata (false) | |
4624 | * | |
4625 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4626 | * delalloc or not. If 'true' data_rwsem of block groups | |
4627 | * is going to be acquired. | |
4628 | * | |
4629 | * | |
4630 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4631 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4632 | * largest available hole the allocator managed to find. | |
4633 | */ | |
18513091 | 4634 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4635 | u64 num_bytes, u64 min_alloc_size, |
4636 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4637 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4638 | { |
ab8d0fc4 | 4639 | struct btrfs_fs_info *fs_info = root->fs_info; |
a12b0dc0 | 4640 | struct find_free_extent_ctl ffe_ctl = {}; |
36af4e07 | 4641 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4642 | u64 flags; |
fec577fb | 4643 | int ret; |
40ab3be1 | 4644 | bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
c2707a25 | 4645 | bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data); |
925baedd | 4646 | |
1b86826d | 4647 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4648 | again: |
0b246afa | 4649 | WARN_ON(num_bytes < fs_info->sectorsize); |
a12b0dc0 NA |
4650 | |
4651 | ffe_ctl.ram_bytes = ram_bytes; | |
4652 | ffe_ctl.num_bytes = num_bytes; | |
a85f05e5 | 4653 | ffe_ctl.min_alloc_size = min_alloc_size; |
a12b0dc0 NA |
4654 | ffe_ctl.empty_size = empty_size; |
4655 | ffe_ctl.flags = flags; | |
4656 | ffe_ctl.delalloc = delalloc; | |
4657 | ffe_ctl.hint_byte = hint_byte; | |
4658 | ffe_ctl.for_treelog = for_treelog; | |
c2707a25 | 4659 | ffe_ctl.for_data_reloc = for_data_reloc; |
a12b0dc0 NA |
4660 | |
4661 | ret = find_free_extent(root, ins, &ffe_ctl); | |
9cfa3e34 | 4662 | if (!ret && !is_data) { |
ab8d0fc4 | 4663 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4664 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4665 | if (!final_tried && ins->offset) { |
4666 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4667 | num_bytes = round_down(num_bytes, |
0b246afa | 4668 | fs_info->sectorsize); |
9e622d6b | 4669 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4670 | ram_bytes = num_bytes; |
9e622d6b MX |
4671 | if (num_bytes == min_alloc_size) |
4672 | final_tried = true; | |
4673 | goto again; | |
ab8d0fc4 | 4674 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4675 | struct btrfs_space_info *sinfo; |
4676 | ||
280c2908 | 4677 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4678 | btrfs_err(fs_info, |
c2707a25 JT |
4679 | "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d", |
4680 | flags, num_bytes, for_treelog, for_data_reloc); | |
53804280 | 4681 | if (sinfo) |
5da6afeb JB |
4682 | btrfs_dump_space_info(fs_info, sinfo, |
4683 | num_bytes, 1); | |
9e622d6b | 4684 | } |
925baedd | 4685 | } |
0f9dd46c JB |
4686 | |
4687 | return ret; | |
e6dcd2dc CM |
4688 | } |
4689 | ||
a0fbf736 NB |
4690 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4691 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4692 | { |
32da5386 | 4693 | struct btrfs_block_group *cache; |
0f9dd46c | 4694 | |
0b246afa | 4695 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4696 | if (!cache) { |
0b246afa JM |
4697 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4698 | start); | |
0f9dd46c JB |
4699 | return -ENOSPC; |
4700 | } | |
1f3c79a2 | 4701 | |
a0fbf736 NB |
4702 | btrfs_add_free_space(cache, start, len); |
4703 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4704 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4705 | ||
fa9c0d79 | 4706 | btrfs_put_block_group(cache); |
a0fbf736 | 4707 | return 0; |
e6dcd2dc CM |
4708 | } |
4709 | ||
f863c502 DS |
4710 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, |
4711 | const struct extent_buffer *eb) | |
e688b725 | 4712 | { |
7ef54d54 | 4713 | struct btrfs_block_group *cache; |
a0fbf736 | 4714 | int ret = 0; |
7ef54d54 | 4715 | |
f863c502 | 4716 | cache = btrfs_lookup_block_group(trans->fs_info, eb->start); |
7ef54d54 | 4717 | if (!cache) { |
7bfc1007 | 4718 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
f863c502 | 4719 | eb->start); |
7ef54d54 NB |
4720 | return -ENOSPC; |
4721 | } | |
4722 | ||
f863c502 | 4723 | ret = pin_down_extent(trans, cache, eb->start, eb->len, 1); |
7ef54d54 | 4724 | btrfs_put_block_group(cache); |
a0fbf736 | 4725 | return ret; |
e688b725 CM |
4726 | } |
4727 | ||
34666705 JB |
4728 | static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr, |
4729 | u64 num_bytes) | |
4730 | { | |
4731 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
4732 | int ret; | |
4733 | ||
4734 | ret = remove_from_free_space_tree(trans, bytenr, num_bytes); | |
4735 | if (ret) | |
4736 | return ret; | |
4737 | ||
4738 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, true); | |
4739 | if (ret) { | |
4740 | ASSERT(!ret); | |
4741 | btrfs_err(fs_info, "update block group failed for %llu %llu", | |
4742 | bytenr, num_bytes); | |
4743 | return ret; | |
4744 | } | |
4745 | ||
4746 | trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes); | |
4747 | return 0; | |
4748 | } | |
4749 | ||
5d4f98a2 | 4750 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4751 | u64 parent, u64 root_objectid, |
4752 | u64 flags, u64 owner, u64 offset, | |
2672a051 | 4753 | struct btrfs_key *ins, int ref_mod, u64 oref_root) |
e6dcd2dc | 4754 | { |
ef89b824 | 4755 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 4756 | struct btrfs_root *extent_root; |
e6dcd2dc | 4757 | int ret; |
e6dcd2dc | 4758 | struct btrfs_extent_item *extent_item; |
d9a620f7 | 4759 | struct btrfs_extent_owner_ref *oref; |
5d4f98a2 | 4760 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4761 | struct btrfs_path *path; |
5d4f98a2 YZ |
4762 | struct extent_buffer *leaf; |
4763 | int type; | |
4764 | u32 size; | |
d9a620f7 | 4765 | const bool simple_quota = (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE); |
26b8003f | 4766 | |
5d4f98a2 YZ |
4767 | if (parent > 0) |
4768 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4769 | else | |
4770 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4771 | |
d9a620f7 BB |
4772 | size = sizeof(*extent_item); |
4773 | if (simple_quota) | |
4774 | size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY); | |
4775 | size += btrfs_extent_inline_ref_size(type); | |
7bb86316 CM |
4776 | |
4777 | path = btrfs_alloc_path(); | |
db5b493a TI |
4778 | if (!path) |
4779 | return -ENOMEM; | |
47e4bb98 | 4780 | |
29cbcf40 JB |
4781 | extent_root = btrfs_extent_root(fs_info, ins->objectid); |
4782 | ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size); | |
79787eaa JM |
4783 | if (ret) { |
4784 | btrfs_free_path(path); | |
4785 | return ret; | |
4786 | } | |
0f9dd46c | 4787 | |
5d4f98a2 YZ |
4788 | leaf = path->nodes[0]; |
4789 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4790 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4791 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4792 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4793 | btrfs_set_extent_flags(leaf, extent_item, | |
4794 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4795 | ||
4796 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
d9a620f7 BB |
4797 | if (simple_quota) { |
4798 | btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY); | |
4799 | oref = (struct btrfs_extent_owner_ref *)(&iref->offset); | |
2672a051 | 4800 | btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root); |
d9a620f7 BB |
4801 | iref = (struct btrfs_extent_inline_ref *)(oref + 1); |
4802 | } | |
5d4f98a2 | 4803 | btrfs_set_extent_inline_ref_type(leaf, iref, type); |
d9a620f7 | 4804 | |
5d4f98a2 YZ |
4805 | if (parent > 0) { |
4806 | struct btrfs_shared_data_ref *ref; | |
4807 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4808 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4809 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4810 | } else { | |
4811 | struct btrfs_extent_data_ref *ref; | |
4812 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4813 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4814 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4815 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4816 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4817 | } | |
47e4bb98 | 4818 | |
50564b65 | 4819 | btrfs_mark_buffer_dirty(trans, path->nodes[0]); |
7bb86316 | 4820 | btrfs_free_path(path); |
f510cfec | 4821 | |
