]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 | 4 | */ |
c1d7c514 | 5 | |
ec6b910f | 6 | #include <linux/sched.h> |
f361bf4a | 7 | #include <linux/sched/signal.h> |
edbd8d4e | 8 | #include <linux/pagemap.h> |
ec44a35c | 9 | #include <linux/writeback.h> |
21af804c | 10 | #include <linux/blkdev.h> |
b7a9f29f | 11 | #include <linux/sort.h> |
4184ea7f | 12 | #include <linux/rcupdate.h> |
817d52f8 | 13 | #include <linux/kthread.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
dff51cd1 | 15 | #include <linux/ratelimit.h> |
b150a4f1 | 16 | #include <linux/percpu_counter.h> |
69fe2d75 | 17 | #include <linux/lockdep.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
602cbe91 | 19 | #include "misc.h" |
995946dd | 20 | #include "tree-log.h" |
fec577fb CM |
21 | #include "disk-io.h" |
22 | #include "print-tree.h" | |
0b86a832 | 23 | #include "volumes.h" |
53b381b3 | 24 | #include "raid56.h" |
925baedd | 25 | #include "locking.h" |
fa9c0d79 | 26 | #include "free-space-cache.h" |
1e144fb8 | 27 | #include "free-space-tree.h" |
6ab0a202 | 28 | #include "sysfs.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
8719aaae | 31 | #include "space-info.h" |
d12ffdd1 | 32 | #include "block-rsv.h" |
86736342 | 33 | #include "delalloc-space.h" |
aac0023c | 34 | #include "block-group.h" |
b0643e59 | 35 | #include "discard.h" |
c57dd1f2 | 36 | #include "rcu-string.h" |
169e0da9 | 37 | #include "zoned.h" |
6143c23c | 38 | #include "dev-replace.h" |
fec577fb | 39 | |
709c0486 AJ |
40 | #undef SCRAMBLE_DELAYED_REFS |
41 | ||
9f9b8e8d | 42 | |
5d4f98a2 | 43 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
44 | struct btrfs_delayed_ref_node *node, u64 parent, |
45 | u64 root_objectid, u64 owner_objectid, | |
46 | u64 owner_offset, int refs_to_drop, | |
47 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
48 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
49 | struct extent_buffer *leaf, | |
50 | struct btrfs_extent_item *ei); | |
51 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
52 | u64 parent, u64 root_objectid, |
53 | u64 flags, u64 owner, u64 offset, | |
54 | struct btrfs_key *ins, int ref_mod); | |
55 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 56 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 57 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
58 | static int find_next_key(struct btrfs_path *path, int level, |
59 | struct btrfs_key *key); | |
6a63209f | 60 | |
32da5386 | 61 | static int block_group_bits(struct btrfs_block_group *cache, u64 bits) |
0f9dd46c JB |
62 | { |
63 | return (cache->flags & bits) == bits; | |
64 | } | |
65 | ||
6f410d1b JB |
66 | int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info, |
67 | u64 start, u64 num_bytes) | |
817d52f8 | 68 | { |
11833d66 | 69 | u64 end = start + num_bytes - 1; |
fe119a6e NB |
70 | set_extent_bits(&fs_info->excluded_extents, start, end, |
71 | EXTENT_UPTODATE); | |
11833d66 YZ |
72 | return 0; |
73 | } | |
817d52f8 | 74 | |
32da5386 | 75 | void btrfs_free_excluded_extents(struct btrfs_block_group *cache) |
11833d66 | 76 | { |
9e715da8 | 77 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 78 | u64 start, end; |
817d52f8 | 79 | |
b3470b5d DS |
80 | start = cache->start; |
81 | end = start + cache->length - 1; | |
11833d66 | 82 | |
fe119a6e NB |
83 | clear_extent_bits(&fs_info->excluded_extents, start, end, |
84 | EXTENT_UPTODATE); | |
817d52f8 JB |
85 | } |
86 | ||
1a4ed8fd | 87 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 88 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
89 | { |
90 | int ret; | |
91 | struct btrfs_key key; | |
31840ae1 | 92 | struct btrfs_path *path; |
e02119d5 | 93 | |
31840ae1 | 94 | path = btrfs_alloc_path(); |
d8926bb3 MF |
95 | if (!path) |
96 | return -ENOMEM; | |
97 | ||
e02119d5 CM |
98 | key.objectid = start; |
99 | key.offset = len; | |
3173a18f | 100 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 101 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 102 | btrfs_free_path(path); |
7bb86316 CM |
103 | return ret; |
104 | } | |
105 | ||
a22285a6 | 106 | /* |
3173a18f | 107 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
108 | * |
109 | * the head node for delayed ref is used to store the sum of all the | |
110 | * reference count modifications queued up in the rbtree. the head | |
111 | * node may also store the extent flags to set. This way you can check | |
112 | * to see what the reference count and extent flags would be if all of | |
113 | * the delayed refs are not processed. | |
114 | */ | |
115 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 116 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 117 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
118 | { |
119 | struct btrfs_delayed_ref_head *head; | |
120 | struct btrfs_delayed_ref_root *delayed_refs; | |
121 | struct btrfs_path *path; | |
122 | struct btrfs_extent_item *ei; | |
123 | struct extent_buffer *leaf; | |
124 | struct btrfs_key key; | |
125 | u32 item_size; | |
126 | u64 num_refs; | |
127 | u64 extent_flags; | |
128 | int ret; | |
129 | ||
3173a18f JB |
130 | /* |
131 | * If we don't have skinny metadata, don't bother doing anything | |
132 | * different | |
133 | */ | |
0b246afa JM |
134 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
135 | offset = fs_info->nodesize; | |
3173a18f JB |
136 | metadata = 0; |
137 | } | |
138 | ||
a22285a6 YZ |
139 | path = btrfs_alloc_path(); |
140 | if (!path) | |
141 | return -ENOMEM; | |
142 | ||
a22285a6 YZ |
143 | if (!trans) { |
144 | path->skip_locking = 1; | |
145 | path->search_commit_root = 1; | |
146 | } | |
639eefc8 FDBM |
147 | |
148 | search_again: | |
149 | key.objectid = bytenr; | |
150 | key.offset = offset; | |
151 | if (metadata) | |
152 | key.type = BTRFS_METADATA_ITEM_KEY; | |
153 | else | |
154 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
155 | ||
0b246afa | 156 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
157 | if (ret < 0) |
158 | goto out_free; | |
159 | ||
3173a18f | 160 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
161 | if (path->slots[0]) { |
162 | path->slots[0]--; | |
163 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
164 | path->slots[0]); | |
165 | if (key.objectid == bytenr && | |
166 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 167 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
168 | ret = 0; |
169 | } | |
3173a18f JB |
170 | } |
171 | ||
a22285a6 YZ |
172 | if (ret == 0) { |
173 | leaf = path->nodes[0]; | |
174 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
175 | if (item_size >= sizeof(*ei)) { | |
176 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
177 | struct btrfs_extent_item); | |
178 | num_refs = btrfs_extent_refs(leaf, ei); | |
179 | extent_flags = btrfs_extent_flags(leaf, ei); | |
180 | } else { | |
ba3c2b19 NB |
181 | ret = -EINVAL; |
182 | btrfs_print_v0_err(fs_info); | |
183 | if (trans) | |
184 | btrfs_abort_transaction(trans, ret); | |
185 | else | |
186 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
187 | ||
188 | goto out_free; | |
a22285a6 | 189 | } |
ba3c2b19 | 190 | |
a22285a6 YZ |
191 | BUG_ON(num_refs == 0); |
192 | } else { | |
193 | num_refs = 0; | |
194 | extent_flags = 0; | |
195 | ret = 0; | |
196 | } | |
197 | ||
198 | if (!trans) | |
199 | goto out; | |
200 | ||
201 | delayed_refs = &trans->transaction->delayed_refs; | |
202 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 203 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
204 | if (head) { |
205 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 206 | refcount_inc(&head->refs); |
a22285a6 YZ |
207 | spin_unlock(&delayed_refs->lock); |
208 | ||
b3b4aa74 | 209 | btrfs_release_path(path); |
a22285a6 | 210 | |
8cc33e5c DS |
211 | /* |
212 | * Mutex was contended, block until it's released and try | |
213 | * again | |
214 | */ | |
a22285a6 YZ |
215 | mutex_lock(&head->mutex); |
216 | mutex_unlock(&head->mutex); | |
d278850e | 217 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 218 | goto search_again; |
a22285a6 | 219 | } |
d7df2c79 | 220 | spin_lock(&head->lock); |
a22285a6 YZ |
221 | if (head->extent_op && head->extent_op->update_flags) |
222 | extent_flags |= head->extent_op->flags_to_set; | |
223 | else | |
224 | BUG_ON(num_refs == 0); | |
225 | ||
d278850e | 226 | num_refs += head->ref_mod; |
d7df2c79 | 227 | spin_unlock(&head->lock); |
a22285a6 YZ |
228 | mutex_unlock(&head->mutex); |
229 | } | |
230 | spin_unlock(&delayed_refs->lock); | |
231 | out: | |
232 | WARN_ON(num_refs == 0); | |
233 | if (refs) | |
234 | *refs = num_refs; | |
235 | if (flags) | |
236 | *flags = extent_flags; | |
237 | out_free: | |
238 | btrfs_free_path(path); | |
239 | return ret; | |
240 | } | |
241 | ||
d8d5f3e1 CM |
242 | /* |
243 | * Back reference rules. Back refs have three main goals: | |
244 | * | |
245 | * 1) differentiate between all holders of references to an extent so that | |
246 | * when a reference is dropped we can make sure it was a valid reference | |
247 | * before freeing the extent. | |
248 | * | |
249 | * 2) Provide enough information to quickly find the holders of an extent | |
250 | * if we notice a given block is corrupted or bad. | |
251 | * | |
252 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
253 | * maintenance. This is actually the same as #2, but with a slightly | |
254 | * different use case. | |
255 | * | |
5d4f98a2 YZ |
256 | * There are two kinds of back refs. The implicit back refs is optimized |
257 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
258 | * back refs of this kind provide information about the block's owner tree | |
259 | * and the pointer's key. These information allow us to find the block by | |
260 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
261 | * referenced by their owner trees. The location of tree block is recorded | |
262 | * in the back refs. Actually the full back refs is generic, and can be | |
263 | * used in all cases the implicit back refs is used. The major shortcoming | |
264 | * of the full back refs is its overhead. Every time a tree block gets | |
265 | * COWed, we have to update back refs entry for all pointers in it. | |
266 | * | |
267 | * For a newly allocated tree block, we use implicit back refs for | |
268 | * pointers in it. This means most tree related operations only involve | |
269 | * implicit back refs. For a tree block created in old transaction, the | |
270 | * only way to drop a reference to it is COW it. So we can detect the | |
271 | * event that tree block loses its owner tree's reference and do the | |
272 | * back refs conversion. | |
273 | * | |
01327610 | 274 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
275 | * |
276 | * The reference count of the block is one and the tree is the block's | |
277 | * owner tree. Nothing to do in this case. | |
278 | * | |
279 | * The reference count of the block is one and the tree is not the | |
280 | * block's owner tree. In this case, full back refs is used for pointers | |
281 | * in the block. Remove these full back refs, add implicit back refs for | |
282 | * every pointers in the new block. | |
283 | * | |
284 | * The reference count of the block is greater than one and the tree is | |
285 | * the block's owner tree. In this case, implicit back refs is used for | |
286 | * pointers in the block. Add full back refs for every pointers in the | |
287 | * block, increase lower level extents' reference counts. The original | |
288 | * implicit back refs are entailed to the new block. | |
289 | * | |
290 | * The reference count of the block is greater than one and the tree is | |
291 | * not the block's owner tree. Add implicit back refs for every pointer in | |
292 | * the new block, increase lower level extents' reference count. | |
293 | * | |
294 | * Back Reference Key composing: | |
295 | * | |
296 | * The key objectid corresponds to the first byte in the extent, | |
297 | * The key type is used to differentiate between types of back refs. | |
298 | * There are different meanings of the key offset for different types | |
299 | * of back refs. | |
300 | * | |
d8d5f3e1 CM |
301 | * File extents can be referenced by: |
302 | * | |
303 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 304 | * - different files inside a single subvolume |
d8d5f3e1 CM |
305 | * - different offsets inside a file (bookend extents in file.c) |
306 | * | |
5d4f98a2 | 307 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
308 | * |
309 | * - Objectid of the subvolume root | |
d8d5f3e1 | 310 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
311 | * - original offset in the file |
312 | * - how many bookend extents | |
d8d5f3e1 | 313 | * |
5d4f98a2 YZ |
314 | * The key offset for the implicit back refs is hash of the first |
315 | * three fields. | |
d8d5f3e1 | 316 | * |
5d4f98a2 | 317 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 318 | * |
5d4f98a2 | 319 | * - number of pointers in the tree leaf |
d8d5f3e1 | 320 | * |
5d4f98a2 YZ |
321 | * The key offset for the implicit back refs is the first byte of |
322 | * the tree leaf | |
d8d5f3e1 | 323 | * |
5d4f98a2 YZ |
324 | * When a file extent is allocated, The implicit back refs is used. |
325 | * the fields are filled in: | |
d8d5f3e1 | 326 | * |
5d4f98a2 | 327 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 328 | * |
5d4f98a2 YZ |
329 | * When a file extent is removed file truncation, we find the |
330 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 331 | * |
5d4f98a2 | 332 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 333 | * |
5d4f98a2 | 334 | * Btree extents can be referenced by: |
d8d5f3e1 | 335 | * |
5d4f98a2 | 336 | * - Different subvolumes |
d8d5f3e1 | 337 | * |
5d4f98a2 YZ |
338 | * Both the implicit back refs and the full back refs for tree blocks |
339 | * only consist of key. The key offset for the implicit back refs is | |
340 | * objectid of block's owner tree. The key offset for the full back refs | |
341 | * is the first byte of parent block. | |
d8d5f3e1 | 342 | * |
5d4f98a2 YZ |
343 | * When implicit back refs is used, information about the lowest key and |
344 | * level of the tree block are required. These information are stored in | |
345 | * tree block info structure. | |
d8d5f3e1 | 346 | */ |
31840ae1 | 347 | |
167ce953 LB |
348 | /* |
349 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 350 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
351 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
352 | */ | |
353 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
354 | struct btrfs_extent_inline_ref *iref, | |
355 | enum btrfs_inline_ref_type is_data) | |
356 | { | |
357 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 358 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
359 | |
360 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
361 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
362 | type == BTRFS_SHARED_DATA_REF_KEY || | |
363 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
364 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 365 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 366 | return type; |
64ecdb64 LB |
367 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
368 | ASSERT(eb->fs_info); | |
369 | /* | |
ea57788e QW |
370 | * Every shared one has parent tree block, |
371 | * which must be aligned to sector size. | |
64ecdb64 LB |
372 | */ |
373 | if (offset && | |
ea57788e | 374 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
375 | return type; |
376 | } | |
167ce953 | 377 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 378 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 379 | return type; |
64ecdb64 LB |
380 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
381 | ASSERT(eb->fs_info); | |
382 | /* | |
ea57788e QW |
383 | * Every shared one has parent tree block, |
384 | * which must be aligned to sector size. | |
64ecdb64 LB |
385 | */ |
386 | if (offset && | |
ea57788e | 387 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
388 | return type; |
389 | } | |
167ce953 LB |
390 | } else { |
391 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
392 | return type; | |
393 | } | |
394 | } | |
395 | ||
396 | btrfs_print_leaf((struct extent_buffer *)eb); | |
ea57788e QW |
397 | btrfs_err(eb->fs_info, |
398 | "eb %llu iref 0x%lx invalid extent inline ref type %d", | |
399 | eb->start, (unsigned long)iref, type); | |
167ce953 LB |
400 | WARN_ON(1); |
401 | ||
402 | return BTRFS_REF_TYPE_INVALID; | |
403 | } | |
404 | ||
0785a9aa | 405 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
406 | { |
407 | u32 high_crc = ~(u32)0; | |
408 | u32 low_crc = ~(u32)0; | |
409 | __le64 lenum; | |
410 | ||