34666705 | 4822 | return alloc_reserved_extent(trans, ins->objectid, ins->offset); |
e6dcd2dc CM |
4823 | } |
4824 | ||
5d4f98a2 | 4825 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4826 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4827 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4828 | { |
9dcdbe01 | 4829 | struct btrfs_fs_info *fs_info = trans->fs_info; |
29cbcf40 | 4830 | struct btrfs_root *extent_root; |
e6dcd2dc | 4831 | int ret; |
5d4f98a2 | 4832 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4833 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4834 | struct btrfs_tree_block_info *block_info; |
4835 | struct btrfs_extent_inline_ref *iref; | |
4836 | struct btrfs_path *path; | |
4837 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4838 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4839 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
21ebfbe7 | 4840 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4841 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4842 | |
4e6bd4e0 NB |
4843 | ref = btrfs_delayed_node_to_tree_ref(node); |
4844 | ||
4e6bd4e0 NB |
4845 | extent_key.objectid = node->bytenr; |
4846 | if (skinny_metadata) { | |
4847 | extent_key.offset = ref->level; | |
4848 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4e6bd4e0 NB |
4849 | } else { |
4850 | extent_key.offset = node->num_bytes; | |
4851 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4852 | size += sizeof(*block_info); |
4e6bd4e0 | 4853 | } |
1c2308f8 | 4854 | |
5d4f98a2 | 4855 | path = btrfs_alloc_path(); |
80ee54bf | 4856 | if (!path) |
d8926bb3 | 4857 | return -ENOMEM; |
56bec294 | 4858 | |
29cbcf40 JB |
4859 | extent_root = btrfs_extent_root(fs_info, extent_key.objectid); |
4860 | ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key, | |
4861 | size); | |
79787eaa | 4862 | if (ret) { |
dd825259 | 4863 | btrfs_free_path(path); |
79787eaa JM |
4864 | return ret; |
4865 | } | |
5d4f98a2 YZ |
4866 | |
4867 | leaf = path->nodes[0]; | |
4868 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4869 | struct btrfs_extent_item); | |
4870 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4871 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4872 | btrfs_set_extent_flags(leaf, extent_item, | |
4873 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4874 | |
3173a18f JB |
4875 | if (skinny_metadata) { |
4876 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4877 | } else { | |
4878 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4879 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4880 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4881 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4882 | } | |
5d4f98a2 | 4883 | |
d4b20733 | 4884 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4885 | btrfs_set_extent_inline_ref_type(leaf, iref, |
4886 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4887 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4888 | } else { |
4889 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4890 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4891 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4892 | } |
4893 | ||
50564b65 | 4894 | btrfs_mark_buffer_dirty(trans, leaf); |
5d4f98a2 YZ |
4895 | btrfs_free_path(path); |
4896 | ||
34666705 | 4897 | return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize); |
5d4f98a2 YZ |
4898 | } |
4899 | ||
4900 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4901 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4902 | u64 offset, u64 ram_bytes, |
4903 | struct btrfs_key *ins) | |
5d4f98a2 | 4904 | { |
76675593 | 4905 | struct btrfs_ref generic_ref = { 0 }; |
457cb1dd BB |
4906 | u64 root_objectid = root->root_key.objectid; |
4907 | u64 owning_root = root_objectid; | |
5d4f98a2 | 4908 | |
457cb1dd | 4909 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4910 | |
2672a051 BB |
4911 | if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root)) |
4912 | owning_root = root->relocation_src_root; | |
4913 | ||
76675593 | 4914 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
457cb1dd BB |
4915 | ins->objectid, ins->offset, 0, owning_root); |
4916 | btrfs_init_data_ref(&generic_ref, root_objectid, owner, | |
f42c5da6 | 4917 | offset, 0, false); |
8a5040f7 | 4918 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
2187374f JB |
4919 | |
4920 | return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes); | |
e6dcd2dc | 4921 | } |
e02119d5 CM |
4922 | |
4923 | /* | |
4924 | * this is used by the tree logging recovery code. It records that | |
4925 | * an extent has been allocated and makes sure to clear the free | |
4926 | * space cache bits as well | |
4927 | */ | |
5d4f98a2 | 4928 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4929 | u64 root_objectid, u64 owner, u64 offset, |
4930 | struct btrfs_key *ins) | |
e02119d5 | 4931 | { |
61da2abf | 4932 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4933 | int ret; |
32da5386 | 4934 | struct btrfs_block_group *block_group; |
ed7a6948 | 4935 | struct btrfs_space_info *space_info; |
cecbb533 BB |
4936 | struct btrfs_squota_delta delta = { |
4937 | .root = root_objectid, | |
4938 | .num_bytes = ins->offset, | |
bd7c1ea3 | 4939 | .generation = trans->transid, |
cecbb533 BB |
4940 | .rsv_bytes = 0, |
4941 | .is_data = true, | |
4942 | .is_inc = true, | |
4943 | }; | |
11833d66 | 4944 | |
8c2a1a30 JB |
4945 | /* |
4946 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4947 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4948 | */ |
0b246afa | 4949 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4950 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4951 | ins->offset); | |
b50c6e25 | 4952 | if (ret) |
8c2a1a30 | 4953 | return ret; |
11833d66 YZ |
4954 | } |
4955 | ||
0b246afa | 4956 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4957 | if (!block_group) |
4958 | return -EINVAL; | |
4959 | ||
ed7a6948 WX |
4960 | space_info = block_group->space_info; |
4961 | spin_lock(&space_info->lock); | |
4962 | spin_lock(&block_group->lock); | |
4963 | space_info->bytes_reserved += ins->offset; | |
4964 | block_group->reserved += ins->offset; | |
4965 | spin_unlock(&block_group->lock); | |
4966 | spin_unlock(&space_info->lock); | |
4967 | ||
ef89b824 | 4968 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
2672a051 | 4969 | offset, ins, 1, root_objectid); |
bd727173 | 4970 | if (ret) |
ab9b2c7b | 4971 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
cecbb533 | 4972 | ret = btrfs_record_squota_delta(fs_info, &delta); |
b50c6e25 | 4973 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4974 | return ret; |
4975 | } | |
4976 | ||
150cce2d DS |
4977 | #ifdef CONFIG_BTRFS_DEBUG |
4978 | /* | |
4979 | * Extra safety check in case the extent tree is corrupted and extent allocator | |
4980 | * chooses to use a tree block which is already used and locked. | |
4981 | */ | |
4982 | static bool check_eb_lock_owner(const struct extent_buffer *eb) | |
4983 | { | |
4984 | if (eb->lock_owner == current->pid) { | |
4985 | btrfs_err_rl(eb->fs_info, | |
4986 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4987 | eb->start, btrfs_header_owner(eb), current->pid); | |
4988 | return true; | |
4989 | } | |
4990 | return false; | |
4991 | } | |
4992 | #else | |
4993 | static bool check_eb_lock_owner(struct extent_buffer *eb) | |
4994 | { | |
4995 | return false; | |
4996 | } | |
4997 | #endif | |
4998 | ||
48a3b636 ES |
4999 | static struct extent_buffer * |
5000 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
9631e4cc JB |
5001 | u64 bytenr, int level, u64 owner, |
5002 | enum btrfs_lock_nesting nest) | |
65b51a00 | 5003 | { |
0b246afa | 5004 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 | 5005 | struct extent_buffer *buf; |
b40130b2 | 5006 | u64 lockdep_owner = owner; |
65b51a00 | 5007 | |
3fbaf258 | 5008 | buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level); |
c871b0f2 LB |
5009 | if (IS_ERR(buf)) |
5010 | return buf; | |
5011 | ||
150cce2d | 5012 | if (check_eb_lock_owner(buf)) { |
b72c3aba QW |
5013 | free_extent_buffer(buf); |
5014 | return ERR_PTR(-EUCLEAN); | |
5015 | } | |
5016 | ||
b40130b2 JB |
5017 | /* |
5018 | * The reloc trees are just snapshots, so we need them to appear to be | |
5019 | * just like any other fs tree WRT lockdep. | |
5020 | * | |
5021 | * The exception however is in replace_path() in relocation, where we | |
5022 | * hold the lock on the original fs root and then search for the reloc | |
5023 | * root. At that point we need to make sure any reloc root buffers are | |
5024 | * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make | |
5025 | * lockdep happy. | |
5026 | */ | |
5027 | if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID && | |
5028 | !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state)) | |
5029 | lockdep_owner = BTRFS_FS_TREE_OBJECTID; | |
5030 | ||
618d1d7d | 5031 | /* btrfs_clear_buffer_dirty() accesses generation field. */ |
cbddcc4f TH |
5032 | btrfs_set_header_generation(buf, trans->transid); |
5033 | ||
e114c545 JB |
5034 | /* |
5035 | * This needs to stay, because we could allocate a freed block from an | |
5036 | * old tree into a new tree, so we need to make sure this new block is | |
5037 | * set to the appropriate level and owner. | |
5038 | */ | |
b40130b2 JB |
5039 | btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level); |
5040 | ||
9631e4cc | 5041 | __btrfs_tree_lock(buf, nest); |
190a8339 | 5042 | btrfs_clear_buffer_dirty(trans, buf); |
3083ee2e | 5043 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
d3575156 | 5044 | clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags); |
b4ce94de | 5045 | |
4db8c528 | 5046 | set_extent_buffer_uptodate(buf); |
b4ce94de | 5047 | |
bc877d28 NB |
5048 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
5049 | btrfs_set_header_level(buf, level); | |
5050 | btrfs_set_header_bytenr(buf, buf->start); | |
5051 | btrfs_set_header_generation(buf, trans->transid); | |
5052 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
5053 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 5054 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 5055 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 5056 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 5057 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
5058 | /* |
5059 | * we allow two log transactions at a time, use different | |
52042d8e | 5060 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 5061 | */ |
656f30db | 5062 | if (buf->log_index == 0) |
fe1a598c DS |
5063 | set_extent_bit(&root->dirty_log_pages, buf->start, |
5064 | buf->start + buf->len - 1, | |
1d126800 | 5065 | EXTENT_DIRTY, NULL); |
8cef4e16 | 5066 | else |
eea8686e DS |
5067 | set_extent_bit(&root->dirty_log_pages, buf->start, |
5068 | buf->start + buf->len - 1, | |
1d126800 | 5069 | EXTENT_NEW, NULL); |
d0c803c4 | 5070 | } else { |
656f30db | 5071 | buf->log_index = -1; |
fe1a598c | 5072 | set_extent_bit(&trans->transaction->dirty_pages, buf->start, |
1d126800 | 5073 | buf->start + buf->len - 1, EXTENT_DIRTY, NULL); |
d0c803c4 | 5074 | } |
b4ce94de | 5075 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
5076 | return buf; |
5077 | } | |
5078 | ||
fec577fb | 5079 | /* |
f0486c68 | 5080 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 5081 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 5082 | */ |
4d75f8a9 | 5083 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
5084 | struct btrfs_root *root, |
5085 | u64 parent, u64 root_objectid, | |
5086 | const struct btrfs_disk_key *key, | |
5087 | int level, u64 hint, | |
9631e4cc | 5088 | u64 empty_size, |
60ea105a | 5089 | u64 reloc_src_root, |
9631e4cc | 5090 | enum btrfs_lock_nesting nest) |
fec577fb | 5091 | { |
0b246afa | 5092 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 5093 | struct btrfs_key ins; |
f0486c68 | 5094 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 5095 | struct extent_buffer *buf; |
67b7859e | 5096 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 5097 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
5098 | u64 flags = 0; |
5099 | int ret; | |
0b246afa JM |
5100 | u32 blocksize = fs_info->nodesize; |
5101 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
60ea105a | 5102 | u64 owning_root; |
fec577fb | 5103 | |
05653ef3 | 5104 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 5105 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 5106 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
9631e4cc | 5107 | level, root_objectid, nest); |
faa2dbf0 JB |
5108 | if (!IS_ERR(buf)) |
5109 | root->alloc_bytenr += blocksize; | |
5110 | return buf; | |
5111 | } | |
05653ef3 | 5112 | #endif |
fccb84c9 | 5113 | |
67f9c220 | 5114 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
5115 | if (IS_ERR(block_rsv)) |
5116 | return ERR_CAST(block_rsv); | |
5117 | ||
18513091 | 5118 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 5119 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
5120 | if (ret) |
5121 | goto out_unuse; | |
55c69072 | 5122 | |
bc877d28 | 5123 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
9631e4cc | 5124 | root_objectid, nest); |
67b7859e OS |
5125 | if (IS_ERR(buf)) { |
5126 | ret = PTR_ERR(buf); | |
5127 | goto out_free_reserved; | |
5128 | } | |
60ea105a | 5129 | owning_root = btrfs_header_owner(buf); |
f0486c68 YZ |
5130 | |
5131 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
5132 | if (parent == 0) | |
5133 | parent = ins.objectid; | |
5134 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
60ea105a | 5135 | owning_root = reloc_src_root; |
f0486c68 YZ |
5136 | } else |
5137 | BUG_ON(parent > 0); | |
5138 | ||
5139 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 5140 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
5141 | if (!extent_op) { |
5142 | ret = -ENOMEM; | |
5143 | goto out_free_buf; | |
5144 | } | |
f0486c68 YZ |
5145 | if (key) |
5146 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
5147 | else | |
5148 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
5149 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
5150 | extent_op->update_key = skinny_metadata ? false : true; |
5151 | extent_op->update_flags = true; | |
b1c79e09 | 5152 | extent_op->level = level; |
f0486c68 | 5153 | |
ed4f255b | 5154 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
60ea105a | 5155 | ins.objectid, ins.offset, parent, owning_root); |
f42c5da6 NB |
5156 | btrfs_init_tree_ref(&generic_ref, level, root_objectid, |
5157 | root->root_key.objectid, false); | |
8a5040f7 | 5158 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 5159 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op); |
67b7859e OS |
5160 | if (ret) |
5161 | goto out_free_delayed; | |
f0486c68 | 5162 | } |
fec577fb | 5163 | return buf; |
67b7859e OS |
5164 | |
5165 | out_free_delayed: | |
5166 | btrfs_free_delayed_extent_op(extent_op); | |
5167 | out_free_buf: | |
19ea40dd | 5168 | btrfs_tree_unlock(buf); |
67b7859e OS |
5169 | free_extent_buffer(buf); |
5170 | out_free_reserved: | |
2ff7e61e | 5171 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 5172 | out_unuse: |
67f9c220 | 5173 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 5174 | return ERR_PTR(ret); |
fec577fb | 5175 | } |
a28ec197 | 5176 | |
2c47e605 YZ |
5177 | struct walk_control { |
5178 | u64 refs[BTRFS_MAX_LEVEL]; | |
5179 | u64 flags[BTRFS_MAX_LEVEL]; | |
5180 | struct btrfs_key update_progress; | |
aea6f028 JB |
5181 | struct btrfs_key drop_progress; |
5182 | int drop_level; | |
2c47e605 YZ |
5183 | int stage; |
5184 | int level; | |
5185 | int shared_level; | |
5186 | int update_ref; | |
5187 | int keep_locks; | |
1c4850e2 YZ |
5188 | int reada_slot; |
5189 | int reada_count; | |
78c52d9e | 5190 | int restarted; |
2c47e605 YZ |
5191 | }; |
5192 | ||
5193 | #define DROP_REFERENCE 1 | |
5194 | #define UPDATE_BACKREF 2 | |
5195 | ||
1c4850e2 YZ |
5196 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
5197 | struct btrfs_root *root, | |
5198 | struct walk_control *wc, | |
5199 | struct btrfs_path *path) | |
6407bf6d | 5200 | { |
0b246afa | 5201 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
5202 | u64 bytenr; |
5203 | u64 generation; | |
5204 | u64 refs; | |
94fcca9f | 5205 | u64 flags; |
5d4f98a2 | 5206 | u32 nritems; |
1c4850e2 YZ |
5207 | struct btrfs_key key; |
5208 | struct extent_buffer *eb; | |
6407bf6d | 5209 | int ret; |
1c4850e2 YZ |
5210 | int slot; |
5211 | int nread = 0; | |
6407bf6d | 5212 | |
1c4850e2 YZ |
5213 | if (path->slots[wc->level] < wc->reada_slot) { |
5214 | wc->reada_count = wc->reada_count * 2 / 3; | |
5215 | wc->reada_count = max(wc->reada_count, 2); | |
5216 | } else { | |
5217 | wc->reada_count = wc->reada_count * 3 / 2; | |
5218 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 5219 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 5220 | } |
7bb86316 | 5221 | |