411 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 412 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 413 | lenum = cpu_to_le64(owner); |
65019df8 | 414 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 415 | lenum = cpu_to_le64(offset); |
65019df8 | 416 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
417 | |
418 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
419 | } | |
420 | ||
421 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
422 | struct btrfs_extent_data_ref *ref) | |
423 | { | |
424 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
425 | btrfs_extent_data_ref_objectid(leaf, ref), | |
426 | btrfs_extent_data_ref_offset(leaf, ref)); | |
427 | } | |
428 | ||
429 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
430 | struct btrfs_extent_data_ref *ref, | |
431 | u64 root_objectid, u64 owner, u64 offset) | |
432 | { | |
433 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
434 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
435 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
436 | return 0; | |
437 | return 1; | |
438 | } | |
439 | ||
440 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
441 | struct btrfs_path *path, |
442 | u64 bytenr, u64 parent, | |
443 | u64 root_objectid, | |
444 | u64 owner, u64 offset) | |
445 | { | |
bd1d53ef | 446 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
447 | struct btrfs_key key; |
448 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 449 | struct extent_buffer *leaf; |
5d4f98a2 | 450 | u32 nritems; |
74493f7a | 451 | int ret; |
5d4f98a2 YZ |
452 | int recow; |
453 | int err = -ENOENT; | |
74493f7a | 454 | |
31840ae1 | 455 | key.objectid = bytenr; |
5d4f98a2 YZ |
456 | if (parent) { |
457 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
458 | key.offset = parent; | |
459 | } else { | |
460 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
461 | key.offset = hash_extent_data_ref(root_objectid, | |
462 | owner, offset); | |
463 | } | |
464 | again: | |
465 | recow = 0; | |
466 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
467 | if (ret < 0) { | |
468 | err = ret; | |
469 | goto fail; | |
470 | } | |
31840ae1 | 471 | |
5d4f98a2 YZ |
472 | if (parent) { |
473 | if (!ret) | |
474 | return 0; | |
5d4f98a2 | 475 | goto fail; |
31840ae1 ZY |
476 | } |
477 | ||
478 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
479 | nritems = btrfs_header_nritems(leaf); |
480 | while (1) { | |
481 | if (path->slots[0] >= nritems) { | |
482 | ret = btrfs_next_leaf(root, path); | |
483 | if (ret < 0) | |
484 | err = ret; | |
485 | if (ret) | |
486 | goto fail; | |
487 | ||
488 | leaf = path->nodes[0]; | |
489 | nritems = btrfs_header_nritems(leaf); | |
490 | recow = 1; | |
491 | } | |
492 | ||
493 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
494 | if (key.objectid != bytenr || | |
495 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
496 | goto fail; | |
497 | ||
498 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
499 | struct btrfs_extent_data_ref); | |
500 | ||
501 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
502 | owner, offset)) { | |
503 | if (recow) { | |
b3b4aa74 | 504 | btrfs_release_path(path); |
5d4f98a2 YZ |
505 | goto again; |
506 | } | |
507 | err = 0; | |
508 | break; | |
509 | } | |
510 | path->slots[0]++; | |
31840ae1 | 511 | } |
5d4f98a2 YZ |
512 | fail: |
513 | return err; | |
31840ae1 ZY |
514 | } |
515 | ||
5d4f98a2 | 516 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
517 | struct btrfs_path *path, |
518 | u64 bytenr, u64 parent, | |
519 | u64 root_objectid, u64 owner, | |
520 | u64 offset, int refs_to_add) | |
31840ae1 | 521 | { |
62b895af | 522 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
523 | struct btrfs_key key; |
524 | struct extent_buffer *leaf; | |
5d4f98a2 | 525 | u32 size; |
31840ae1 ZY |
526 | u32 num_refs; |
527 | int ret; | |
74493f7a | 528 | |
74493f7a | 529 | key.objectid = bytenr; |
5d4f98a2 YZ |
530 | if (parent) { |
531 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
532 | key.offset = parent; | |
533 | size = sizeof(struct btrfs_shared_data_ref); | |
534 | } else { | |
535 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
536 | key.offset = hash_extent_data_ref(root_objectid, | |
537 | owner, offset); | |
538 | size = sizeof(struct btrfs_extent_data_ref); | |
539 | } | |
74493f7a | 540 | |
5d4f98a2 YZ |
541 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
542 | if (ret && ret != -EEXIST) | |
543 | goto fail; | |
544 | ||
545 | leaf = path->nodes[0]; | |
546 | if (parent) { | |
547 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 548 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
549 | struct btrfs_shared_data_ref); |
550 | if (ret == 0) { | |
551 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
552 | } else { | |
553 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
554 | num_refs += refs_to_add; | |
555 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 556 | } |
5d4f98a2 YZ |
557 | } else { |
558 | struct btrfs_extent_data_ref *ref; | |
559 | while (ret == -EEXIST) { | |
560 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
561 | struct btrfs_extent_data_ref); | |
562 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
563 | owner, offset)) | |
564 | break; | |
b3b4aa74 | 565 | btrfs_release_path(path); |
5d4f98a2 YZ |
566 | key.offset++; |
567 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
568 | size); | |
569 | if (ret && ret != -EEXIST) | |
570 | goto fail; | |
31840ae1 | 571 | |
5d4f98a2 YZ |
572 | leaf = path->nodes[0]; |
573 | } | |
574 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
575 | struct btrfs_extent_data_ref); | |
576 | if (ret == 0) { | |
577 | btrfs_set_extent_data_ref_root(leaf, ref, | |
578 | root_objectid); | |
579 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
580 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
581 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
582 | } else { | |
583 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
584 | num_refs += refs_to_add; | |
585 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 586 | } |
31840ae1 | 587 | } |
5d4f98a2 YZ |
588 | btrfs_mark_buffer_dirty(leaf); |
589 | ret = 0; | |
590 | fail: | |
b3b4aa74 | 591 | btrfs_release_path(path); |
7bb86316 | 592 | return ret; |
74493f7a CM |
593 | } |
594 | ||
5d4f98a2 | 595 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 596 | struct btrfs_path *path, |
fcebe456 | 597 | int refs_to_drop, int *last_ref) |
31840ae1 | 598 | { |
5d4f98a2 YZ |
599 | struct btrfs_key key; |
600 | struct btrfs_extent_data_ref *ref1 = NULL; | |
601 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 602 | struct extent_buffer *leaf; |
5d4f98a2 | 603 | u32 num_refs = 0; |
31840ae1 ZY |
604 | int ret = 0; |
605 | ||
606 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
607 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
608 | ||
609 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
610 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
611 | struct btrfs_extent_data_ref); | |
612 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
613 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
614 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
615 | struct btrfs_shared_data_ref); | |
616 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 617 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
618 | btrfs_print_v0_err(trans->fs_info); |
619 | btrfs_abort_transaction(trans, -EINVAL); | |
620 | return -EINVAL; | |
5d4f98a2 YZ |
621 | } else { |
622 | BUG(); | |
623 | } | |
624 | ||
56bec294 CM |
625 | BUG_ON(num_refs < refs_to_drop); |
626 | num_refs -= refs_to_drop; | |
5d4f98a2 | 627 | |
31840ae1 | 628 | if (num_refs == 0) { |
e9f6290d | 629 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 630 | *last_ref = 1; |
31840ae1 | 631 | } else { |
5d4f98a2 YZ |
632 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
633 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
634 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
635 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
636 | btrfs_mark_buffer_dirty(leaf); |
637 | } | |
31840ae1 ZY |
638 | return ret; |
639 | } | |
640 | ||
9ed0dea0 | 641 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 642 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 643 | { |
5d4f98a2 YZ |
644 | struct btrfs_key key; |
645 | struct extent_buffer *leaf; | |
646 | struct btrfs_extent_data_ref *ref1; | |
647 | struct btrfs_shared_data_ref *ref2; | |
648 | u32 num_refs = 0; | |
3de28d57 | 649 | int type; |
5d4f98a2 YZ |
650 | |
651 | leaf = path->nodes[0]; | |
652 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
653 | |
654 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 655 | if (iref) { |
3de28d57 LB |
656 | /* |
657 | * If type is invalid, we should have bailed out earlier than | |
658 | * this call. | |
659 | */ | |
660 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
661 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
662 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
663 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
664 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
665 | } else { | |
666 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
667 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
668 | } | |
669 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
670 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
671 | struct btrfs_extent_data_ref); | |
672 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
673 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
674 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
675 | struct btrfs_shared_data_ref); | |
676 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
677 | } else { |
678 | WARN_ON(1); | |
679 | } | |
680 | return num_refs; | |
681 | } | |
15916de8 | 682 | |
5d4f98a2 | 683 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
684 | struct btrfs_path *path, |
685 | u64 bytenr, u64 parent, | |
686 | u64 root_objectid) | |
1f3c79a2 | 687 | { |
b8582eea | 688 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 689 | struct btrfs_key key; |
1f3c79a2 | 690 | int ret; |
1f3c79a2 | 691 | |
5d4f98a2 YZ |
692 | key.objectid = bytenr; |
693 | if (parent) { | |
694 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
695 | key.offset = parent; | |
696 | } else { | |
697 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
698 | key.offset = root_objectid; | |
1f3c79a2 LH |
699 | } |
700 | ||
5d4f98a2 YZ |
701 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
702 | if (ret > 0) | |
703 | ret = -ENOENT; | |
5d4f98a2 | 704 | return ret; |
1f3c79a2 LH |
705 | } |
706 | ||
5d4f98a2 | 707 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
708 | struct btrfs_path *path, |
709 | u64 bytenr, u64 parent, | |
710 | u64 root_objectid) | |
31840ae1 | 711 | { |
5d4f98a2 | 712 | struct btrfs_key key; |
31840ae1 | 713 | int ret; |
31840ae1 | 714 | |
5d4f98a2 YZ |
715 | key.objectid = bytenr; |
716 | if (parent) { | |
717 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
718 | key.offset = parent; | |
719 | } else { | |
720 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
721 | key.offset = root_objectid; | |
722 | } | |
723 | ||
10728404 | 724 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 725 | path, &key, 0); |
b3b4aa74 | 726 | btrfs_release_path(path); |
31840ae1 ZY |
727 | return ret; |
728 | } | |
729 | ||
5d4f98a2 | 730 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 731 | { |
5d4f98a2 YZ |
732 | int type; |
733 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
734 | if (parent > 0) | |
735 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
736 | else | |
737 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
738 | } else { | |
739 | if (parent > 0) | |
740 | type = BTRFS_SHARED_DATA_REF_KEY; | |
741 | else | |
742 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
743 | } | |
744 | return type; | |
31840ae1 | 745 | } |
56bec294 | 746 | |
2c47e605 YZ |
747 | static int find_next_key(struct btrfs_path *path, int level, |
748 | struct btrfs_key *key) | |
56bec294 | 749 | |
02217ed2 | 750 | { |
2c47e605 | 751 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
752 | if (!path->nodes[level]) |
753 | break; | |
5d4f98a2 YZ |
754 | if (path->slots[level] + 1 >= |
755 | btrfs_header_nritems(path->nodes[level])) | |
756 | continue; | |
757 | if (level == 0) | |
758 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
759 | path->slots[level] + 1); | |
760 | else | |
761 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
762 | path->slots[level] + 1); | |
763 | return 0; | |
764 | } | |
765 | return 1; | |
766 | } | |
037e6390 | 767 | |
5d4f98a2 YZ |
768 | /* |
769 | * look for inline back ref. if back ref is found, *ref_ret is set | |
770 | * to the address of inline back ref, and 0 is returned. | |
771 | * | |
772 | * if back ref isn't found, *ref_ret is set to the address where it | |
773 | * should be inserted, and -ENOENT is returned. | |
774 | * | |
775 | * if insert is true and there are too many inline back refs, the path | |
776 | * points to the extent item, and -EAGAIN is returned. | |
777 | * | |
778 | * NOTE: inline back refs are ordered in the same way that back ref | |
779 | * items in the tree are ordered. | |
780 | */ | |
781 | static noinline_for_stack | |
782 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
783 | struct btrfs_path *path, |
784 | struct btrfs_extent_inline_ref **ref_ret, | |
785 | u64 bytenr, u64 num_bytes, | |
786 | u64 parent, u64 root_objectid, | |
787 | u64 owner, u64 offset, int insert) | |
788 | { | |
867cc1fb | 789 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 790 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
791 | struct btrfs_key key; |
792 | struct extent_buffer *leaf; | |
793 | struct btrfs_extent_item *ei; | |
794 | struct btrfs_extent_inline_ref *iref; | |
795 | u64 flags; | |
796 | u64 item_size; | |
797 | unsigned long ptr; | |
798 | unsigned long end; | |
799 | int extra_size; | |
800 | int type; | |
801 | int want; | |
802 | int ret; | |
803 | int err = 0; | |
0b246afa | 804 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 805 | int needed; |
26b8003f | 806 | |
db94535d | 807 | key.objectid = bytenr; |
31840ae1 | 808 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 809 | key.offset = num_bytes; |
31840ae1 | 810 | |
5d4f98a2 YZ |
811 | want = extent_ref_type(parent, owner); |
812 | if (insert) { | |
813 | extra_size = btrfs_extent_inline_ref_size(want); | |
9a664971 | 814 | path->search_for_extension = 1; |
85d4198e | 815 | path->keep_locks = 1; |
5d4f98a2 YZ |
816 | } else |
817 | extra_size = -1; | |
3173a18f JB |
818 | |
819 | /* | |
16d1c062 NB |
820 | * Owner is our level, so we can just add one to get the level for the |
821 | * block we are interested in. | |
3173a18f JB |
822 | */ |
823 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
824 | key.type = BTRFS_METADATA_ITEM_KEY; | |
825 | key.offset = owner; | |
826 | } | |
827 | ||
828 | again: | |
5d4f98a2 | 829 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 830 | if (ret < 0) { |
5d4f98a2 YZ |
831 | err = ret; |
832 | goto out; | |
833 | } | |
3173a18f JB |
834 | |
835 | /* | |
836 | * We may be a newly converted file system which still has the old fat | |
837 | * extent entries for metadata, so try and see if we have one of those. | |
838 | */ | |
839 | if (ret > 0 && skinny_metadata) { | |
840 | skinny_metadata = false; | |
841 | if (path->slots[0]) { | |
842 | path->slots[0]--; | |
843 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
844 | path->slots[0]); | |
845 | if (key.objectid == bytenr && | |
846 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
847 | key.offset == num_bytes) | |
848 | ret = 0; | |
849 | } | |
850 | if (ret) { | |
9ce49a0b | 851 | key.objectid = bytenr; |
3173a18f JB |
852 | key.type = BTRFS_EXTENT_ITEM_KEY; |
853 | key.offset = num_bytes; | |
854 | btrfs_release_path(path); | |
855 | goto again; | |
856 | } | |
857 | } | |
858 | ||
79787eaa JM |
859 | if (ret && !insert) { |
860 | err = -ENOENT; | |
861 | goto out; | |
fae7f21c | 862 | } else if (WARN_ON(ret)) { |
492104c8 | 863 | err = -EIO; |
492104c8 | 864 | goto out; |
79787eaa | 865 | } |
5d4f98a2 YZ |
866 | |
867 | leaf = path->nodes[0]; | |
868 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 869 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
870 | err = -EINVAL; |
871 | btrfs_print_v0_err(fs_info); | |
872 | btrfs_abort_transaction(trans, err); | |
873 | goto out; | |
874 | } | |
5d4f98a2 | 875 | |
5d4f98a2 YZ |
876 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
877 | flags = btrfs_extent_flags(leaf, ei); | |
878 | ||
879 | ptr = (unsigned long)(ei + 1); | |
880 | end = (unsigned long)ei + item_size; | |
881 | ||
3173a18f | 882 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
883 | ptr += sizeof(struct btrfs_tree_block_info); |
884 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
885 | } |
886 | ||
3de28d57 LB |
887 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
888 | needed = BTRFS_REF_TYPE_DATA; | |
889 | else | |
890 | needed = BTRFS_REF_TYPE_BLOCK; | |
891 | ||
5d4f98a2 YZ |
892 | err = -ENOENT; |
893 | while (1) { | |
894 | if (ptr >= end) { | |
895 | WARN_ON(ptr > end); | |
896 | break; | |
897 | } | |
898 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
899 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
900 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 901 | err = -EUCLEAN; |
3de28d57 LB |
902 | goto out; |
903 | } | |
904 | ||
5d4f98a2 YZ |
905 | if (want < type) |
906 | break; | |
907 | if (want > type) { | |
908 | ptr += btrfs_extent_inline_ref_size(type); | |
909 | continue; | |
910 | } | |
911 | ||
912 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
913 | struct btrfs_extent_data_ref *dref; | |
914 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
915 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
916 | owner, offset)) { | |
917 | err = 0; | |
918 | break; | |
919 | } | |
920 | if (hash_extent_data_ref_item(leaf, dref) < | |
921 | hash_extent_data_ref(root_objectid, owner, offset)) | |
922 | break; | |
923 | } else { | |
924 | u64 ref_offset; | |
925 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
926 | if (parent > 0) { | |
927 | if (parent == ref_offset) { | |
928 | err = 0; | |
929 | break; | |
930 | } | |
931 | if (ref_offset < parent) | |
932 | break; | |
933 | } else { | |
934 | if (root_objectid == ref_offset) { | |
935 | err = 0; | |
936 | break; | |
937 | } | |
938 | if (ref_offset < root_objectid) | |
939 | break; | |
940 | } | |
941 | } | |
942 | ptr += btrfs_extent_inline_ref_size(type); | |
943 | } | |
944 | if (err == -ENOENT && insert) { | |
945 | if (item_size + extra_size >= | |
946 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
947 | err = -EAGAIN; | |
948 | goto out; | |
949 | } | |
950 | /* | |
951 | * To add new inline back ref, we have to make sure | |
952 | * there is no corresponding back ref item. | |
953 | * For simplicity, we just do not add new inline back | |
954 | * ref if there is any kind of item for this block | |
955 | */ | |
2c47e605 YZ |
956 | if (find_next_key(path, 0, &key) == 0 && |
957 | key.objectid == bytenr && | |
85d4198e | 958 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
959 | err = -EAGAIN; |
960 | goto out; | |
961 | } | |
962 | } | |
963 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
964 | out: | |
85d4198e | 965 | if (insert) { |
5d4f98a2 | 966 | path->keep_locks = 0; |
9a664971 | 967 | path->search_for_extension = 0; |
5d4f98a2 YZ |
968 | btrfs_unlock_up_safe(path, 1); |
969 | } | |
970 | return err; | |
971 | } | |
972 | ||
973 | /* | |
974 | * helper to add new inline back ref | |
975 | */ | |
976 | static noinline_for_stack | |
87bde3cd | 977 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
978 | struct btrfs_path *path, |
979 | struct btrfs_extent_inline_ref *iref, | |
980 | u64 parent, u64 root_objectid, | |
981 | u64 owner, u64 offset, int refs_to_add, | |
982 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
983 | { |
984 | struct extent_buffer *leaf; | |
985 | struct btrfs_extent_item *ei; | |
986 | unsigned long ptr; | |
987 | unsigned long end; | |
988 | unsigned long item_offset; | |
989 | u64 refs; | |
990 | int size; | |
991 | int type; | |
5d4f98a2 YZ |
992 | |
993 | leaf = path->nodes[0]; | |
994 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
995 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
996 | ||
997 | type = extent_ref_type(parent, owner); | |
998 | size = btrfs_extent_inline_ref_size(type); | |
999 | ||
c71dd880 | 1000 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1001 | |
1002 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1003 | refs = btrfs_extent_refs(leaf, ei); | |
1004 | refs += refs_to_add; | |
1005 | btrfs_set_extent_refs(leaf, ei, refs); | |
1006 | if (extent_op) | |
1007 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1008 | ||
1009 | ptr = (unsigned long)ei + item_offset; | |
1010 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1011 | if (ptr < end - size) | |
1012 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1013 | end - size - ptr); | |
1014 | ||
1015 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1016 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1017 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1018 | struct btrfs_extent_data_ref *dref; | |
1019 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1020 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1021 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1022 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1023 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1024 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1025 | struct btrfs_shared_data_ref *sref; | |
1026 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1027 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1028 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1029 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1030 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1031 | } else { | |
1032 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1033 | } | |
1034 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1035 | } |
1036 | ||
1037 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1038 | struct btrfs_path *path, |
1039 | struct btrfs_extent_inline_ref **ref_ret, | |
1040 | u64 bytenr, u64 num_bytes, u64 parent, | |
1041 | u64 root_objectid, u64 owner, u64 offset) | |
1042 | { | |
1043 | int ret; | |
1044 | ||
867cc1fb NB |
1045 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1046 | num_bytes, parent, root_objectid, | |
1047 | owner, offset, 0); | |
5d4f98a2 | 1048 | if (ret != -ENOENT) |
54aa1f4d | 1049 | return ret; |
5d4f98a2 | 1050 | |
b3b4aa74 | 1051 | btrfs_release_path(path); |
5d4f98a2 YZ |
1052 | *ref_ret = NULL; |
1053 | ||
1054 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1055 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1056 | root_objectid); | |
5d4f98a2 | 1057 | } else { |
bd1d53ef NB |
1058 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1059 | root_objectid, owner, offset); | |
b9473439 | 1060 | } |
5d4f98a2 YZ |
1061 | return ret; |
1062 | } | |
31840ae1 | 1063 | |
5d4f98a2 YZ |
1064 | /* |
1065 | * helper to update/remove inline back ref | |
1066 | */ | |
1067 | static noinline_for_stack | |
61a18f1c | 1068 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1069 | struct btrfs_extent_inline_ref *iref, |
1070 | int refs_to_mod, | |
fcebe456 JB |
1071 | struct btrfs_delayed_extent_op *extent_op, |
1072 | int *last_ref) | |
5d4f98a2 | 1073 | { |
61a18f1c | 1074 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1075 | struct btrfs_extent_item *ei; |
1076 | struct btrfs_extent_data_ref *dref = NULL; | |
1077 | struct btrfs_shared_data_ref *sref = NULL; | |
1078 | unsigned long ptr; | |
1079 | unsigned long end; | |
1080 | u32 item_size; | |
1081 | int size; | |
1082 | int type; | |
5d4f98a2 YZ |
1083 | u64 refs; |
1084 | ||
5d4f98a2 YZ |
1085 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1086 | refs = btrfs_extent_refs(leaf, ei); | |
1087 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1088 | refs += refs_to_mod; | |
1089 | btrfs_set_extent_refs(leaf, ei, refs); | |
1090 | if (extent_op) | |
1091 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1092 | ||
3de28d57 LB |
1093 | /* |
1094 | * If type is invalid, we should have bailed out after | |
1095 | * lookup_inline_extent_backref(). | |
1096 | */ | |
1097 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1098 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1099 | |
1100 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1101 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1102 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1103 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1104 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1105 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1106 | } else { | |
1107 | refs = 1; | |
1108 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1109 | } |
31840ae1 | 1110 | |
5d4f98a2 YZ |
1111 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1112 | refs += refs_to_mod; | |
1113 | ||
1114 | if (refs > 0) { | |
1115 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1116 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1117 | else | |
1118 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1119 | } else { | |
fcebe456 | 1120 | *last_ref = 1; |
5d4f98a2 YZ |
1121 | size = btrfs_extent_inline_ref_size(type); |
1122 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1123 | ptr = (unsigned long)iref; | |
1124 | end = (unsigned long)ei + item_size; | |
1125 | if (ptr + size < end) | |
1126 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1127 | end - ptr - size); | |
1128 | item_size -= size; | |
78ac4f9e | 1129 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1130 | } |
1131 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1132 | } |
1133 | ||
1134 | static noinline_for_stack | |
1135 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1136 | struct btrfs_path *path, |
1137 | u64 bytenr, u64 num_bytes, u64 parent, | |
1138 | u64 root_objectid, u64 owner, | |
1139 | u64 offset, int refs_to_add, | |
1140 | struct btrfs_delayed_extent_op *extent_op) | |
1141 | { | |
1142 | struct btrfs_extent_inline_ref *iref; | |
1143 | int ret; | |
1144 | ||
867cc1fb NB |
1145 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1146 | num_bytes, parent, root_objectid, | |
1147 | owner, offset, 1); | |
5d4f98a2 | 1148 | if (ret == 0) { |
07cce5cf QW |
1149 | /* |
1150 | * We're adding refs to a tree block we already own, this | |
1151 | * should not happen at all. | |
1152 | */ | |
1153 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1154 | btrfs_crit(trans->fs_info, | |
1155 | "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu", | |
1156 | bytenr, num_bytes, root_objectid); | |
1157 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
1158 | WARN_ON(1); | |
1159 | btrfs_crit(trans->fs_info, | |
1160 | "path->slots[0]=%d path->nodes[0]:", path->slots[0]); | |
1161 | btrfs_print_leaf(path->nodes[0]); | |
1162 | } | |
1163 | return -EUCLEAN; | |
1164 | } | |
61a18f1c NB |
1165 | update_inline_extent_backref(path, iref, refs_to_add, |
1166 | extent_op, NULL); | |
5d4f98a2 | 1167 | } else if (ret == -ENOENT) { |
a639cdeb | 1168 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1169 | root_objectid, owner, offset, |
1170 | refs_to_add, extent_op); | |
1171 | ret = 0; | |
771ed689 | 1172 | } |
5d4f98a2 YZ |
1173 | return ret; |
1174 | } | |
31840ae1 | 1175 | |
5d4f98a2 | 1176 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1177 | struct btrfs_path *path, |
1178 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1179 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1180 | { |
143bede5 | 1181 | int ret = 0; |
b9473439 | 1182 | |
5d4f98a2 YZ |
1183 | BUG_ON(!is_data && refs_to_drop != 1); |
1184 | if (iref) { | |
61a18f1c NB |
1185 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1186 | last_ref); | |
5d4f98a2 | 1187 | } else if (is_data) { |
e9f6290d | 1188 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1189 | last_ref); |
5d4f98a2 | 1190 | } else { |
fcebe456 | 1191 | *last_ref = 1; |
87cc7a8a | 1192 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1193 | } |
1194 | return ret; | |
1195 | } | |
1196 | ||
d04c6b88 JM |
1197 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1198 | u64 *discarded_bytes) | |
5d4f98a2 | 1199 | { |
86557861 JM |
1200 | int j, ret = 0; |
1201 | u64 bytes_left, end; | |
4d89d377 | 1202 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1203 | |
4d89d377 JM |
1204 | if (WARN_ON(start != aligned_start)) { |
1205 | len -= aligned_start - start; | |
1206 | len = round_down(len, 1 << 9); | |
1207 | start = aligned_start; | |
1208 | } | |
d04c6b88 | 1209 | |
4d89d377 | 1210 | *discarded_bytes = 0; |
86557861 JM |
1211 | |
1212 | if (!len) | |
1213 | return 0; | |
1214 | ||
1215 | end = start + len; | |
1216 | bytes_left = len; | |
1217 | ||
1218 | /* Skip any superblocks on this device. */ | |
1219 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1220 | u64 sb_start = btrfs_sb_offset(j); | |
1221 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1222 | u64 size = sb_start - start; | |
1223 | ||
1224 | if (!in_range(sb_start, start, bytes_left) && | |
1225 | !in_range(sb_end, start, bytes_left) && | |
1226 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1227 | continue; | |
1228 | ||
1229 | /* | |
1230 | * Superblock spans beginning of range. Adjust start and | |
1231 | * try again. | |
1232 | */ | |
1233 | if (sb_start <= start) { | |
1234 | start += sb_end - start; | |
1235 | if (start > end) { | |
1236 | bytes_left = 0; | |
1237 | break; | |
1238 | } | |
1239 | bytes_left = end - start; | |
1240 | continue; | |
1241 | } | |
1242 | ||
1243 | if (size) { | |
1244 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1245 | GFP_NOFS, 0); | |
1246 | if (!ret) | |
1247 | *discarded_bytes += size; | |
1248 | else if (ret != -EOPNOTSUPP) | |
1249 | return ret; | |
1250 | } | |
1251 | ||
1252 | start = sb_end; | |
1253 | if (start > end) { | |
1254 | bytes_left = 0; | |
1255 | break; | |
1256 | } | |
1257 | bytes_left = end - start; | |
1258 | } | |
1259 | ||
1260 | if (bytes_left) { | |
1261 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1262 | GFP_NOFS, 0); |
1263 | if (!ret) | |
86557861 | 1264 | *discarded_bytes += bytes_left; |
4d89d377 | 1265 | } |
d04c6b88 | 1266 | return ret; |
5d4f98a2 | 1267 | } |
5d4f98a2 | 1268 | |
6143c23c NA |
1269 | static int do_discard_extent(struct btrfs_bio_stripe *stripe, u64 *bytes) |
1270 | { | |
1271 | struct btrfs_device *dev = stripe->dev; | |
1272 | struct btrfs_fs_info *fs_info = dev->fs_info; | |
1273 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1274 | u64 phys = stripe->physical; | |
1275 | u64 len = stripe->length; | |
1276 | u64 discarded = 0; | |
1277 | int ret = 0; | |
1278 | ||
1279 | /* Zone reset on a zoned filesystem */ | |
1280 | if (btrfs_can_zone_reset(dev, phys, len)) { | |
1281 | u64 src_disc; | |
1282 | ||
1283 | ret = btrfs_reset_device_zone(dev, phys, len, &discarded); | |
1284 | if (ret) | |
1285 | goto out; | |
1286 | ||
1287 | if (!btrfs_dev_replace_is_ongoing(dev_replace) || | |
1288 | dev != dev_replace->srcdev) | |
1289 | goto out; | |
1290 | ||
1291 | src_disc = discarded; | |
1292 | ||
1293 | /* Send to replace target as well */ | |
1294 | ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len, | |
1295 | &discarded); | |
1296 | discarded += src_disc; | |
1297 | } else if (blk_queue_discard(bdev_get_queue(stripe->dev->bdev))) { | |
1298 | ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded); | |
1299 | } else { | |
1300 | ret = 0; | |
1301 | *bytes = 0; | |
1302 | } | |
1303 | ||
1304 | out: | |
1305 | *bytes = discarded; | |
1306 | return ret; | |
1307 | } | |
1308 | ||
2ff7e61e | 1309 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1310 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1311 | { |
6b7faadd | 1312 | int ret = 0; |
5378e607 | 1313 | u64 discarded_bytes = 0; |
6b7faadd QW |
1314 | u64 end = bytenr + num_bytes; |
1315 | u64 cur = bytenr; | |
a1d3c478 | 1316 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1317 | |
e244a0ae | 1318 | |
2999241d FM |
1319 | /* |
1320 | * Avoid races with device replace and make sure our bbio has devices | |
1321 | * associated to its stripes that don't go away while we are discarding. | |
1322 | */ | |
0b246afa | 1323 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd QW |
1324 | while (cur < end) { |
1325 | struct btrfs_bio_stripe *stripe; | |
5d4f98a2 YZ |
1326 | int i; |
1327 | ||
6b7faadd QW |
1328 | num_bytes = end - cur; |
1329 | /* Tell the block device(s) that the sectors can be discarded */ | |
1330 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
1331 | &num_bytes, &bbio, 0); | |
1332 | /* | |
1333 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1334 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1335 | * thus we can't continue anyway. | |
1336 | */ | |
1337 | if (ret < 0) | |
1338 | goto out; | |
5d4f98a2 | 1339 | |