1c4850e2 YZ |
5222 | eb = path->nodes[wc->level]; |
5223 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 5224 | |
1c4850e2 YZ |
5225 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5226 | if (nread >= wc->reada_count) | |
5227 | break; | |
bd56b302 | 5228 | |
2dd3e67b | 5229 | cond_resched(); |
1c4850e2 YZ |
5230 | bytenr = btrfs_node_blockptr(eb, slot); |
5231 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5232 | |
1c4850e2 YZ |
5233 | if (slot == path->slots[wc->level]) |
5234 | goto reada; | |
5d4f98a2 | 5235 | |
1c4850e2 YZ |
5236 | if (wc->stage == UPDATE_BACKREF && |
5237 | generation <= root->root_key.offset) | |
bd56b302 CM |
5238 | continue; |
5239 | ||
94fcca9f | 5240 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 5241 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f | 5242 | wc->level - 1, 1, &refs, |
d8ba2a91 | 5243 | &flags, NULL); |
79787eaa JM |
5244 | /* We don't care about errors in readahead. */ |
5245 | if (ret < 0) | |
5246 | continue; | |
94fcca9f YZ |
5247 | BUG_ON(refs == 0); |
5248 | ||
1c4850e2 | 5249 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5250 | if (refs == 1) |
5251 | goto reada; | |
bd56b302 | 5252 | |
94fcca9f YZ |
5253 | if (wc->level == 1 && |
5254 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5255 | continue; | |
1c4850e2 YZ |
5256 | if (!wc->update_ref || |
5257 | generation <= root->root_key.offset) | |
5258 | continue; | |
5259 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5260 | ret = btrfs_comp_cpu_keys(&key, | |
5261 | &wc->update_progress); | |
5262 | if (ret < 0) | |
5263 | continue; | |
94fcca9f YZ |
5264 | } else { |
5265 | if (wc->level == 1 && | |
5266 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5267 | continue; | |
6407bf6d | 5268 | } |
1c4850e2 | 5269 | reada: |
bfb484d9 | 5270 | btrfs_readahead_node_child(eb, slot); |
1c4850e2 | 5271 | nread++; |
20524f02 | 5272 | } |
1c4850e2 | 5273 | wc->reada_slot = slot; |
20524f02 | 5274 | } |
2c47e605 | 5275 | |
f82d02d9 | 5276 | /* |
2c016dc2 | 5277 | * helper to process tree block while walking down the tree. |
2c47e605 | 5278 | * |
2c47e605 YZ |
5279 | * when wc->stage == UPDATE_BACKREF, this function updates |
5280 | * back refs for pointers in the block. | |
5281 | * | |
5282 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5283 | */ |
2c47e605 | 5284 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5285 | struct btrfs_root *root, |
2c47e605 | 5286 | struct btrfs_path *path, |
94fcca9f | 5287 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5288 | { |
2ff7e61e | 5289 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
5290 | int level = wc->level; |
5291 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5292 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5293 | int ret; |
5294 | ||
2c47e605 YZ |
5295 | if (wc->stage == UPDATE_BACKREF && |
5296 | btrfs_header_owner(eb) != root->root_key.objectid) | |
5297 | return 1; | |
f82d02d9 | 5298 | |
2c47e605 YZ |
5299 | /* |
5300 | * when reference count of tree block is 1, it won't increase | |
5301 | * again. once full backref flag is set, we never clear it. | |
5302 | */ | |
94fcca9f YZ |
5303 | if (lookup_info && |
5304 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5305 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 5306 | BUG_ON(!path->locks[level]); |
2ff7e61e | 5307 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5308 | eb->start, level, 1, |
2c47e605 | 5309 | &wc->refs[level], |
d8ba2a91 JB |
5310 | &wc->flags[level], |
5311 | NULL); | |
79787eaa JM |
5312 | BUG_ON(ret == -ENOMEM); |
5313 | if (ret) | |
5314 | return ret; | |
2c47e605 YZ |
5315 | BUG_ON(wc->refs[level] == 0); |
5316 | } | |
5d4f98a2 | 5317 | |
2c47e605 YZ |
5318 | if (wc->stage == DROP_REFERENCE) { |
5319 | if (wc->refs[level] > 1) | |
5320 | return 1; | |
f82d02d9 | 5321 | |
2c47e605 | 5322 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 5323 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5324 | path->locks[level] = 0; |
5325 | } | |
5326 | return 0; | |
5327 | } | |
f82d02d9 | 5328 | |
2c47e605 YZ |
5329 | /* wc->stage == UPDATE_BACKREF */ |
5330 | if (!(wc->flags[level] & flag)) { | |
5331 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 5332 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 5333 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 5334 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5335 | BUG_ON(ret); /* -ENOMEM */ |
4aec05fa | 5336 | ret = btrfs_set_disk_extent_flags(trans, eb, flag); |
79787eaa | 5337 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
5338 | wc->flags[level] |= flag; |
5339 | } | |
5340 | ||
5341 | /* | |
5342 | * the block is shared by multiple trees, so it's not good to | |
5343 | * keep the tree lock | |
5344 | */ | |
5345 | if (path->locks[level] && level > 0) { | |
bd681513 | 5346 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5347 | path->locks[level] = 0; |
5348 | } | |
5349 | return 0; | |
5350 | } | |
5351 | ||
78c52d9e JB |
5352 | /* |
5353 | * This is used to verify a ref exists for this root to deal with a bug where we | |
5354 | * would have a drop_progress key that hadn't been updated properly. | |
5355 | */ | |
5356 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
5357 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
5358 | int level) | |
5359 | { | |
5360 | struct btrfs_path *path; | |
5361 | struct btrfs_extent_inline_ref *iref; | |
5362 | int ret; | |
5363 | ||
5364 | path = btrfs_alloc_path(); | |
5365 | if (!path) | |
5366 | return -ENOMEM; | |
5367 | ||
5368 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
5369 | root->fs_info->nodesize, parent, | |
5370 | root->root_key.objectid, level, 0); | |
5371 | btrfs_free_path(path); | |
5372 | if (ret == -ENOENT) | |
5373 | return 0; | |
5374 | if (ret < 0) | |
5375 | return ret; | |
5376 | return 1; | |
5377 | } | |
5378 | ||
1c4850e2 | 5379 | /* |
2c016dc2 | 5380 | * helper to process tree block pointer. |
1c4850e2 YZ |
5381 | * |
5382 | * when wc->stage == DROP_REFERENCE, this function checks | |
5383 | * reference count of the block pointed to. if the block | |
5384 | * is shared and we need update back refs for the subtree | |
5385 | * rooted at the block, this function changes wc->stage to | |
5386 | * UPDATE_BACKREF. if the block is shared and there is no | |
5387 | * need to update back, this function drops the reference | |
5388 | * to the block. | |
5389 | * | |
5390 | * NOTE: return value 1 means we should stop walking down. | |
5391 | */ | |
5392 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5393 | struct btrfs_root *root, | |
5394 | struct btrfs_path *path, | |
94fcca9f | 5395 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 5396 | { |
0b246afa | 5397 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
5398 | u64 bytenr; |
5399 | u64 generation; | |
5400 | u64 parent; | |
d8ba2a91 | 5401 | u64 owner_root = 0; |
789d6a3a | 5402 | struct btrfs_tree_parent_check check = { 0 }; |
1c4850e2 | 5403 | struct btrfs_key key; |
ffd4bb2a | 5404 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
5405 | struct extent_buffer *next; |
5406 | int level = wc->level; | |
5407 | int reada = 0; | |
5408 | int ret = 0; | |
1152651a | 5409 | bool need_account = false; |
1c4850e2 YZ |
5410 | |
5411 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5412 | path->slots[level]); | |
5413 | /* | |
5414 | * if the lower level block was created before the snapshot | |
5415 | * was created, we know there is no need to update back refs | |
5416 | * for the subtree | |
5417 | */ | |
5418 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5419 | generation <= root->root_key.offset) { |
5420 | *lookup_info = 1; | |
1c4850e2 | 5421 | return 1; |
94fcca9f | 5422 | } |
1c4850e2 YZ |
5423 | |
5424 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
789d6a3a QW |
5425 | |
5426 | check.level = level - 1; | |
5427 | check.transid = generation; | |
5428 | check.owner_root = root->root_key.objectid; | |
5429 | check.has_first_key = true; | |
5430 | btrfs_node_key_to_cpu(path->nodes[level], &check.first_key, | |
581c1760 | 5431 | path->slots[level]); |
1c4850e2 | 5432 | |
0b246afa | 5433 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5434 | if (!next) { |
3fbaf258 JB |
5435 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
5436 | root->root_key.objectid, level - 1); | |
c871b0f2 LB |
5437 | if (IS_ERR(next)) |
5438 | return PTR_ERR(next); | |