6b7faadd | 1340 | stripe = bbio->stripes; |
a1d3c478 | 1341 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1342 | u64 bytes; |
38b5f68e | 1343 | |
627e0873 FM |
1344 | if (!stripe->dev->bdev) { |
1345 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1346 | continue; | |
1347 | } | |
dcba6e48 | 1348 | |
6143c23c | 1349 | ret = do_discard_extent(stripe, &bytes); |
6b7faadd | 1350 | if (!ret) { |
d04c6b88 | 1351 | discarded_bytes += bytes; |
6b7faadd QW |
1352 | } else if (ret != -EOPNOTSUPP) { |
1353 | /* | |
1354 | * Logic errors or -ENOMEM, or -EIO, but | |
1355 | * unlikely to happen. | |
1356 | * | |
1357 | * And since there are two loops, explicitly | |
1358 | * go to out to avoid confusion. | |
1359 | */ | |
1360 | btrfs_put_bbio(bbio); | |
1361 | goto out; | |
1362 | } | |
d5e2003c JB |
1363 | |
1364 | /* | |
1365 | * Just in case we get back EOPNOTSUPP for some reason, | |
1366 | * just ignore the return value so we don't screw up | |
1367 | * people calling discard_extent. | |
1368 | */ | |
1369 | ret = 0; | |
5d4f98a2 | 1370 | } |
6e9606d2 | 1371 | btrfs_put_bbio(bbio); |
6b7faadd | 1372 | cur += num_bytes; |
5d4f98a2 | 1373 | } |
6b7faadd | 1374 | out: |
0b246afa | 1375 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1376 | |
1377 | if (actual_bytes) | |
1378 | *actual_bytes = discarded_bytes; | |
1379 | ||
5d4f98a2 | 1380 | |
53b381b3 DW |
1381 | if (ret == -EOPNOTSUPP) |
1382 | ret = 0; | |
5d4f98a2 | 1383 | return ret; |
5d4f98a2 YZ |
1384 | } |
1385 | ||
79787eaa | 1386 | /* Can return -ENOMEM */ |
5d4f98a2 | 1387 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1388 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1389 | { |
82fa113f | 1390 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 1391 | int ret; |
66d7e7f0 | 1392 | |
82fa113f QW |
1393 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1394 | generic_ref->action); | |
1395 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1396 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1397 | |
82fa113f | 1398 | if (generic_ref->type == BTRFS_REF_METADATA) |
2187374f | 1399 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL); |
82fa113f | 1400 | else |
2187374f | 1401 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0); |
d7eae340 | 1402 | |
82fa113f | 1403 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1404 | |
5d4f98a2 YZ |
1405 | return ret; |
1406 | } | |
1407 | ||
bd3c685e NB |
1408 | /* |
1409 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1410 | * | |
07cce5cf QW |
1411 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1412 | * how it works. | |
1413 | * | |
bd3c685e NB |
1414 | * @trans: Handle of transaction |
1415 | * | |
1416 | * @node: The delayed ref node used to get the bytenr/length for | |
1417 | * extent whose references are incremented. | |
1418 | * | |
1419 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1420 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1421 | * bytenr of the parent block. Since new extents are always | |
1422 | * created with indirect references, this will only be the case | |
1423 | * when relocating a shared extent. In that case, root_objectid | |
1424 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1425 | * be 0 | |
1426 | * | |
1427 | * @root_objectid: The id of the root where this modification has originated, | |
1428 | * this can be either one of the well-known metadata trees or | |
1429 | * the subvolume id which references this extent. | |
1430 | * | |
1431 | * @owner: For data extents it is the inode number of the owning file. | |
1432 | * For metadata extents this parameter holds the level in the | |
1433 | * tree of the extent. | |
1434 | * | |
1435 | * @offset: For metadata extents the offset is ignored and is currently | |
1436 | * always passed as 0. For data extents it is the fileoffset | |
1437 | * this extent belongs to. | |
1438 | * | |
1439 | * @refs_to_add Number of references to add | |
1440 | * | |
1441 | * @extent_op Pointer to a structure, holding information necessary when | |
1442 | * updating a tree block's flags | |
1443 | * | |
1444 | */ | |
5d4f98a2 | 1445 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1446 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1447 | u64 parent, u64 root_objectid, |
1448 | u64 owner, u64 offset, int refs_to_add, | |
1449 | struct btrfs_delayed_extent_op *extent_op) | |
1450 | { | |
1451 | struct btrfs_path *path; | |
1452 | struct extent_buffer *leaf; | |
1453 | struct btrfs_extent_item *item; | |
fcebe456 | 1454 | struct btrfs_key key; |
c682f9b3 QW |
1455 | u64 bytenr = node->bytenr; |
1456 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1457 | u64 refs; |
1458 | int ret; | |
5d4f98a2 YZ |
1459 | |
1460 | path = btrfs_alloc_path(); | |
1461 | if (!path) | |
1462 | return -ENOMEM; | |
1463 | ||
5d4f98a2 | 1464 | /* this will setup the path even if it fails to insert the back ref */ |
a639cdeb NB |
1465 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1466 | parent, root_objectid, owner, | |
1467 | offset, refs_to_add, extent_op); | |
0ed4792a | 1468 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1469 | goto out; |
fcebe456 JB |
1470 | |
1471 | /* | |
1472 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1473 | * inline extent ref, so just update the reference count and add a | |
1474 | * normal backref. | |
1475 | */ | |
5d4f98a2 | 1476 | leaf = path->nodes[0]; |
fcebe456 | 1477 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1478 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1479 | refs = btrfs_extent_refs(leaf, item); | |
1480 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1481 | if (extent_op) | |
1482 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1483 | |
5d4f98a2 | 1484 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1485 | btrfs_release_path(path); |
56bec294 | 1486 | |
56bec294 | 1487 | /* now insert the actual backref */ |
65cd6d9e NB |
1488 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
1489 | BUG_ON(refs_to_add != 1); | |
1490 | ret = insert_tree_block_ref(trans, path, bytenr, parent, | |
1491 | root_objectid); | |
1492 | } else { | |
1493 | ret = insert_extent_data_ref(trans, path, bytenr, parent, | |
1494 | root_objectid, owner, offset, | |
1495 | refs_to_add); | |
1496 | } | |
79787eaa | 1497 | if (ret) |
66642832 | 1498 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1499 | out: |
56bec294 | 1500 | btrfs_free_path(path); |
30d133fc | 1501 | return ret; |
56bec294 CM |
1502 | } |
1503 | ||
5d4f98a2 | 1504 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1505 | struct btrfs_delayed_ref_node *node, |
1506 | struct btrfs_delayed_extent_op *extent_op, | |
1507 | int insert_reserved) | |
56bec294 | 1508 | { |
5d4f98a2 YZ |
1509 | int ret = 0; |
1510 | struct btrfs_delayed_data_ref *ref; | |
1511 | struct btrfs_key ins; | |
1512 | u64 parent = 0; | |
1513 | u64 ref_root = 0; | |
1514 | u64 flags = 0; | |
1515 | ||
1516 | ins.objectid = node->bytenr; | |
1517 | ins.offset = node->num_bytes; | |
1518 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1519 | ||
1520 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1521 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1522 | |
5d4f98a2 YZ |
1523 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1524 | parent = ref->parent; | |
fcebe456 | 1525 | ref_root = ref->root; |
5d4f98a2 YZ |
1526 | |
1527 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1528 | if (extent_op) |
5d4f98a2 | 1529 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1530 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1531 | flags, ref->objectid, | |
1532 | ref->offset, &ins, | |
1533 | node->ref_mod); | |
5d4f98a2 | 1534 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1535 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1536 | ref->objectid, ref->offset, | |
1537 | node->ref_mod, extent_op); | |
5d4f98a2 | 1538 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1539 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1540 | ref_root, ref->objectid, |
1541 | ref->offset, node->ref_mod, | |
c682f9b3 | 1542 | extent_op); |
5d4f98a2 YZ |
1543 | } else { |
1544 | BUG(); | |
1545 | } | |
1546 | return ret; | |
1547 | } | |
1548 | ||
1549 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1550 | struct extent_buffer *leaf, | |
1551 | struct btrfs_extent_item *ei) | |
1552 | { | |
1553 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1554 | if (extent_op->update_flags) { | |
1555 | flags |= extent_op->flags_to_set; | |
1556 | btrfs_set_extent_flags(leaf, ei, flags); | |
1557 | } | |
1558 | ||
1559 | if (extent_op->update_key) { | |
1560 | struct btrfs_tree_block_info *bi; | |
1561 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1562 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1563 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1564 | } | |
1565 | } | |
1566 | ||
1567 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1568 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1569 | struct btrfs_delayed_extent_op *extent_op) |
1570 | { | |
20b9a2d6 | 1571 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1572 | struct btrfs_key key; |
1573 | struct btrfs_path *path; | |
1574 | struct btrfs_extent_item *ei; | |
1575 | struct extent_buffer *leaf; | |
1576 | u32 item_size; | |
56bec294 | 1577 | int ret; |
5d4f98a2 | 1578 | int err = 0; |
b1c79e09 | 1579 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1580 | |
bf31f87f | 1581 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1582 | return 0; |
1583 | ||
0b246afa | 1584 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1585 | metadata = 0; |
1586 | ||
5d4f98a2 YZ |
1587 | path = btrfs_alloc_path(); |
1588 | if (!path) | |
1589 | return -ENOMEM; | |
1590 | ||
d278850e | 1591 | key.objectid = head->bytenr; |
5d4f98a2 | 1592 | |
3173a18f | 1593 | if (metadata) { |
3173a18f | 1594 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1595 | key.offset = extent_op->level; |
3173a18f JB |
1596 | } else { |
1597 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1598 | key.offset = head->num_bytes; |
3173a18f JB |
1599 | } |
1600 | ||
1601 | again: | |
0b246afa | 1602 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1603 | if (ret < 0) { |
1604 | err = ret; | |
1605 | goto out; | |
1606 | } | |
1607 | if (ret > 0) { | |
3173a18f | 1608 | if (metadata) { |
55994887 FDBM |
1609 | if (path->slots[0] > 0) { |
1610 | path->slots[0]--; | |
1611 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1612 | path->slots[0]); | |
d278850e | 1613 | if (key.objectid == head->bytenr && |
55994887 | 1614 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1615 | key.offset == head->num_bytes) |
55994887 FDBM |
1616 | ret = 0; |
1617 | } | |
1618 | if (ret > 0) { | |
1619 | btrfs_release_path(path); | |
1620 | metadata = 0; | |
3173a18f | 1621 | |
d278850e JB |
1622 | key.objectid = head->bytenr; |
1623 | key.offset = head->num_bytes; | |
55994887 FDBM |
1624 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1625 | goto again; | |
1626 | } | |
1627 | } else { | |
1628 | err = -EIO; | |
1629 | goto out; | |
3173a18f | 1630 | } |
5d4f98a2 YZ |
1631 | } |
1632 | ||
1633 | leaf = path->nodes[0]; | |
1634 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1635 | |
6d8ff4e4 | 1636 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1637 | err = -EINVAL; |
1638 | btrfs_print_v0_err(fs_info); | |
1639 | btrfs_abort_transaction(trans, err); | |
1640 | goto out; | |
1641 | } | |
1642 | ||
5d4f98a2 YZ |
1643 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1644 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1645 | |
5d4f98a2 YZ |
1646 | btrfs_mark_buffer_dirty(leaf); |
1647 | out: | |
1648 | btrfs_free_path(path); | |
1649 | return err; | |
56bec294 CM |
1650 | } |
1651 | ||
5d4f98a2 | 1652 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1653 | struct btrfs_delayed_ref_node *node, |
1654 | struct btrfs_delayed_extent_op *extent_op, | |
1655 | int insert_reserved) | |
56bec294 CM |
1656 | { |
1657 | int ret = 0; | |
5d4f98a2 | 1658 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1659 | u64 parent = 0; |
1660 | u64 ref_root = 0; | |
56bec294 | 1661 | |
5d4f98a2 | 1662 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1663 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1664 | |
5d4f98a2 YZ |
1665 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1666 | parent = ref->parent; | |
fcebe456 | 1667 | ref_root = ref->root; |
5d4f98a2 | 1668 | |
02794222 | 1669 | if (node->ref_mod != 1) { |
f97806f2 | 1670 | btrfs_err(trans->fs_info, |
02794222 LB |
1671 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1672 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1673 | parent); | |
1674 | return -EIO; | |
1675 | } | |
5d4f98a2 | 1676 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1677 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1678 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1679 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1680 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1681 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1682 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1683 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1684 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1685 | } else { |
1686 | BUG(); | |
1687 | } | |
56bec294 CM |
1688 | return ret; |
1689 | } | |
1690 | ||
1691 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1692 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1693 | struct btrfs_delayed_ref_node *node, |
1694 | struct btrfs_delayed_extent_op *extent_op, | |
1695 | int insert_reserved) | |
56bec294 | 1696 | { |
79787eaa JM |
1697 | int ret = 0; |
1698 | ||
bf31f87f | 1699 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1700 | if (insert_reserved) |
b25c36f8 | 1701 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1702 | return 0; |
857cc2fc | 1703 | } |
79787eaa | 1704 | |
5d4f98a2 YZ |
1705 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1706 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1707 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1708 | insert_reserved); |
1709 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1710 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1711 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1712 | insert_reserved); |
1713 | else | |
1714 | BUG(); | |
80ee54bf | 1715 | if (ret && insert_reserved) |
b25c36f8 | 1716 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
5d4f98a2 | 1717 | return ret; |
56bec294 CM |
1718 | } |
1719 | ||
c6fc2454 | 1720 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1721 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1722 | { | |
cffc3374 FM |
1723 | struct btrfs_delayed_ref_node *ref; |
1724 | ||
e3d03965 | 1725 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1726 | return NULL; |
d7df2c79 | 1727 | |
cffc3374 FM |
1728 | /* |
1729 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1730 | * This is to prevent a ref count from going down to zero, which deletes | |
1731 | * the extent item from the extent tree, when there still are references | |
1732 | * to add, which would fail because they would not find the extent item. | |
1733 | */ | |
1d57ee94 WX |
1734 | if (!list_empty(&head->ref_add_list)) |
1735 | return list_first_entry(&head->ref_add_list, | |
1736 | struct btrfs_delayed_ref_node, add_list); | |
1737 | ||
e3d03965 | 1738 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1739 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1740 | ASSERT(list_empty(&ref->add_list)); |
1741 | return ref; | |
56bec294 CM |
1742 | } |
1743 | ||
2eadaa22 JB |
1744 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1745 | struct btrfs_delayed_ref_head *head) | |
1746 | { | |
1747 | spin_lock(&delayed_refs->lock); | |
1748 | head->processing = 0; | |
1749 | delayed_refs->num_heads_ready++; | |
1750 | spin_unlock(&delayed_refs->lock); | |
1751 | btrfs_delayed_ref_unlock(head); | |
1752 | } | |
1753 | ||
bedc6617 JB |
1754 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1755 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1756 | { |
1757 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1758 | |
1759 | if (!extent_op) | |
bedc6617 JB |
1760 | return NULL; |
1761 | ||
b00e6250 | 1762 | if (head->must_insert_reserved) { |
bedc6617 | 1763 | head->extent_op = NULL; |
b00e6250 | 1764 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1765 | return NULL; |
b00e6250 | 1766 | } |
bedc6617 JB |
1767 | return extent_op; |
1768 | } | |
1769 | ||
1770 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1771 | struct btrfs_delayed_ref_head *head) | |