1c4850e2 YZ |
5439 | reada = 1; |
5440 | } | |
5441 | btrfs_tree_lock(next); | |
1c4850e2 | 5442 | |
2ff7e61e | 5443 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f | 5444 | &wc->refs[level - 1], |
d8ba2a91 JB |
5445 | &wc->flags[level - 1], |
5446 | &owner_root); | |
4867268c JB |
5447 | if (ret < 0) |
5448 | goto out_unlock; | |
79787eaa | 5449 | |
c2cf52eb | 5450 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5451 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5452 | ret = -EIO; |
5453 | goto out_unlock; | |
c2cf52eb | 5454 | } |
94fcca9f | 5455 | *lookup_info = 0; |
1c4850e2 | 5456 | |
94fcca9f | 5457 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5458 | if (wc->refs[level - 1] > 1) { |
1152651a | 5459 | need_account = true; |
94fcca9f YZ |
5460 | if (level == 1 && |
5461 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5462 | goto skip; | |
5463 | ||
1c4850e2 YZ |
5464 | if (!wc->update_ref || |
5465 | generation <= root->root_key.offset) | |
5466 | goto skip; | |
5467 | ||
5468 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5469 | path->slots[level]); | |
5470 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5471 | if (ret < 0) | |
5472 | goto skip; | |
5473 | ||
5474 | wc->stage = UPDATE_BACKREF; | |
5475 | wc->shared_level = level - 1; | |
5476 | } | |
94fcca9f YZ |
5477 | } else { |
5478 | if (level == 1 && | |
5479 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5480 | goto skip; | |
1c4850e2 YZ |
5481 | } |
5482 | ||
b9fab919 | 5483 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5484 | btrfs_tree_unlock(next); |
5485 | free_extent_buffer(next); | |
5486 | next = NULL; | |
94fcca9f | 5487 | *lookup_info = 1; |
1c4850e2 YZ |
5488 | } |
5489 | ||
5490 | if (!next) { | |
5491 | if (reada && level == 1) | |
5492 | reada_walk_down(trans, root, wc, path); | |
789d6a3a | 5493 | next = read_tree_block(fs_info, bytenr, &check); |
64c043de LB |
5494 | if (IS_ERR(next)) { |
5495 | return PTR_ERR(next); | |
5496 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5497 | free_extent_buffer(next); |
97d9a8a4 | 5498 | return -EIO; |
416bc658 | 5499 | } |
1c4850e2 | 5500 | btrfs_tree_lock(next); |
1c4850e2 YZ |
5501 | } |
5502 | ||
5503 | level--; | |
4867268c JB |
5504 | ASSERT(level == btrfs_header_level(next)); |
5505 | if (level != btrfs_header_level(next)) { | |
5506 | btrfs_err(root->fs_info, "mismatched level"); | |
5507 | ret = -EIO; | |
5508 | goto out_unlock; | |
5509 | } | |
1c4850e2 YZ |
5510 | path->nodes[level] = next; |
5511 | path->slots[level] = 0; | |
ac5887c8 | 5512 | path->locks[level] = BTRFS_WRITE_LOCK; |
1c4850e2 YZ |
5513 | wc->level = level; |
5514 | if (wc->level == 1) | |
5515 | wc->reada_slot = 0; | |
5516 | return 0; | |
5517 | skip: | |
5518 | wc->refs[level - 1] = 0; | |
5519 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5520 | if (wc->stage == DROP_REFERENCE) { |
5521 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5522 | parent = path->nodes[level]->start; | |
5523 | } else { | |
4867268c | 5524 | ASSERT(root->root_key.objectid == |
94fcca9f | 5525 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5526 | if (root->root_key.objectid != |
5527 | btrfs_header_owner(path->nodes[level])) { | |
5528 | btrfs_err(root->fs_info, | |
5529 | "mismatched block owner"); | |
5530 | ret = -EIO; | |
5531 | goto out_unlock; | |
5532 | } | |
94fcca9f YZ |
5533 | parent = 0; |
5534 | } | |
1c4850e2 | 5535 | |
78c52d9e JB |
5536 | /* |
5537 | * If we had a drop_progress we need to verify the refs are set | |
5538 | * as expected. If we find our ref then we know that from here | |
5539 | * on out everything should be correct, and we can clear the | |
5540 | * ->restarted flag. | |
5541 | */ | |
5542 | if (wc->restarted) { | |
5543 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5544 | level - 1); | |
5545 | if (ret < 0) | |
5546 | goto out_unlock; | |
5547 | if (ret == 0) | |
5548 | goto no_delete; | |
5549 | ret = 0; | |
5550 | wc->restarted = 0; | |
5551 | } | |
5552 | ||
2cd86d30 QW |
5553 | /* |
5554 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5555 | * already accounted them at merge time (replace_path), | |
5556 | * thus we could skip expensive subtree trace here. | |
5557 | */ | |
5558 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5559 | need_account) { | |
deb40627 | 5560 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5561 | generation, level - 1); |
1152651a | 5562 | if (ret) { |
0b246afa | 5563 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5564 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5565 | ret); | |
1152651a MF |
5566 | } |
5567 | } | |
aea6f028 JB |
5568 | |
5569 | /* | |
5570 | * We need to update the next key in our walk control so we can | |
5571 | * update the drop_progress key accordingly. We don't care if | |
5572 | * find_next_key doesn't find a key because that means we're at | |
5573 | * the end and are going to clean up now. | |
5574 | */ | |
5575 | wc->drop_level = level; | |
5576 | find_next_key(path, level, &wc->drop_progress); | |
5577 | ||
ffd4bb2a | 5578 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
d8ba2a91 | 5579 | fs_info->nodesize, parent, owner_root); |
f42c5da6 NB |
5580 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid, |
5581 | 0, false); | |
ffd4bb2a | 5582 | ret = btrfs_free_extent(trans, &ref); |
4867268c JB |
5583 | if (ret) |
5584 | goto out_unlock; | |
1c4850e2 | 5585 | } |
78c52d9e | 5586 | no_delete: |
4867268c JB |
5587 | *lookup_info = 1; |
5588 | ret = 1; | |
5589 | ||
5590 | out_unlock: | |
1c4850e2 YZ |
5591 | btrfs_tree_unlock(next); |
5592 | free_extent_buffer(next); | |
4867268c JB |
5593 | |
5594 | return ret; | |
1c4850e2 YZ |
5595 | } |
5596 | ||
2c47e605 | 5597 | /* |
2c016dc2 | 5598 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5599 | * |
5600 | * when wc->stage == DROP_REFERENCE, this function drops | |
5601 | * reference count on the block. | |
5602 | * | |
5603 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5604 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5605 | * to UPDATE_BACKREF previously while processing the block. | |
5606 | * | |
5607 | * NOTE: return value 1 means we should stop walking up. | |
5608 | */ | |
5609 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5610 | struct btrfs_root *root, | |
5611 | struct btrfs_path *path, | |
5612 | struct walk_control *wc) | |
5613 | { | |
0b246afa | 5614 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5615 | int ret; |
2c47e605 YZ |
5616 | int level = wc->level; |
5617 | struct extent_buffer *eb = path->nodes[level]; | |
5618 | u64 parent = 0; | |
5619 | ||
5620 | if (wc->stage == UPDATE_BACKREF) { | |
5621 | BUG_ON(wc->shared_level < level); | |
5622 | if (level < wc->shared_level) | |
5623 | goto out; | |
5624 | ||
2c47e605 YZ |
5625 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5626 | if (ret > 0) | |
5627 | wc->update_ref = 0; | |
5628 | ||
5629 | wc->stage = DROP_REFERENCE; | |
5630 | wc->shared_level = -1; | |
5631 | path->slots[level] = 0; | |
5632 | ||
5633 | /* | |
5634 | * check reference count again if the block isn't locked. | |
5635 | * we should start walking down the tree again if reference | |
5636 | * count is one. | |
5637 | */ | |
5638 | if (!path->locks[level]) { | |
5639 | BUG_ON(level == 0); | |
5640 | btrfs_tree_lock(eb); | |
ac5887c8 | 5641 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5642 | |
2ff7e61e | 5643 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5644 | eb->start, level, 1, |
2c47e605 | 5645 | &wc->refs[level], |
d8ba2a91 JB |
5646 | &wc->flags[level], |
5647 | NULL); | |
79787eaa JM |
5648 | if (ret < 0) { |
5649 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5650 | path->locks[level] = 0; |
79787eaa JM |
5651 | return ret; |
5652 | } | |
2c47e605 YZ |
5653 | BUG_ON(wc->refs[level] == 0); |
5654 | if (wc->refs[level] == 1) { | |
bd681513 | 5655 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5656 | path->locks[level] = 0; |
2c47e605 YZ |
5657 | return 1; |
5658 | } | |
f82d02d9 | 5659 | } |
2c47e605 | 5660 | } |
f82d02d9 | 5661 | |
2c47e605 YZ |
5662 | /* wc->stage == DROP_REFERENCE */ |
5663 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5664 | |
2c47e605 YZ |
5665 | if (wc->refs[level] == 1) { |
5666 | if (level == 0) { | |
5667 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5668 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5669 | else |
e339a6b0 | 5670 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5671 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5672 | if (is_fstree(root->root_key.objectid)) { |
5673 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5674 | if (ret) { | |
5675 | btrfs_err_rl(fs_info, | |
5676 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5677 | ret); |
c4140cbf | 5678 | } |
1152651a | 5679 | } |
2c47e605 | 5680 | } |
190a8339 | 5681 | /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */ |
d3fb6615 | 5682 | if (!path->locks[level]) { |
2c47e605 | 5683 | btrfs_tree_lock(eb); |
ac5887c8 | 5684 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5685 | } |
190a8339 | 5686 | btrfs_clear_buffer_dirty(trans, eb); |
2c47e605 YZ |
5687 | } |
5688 | ||
5689 | if (eb == root->node) { | |
5690 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5691 | parent = eb->start; | |
65c6e82b QW |
5692 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5693 | goto owner_mismatch; | |
2c47e605 YZ |
5694 | } else { |
5695 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5696 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5697 | else if (root->root_key.objectid != |
5698 | btrfs_header_owner(path->nodes[level + 1])) | |
5699 | goto owner_mismatch; | |
f82d02d9 | 5700 | } |
f82d02d9 | 5701 | |
7a163608 FM |
5702 | btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent, |
5703 | wc->refs[level] == 1); | |
2c47e605 YZ |
5704 | out: |
5705 | wc->refs[level] = 0; | |
5706 | wc->flags[level] = 0; | |
f0486c68 | 5707 | return 0; |
65c6e82b QW |
5708 | |
5709 | owner_mismatch: | |
5710 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5711 | btrfs_header_owner(eb), root->root_key.objectid); | |
5712 | return -EUCLEAN; | |
2c47e605 YZ |
5713 | } |
5714 | ||
5715 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5716 | struct btrfs_root *root, | |
5717 | struct btrfs_path *path, | |
5718 | struct walk_control *wc) | |
5719 | { | |
2c47e605 | 5720 | int level = wc->level; |
94fcca9f | 5721 | int lookup_info = 1; |
4e194384 | 5722 | int ret = 0; |
2c47e605 YZ |
5723 | |
5724 | while (level >= 0) { | |
94fcca9f | 5725 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
4e194384 | 5726 | if (ret) |
2c47e605 YZ |
5727 | break; |
5728 | ||
5729 | if (level == 0) | |
5730 | break; | |
5731 | ||
7a7965f8 YZ |
5732 | if (path->slots[level] >= |
5733 | btrfs_header_nritems(path->nodes[level])) | |
5734 | break; | |
5735 | ||
94fcca9f | 5736 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5737 | if (ret > 0) { |
5738 | path->slots[level]++; | |
5739 | continue; | |
90d2c51d | 5740 | } else if (ret < 0) |
4e194384 | 5741 | break; |
1c4850e2 | 5742 | level = wc->level; |
f82d02d9 | 5743 | } |
4e194384 | 5744 | return (ret == 1) ? 0 : ret; |
f82d02d9 YZ |
5745 | } |
5746 | ||
d397712b | 5747 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5748 | struct btrfs_root *root, |
f82d02d9 | 5749 | struct btrfs_path *path, |
2c47e605 | 5750 | struct walk_control *wc, int max_level) |
20524f02 | 5751 | { |
2c47e605 | 5752 | int level = wc->level; |
20524f02 | 5753 | int ret; |
9f3a7427 | 5754 | |
2c47e605 YZ |
5755 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5756 | while (level < max_level && path->nodes[level]) { | |
5757 | wc->level = level; | |
5758 | if (path->slots[level] + 1 < | |
5759 | btrfs_header_nritems(path->nodes[level])) { | |
5760 | path->slots[level]++; | |
20524f02 CM |
5761 | return 0; |
5762 | } else { | |
2c47e605 YZ |
5763 | ret = walk_up_proc(trans, root, path, wc); |
5764 | if (ret > 0) | |
5765 | return 0; | |
65c6e82b QW |
5766 | if (ret < 0) |
5767 | return ret; | |
bd56b302 | 5768 | |
2c47e605 | 5769 | if (path->locks[level]) { |
bd681513 CM |
5770 | btrfs_tree_unlock_rw(path->nodes[level], |
5771 | path->locks[level]); | |
2c47e605 | 5772 | path->locks[level] = 0; |
f82d02d9 | 5773 | } |
2c47e605 YZ |
5774 | free_extent_buffer(path->nodes[level]); |
5775 | path->nodes[level] = NULL; | |
5776 | level++; | |
20524f02 CM |
5777 | } |
5778 | } | |
5779 | return 1; | |
5780 | } | |
5781 | ||
9aca1d51 | 5782 | /* |
2c47e605 YZ |
5783 | * drop a subvolume tree. |
5784 | * | |
5785 | * this function traverses the tree freeing any blocks that only | |
5786 | * referenced by the tree. | |
5787 | * | |
5788 | * when a shared tree block is found. this function decreases its | |
5789 | * reference count by one. if update_ref is true, this function | |
5790 | * also make sure backrefs for the shared block and all lower level | |
5791 | * blocks are properly updated. | |
9d1a2a3a DS |
5792 | * |
5793 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5794 | */ |
0078a9f9 | 5795 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5796 | { |
12a824dc FM |
5797 | const bool is_reloc_root = (root->root_key.objectid == |
5798 | BTRFS_TREE_RELOC_OBJECTID); | |
ab8d0fc4 | 5799 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5800 | struct btrfs_path *path; |
2c47e605 | 5801 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5802 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5803 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5804 | struct walk_control *wc; |
5805 | struct btrfs_key key; | |
5806 | int err = 0; | |
5807 | int ret; | |
5808 | int level; | |
d29a9f62 | 5809 | bool root_dropped = false; |
b4be6aef | 5810 | bool unfinished_drop = false; |
20524f02 | 5811 | |
4fd786e6 | 5812 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5813 | |
5caf2a00 | 5814 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5815 | if (!path) { |
5816 | err = -ENOMEM; | |
5817 | goto out; | |
5818 | } | |
20524f02 | 5819 | |
2c47e605 | 5820 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5821 | if (!wc) { |
5822 | btrfs_free_path(path); | |
cb1b69f4 TI |
5823 | err = -ENOMEM; |
5824 | goto out; | |
38a1a919 | 5825 | } |
2c47e605 | 5826 | |
f3e3d9cc QW |
5827 | /* |
5828 | * Use join to avoid potential EINTR from transaction start. See | |
5829 | * wait_reserve_ticket and the whole reservation callchain. | |
5830 | */ | |
5831 | if (for_reloc) | |
5832 | trans = btrfs_join_transaction(tree_root); | |
5833 | else | |
5834 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5835 | if (IS_ERR(trans)) { |
5836 | err = PTR_ERR(trans); | |
5837 | goto out_free; | |
5838 | } | |
98d5dc13 | 5839 | |
0568e82d JB |
5840 | err = btrfs_run_delayed_items(trans); |
5841 | if (err) | |
5842 | goto out_end_trans; | |
5843 | ||
83354f07 JB |
5844 | /* |
5845 | * This will help us catch people modifying the fs tree while we're | |
5846 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5847 | * dropped as we unlock the root node and parent nodes as we walk down | |
5848 | * the tree, assuming nothing will change. If something does change | |
5849 | * then we'll have stale information and drop references to blocks we've | |
5850 | * already dropped. | |
5851 | */ | |
5852 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
b4be6aef JB |
5853 | unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state); |
5854 | ||
9f3a7427 | 5855 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5856 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5857 | path->nodes[level] = btrfs_lock_root_node(root); |
9f3a7427 | 5858 | path->slots[level] = 0; |
ac5887c8 | 5859 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5860 | memset(&wc->update_progress, 0, |
5861 | sizeof(wc->update_progress)); | |
9f3a7427 | 5862 | } else { |
9f3a7427 | 5863 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5864 | memcpy(&wc->update_progress, &key, |
5865 | sizeof(wc->update_progress)); | |
5866 | ||
c8422684 | 5867 | level = btrfs_root_drop_level(root_item); |
2c47e605 | 5868 | BUG_ON(level == 0); |
6702ed49 | 5869 | path->lowest_level = level; |
2c47e605 YZ |
5870 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5871 | path->lowest_level = 0; | |
5872 | if (ret < 0) { | |
5873 | err = ret; | |
79787eaa | 5874 | goto out_end_trans; |
9f3a7427 | 5875 | } |
1c4850e2 | 5876 | WARN_ON(ret > 0); |
2c47e605 | 5877 | |
7d9eb12c CM |
5878 | /* |
5879 | * unlock our path, this is safe because only this | |
5880 | * function is allowed to delete this snapshot | |
5881 | */ | |
5d4f98a2 | 5882 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5883 | |
5884 | level = btrfs_header_level(root->node); | |
5885 | while (1) { | |
5886 | btrfs_tree_lock(path->nodes[level]); | |
ac5887c8 | 5887 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5888 | |