1772 | { | |
1773 | struct btrfs_delayed_extent_op *extent_op; | |
1774 | int ret; | |
1775 | ||
1776 | extent_op = cleanup_extent_op(head); | |
1777 | if (!extent_op) | |
1778 | return 0; | |
1779 | head->extent_op = NULL; | |
b00e6250 | 1780 | spin_unlock(&head->lock); |
20b9a2d6 | 1781 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1782 | btrfs_free_delayed_extent_op(extent_op); |
1783 | return ret ? ret : 1; | |
1784 | } | |
1785 | ||
31890da0 JB |
1786 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1787 | struct btrfs_delayed_ref_root *delayed_refs, | |
1788 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1789 | { |
ba2c4d4e | 1790 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 | 1791 | |
81e75ac7 JB |
1792 | /* |
1793 | * We had csum deletions accounted for in our delayed refs rsv, we need | |
1794 | * to drop the csum leaves for this update from our delayed_refs_rsv. | |
1795 | */ | |
1796 | if (head->total_ref_mod < 0 && head->is_data) { | |
1797 | spin_lock(&delayed_refs->lock); | |
1798 | delayed_refs->pending_csums -= head->num_bytes; | |
1799 | spin_unlock(&delayed_refs->lock); | |
1800 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes); | |
1801 | } | |
1802 | ||
1803 | /* | |
1804 | * We were dropping refs, or had a new ref and dropped it, and thus must | |
1805 | * adjust down our total_bytes_pinned, the space may or may not have | |
1806 | * been pinned and so is accounted for properly in the pinned space by | |
1807 | * now. | |
1808 | */ | |
1809 | if (head->total_ref_mod < 0 || | |
1810 | (head->total_ref_mod == 0 && head->must_insert_reserved)) { | |
2187374f | 1811 | u64 flags = btrfs_ref_head_to_space_flags(head); |
07c47775 | 1812 | |
2187374f | 1813 | btrfs_mod_total_bytes_pinned(fs_info, flags, -head->num_bytes); |
07c47775 JB |
1814 | } |
1815 | ||
ba2c4d4e | 1816 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1817 | } |
1818 | ||
194ab0bc | 1819 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1820 | struct btrfs_delayed_ref_head *head) |
1821 | { | |
f9871edd NB |
1822 | |
1823 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1824 | struct btrfs_delayed_ref_root *delayed_refs; |
1825 | int ret; | |
1826 | ||
1827 | delayed_refs = &trans->transaction->delayed_refs; | |
1828 | ||
bedc6617 | 1829 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1830 | if (ret < 0) { |
1831 | unselect_delayed_ref_head(delayed_refs, head); | |
1832 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1833 | return ret; | |
1834 | } else if (ret) { | |
1835 | return ret; | |
1836 | } | |
1837 | ||
1838 | /* | |
1839 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1840 | * and then re-check to make sure nobody got added. | |
1841 | */ | |
1842 | spin_unlock(&head->lock); | |
1843 | spin_lock(&delayed_refs->lock); | |
1844 | spin_lock(&head->lock); | |
e3d03965 | 1845 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1846 | spin_unlock(&head->lock); |
1847 | spin_unlock(&delayed_refs->lock); | |
1848 | return 1; | |
1849 | } | |
d7baffda | 1850 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1851 | spin_unlock(&head->lock); |
1e7a1421 | 1852 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1853 | |
c1103f7a | 1854 | if (head->must_insert_reserved) { |
b25c36f8 | 1855 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1856 | if (head->is_data) { |
40e046ac FM |
1857 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1858 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1859 | } |
1860 | } | |
1861 | ||
31890da0 | 1862 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1863 | |
1864 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1865 | btrfs_delayed_ref_unlock(head); |
d278850e | 1866 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1867 | return 0; |
1868 | } | |
1869 | ||
b1cdbcb5 NB |
1870 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1871 | struct btrfs_trans_handle *trans) | |
1872 | { | |
1873 | struct btrfs_delayed_ref_root *delayed_refs = | |
1874 | &trans->transaction->delayed_refs; | |
1875 | struct btrfs_delayed_ref_head *head = NULL; | |
1876 | int ret; | |
1877 | ||
1878 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1879 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1880 | if (!head) { |
1881 | spin_unlock(&delayed_refs->lock); | |
1882 | return head; | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * Grab the lock that says we are going to process all the refs for | |
1887 | * this head | |
1888 | */ | |
9e920a6f | 1889 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1890 | spin_unlock(&delayed_refs->lock); |
1891 | ||
1892 | /* | |
1893 | * We may have dropped the spin lock to get the head mutex lock, and | |
1894 | * that might have given someone else time to free the head. If that's | |
1895 | * true, it has been removed from our list and we can move on. | |
1896 | */ | |
1897 | if (ret == -EAGAIN) | |
1898 | head = ERR_PTR(-EAGAIN); | |
1899 | ||
1900 | return head; | |
1901 | } | |
1902 | ||
e7261386 NB |
1903 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1904 | struct btrfs_delayed_ref_head *locked_ref, | |
1905 | unsigned long *run_refs) | |
1906 | { | |
1907 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1908 | struct btrfs_delayed_ref_root *delayed_refs; | |
1909 | struct btrfs_delayed_extent_op *extent_op; | |
1910 | struct btrfs_delayed_ref_node *ref; | |
1911 | int must_insert_reserved = 0; | |
1912 | int ret; | |
1913 | ||
1914 | delayed_refs = &trans->transaction->delayed_refs; | |
1915 | ||
0110a4c4 NB |
1916 | lockdep_assert_held(&locked_ref->mutex); |
1917 | lockdep_assert_held(&locked_ref->lock); | |
1918 | ||
e7261386 NB |
1919 | while ((ref = select_delayed_ref(locked_ref))) { |
1920 | if (ref->seq && | |
1921 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1922 | spin_unlock(&locked_ref->lock); | |
1923 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1924 | return -EAGAIN; | |
1925 | } | |
1926 | ||
1927 | (*run_refs)++; | |
1928 | ref->in_tree = 0; | |
1929 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1930 | RB_CLEAR_NODE(&ref->ref_node); | |
1931 | if (!list_empty(&ref->add_list)) | |
1932 | list_del(&ref->add_list); | |
1933 | /* | |
1934 | * When we play the delayed ref, also correct the ref_mod on | |
1935 | * head | |
1936 | */ | |
1937 | switch (ref->action) { | |
1938 | case BTRFS_ADD_DELAYED_REF: | |
1939 | case BTRFS_ADD_DELAYED_EXTENT: | |
1940 | locked_ref->ref_mod -= ref->ref_mod; | |
1941 | break; | |
1942 | case BTRFS_DROP_DELAYED_REF: | |
1943 | locked_ref->ref_mod += ref->ref_mod; | |
1944 | break; | |
1945 | default: | |
1946 | WARN_ON(1); | |
1947 | } | |
1948 | atomic_dec(&delayed_refs->num_entries); | |
1949 | ||
1950 | /* | |
1951 | * Record the must_insert_reserved flag before we drop the | |
1952 | * spin lock. | |
1953 | */ | |
1954 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1955 | locked_ref->must_insert_reserved = 0; | |
1956 | ||
1957 | extent_op = locked_ref->extent_op; | |
1958 | locked_ref->extent_op = NULL; | |
1959 | spin_unlock(&locked_ref->lock); | |
1960 | ||
1961 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1962 | must_insert_reserved); | |
1963 | ||
1964 | btrfs_free_delayed_extent_op(extent_op); | |
1965 | if (ret) { | |
1966 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1967 | btrfs_put_delayed_ref(ref); | |
1968 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1969 | ret); | |
1970 | return ret; | |
1971 | } | |
1972 | ||
1973 | btrfs_put_delayed_ref(ref); | |
1974 | cond_resched(); | |
1975 | ||
1976 | spin_lock(&locked_ref->lock); | |
1977 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1978 | } | |
1979 | ||
1980 | return 0; | |
1981 | } | |
1982 | ||
79787eaa JM |
1983 | /* |
1984 | * Returns 0 on success or if called with an already aborted transaction. | |
1985 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1986 | */ | |
d7df2c79 | 1987 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 1988 | unsigned long nr) |
56bec294 | 1989 | { |
0a1e458a | 1990 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 1991 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 1992 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 1993 | ktime_t start = ktime_get(); |
56bec294 | 1994 | int ret; |
d7df2c79 | 1995 | unsigned long count = 0; |
0a2b2a84 | 1996 | unsigned long actual_count = 0; |
56bec294 CM |
1997 | |
1998 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 1999 | do { |
56bec294 | 2000 | if (!locked_ref) { |
b1cdbcb5 | 2001 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2002 | if (IS_ERR_OR_NULL(locked_ref)) { |
2003 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2004 | continue; | |
2005 | } else { | |
2006 | break; | |
2007 | } | |
56bec294 | 2008 | } |
0110a4c4 | 2009 | count++; |
56bec294 | 2010 | } |
2c3cf7d5 FM |
2011 | /* |
2012 | * We need to try and merge add/drops of the same ref since we | |
2013 | * can run into issues with relocate dropping the implicit ref | |
2014 | * and then it being added back again before the drop can | |
2015 | * finish. If we merged anything we need to re-loop so we can | |
2016 | * get a good ref. | |
2017 | * Or we can get node references of the same type that weren't | |
2018 | * merged when created due to bumps in the tree mod seq, and | |
2019 | * we need to merge them to prevent adding an inline extent | |
2020 | * backref before dropping it (triggering a BUG_ON at | |
2021 | * insert_inline_extent_backref()). | |
2022 | */ | |
d7df2c79 | 2023 | spin_lock(&locked_ref->lock); |
be97f133 | 2024 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2025 | |
0110a4c4 NB |
2026 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2027 | &actual_count); | |
2028 | if (ret < 0 && ret != -EAGAIN) { | |
2029 | /* | |
2030 | * Error, btrfs_run_delayed_refs_for_head already | |
2031 | * unlocked everything so just bail out | |
2032 | */ | |
2033 | return ret; | |
2034 | } else if (!ret) { | |
2035 | /* | |
2036 | * Success, perform the usual cleanup of a processed | |
2037 | * head | |
2038 | */ | |
f9871edd | 2039 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2040 | if (ret > 0 ) { |
b00e6250 JB |
2041 | /* We dropped our lock, we need to loop. */ |
2042 | ret = 0; | |
d7df2c79 | 2043 | continue; |
194ab0bc JB |
2044 | } else if (ret) { |
2045 | return ret; | |
5d4f98a2 | 2046 | } |
22cd2e7d | 2047 | } |
1ce7a5ec | 2048 | |
b00e6250 | 2049 | /* |
0110a4c4 NB |
2050 | * Either success case or btrfs_run_delayed_refs_for_head |
2051 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2052 | */ |
b00e6250 | 2053 | |
0110a4c4 | 2054 | locked_ref = NULL; |
c3e69d58 | 2055 | cond_resched(); |
0110a4c4 | 2056 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2057 | |
2058 | /* | |
2059 | * We don't want to include ref heads since we can have empty ref heads | |
2060 | * and those will drastically skew our runtime down since we just do | |
2061 | * accounting, no actual extent tree updates. | |
2062 | */ | |
2063 | if (actual_count > 0) { | |
2064 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2065 | u64 avg; | |
2066 | ||
2067 | /* | |
2068 | * We weigh the current average higher than our current runtime | |
2069 | * to avoid large swings in the average. | |
2070 | */ | |
2071 | spin_lock(&delayed_refs->lock); | |
2072 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2073 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2074 | spin_unlock(&delayed_refs->lock); |
2075 | } | |
d7df2c79 | 2076 | return 0; |
c3e69d58 CM |
2077 | } |
2078 | ||
709c0486 AJ |
2079 | #ifdef SCRAMBLE_DELAYED_REFS |
2080 | /* | |
2081 | * Normally delayed refs get processed in ascending bytenr order. This | |
2082 | * correlates in most cases to the order added. To expose dependencies on this | |
2083 | * order, we start to process the tree in the middle instead of the beginning | |
2084 | */ | |
2085 | static u64 find_middle(struct rb_root *root) | |
2086 | { | |
2087 | struct rb_node *n = root->rb_node; | |
2088 | struct btrfs_delayed_ref_node *entry; | |
2089 | int alt = 1; | |
2090 | u64 middle; | |
2091 | u64 first = 0, last = 0; | |
2092 | ||
2093 | n = rb_first(root); | |
2094 | if (n) { | |
2095 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2096 | first = entry->bytenr; | |
2097 | } | |
2098 | n = rb_last(root); | |
2099 | if (n) { | |
2100 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2101 | last = entry->bytenr; | |
2102 | } | |
2103 | n = root->rb_node; | |
2104 | ||
2105 | while (n) { | |
2106 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2107 | WARN_ON(!entry->in_tree); | |
2108 | ||
2109 | middle = entry->bytenr; | |
2110 | ||
2111 | if (alt) | |
2112 | n = n->rb_left; | |
2113 | else | |
2114 | n = n->rb_right; | |
2115 | ||
2116 | alt = 1 - alt; | |
2117 | } | |
2118 | return middle; | |
2119 | } | |
2120 | #endif | |
2121 | ||
c3e69d58 CM |
2122 | /* |
2123 | * this starts processing the delayed reference count updates and | |
2124 | * extent insertions we have queued up so far. count can be | |
2125 | * 0, which means to process everything in the tree at the start | |
2126 | * of the run (but not newly added entries), or it can be some target | |
2127 | * number you'd like to process. | |
79787eaa JM |
2128 | * |
2129 | * Returns 0 on success or if called with an aborted transaction | |
2130 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2131 | */ |
2132 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2133 | unsigned long count) |
c3e69d58 | 2134 | { |
c79a70b1 | 2135 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2136 | struct rb_node *node; |
2137 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2138 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2139 | int ret; |
2140 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2141 | |
79787eaa | 2142 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2143 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2144 | return 0; |
2145 | ||
0b246afa | 2146 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2147 | return 0; |
2148 | ||
c3e69d58 | 2149 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2150 | if (count == 0) |
61a56a99 | 2151 | count = delayed_refs->num_heads_ready; |
bb721703 | 2152 | |
c3e69d58 | 2153 | again: |
709c0486 AJ |
2154 | #ifdef SCRAMBLE_DELAYED_REFS |
2155 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2156 | #endif | |
0a1e458a | 2157 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2158 | if (ret < 0) { |
66642832 | 2159 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2160 | return ret; |
eb099670 | 2161 | } |
c3e69d58 | 2162 | |
56bec294 | 2163 | if (run_all) { |
119e80df | 2164 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2165 | |
d7df2c79 | 2166 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2167 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2168 | if (!node) { |
2169 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2170 | goto out; |
d7df2c79 | 2171 | } |
d278850e JB |
2172 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2173 | href_node); | |
2174 | refcount_inc(&head->refs); | |
2175 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2176 | |
d278850e JB |
2177 | /* Mutex was contended, block until it's released and retry. */ |
2178 | mutex_lock(&head->mutex); | |
2179 | mutex_unlock(&head->mutex); | |
56bec294 | 2180 | |
d278850e | 2181 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2182 | cond_resched(); |
56bec294 | 2183 | goto again; |
5f39d397 | 2184 | } |
54aa1f4d | 2185 | out: |
a28ec197 CM |
2186 | return 0; |
2187 | } | |
2188 | ||
5d4f98a2 | 2189 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
42c9d0b5 | 2190 | struct extent_buffer *eb, u64 flags, |
b1c79e09 | 2191 | int level, int is_data) |
5d4f98a2 YZ |
2192 | { |
2193 | struct btrfs_delayed_extent_op *extent_op; | |
2194 | int ret; | |
2195 | ||
78a6184a | 2196 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2197 | if (!extent_op) |
2198 | return -ENOMEM; | |
2199 | ||
2200 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2201 | extent_op->update_flags = true; |
2202 | extent_op->update_key = false; | |
2203 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2204 | extent_op->level = level; |
5d4f98a2 | 2205 | |
42c9d0b5 | 2206 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2207 | if (ret) |
78a6184a | 2208 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2209 | return ret; |
2210 | } | |
2211 | ||
e4c3b2dc | 2212 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2213 | struct btrfs_path *path, |
2214 | u64 objectid, u64 offset, u64 bytenr) | |
2215 | { | |