2ff7e61e | 5889 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5890 | path->nodes[level]->start, |
3173a18f | 5891 | level, 1, &wc->refs[level], |
d8ba2a91 | 5892 | &wc->flags[level], NULL); |
79787eaa JM |
5893 | if (ret < 0) { |
5894 | err = ret; | |
5895 | goto out_end_trans; | |
5896 | } | |
2c47e605 YZ |
5897 | BUG_ON(wc->refs[level] == 0); |
5898 | ||
c8422684 | 5899 | if (level == btrfs_root_drop_level(root_item)) |
2c47e605 YZ |
5900 | break; |
5901 | ||
5902 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5903 | path->locks[level] = 0; |
2c47e605 YZ |
5904 | WARN_ON(wc->refs[level] != 1); |
5905 | level--; | |
5906 | } | |
9f3a7427 | 5907 | } |
2c47e605 | 5908 | |
78c52d9e | 5909 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5910 | wc->level = level; |
5911 | wc->shared_level = -1; | |
5912 | wc->stage = DROP_REFERENCE; | |
5913 | wc->update_ref = update_ref; | |
5914 | wc->keep_locks = 0; | |
0b246afa | 5915 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5916 | |
d397712b | 5917 | while (1) { |
9d1a2a3a | 5918 | |
2c47e605 YZ |
5919 | ret = walk_down_tree(trans, root, path, wc); |
5920 | if (ret < 0) { | |
9a93b5a3 | 5921 | btrfs_abort_transaction(trans, ret); |
2c47e605 | 5922 | err = ret; |
20524f02 | 5923 | break; |
2c47e605 | 5924 | } |
9aca1d51 | 5925 | |
2c47e605 YZ |
5926 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5927 | if (ret < 0) { | |
9a93b5a3 | 5928 | btrfs_abort_transaction(trans, ret); |
2c47e605 | 5929 | err = ret; |
20524f02 | 5930 | break; |
2c47e605 YZ |
5931 | } |
5932 | ||
5933 | if (ret > 0) { | |
5934 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5935 | break; |
5936 | } | |
2c47e605 YZ |
5937 | |
5938 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5939 | wc->drop_level = wc->level; |
5940 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5941 | &wc->drop_progress, | |
5942 | path->slots[wc->drop_level]); | |
5943 | } | |
5944 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5945 | &wc->drop_progress); | |
c8422684 | 5946 | btrfs_set_root_drop_level(root_item, wc->drop_level); |
2c47e605 YZ |
5947 | |
5948 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5949 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5950 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5951 | ret = btrfs_update_root(trans, tree_root, |
5952 | &root->root_key, | |
5953 | root_item); | |
79787eaa | 5954 | if (ret) { |
66642832 | 5955 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5956 | err = ret; |
5957 | goto out_end_trans; | |
5958 | } | |
2c47e605 | 5959 | |
12a824dc FM |
5960 | if (!is_reloc_root) |
5961 | btrfs_set_last_root_drop_gen(fs_info, trans->transid); | |
5962 | ||
3a45bb20 | 5963 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5964 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5965 | btrfs_debug(fs_info, |
5966 | "drop snapshot early exit"); | |
3c8f2422 JB |
5967 | err = -EAGAIN; |
5968 | goto out_free; | |
5969 | } | |
5970 | ||
18d3bff4 JB |
5971 | /* |
5972 | * Use join to avoid potential EINTR from transaction | |
5973 | * start. See wait_reserve_ticket and the whole | |
5974 | * reservation callchain. | |
5975 | */ | |
5976 | if (for_reloc) | |
5977 | trans = btrfs_join_transaction(tree_root); | |
5978 | else | |
5979 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5980 | if (IS_ERR(trans)) { |
5981 | err = PTR_ERR(trans); | |
5982 | goto out_free; | |
5983 | } | |
c3e69d58 | 5984 | } |
20524f02 | 5985 | } |
b3b4aa74 | 5986 | btrfs_release_path(path); |
79787eaa JM |
5987 | if (err) |
5988 | goto out_end_trans; | |
2c47e605 | 5989 | |
ab9ce7d4 | 5990 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5991 | if (ret) { |
66642832 | 5992 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5993 | err = ret; |
79787eaa JM |
5994 | goto out_end_trans; |
5995 | } | |
2c47e605 | 5996 | |
12a824dc | 5997 | if (!is_reloc_root) { |
cb517eab MX |
5998 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5999 | NULL, NULL); | |
79787eaa | 6000 | if (ret < 0) { |
66642832 | 6001 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
6002 | err = ret; |
6003 | goto out_end_trans; | |
6004 | } else if (ret > 0) { | |
84cd948c JB |
6005 | /* if we fail to delete the orphan item this time |
6006 | * around, it'll get picked up the next time. | |
6007 | * | |
6008 | * The most common failure here is just -ENOENT. | |
6009 | */ | |
6010 | btrfs_del_orphan_item(trans, tree_root, | |
6011 | root->root_key.objectid); | |
76dda93c YZ |
6012 | } |
6013 | } | |
6014 | ||
a3cf0e43 QW |
6015 | /* |
6016 | * This subvolume is going to be completely dropped, and won't be | |
6017 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
6018 | * commit transaction time. So free it here manually. | |
6019 | */ | |
6020 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
6021 | btrfs_qgroup_free_meta_all_pertrans(root); | |
6022 | ||
fc7cbcd4 | 6023 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 6024 | btrfs_add_dropped_root(trans, root); |
8c38938c | 6025 | else |
00246528 | 6026 | btrfs_put_root(root); |
d29a9f62 | 6027 | root_dropped = true; |
79787eaa | 6028 | out_end_trans: |
12a824dc FM |
6029 | if (!is_reloc_root) |
6030 | btrfs_set_last_root_drop_gen(fs_info, trans->transid); | |
6031 | ||
3a45bb20 | 6032 | btrfs_end_transaction_throttle(trans); |
79787eaa | 6033 | out_free: |
2c47e605 | 6034 | kfree(wc); |
5caf2a00 | 6035 | btrfs_free_path(path); |
cb1b69f4 | 6036 | out: |
b4be6aef JB |
6037 | /* |
6038 | * We were an unfinished drop root, check to see if there are any | |
6039 | * pending, and if not clear and wake up any waiters. | |
6040 | */ | |
6041 | if (!err && unfinished_drop) | |
6042 | btrfs_maybe_wake_unfinished_drop(fs_info); | |
6043 | ||
d29a9f62 JB |
6044 | /* |
6045 | * So if we need to stop dropping the snapshot for whatever reason we | |
6046 | * need to make sure to add it back to the dead root list so that we | |
6047 | * keep trying to do the work later. This also cleans up roots if we | |
6048 | * don't have it in the radix (like when we recover after a power fail | |
6049 | * or unmount) so we don't leak memory. | |
6050 | */ | |
897ca819 | 6051 | if (!for_reloc && !root_dropped) |
d29a9f62 | 6052 | btrfs_add_dead_root(root); |
2c536799 | 6053 | return err; |
20524f02 | 6054 | } |
9078a3e1 | 6055 | |
2c47e605 YZ |
6056 | /* |
6057 | * drop subtree rooted at tree block 'node'. | |
6058 | * | |
6059 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 6060 | * only used by relocation code |
2c47e605 | 6061 | */ |
f82d02d9 YZ |
6062 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6063 | struct btrfs_root *root, | |
6064 | struct extent_buffer *node, | |
6065 | struct extent_buffer *parent) | |
6066 | { | |
0b246afa | 6067 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 6068 | struct btrfs_path *path; |
2c47e605 | 6069 | struct walk_control *wc; |
f82d02d9 YZ |
6070 | int level; |
6071 | int parent_level; | |
6072 | int ret = 0; | |
6073 | int wret; | |
6074 | ||
2c47e605 YZ |
6075 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6076 | ||
f82d02d9 | 6077 | path = btrfs_alloc_path(); |
db5b493a TI |
6078 | if (!path) |
6079 | return -ENOMEM; | |
f82d02d9 | 6080 | |
2c47e605 | 6081 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6082 | if (!wc) { |
6083 | btrfs_free_path(path); | |
6084 | return -ENOMEM; | |
6085 | } | |
2c47e605 | 6086 | |
49d0c642 | 6087 | btrfs_assert_tree_write_locked(parent); |
f82d02d9 | 6088 | parent_level = btrfs_header_level(parent); |
67439dad | 6089 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
6090 | path->nodes[parent_level] = parent; |
6091 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6092 | ||
49d0c642 | 6093 | btrfs_assert_tree_write_locked(node); |
f82d02d9 | 6094 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6095 | path->nodes[level] = node; |
6096 | path->slots[level] = 0; | |
ac5887c8 | 6097 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
6098 | |
6099 | wc->refs[parent_level] = 1; | |
6100 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6101 | wc->level = level; | |
6102 | wc->shared_level = -1; | |
6103 | wc->stage = DROP_REFERENCE; | |
6104 | wc->update_ref = 0; | |
6105 | wc->keep_locks = 1; | |
0b246afa | 6106 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
6107 | |
6108 | while (1) { | |
2c47e605 YZ |
6109 | wret = walk_down_tree(trans, root, path, wc); |