2216 | struct btrfs_delayed_ref_head *head; | |
2217 | struct btrfs_delayed_ref_node *ref; | |
2218 | struct btrfs_delayed_data_ref *data_ref; | |
2219 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2220 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2221 | struct rb_node *node; |
5d4f98a2 YZ |
2222 | int ret = 0; |
2223 | ||
998ac6d2 | 2224 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2225 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2226 | if (cur_trans) |
2227 | refcount_inc(&cur_trans->use_count); | |
2228 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2229 | if (!cur_trans) |
2230 | return 0; | |
2231 | ||
2232 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2233 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2234 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2235 | if (!head) { |
2236 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2237 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2238 | return 0; |
2239 | } | |
5d4f98a2 YZ |
2240 | |
2241 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2242 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2243 | spin_unlock(&delayed_refs->lock); |
2244 | ||
b3b4aa74 | 2245 | btrfs_release_path(path); |
5d4f98a2 | 2246 | |
8cc33e5c DS |
2247 | /* |
2248 | * Mutex was contended, block until it's released and let | |
2249 | * caller try again | |
2250 | */ | |
5d4f98a2 YZ |
2251 | mutex_lock(&head->mutex); |
2252 | mutex_unlock(&head->mutex); | |
d278850e | 2253 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2254 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2255 | return -EAGAIN; |
2256 | } | |
d7df2c79 | 2257 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2258 | |
d7df2c79 | 2259 | spin_lock(&head->lock); |
0e0adbcf JB |
2260 | /* |
2261 | * XXX: We should replace this with a proper search function in the | |
2262 | * future. | |
2263 | */ | |
e3d03965 LB |
2264 | for (node = rb_first_cached(&head->ref_tree); node; |
2265 | node = rb_next(node)) { | |
0e0adbcf | 2266 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2267 | /* If it's a shared ref we know a cross reference exists */ |
2268 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2269 | ret = 1; | |
2270 | break; | |
2271 | } | |
5d4f98a2 | 2272 | |
d7df2c79 | 2273 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2274 | |
d7df2c79 JB |
2275 | /* |
2276 | * If our ref doesn't match the one we're currently looking at | |
2277 | * then we have a cross reference. | |
2278 | */ | |
2279 | if (data_ref->root != root->root_key.objectid || | |
2280 | data_ref->objectid != objectid || | |
2281 | data_ref->offset != offset) { | |
2282 | ret = 1; | |
2283 | break; | |
2284 | } | |
5d4f98a2 | 2285 | } |
d7df2c79 | 2286 | spin_unlock(&head->lock); |
5d4f98a2 | 2287 | mutex_unlock(&head->mutex); |
998ac6d2 | 2288 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2289 | return ret; |
2290 | } | |
2291 | ||
e4c3b2dc | 2292 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2293 | struct btrfs_path *path, |
a84d5d42 BB |
2294 | u64 objectid, u64 offset, u64 bytenr, |
2295 | bool strict) | |
be20aa9d | 2296 | { |
0b246afa JM |
2297 | struct btrfs_fs_info *fs_info = root->fs_info; |
2298 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2299 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2300 | struct btrfs_extent_data_ref *ref; |
2301 | struct btrfs_extent_inline_ref *iref; | |
2302 | struct btrfs_extent_item *ei; | |
f321e491 | 2303 | struct btrfs_key key; |
5d4f98a2 | 2304 | u32 item_size; |
3de28d57 | 2305 | int type; |
be20aa9d | 2306 | int ret; |
925baedd | 2307 | |
be20aa9d | 2308 | key.objectid = bytenr; |
31840ae1 | 2309 | key.offset = (u64)-1; |
f321e491 | 2310 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2311 | |
be20aa9d CM |
2312 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2313 | if (ret < 0) | |
2314 | goto out; | |
79787eaa | 2315 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2316 | |
2317 | ret = -ENOENT; | |
2318 | if (path->slots[0] == 0) | |
31840ae1 | 2319 | goto out; |
be20aa9d | 2320 | |
31840ae1 | 2321 | path->slots[0]--; |
f321e491 | 2322 | leaf = path->nodes[0]; |
5d4f98a2 | 2323 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2324 | |
5d4f98a2 | 2325 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2326 | goto out; |
f321e491 | 2327 | |
5d4f98a2 YZ |
2328 | ret = 1; |
2329 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2330 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2331 | |
a6bd9cd1 | 2332 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2333 | if (item_size != sizeof(*ei) + |
2334 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2335 | goto out; | |
be20aa9d | 2336 | |
a84d5d42 BB |
2337 | /* |
2338 | * If extent created before last snapshot => it's shared unless the | |
2339 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2340 | */ | |
2341 | if (!strict && | |
2342 | (btrfs_extent_generation(leaf, ei) <= | |
2343 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2344 | goto out; |
2345 | ||
2346 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2347 | |
a6bd9cd1 | 2348 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2349 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2350 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2351 | goto out; |
2352 | ||
2353 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2354 | if (btrfs_extent_refs(leaf, ei) != | |
2355 | btrfs_extent_data_ref_count(leaf, ref) || | |
2356 | btrfs_extent_data_ref_root(leaf, ref) != | |
2357 | root->root_key.objectid || | |
2358 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2359 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2360 | goto out; | |
2361 | ||
2362 | ret = 0; | |
2363 | out: | |
2364 | return ret; | |
2365 | } | |
2366 | ||
e4c3b2dc | 2367 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
a84d5d42 | 2368 | u64 bytenr, bool strict) |
5d4f98a2 YZ |
2369 | { |
2370 | struct btrfs_path *path; | |
2371 | int ret; | |
5d4f98a2 YZ |
2372 | |
2373 | path = btrfs_alloc_path(); | |
2374 | if (!path) | |
9132c4ff | 2375 | return -ENOMEM; |
5d4f98a2 YZ |
2376 | |
2377 | do { | |
e4c3b2dc | 2378 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2379 | offset, bytenr, strict); |
5d4f98a2 | 2380 | if (ret && ret != -ENOENT) |
f321e491 | 2381 | goto out; |
80ff3856 | 2382 | |
380fd066 MT |
2383 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2384 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2385 | |
be20aa9d | 2386 | out: |
80ff3856 | 2387 | btrfs_free_path(path); |
f0486c68 YZ |
2388 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2389 | WARN_ON(ret > 0); | |
f321e491 | 2390 | return ret; |
be20aa9d | 2391 | } |
c5739bba | 2392 | |
5d4f98a2 | 2393 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2394 | struct btrfs_root *root, |
5d4f98a2 | 2395 | struct extent_buffer *buf, |
e339a6b0 | 2396 | int full_backref, int inc) |
31840ae1 | 2397 | { |
0b246afa | 2398 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2399 | u64 bytenr; |
5d4f98a2 YZ |
2400 | u64 num_bytes; |
2401 | u64 parent; | |
31840ae1 | 2402 | u64 ref_root; |
31840ae1 | 2403 | u32 nritems; |
31840ae1 ZY |
2404 | struct btrfs_key key; |
2405 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2406 | struct btrfs_ref generic_ref = { 0 }; |
2407 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2408 | int i; |
82fa113f | 2409 | int action; |
31840ae1 ZY |
2410 | int level; |
2411 | int ret = 0; | |
fccb84c9 | 2412 | |
0b246afa | 2413 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2414 | return 0; |
fccb84c9 | 2415 | |
31840ae1 | 2416 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2417 | nritems = btrfs_header_nritems(buf); |
2418 | level = btrfs_header_level(buf); | |
2419 | ||
92a7cc42 | 2420 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2421 | return 0; |
31840ae1 | 2422 | |
5d4f98a2 YZ |
2423 | if (full_backref) |
2424 | parent = buf->start; | |
2425 | else | |
2426 | parent = 0; | |
82fa113f QW |
2427 | if (inc) |
2428 | action = BTRFS_ADD_DELAYED_REF; | |
2429 | else | |
2430 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2431 | |
2432 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2433 | if (level == 0) { |
5d4f98a2 | 2434 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2435 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2436 | continue; |
5d4f98a2 | 2437 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2438 | struct btrfs_file_extent_item); |
2439 | if (btrfs_file_extent_type(buf, fi) == | |
2440 | BTRFS_FILE_EXTENT_INLINE) | |
2441 | continue; | |
2442 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2443 | if (bytenr == 0) | |
2444 | continue; | |
5d4f98a2 YZ |
2445 | |
2446 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2447 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2448 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2449 | num_bytes, parent); | |
2450 | generic_ref.real_root = root->root_key.objectid; | |
2451 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2452 | key.offset); | |
2453 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2454 | if (inc) |
82fa113f | 2455 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2456 | else |
ffd4bb2a | 2457 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2458 | if (ret) |
2459 | goto fail; | |
2460 | } else { | |
5d4f98a2 | 2461 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2462 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2463 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2464 | num_bytes, parent); | |
2465 | generic_ref.real_root = root->root_key.objectid; | |
2466 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2467 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2468 | if (inc) |
82fa113f | 2469 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2470 | else |
ffd4bb2a | 2471 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2472 | if (ret) |
2473 | goto fail; | |
2474 | } | |
2475 | } | |
2476 | return 0; | |
2477 | fail: | |
5d4f98a2 YZ |
2478 | return ret; |
2479 | } | |
2480 | ||
2481 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2482 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2483 | { |
e339a6b0 | 2484 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2485 | } |
2486 | ||
2487 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2488 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2489 | { |
e339a6b0 | 2490 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2491 | } |
2492 | ||
2ff7e61e | 2493 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 | 2494 | { |
32da5386 | 2495 | struct btrfs_block_group *block_group; |
d2fb3437 YZ |
2496 | int readonly = 0; |
2497 | ||
0b246afa | 2498 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2499 | if (!block_group || block_group->ro) |
2500 | readonly = 1; | |
2501 | if (block_group) | |
fa9c0d79 | 2502 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2503 | return readonly; |
2504 | } | |
2505 | ||
1b86826d | 2506 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2507 | { |
0b246afa | 2508 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2509 | u64 flags; |
53b381b3 | 2510 | u64 ret; |
9ed74f2d | 2511 | |
b742bb82 YZ |
2512 | if (data) |
2513 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2514 | else if (root == fs_info->chunk_root) |
b742bb82 | 2515 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2516 | else |
b742bb82 | 2517 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2518 | |
878d7b67 | 2519 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2520 | return ret; |
6a63209f | 2521 | } |
9ed74f2d | 2522 | |
2ff7e61e | 2523 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2524 | { |
32da5386 | 2525 | struct btrfs_block_group *cache; |
d2fb3437 | 2526 | u64 bytenr; |
0f9dd46c | 2527 | |
0b246afa JM |
2528 | spin_lock(&fs_info->block_group_cache_lock); |
2529 | bytenr = fs_info->first_logical_byte; | |
2530 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2531 | |
2532 | if (bytenr < (u64)-1) | |
2533 | return bytenr; | |
2534 | ||
0b246afa | 2535 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2536 | if (!cache) |
a061fc8d | 2537 | return 0; |
0f9dd46c | 2538 | |
b3470b5d | 2539 | bytenr = cache->start; |
fa9c0d79 | 2540 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2541 | |
2542 | return bytenr; | |
a061fc8d CM |
2543 | } |
2544 | ||
6690d071 NB |
2545 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2546 | struct btrfs_block_group *cache, | |
f0486c68 | 2547 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2548 | { |
fdf08605 DS |
2549 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2550 | ||
11833d66 YZ |
2551 | spin_lock(&cache->space_info->lock); |
2552 | spin_lock(&cache->lock); | |
2553 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2554 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2555 | num_bytes); | |
11833d66 YZ |
2556 | if (reserved) { |
2557 | cache->reserved -= num_bytes; | |
2558 | cache->space_info->bytes_reserved -= num_bytes; | |
2559 | } | |
2560 | spin_unlock(&cache->lock); | |
2561 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2562 | |
2187374f | 2563 | __btrfs_mod_total_bytes_pinned(cache->space_info, num_bytes); |
fe119a6e | 2564 | set_extent_dirty(&trans->transaction->pinned_extents, bytenr, |
f0486c68 YZ |
2565 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2566 | return 0; | |
2567 | } | |
68b38550 | 2568 | |
b25c36f8 | 2569 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2570 | u64 bytenr, u64 num_bytes, int reserved) |
2571 | { | |
32da5386 | 2572 | struct btrfs_block_group *cache; |
68b38550 | 2573 | |
b25c36f8 | 2574 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2575 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2576 | |
6690d071 | 2577 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2578 | |
2579 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2580 | return 0; |
2581 | } | |
2582 | ||
f0486c68 | 2583 | /* |
e688b725 CM |
2584 | * this function must be called within transaction |
2585 | */ | |
9fce5704 | 2586 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
e688b725 CM |
2587 | u64 bytenr, u64 num_bytes) |
2588 | { | |
32da5386 | 2589 | struct btrfs_block_group *cache; |
b50c6e25 | 2590 | int ret; |
e688b725 | 2591 | |
9fce5704 | 2592 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
b50c6e25 JB |
2593 | if (!cache) |
2594 | return -EINVAL; | |
e688b725 CM |
2595 | |
2596 | /* | |
2597 | * pull in the free space cache (if any) so that our pin | |
2598 | * removes the free space from the cache. We have load_only set | |
2599 | * to one because the slow code to read in the free extents does check | |
2600 | * the pinned extents. | |
2601 | */ | |
676f1f75 | 2602 | btrfs_cache_block_group(cache, 1); |
9ad6d91f FM |
2603 | /* |
2604 | * Make sure we wait until the cache is completely built in case it is | |
2605 | * missing or is invalid and therefore needs to be rebuilt. | |
2606 | */ | |
2607 | ret = btrfs_wait_block_group_cache_done(cache); | |
2608 | if (ret) | |
2609 | goto out; | |
e688b725 | 2610 | |
6690d071 | 2611 | pin_down_extent(trans, cache, bytenr, num_bytes, 0); |
e688b725 CM |
2612 | |
2613 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2614 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
9ad6d91f | 2615 | out: |
e688b725 | 2616 | btrfs_put_block_group(cache); |
b50c6e25 | 2617 | return ret; |
e688b725 CM |
2618 | } |
2619 | ||
2ff7e61e JM |
2620 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2621 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2622 | { |
2623 | int ret; | |
32da5386 | 2624 | struct btrfs_block_group *block_group; |
8c2a1a30 | 2625 | |
0b246afa | 2626 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2627 | if (!block_group) |
2628 | return -EINVAL; | |
2629 | ||
9ad6d91f FM |
2630 | btrfs_cache_block_group(block_group, 1); |
2631 | /* | |
2632 | * Make sure we wait until the cache is completely built in case it is | |
2633 | * missing or is invalid and therefore needs to be rebuilt. | |
2634 | */ | |
2635 | ret = btrfs_wait_block_group_cache_done(block_group); | |
2636 | if (ret) | |
2637 | goto out; | |
8c2a1a30 | 2638 | |
9ad6d91f FM |
2639 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2640 | out: | |
8c2a1a30 JB |
2641 | btrfs_put_block_group(block_group); |
2642 | return ret; | |
2643 | } | |
2644 | ||
bcdc428c | 2645 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2646 | { |
bcdc428c | 2647 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2648 | struct btrfs_file_extent_item *item; |
2649 | struct btrfs_key key; | |
2650 | int found_type; | |
2651 | int i; | |
b89311ef | 2652 | int ret = 0; |
8c2a1a30 | 2653 | |
2ff7e61e | 2654 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2655 | return 0; |
2656 | ||
2657 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2658 | btrfs_item_key_to_cpu(eb, &key, i); | |
2659 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2660 | continue; | |
2661 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2662 | found_type = btrfs_file_extent_type(eb, item); | |
2663 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2664 | continue; | |
2665 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2666 | continue; | |
2667 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2668 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2669 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2670 | if (ret) | |
2671 | break; | |
8c2a1a30 JB |
2672 | } |
2673 | ||
b89311ef | 2674 | return ret; |
8c2a1a30 JB |
2675 | } |
2676 | ||
9cfa3e34 | 2677 | static void |
32da5386 | 2678 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2679 | { |
2680 | atomic_inc(&bg->reservations); | |
2681 | } | |
2682 | ||
c759c4e1 JB |
2683 | /* |
2684 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2685 | * what it should be based on the mount options. | |
2686 | */ | |
2687 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2688 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2689 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2690 | { |
2691 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2692 | |
2693 | *empty_cluster = 0; | |
2694 | if (btrfs_mixed_space_info(space_info)) | |
2695 | return ret; | |
2696 | ||
c759c4e1 | 2697 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2698 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2699 | if (btrfs_test_opt(fs_info, SSD)) |
2700 | *empty_cluster = SZ_2M; | |
2701 | else | |
ee22184b | 2702 | *empty_cluster = SZ_64K; |
583b7231 HK |
2703 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2704 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2705 | *empty_cluster = SZ_2M; | |
0b246afa | 2706 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2707 | } |
2708 | ||
2709 | return ret; | |
2710 | } | |
2711 | ||
2ff7e61e JM |
2712 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2713 | u64 start, u64 end, | |
678886bd | 2714 | const bool return_free_space) |
ccd467d6 | 2715 | { |
32da5386 | 2716 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2717 | struct btrfs_space_info *space_info; |
2718 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2719 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2720 | u64 len; |
c759c4e1 JB |
2721 | u64 total_unpinned = 0; |
2722 | u64 empty_cluster = 0; | |
7b398f8e | 2723 | bool readonly; |
ccd467d6 | 2724 | |
11833d66 | 2725 | while (start <= end) { |
7b398f8e | 2726 | readonly = false; |
11833d66 | 2727 | if (!cache || |
b3470b5d | 2728 | start >= cache->start + cache->length) { |
11833d66 YZ |
2729 | if (cache) |
2730 | btrfs_put_block_group(cache); | |
c759c4e1 | 2731 | total_unpinned = 0; |
11833d66 | 2732 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2733 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2734 | |
2ff7e61e | 2735 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2736 | cache->space_info, |
2737 | &empty_cluster); | |
2738 | empty_cluster <<= 1; | |
11833d66 YZ |
2739 | } |
2740 | ||
b3470b5d | 2741 | len = cache->start + cache->length - start; |
11833d66 YZ |
2742 | len = min(len, end + 1 - start); |
2743 | ||
2ca08c56 | 2744 | down_read(&fs_info->commit_root_sem); |
9076dbd5 JB |
2745 | if (start < cache->last_byte_to_unpin && return_free_space) { |
2746 | u64 add_len = min(len, cache->last_byte_to_unpin - start); | |
2747 | ||
2748 | btrfs_add_free_space(cache, start, add_len); | |
11833d66 | 2749 | } |
2ca08c56 | 2750 | up_read(&fs_info->commit_root_sem); |
11833d66 | 2751 | |
f0486c68 | 2752 | start += len; |
c759c4e1 | 2753 | total_unpinned += len; |
7b398f8e | 2754 | space_info = cache->space_info; |
f0486c68 | 2755 | |
c759c4e1 JB |
2756 | /* |
2757 | * If this space cluster has been marked as fragmented and we've | |
2758 | * unpinned enough in this block group to potentially allow a | |
2759 | * cluster to be created inside of it go ahead and clear the | |
2760 | * fragmented check. | |
2761 | */ | |
2762 | if (cluster && cluster->fragmented && | |
2763 | total_unpinned > empty_cluster) { | |
2764 | spin_lock(&cluster->lock); | |
2765 | cluster->fragmented = 0; | |
2766 | spin_unlock(&cluster->lock); | |
2767 | } | |
2768 | ||
7b398f8e | 2769 | spin_lock(&space_info->lock); |
11833d66 YZ |
2770 | spin_lock(&cache->lock); |
2771 | cache->pinned -= len; | |
bb96c4e5 | 2772 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2773 | space_info->max_extent_size = 0; |
2187374f | 2774 | __btrfs_mod_total_bytes_pinned(space_info, -len); |
7b398f8e JB |
2775 | if (cache->ro) { |
2776 | space_info->bytes_readonly += len; | |
2777 | readonly = true; | |
169e0da9 NA |
2778 | } else if (btrfs_is_zoned(fs_info)) { |
2779 | /* Need reset before reusing in a zoned block group */ | |
2780 | space_info->bytes_zone_unusable += len; | |
2781 | readonly = true; | |
7b398f8e | 2782 | } |
11833d66 | 2783 | spin_unlock(&cache->lock); |
957780eb JB |
2784 | if (!readonly && return_free_space && |
2785 | global_rsv->space_info == space_info) { | |
2786 | u64 to_add = len; | |
92ac58ec | 2787 | |
7b398f8e JB |
2788 | spin_lock(&global_rsv->lock); |
2789 | if (!global_rsv->full) { | |
957780eb JB |
2790 | to_add = min(len, global_rsv->size - |
2791 | global_rsv->reserved); | |
2792 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2793 | btrfs_space_info_update_bytes_may_use(fs_info, |
2794 | space_info, to_add); | |
7b398f8e JB |
2795 | if (global_rsv->reserved >= global_rsv->size) |
2796 | global_rsv->full = 1; | |
957780eb | 2797 | len -= to_add; |
7b398f8e JB |
2798 | } |
2799 | spin_unlock(&global_rsv->lock); | |
2800 | } | |
2732798c JB |
2801 | /* Add to any tickets we may have */ |
2802 | if (!readonly && return_free_space && len) | |
2803 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2804 | spin_unlock(&space_info->lock); |
ccd467d6 | 2805 | } |
11833d66 YZ |
2806 | |
2807 | if (cache) | |
2808 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2809 | return 0; |
2810 | } | |
2811 | ||
5ead2dd0 | 2812 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2813 | { |
5ead2dd0 | 2814 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2815 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2816 | struct list_head *deleted_bgs; |
11833d66 | 2817 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2818 | u64 start; |
2819 | u64 end; | |
a28ec197 | 2820 | int ret; |
a28ec197 | 2821 | |
fe119a6e | 2822 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2823 | |
bf31f87f | 2824 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2825 | struct extent_state *cached_state = NULL; |
2826 | ||
d4b450cd | 2827 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2828 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2829 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2830 | if (ret) { |
2831 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2832 | break; |
d4b450cd | 2833 | } |
1f3c79a2 | 2834 | |
46b27f50 | 2835 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2836 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2837 | end + 1 - start, NULL); |
1f3c79a2 | 2838 | |
0e6ec385 | 2839 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2840 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2841 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2842 | free_extent_state(cached_state); |
b9473439 | 2843 | cond_resched(); |
a28ec197 | 2844 | } |
817d52f8 | 2845 | |
a2309300 DZ |
2846 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2847 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2848 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2849 | } |
b0643e59 | 2850 | |
e33e17ee JM |
2851 | /* |
2852 | * Transaction is finished. We don't need the lock anymore. We | |
2853 | * do need to clean up the block groups in case of a transaction | |
2854 | * abort. | |
2855 | */ | |
2856 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2857 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2858 | u64 trimmed = 0; | |
2859 | ||
2860 | ret = -EROFS; | |
bf31f87f | 2861 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2862 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2863 | block_group->start, |
2864 | block_group->length, | |
e33e17ee JM |
2865 | &trimmed); |
2866 | ||
2867 | list_del_init(&block_group->bg_list); | |
6b7304af | 2868 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2869 | btrfs_put_block_group(block_group); |
2870 | ||
2871 | if (ret) { | |
2872 | const char *errstr = btrfs_decode_error(ret); | |
2873 | btrfs_warn(fs_info, | |
913e1535 | 2874 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2875 | ret, errstr); |
2876 | } | |
2877 | } | |
2878 | ||
e20d96d6 CM |
2879 | return 0; |
2880 | } | |
2881 | ||
1c2a07f5 QW |
2882 | /* |
2883 | * Drop one or more refs of @node. | |
2884 | * | |
2885 | * 1. Locate the extent refs. | |
2886 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
2887 | * Locate it, then reduce the refs number or remove the ref line completely. | |
2888 | * | |
2889 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
2890 | * | |
2891 | * Inline backref case: | |
2892 | * | |
2893 | * in extent tree we have: | |
2894 | * | |
2895 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2896 | * refs 2 gen 6 flags DATA | |
2897 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2898 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
2899 | * | |
2900 | * This function gets called with: | |
2901 | * | |
2902 | * node->bytenr = 13631488 | |
2903 | * node->num_bytes = 1048576 | |
2904 | * root_objectid = FS_TREE | |
2905 | * owner_objectid = 257 | |
2906 | * owner_offset = 0 | |
2907 | * refs_to_drop = 1 | |
2908 | * | |
2909 | * Then we should get some like: | |
2910 | * | |
2911 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2912 | * refs 1 gen 6 flags DATA | |
2913 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2914 | * | |
2915 | * Keyed backref case: | |
2916 | * | |
2917 | * in extent tree we have: | |
2918 | * | |
2919 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2920 | * refs 754 gen 6 flags DATA | |
2921 | * [...] | |
2922 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
2923 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
2924 | * | |
2925 | * This function get called with: | |
2926 | * | |
2927 | * node->bytenr = 13631488 | |
2928 | * node->num_bytes = 1048576 | |
2929 | * root_objectid = FS_TREE | |
2930 | * owner_objectid = 866 | |
2931 | * owner_offset = 0 | |
2932 | * refs_to_drop = 1 | |
2933 | * | |
2934 | * Then we should get some like: | |
2935 | * | |
2936 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2937 | * refs 753 gen 6 flags DATA | |
2938 | * | |
2939 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
2940 | */ | |
5d4f98a2 | 2941 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2942 | struct btrfs_delayed_ref_node *node, u64 parent, |
2943 | u64 root_objectid, u64 owner_objectid, | |
2944 | u64 owner_offset, int refs_to_drop, | |
2945 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2946 | { |
e72cb923 | 2947 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2948 | struct btrfs_key key; |
5d4f98a2 | 2949 | struct btrfs_path *path; |
1261ec42 | 2950 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2951 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2952 | struct btrfs_extent_item *ei; |
2953 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2954 | int ret; |
5d4f98a2 | 2955 | int is_data; |
952fccac CM |
2956 | int extent_slot = 0; |
2957 | int found_extent = 0; | |
2958 | int num_to_del = 1; | |
5d4f98a2 YZ |
2959 | u32 item_size; |
2960 | u64 refs; | |
c682f9b3 QW |
2961 | u64 bytenr = node->bytenr; |
2962 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2963 | int last_ref = 0; |
0b246afa | 2964 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 2965 | |
5caf2a00 | 2966 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2967 | if (!path) |
2968 | return -ENOMEM; | |
5f26f772 | 2969 | |
5d4f98a2 | 2970 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
1c2a07f5 QW |
2971 | |
2972 | if (!is_data && refs_to_drop != 1) { | |
2973 | btrfs_crit(info, | |
2974 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
2975 | node->bytenr, refs_to_drop); | |
2976 | ret = -EINVAL; | |
2977 | btrfs_abort_transaction(trans, ret); | |
2978 | goto out; | |
2979 | } | |
5d4f98a2 | 2980 | |
3173a18f | 2981 | if (is_data) |
897ca819 | 2982 | skinny_metadata = false; |
3173a18f | 2983 | |
fbe4801b NB |
2984 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
2985 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 2986 | owner_offset); |
7bb86316 | 2987 | if (ret == 0) { |
1c2a07f5 QW |
2988 | /* |
2989 | * Either the inline backref or the SHARED_DATA_REF/ | |
2990 | * SHARED_BLOCK_REF is found | |
2991 | * | |
2992 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
2993 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
2994 | */ | |
952fccac | 2995 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
2996 | while (extent_slot >= 0) { |
2997 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 2998 | extent_slot); |
5d4f98a2 | 2999 | if (key.objectid != bytenr) |
952fccac | 3000 | break; |
5d4f98a2 YZ |
3001 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3002 | key.offset == num_bytes) { | |
952fccac CM |
3003 | found_extent = 1; |
3004 | break; | |
3005 | } | |
3173a18f JB |
3006 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3007 | key.offset == owner_objectid) { | |
3008 | found_extent = 1; | |
3009 | break; | |
3010 | } | |
1c2a07f5 QW |
3011 | |
3012 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
3013 | if (path->slots[0] - extent_slot > 5) |
3014 | break; | |
5d4f98a2 | 3015 | extent_slot--; |
952fccac | 3016 | } |
a79865c6 | 3017 | |
31840ae1 | 3018 | if (!found_extent) { |
1c2a07f5 QW |
3019 | if (iref) { |
3020 | btrfs_crit(info, | |
3021 | "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref"); | |
3022 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3023 | goto err_dump; | |
3024 | } | |
3025 | /* Must be SHARED_* item, remove the backref first */ | |
87cc7a8a | 3026 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3027 | refs_to_drop, |
fcebe456 | 3028 | is_data, &last_ref); |
005d6427 | 3029 | if (ret) { |
66642832 | 3030 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3031 | goto out; |
3032 | } | |
b3b4aa74 | 3033 | btrfs_release_path(path); |
5d4f98a2 | 3034 | |
1c2a07f5 | 3035 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3036 | key.objectid = bytenr; |
3037 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3038 | key.offset = num_bytes; | |
3039 | ||
3173a18f JB |
3040 | if (!is_data && skinny_metadata) { |
3041 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3042 | key.offset = owner_objectid; | |
3043 | } | |
3044 | ||
31840ae1 ZY |
3045 | ret = btrfs_search_slot(trans, extent_root, |
3046 | &key, path, -1, 1); | |
3173a18f JB |
3047 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3048 | /* | |
3049 | * Couldn't find our skinny metadata item, | |
3050 | * see if we have ye olde extent item. | |
3051 | */ | |
3052 | path->slots[0]--; | |
3053 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3054 | path->slots[0]); | |
3055 | if (key.objectid == bytenr && | |
3056 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3057 | key.offset == num_bytes) | |
3058 | ret = 0; | |
3059 | } | |
3060 | ||
3061 | if (ret > 0 && skinny_metadata) { | |
3062 | skinny_metadata = false; | |
9ce49a0b | 3063 | key.objectid = bytenr; |
3173a18f JB |
3064 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3065 | key.offset = num_bytes; | |
3066 | btrfs_release_path(path); | |
3067 | ret = btrfs_search_slot(trans, extent_root, | |
3068 | &key, path, -1, 1); | |
3069 | } | |
3070 | ||
f3465ca4 | 3071 | if (ret) { |
5d163e0e JM |
3072 | btrfs_err(info, |
3073 | "umm, got %d back from search, was looking for %llu", | |
3074 | ret, bytenr); | |
b783e62d | 3075 | if (ret > 0) |
a4f78750 | 3076 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3077 | } |
005d6427 | 3078 | if (ret < 0) { |
66642832 | 3079 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3080 | goto out; |
3081 | } | |
31840ae1 ZY |
3082 | extent_slot = path->slots[0]; |
3083 | } | |