6110 | if (wret < 0) { | |
f82d02d9 | 6111 | ret = wret; |
f82d02d9 | 6112 | break; |
2c47e605 | 6113 | } |
f82d02d9 | 6114 | |
2c47e605 | 6115 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6116 | if (wret < 0) |
6117 | ret = wret; | |
6118 | if (wret != 0) | |
6119 | break; | |
6120 | } | |
6121 | ||
2c47e605 | 6122 | kfree(wc); |
f82d02d9 YZ |
6123 | btrfs_free_path(path); |
6124 | return ret; | |
6125 | } | |
6126 | ||
2ff7e61e JM |
6127 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
6128 | u64 start, u64 end) | |
acce952b | 6129 | { |
2ff7e61e | 6130 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 6131 | } |
6132 | ||
499f377f JM |
6133 | /* |
6134 | * It used to be that old block groups would be left around forever. | |
6135 | * Iterating over them would be enough to trim unused space. Since we | |
6136 | * now automatically remove them, we also need to iterate over unallocated | |
6137 | * space. | |
6138 | * | |
6139 | * We don't want a transaction for this since the discard may take a | |
6140 | * substantial amount of time. We don't require that a transaction be | |
6141 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
6142 | * to ensure that we're not discarding chunks that were released or |
6143 | * allocated in the current transaction. | |
499f377f JM |
6144 | * |
6145 | * Holding the chunks lock will prevent other threads from allocating | |
6146 | * or releasing chunks, but it won't prevent a running transaction | |
6147 | * from committing and releasing the memory that the pending chunks | |
6148 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
6149 | * transaction and hold the commit root sem. We only need to hold |
6150 | * it while performing the free space search since we have already | |
6151 | * held back allocations. | |
499f377f | 6152 | */ |
8103d10b | 6153 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 6154 | { |
37f85ec3 | 6155 | u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0; |
499f377f JM |
6156 | int ret; |
6157 | ||
6158 | *trimmed = 0; | |
6159 | ||
0be88e36 | 6160 | /* Discard not supported = nothing to do. */ |
70200574 | 6161 | if (!bdev_max_discard_sectors(device->bdev)) |
0be88e36 JM |
6162 | return 0; |
6163 | ||
52042d8e | 6164 | /* Not writable = nothing to do. */ |
ebbede42 | 6165 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
6166 | return 0; |
6167 | ||
6168 | /* No free space = nothing to do. */ | |
6169 | if (device->total_bytes <= device->bytes_used) | |
6170 | return 0; | |
6171 | ||
6172 | ret = 0; | |
6173 | ||
6174 | while (1) { | |
fb456252 | 6175 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
6176 | u64 bytes; |
6177 | ||
6178 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
6179 | if (ret) | |
fee7acc3 | 6180 | break; |
499f377f | 6181 | |
929be17a NB |
6182 | find_first_clear_extent_bit(&device->alloc_state, start, |
6183 | &start, &end, | |
6184 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 6185 | |
c57dd1f2 QW |
6186 | /* Check if there are any CHUNK_* bits left */ |
6187 | if (start > device->total_bytes) { | |
6188 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
6189 | btrfs_warn_in_rcu(fs_info, | |
6190 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
6191 | start, end - start + 1, | |
cb3e217b | 6192 | btrfs_dev_name(device), |
c57dd1f2 QW |
6193 | device->total_bytes); |
6194 | mutex_unlock(&fs_info->chunk_mutex); | |
6195 | ret = 0; | |
6196 | break; | |
6197 | } | |
6198 | ||
37f85ec3 QW |
6199 | /* Ensure we skip the reserved space on each device. */ |
6200 | start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); | |
53460a45 | 6201 | |
929be17a NB |
6202 | /* |
6203 | * If find_first_clear_extent_bit find a range that spans the | |
6204 | * end of the device it will set end to -1, in this case it's up | |
6205 | * to the caller to trim the value to the size of the device. | |
6206 | */ | |
6207 | end = min(end, device->total_bytes - 1); | |
53460a45 | 6208 | |
929be17a | 6209 | len = end - start + 1; |
499f377f | 6210 | |
929be17a NB |
6211 | /* We didn't find any extents */ |
6212 | if (!len) { | |
499f377f | 6213 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 6214 | ret = 0; |
499f377f JM |
6215 | break; |
6216 | } | |
6217 | ||
929be17a NB |
6218 | ret = btrfs_issue_discard(device->bdev, start, len, |
6219 | &bytes); | |
6220 | if (!ret) | |
0acd32c2 | 6221 | set_extent_bit(&device->alloc_state, start, |
1d126800 | 6222 | start + bytes - 1, CHUNK_TRIMMED, NULL); |
499f377f JM |
6223 | mutex_unlock(&fs_info->chunk_mutex); |
6224 | ||
6225 | if (ret) | |
6226 | break; | |
6227 | ||
6228 | start += len; | |
6229 | *trimmed += bytes; | |
6230 | ||
6231 | if (fatal_signal_pending(current)) { | |
6232 | ret = -ERESTARTSYS; | |
6233 | break; | |
6234 | } | |
6235 | ||
6236 | cond_resched(); | |
6237 | } | |
6238 | ||
6239 | return ret; | |
6240 | } | |
6241 | ||
93bba24d QW |
6242 | /* |
6243 | * Trim the whole filesystem by: | |
6244 | * 1) trimming the free space in each block group | |
6245 | * 2) trimming the unallocated space on each device | |
6246 | * | |
6247 | * This will also continue trimming even if a block group or device encounters | |
6248 | * an error. The return value will be the last error, or 0 if nothing bad | |
6249 | * happens. | |
6250 | */ | |
2ff7e61e | 6251 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 6252 | { |
23608d51 | 6253 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
32da5386 | 6254 | struct btrfs_block_group *cache = NULL; |
499f377f | 6255 | struct btrfs_device *device; |
f7039b1d | 6256 | u64 group_trimmed; |
07301df7 | 6257 | u64 range_end = U64_MAX; |
f7039b1d LD |
6258 | u64 start; |
6259 | u64 end; | |
6260 | u64 trimmed = 0; | |
93bba24d QW |
6261 | u64 bg_failed = 0; |
6262 | u64 dev_failed = 0; | |
6263 | int bg_ret = 0; | |
6264 | int dev_ret = 0; | |
f7039b1d LD |
6265 | int ret = 0; |
6266 | ||
f981fec1 JB |
6267 | if (range->start == U64_MAX) |
6268 | return -EINVAL; | |
6269 | ||
07301df7 QW |
6270 | /* |
6271 | * Check range overflow if range->len is set. | |
6272 | * The default range->len is U64_MAX. | |
6273 | */ | |
6274 | if (range->len != U64_MAX && | |
6275 | check_add_overflow(range->start, range->len, &range_end)) | |
6276 | return -EINVAL; | |
6277 | ||
6ba9fc8e | 6278 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 6279 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 6280 | if (cache->start >= range_end) { |
f7039b1d LD |
6281 | btrfs_put_block_group(cache); |
6282 | break; | |
6283 | } | |
6284 | ||
b3470b5d DS |
6285 | start = max(range->start, cache->start); |
6286 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
6287 | |
6288 | if (end - start >= range->minlen) { | |
32da5386 | 6289 | if (!btrfs_block_group_done(cache)) { |
ced8ecf0 | 6290 | ret = btrfs_cache_block_group(cache, true); |
1be41b78 | 6291 | if (ret) { |
93bba24d QW |
6292 | bg_failed++; |
6293 | bg_ret = ret; | |
6294 | continue; | |
1be41b78 | 6295 | } |
f7039b1d LD |
6296 | } |
6297 | ret = btrfs_trim_block_group(cache, | |
6298 | &group_trimmed, | |
6299 | start, | |
6300 | end, | |
6301 | range->minlen); | |
6302 | ||
6303 | trimmed += group_trimmed; | |
6304 | if (ret) { | |
93bba24d QW |
6305 | bg_failed++; |
6306 | bg_ret = ret; | |
6307 | continue; | |
f7039b1d LD |
6308 | } |
6309 | } | |
f7039b1d LD |
6310 | } |
6311 | ||
93bba24d QW |
6312 | if (bg_failed) |
6313 | btrfs_warn(fs_info, | |
6314 | "failed to trim %llu block group(s), last error %d", | |
6315 | bg_failed, bg_ret); | |
23608d51 AJ |
6316 | |
6317 | mutex_lock(&fs_devices->device_list_mutex); | |
6318 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
16a200f6 AJ |
6319 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
6320 | continue; | |
6321 | ||
8103d10b | 6322 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
6323 | if (ret) { |
6324 | dev_failed++; | |
6325 | dev_ret = ret; | |
499f377f | 6326 | break; |
93bba24d | 6327 | } |
499f377f JM |
6328 | |
6329 | trimmed += group_trimmed; | |
6330 | } | |
23608d51 | 6331 | mutex_unlock(&fs_devices->device_list_mutex); |
499f377f | 6332 | |
93bba24d QW |
6333 | if (dev_failed) |
6334 | btrfs_warn(fs_info, | |
6335 | "failed to trim %llu device(s), last error %d", | |
6336 | dev_failed, dev_ret); | |
f7039b1d | 6337 | range->len = trimmed; |
93bba24d QW |
6338 | if (bg_ret) |
6339 | return bg_ret; | |
6340 | return dev_ret; | |
f7039b1d | 6341 | } |