fae7f21c | 3084 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3085 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3086 | btrfs_err(info, |
3087 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3088 | bytenr, parent, root_objectid, owner_objectid, |
3089 | owner_offset); | |
66642832 | 3090 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3091 | goto out; |
79787eaa | 3092 | } else { |
66642832 | 3093 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3094 | goto out; |
7bb86316 | 3095 | } |
5f39d397 CM |
3096 | |
3097 | leaf = path->nodes[0]; | |
5d4f98a2 | 3098 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3099 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3100 | ret = -EINVAL; |
3101 | btrfs_print_v0_err(info); | |
3102 | btrfs_abort_transaction(trans, ret); | |
3103 | goto out; | |
3104 | } | |
952fccac | 3105 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3106 | struct btrfs_extent_item); |
3173a18f JB |
3107 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3108 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3109 | struct btrfs_tree_block_info *bi; |
1c2a07f5 QW |
3110 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
3111 | btrfs_crit(info, | |
cad69d13 | 3112 | "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu", |
1c2a07f5 QW |
3113 | key.objectid, key.type, key.offset, |
3114 | owner_objectid, item_size, | |
3115 | sizeof(*ei) + sizeof(*bi)); | |
3116 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3117 | goto err_dump; | |
3118 | } | |
5d4f98a2 YZ |
3119 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3120 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3121 | } | |
56bec294 | 3122 | |
5d4f98a2 | 3123 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3124 | if (refs < refs_to_drop) { |
1c2a07f5 QW |
3125 | btrfs_crit(info, |
3126 | "trying to drop %d refs but we only have %llu for bytenr %llu", | |
5d163e0e | 3127 | refs_to_drop, refs, bytenr); |
1c2a07f5 QW |
3128 | btrfs_abort_transaction(trans, -EUCLEAN); |
3129 | goto err_dump; | |
32b02538 | 3130 | } |
56bec294 | 3131 | refs -= refs_to_drop; |
5f39d397 | 3132 | |
5d4f98a2 YZ |
3133 | if (refs > 0) { |
3134 | if (extent_op) | |
3135 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3136 | /* | |
3137 | * In the case of inline back ref, reference count will | |
3138 | * be updated by remove_extent_backref | |
952fccac | 3139 | */ |
5d4f98a2 | 3140 | if (iref) { |
1c2a07f5 QW |
3141 | if (!found_extent) { |
3142 | btrfs_crit(info, | |
3143 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found"); | |
3144 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3145 | goto err_dump; | |
3146 | } | |
5d4f98a2 YZ |
3147 | } else { |
3148 | btrfs_set_extent_refs(leaf, ei, refs); | |
3149 | btrfs_mark_buffer_dirty(leaf); | |
3150 | } | |
3151 | if (found_extent) { | |
87cc7a8a NB |
3152 | ret = remove_extent_backref(trans, path, iref, |
3153 | refs_to_drop, is_data, | |
3154 | &last_ref); | |
005d6427 | 3155 | if (ret) { |
66642832 | 3156 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3157 | goto out; |
3158 | } | |
952fccac | 3159 | } |
5d4f98a2 | 3160 | } else { |
1c2a07f5 | 3161 | /* In this branch refs == 1 */ |
5d4f98a2 | 3162 | if (found_extent) { |
1c2a07f5 QW |
3163 | if (is_data && refs_to_drop != |
3164 | extent_data_ref_count(path, iref)) { | |
3165 | btrfs_crit(info, | |
3166 | "invalid refs_to_drop, current refs %u refs_to_drop %u", | |
3167 | extent_data_ref_count(path, iref), | |
3168 | refs_to_drop); | |
3169 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3170 | goto err_dump; | |
3171 | } | |
5d4f98a2 | 3172 | if (iref) { |
1c2a07f5 QW |
3173 | if (path->slots[0] != extent_slot) { |
3174 | btrfs_crit(info, | |
3175 | "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref", | |
3176 | key.objectid, key.type, | |
3177 | key.offset); | |
3178 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3179 | goto err_dump; | |
3180 | } | |
5d4f98a2 | 3181 | } else { |
1c2a07f5 QW |
3182 | /* |
3183 | * No inline ref, we must be at SHARED_* item, | |
3184 | * And it's single ref, it must be: | |
3185 | * | extent_slot ||extent_slot + 1| | |
3186 | * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ] | |
3187 | */ | |
3188 | if (path->slots[0] != extent_slot + 1) { | |
3189 | btrfs_crit(info, | |
3190 | "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM"); | |
3191 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3192 | goto err_dump; | |
3193 | } | |
5d4f98a2 YZ |
3194 | path->slots[0] = extent_slot; |
3195 | num_to_del = 2; | |
3196 | } | |
78fae27e | 3197 | } |
b9473439 | 3198 | |
fcebe456 | 3199 | last_ref = 1; |
952fccac CM |
3200 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3201 | num_to_del); | |
005d6427 | 3202 | if (ret) { |
66642832 | 3203 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3204 | goto out; |
3205 | } | |
b3b4aa74 | 3206 | btrfs_release_path(path); |
21af804c | 3207 | |
5d4f98a2 | 3208 | if (is_data) { |
40e046ac FM |
3209 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3210 | num_bytes); | |
005d6427 | 3211 | if (ret) { |
66642832 | 3212 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3213 | goto out; |
3214 | } | |
459931ec CM |
3215 | } |
3216 | ||
e7355e50 | 3217 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3218 | if (ret) { |
66642832 | 3219 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3220 | goto out; |
3221 | } | |
3222 | ||
ade4b516 | 3223 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3224 | if (ret) { |
66642832 | 3225 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3226 | goto out; |
3227 | } | |
a28ec197 | 3228 | } |
fcebe456 JB |
3229 | btrfs_release_path(path); |
3230 | ||
79787eaa | 3231 | out: |
5caf2a00 | 3232 | btrfs_free_path(path); |
a28ec197 | 3233 | return ret; |
1c2a07f5 QW |
3234 | err_dump: |
3235 | /* | |
3236 | * Leaf dump can take up a lot of log buffer, so we only do full leaf | |
3237 | * dump for debug build. | |
3238 | */ | |
3239 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
3240 | btrfs_crit(info, "path->slots[0]=%d extent_slot=%d", | |
3241 | path->slots[0], extent_slot); | |
3242 | btrfs_print_leaf(path->nodes[0]); | |
3243 | } | |
3244 | ||
3245 | btrfs_free_path(path); | |
3246 | return -EUCLEAN; | |
a28ec197 CM |
3247 | } |
3248 | ||
1887be66 | 3249 | /* |
f0486c68 | 3250 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3251 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3252 | * a given extent, and if there are no other delayed refs to be processed, it | |
3253 | * removes it from the tree. | |
3254 | */ | |
3255 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3256 | u64 bytenr) |
1887be66 CM |
3257 | { |
3258 | struct btrfs_delayed_ref_head *head; | |
3259 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3260 | int ret = 0; |
1887be66 CM |
3261 | |
3262 | delayed_refs = &trans->transaction->delayed_refs; | |
3263 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3264 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3265 | if (!head) |
cf93da7b | 3266 | goto out_delayed_unlock; |
1887be66 | 3267 | |
d7df2c79 | 3268 | spin_lock(&head->lock); |
e3d03965 | 3269 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3270 | goto out; |
3271 | ||
bedc6617 JB |
3272 | if (cleanup_extent_op(head) != NULL) |
3273 | goto out; | |
5d4f98a2 | 3274 | |
1887be66 CM |
3275 | /* |
3276 | * waiting for the lock here would deadlock. If someone else has it | |
3277 | * locked they are already in the process of dropping it anyway | |
3278 | */ | |
3279 | if (!mutex_trylock(&head->mutex)) | |
3280 | goto out; | |
3281 | ||
d7baffda | 3282 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3283 | head->processing = 0; |
d7baffda | 3284 | |
d7df2c79 | 3285 | spin_unlock(&head->lock); |
1887be66 CM |
3286 | spin_unlock(&delayed_refs->lock); |
3287 | ||
f0486c68 YZ |
3288 | BUG_ON(head->extent_op); |
3289 | if (head->must_insert_reserved) | |
3290 | ret = 1; | |
3291 | ||
31890da0 | 3292 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3293 | mutex_unlock(&head->mutex); |
d278850e | 3294 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3295 | return ret; |
1887be66 | 3296 | out: |
d7df2c79 | 3297 | spin_unlock(&head->lock); |
cf93da7b CM |
3298 | |
3299 | out_delayed_unlock: | |
1887be66 CM |
3300 | spin_unlock(&delayed_refs->lock); |
3301 | return 0; | |
3302 | } | |
3303 | ||
f0486c68 YZ |
3304 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3305 | struct btrfs_root *root, | |
3306 | struct extent_buffer *buf, | |
5581a51a | 3307 | u64 parent, int last_ref) |
f0486c68 | 3308 | { |
0b246afa | 3309 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3310 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
3311 | int ret; |
3312 | ||
ed4f255b QW |
3313 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3314 | buf->start, buf->len, parent); | |
3315 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3316 | root->root_key.objectid); | |
3317 | ||
f0486c68 | 3318 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
8a5040f7 | 3319 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 3320 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL); |
79787eaa | 3321 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
3322 | } |
3323 | ||
0a16c7d7 | 3324 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3325 | struct btrfs_block_group *cache; |
6219872d | 3326 | |
f0486c68 | 3327 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3328 | ret = check_ref_cleanup(trans, buf->start); |
d3575156 NA |
3329 | if (!ret) { |
3330 | btrfs_redirty_list_add(trans->transaction, buf); | |
37be25bc | 3331 | goto out; |
d3575156 | 3332 | } |
f0486c68 YZ |
3333 | } |
3334 | ||
0b246afa | 3335 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3336 | |
f0486c68 | 3337 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3338 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3339 | btrfs_put_block_group(cache); |
37be25bc | 3340 | goto out; |
f0486c68 YZ |
3341 | } |
3342 | ||
d3575156 NA |
3343 | if (btrfs_is_zoned(fs_info)) { |
3344 | btrfs_redirty_list_add(trans->transaction, buf); | |
3345 | pin_down_extent(trans, cache, buf->start, buf->len, 1); | |
3346 | btrfs_put_block_group(cache); | |
3347 | goto out; | |
3348 | } | |
3349 | ||
f0486c68 YZ |
3350 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); |
3351 | ||
3352 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3353 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3354 | btrfs_put_block_group(cache); |
71ff6437 | 3355 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3356 | } |
3357 | out: | |
0a16c7d7 OS |
3358 | if (last_ref) { |
3359 | /* | |
3360 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3361 | * matter anymore. | |
3362 | */ | |
3363 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3364 | } | |
f0486c68 YZ |
3365 | } |
3366 | ||
79787eaa | 3367 | /* Can return -ENOMEM */ |
ffd4bb2a | 3368 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3369 | { |
ffd4bb2a | 3370 | struct btrfs_fs_info *fs_info = trans->fs_info; |
925baedd CM |
3371 | int ret; |
3372 | ||
f5ee5c9a | 3373 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3374 | return 0; |
fccb84c9 | 3375 | |
56bec294 CM |
3376 | /* |
3377 | * tree log blocks never actually go into the extent allocation | |
3378 | * tree, just update pinning info and exit early. | |
56bec294 | 3379 | */ |
ffd4bb2a QW |
3380 | if ((ref->type == BTRFS_REF_METADATA && |
3381 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3382 | (ref->type == BTRFS_REF_DATA && | |
3383 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3384 | /* unlocks the pinned mutex */ |
b25c36f8 | 3385 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
56bec294 | 3386 | ret = 0; |
ffd4bb2a | 3387 | } else if (ref->type == BTRFS_REF_METADATA) { |
2187374f | 3388 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL); |
5d4f98a2 | 3389 | } else { |
2187374f | 3390 | ret = btrfs_add_delayed_data_ref(trans, ref, 0); |
56bec294 | 3391 | } |
d7eae340 | 3392 | |
ffd4bb2a QW |
3393 | if (!((ref->type == BTRFS_REF_METADATA && |
3394 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3395 | (ref->type == BTRFS_REF_DATA && | |
3396 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3397 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3398 | |
925baedd CM |
3399 | return ret; |
3400 | } | |
3401 | ||
817d52f8 | 3402 | enum btrfs_loop_type { |
f262fa8d DS |
3403 | LOOP_CACHING_NOWAIT, |
3404 | LOOP_CACHING_WAIT, | |
3405 | LOOP_ALLOC_CHUNK, | |
3406 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3407 | }; |
3408 | ||
e570fd27 | 3409 | static inline void |
32da5386 | 3410 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3411 | int delalloc) |
3412 | { | |
3413 | if (delalloc) | |
3414 | down_read(&cache->data_rwsem); | |
3415 | } | |
3416 | ||
32da5386 | 3417 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3418 | int delalloc) |
3419 | { | |
3420 | btrfs_get_block_group(cache); | |
3421 | if (delalloc) | |
3422 | down_read(&cache->data_rwsem); | |
3423 | } | |
3424 | ||
32da5386 DS |
3425 | static struct btrfs_block_group *btrfs_lock_cluster( |
3426 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3427 | struct btrfs_free_cluster *cluster, |
3428 | int delalloc) | |
c142c6a4 | 3429 | __acquires(&cluster->refill_lock) |
e570fd27 | 3430 | { |
32da5386 | 3431 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3432 | |
e570fd27 | 3433 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3434 | while (1) { |
3435 | used_bg = cluster->block_group; | |
3436 | if (!used_bg) | |
3437 | return NULL; | |
3438 | ||
3439 | if (used_bg == block_group) | |
e570fd27 MX |
3440 | return used_bg; |
3441 | ||
6719afdc | 3442 | btrfs_get_block_group(used_bg); |
e570fd27 | 3443 | |
6719afdc GU |
3444 | if (!delalloc) |
3445 | return used_bg; | |
e570fd27 | 3446 | |
6719afdc GU |
3447 | if (down_read_trylock(&used_bg->data_rwsem)) |
3448 | return used_bg; | |
e570fd27 | 3449 | |
6719afdc | 3450 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3451 | |
e321f8a8 LB |
3452 | /* We should only have one-level nested. */ |
3453 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3454 | |
6719afdc GU |
3455 | spin_lock(&cluster->refill_lock); |
3456 | if (used_bg == cluster->block_group) | |
3457 | return used_bg; | |
e570fd27 | 3458 | |
6719afdc GU |
3459 | up_read(&used_bg->data_rwsem); |
3460 | btrfs_put_block_group(used_bg); | |
3461 | } | |
e570fd27 MX |
3462 | } |
3463 | ||
3464 | static inline void | |
32da5386 | 3465 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3466 | int delalloc) |
3467 | { | |
3468 | if (delalloc) | |
3469 | up_read(&cache->data_rwsem); | |
3470 | btrfs_put_block_group(cache); | |
3471 | } | |
3472 | ||
cb2f96f8 NA |
3473 | enum btrfs_extent_allocation_policy { |
3474 | BTRFS_EXTENT_ALLOC_CLUSTERED, | |
2eda5708 | 3475 | BTRFS_EXTENT_ALLOC_ZONED, |
cb2f96f8 NA |
3476 | }; |
3477 | ||
b4bd745d QW |
3478 | /* |
3479 | * Structure used internally for find_free_extent() function. Wraps needed | |
3480 | * parameters. | |
3481 | */ | |
3482 | struct find_free_extent_ctl { | |
3483 | /* Basic allocation info */ | |
b4bd745d QW |
3484 | u64 num_bytes; |
3485 | u64 empty_size; | |
3486 | u64 flags; | |
3487 | int delalloc; | |
3488 | ||
3489 | /* Where to start the search inside the bg */ | |
3490 | u64 search_start; | |
3491 | ||
3492 | /* For clustered allocation */ | |
3493 | u64 empty_cluster; | |
c10859be NA |
3494 | struct btrfs_free_cluster *last_ptr; |
3495 | bool use_cluster; | |
b4bd745d QW |
3496 | |
3497 | bool have_caching_bg; | |
3498 | bool orig_have_caching_bg; | |
3499 | ||
3500 | /* RAID index, converted from flags */ | |
3501 | int index; | |
3502 | ||
e72d79d6 QW |
3503 | /* |
3504 | * Current loop number, check find_free_extent_update_loop() for details | |
3505 | */ | |
b4bd745d QW |
3506 | int loop; |
3507 | ||
3508 | /* | |
3509 | * Whether we're refilling a cluster, if true we need to re-search | |
3510 | * current block group but don't try to refill the cluster again. | |
3511 | */ | |
3512 | bool retry_clustered; | |
3513 | ||
3514 | /* | |
3515 | * Whether we're updating free space cache, if true we need to re-search | |
3516 | * current block group but don't try updating free space cache again. | |
3517 | */ | |
3518 | bool retry_unclustered; | |
3519 | ||
3520 | /* If current block group is cached */ | |
3521 | int cached; | |
3522 | ||
3523 | /* Max contiguous hole found */ | |
3524 | u64 max_extent_size; | |
3525 | ||
3526 | /* Total free space from free space cache, not always contiguous */ | |
3527 | u64 total_free_space; | |
3528 | ||
3529 | /* Found result */ | |
3530 | u64 found_offset; | |
cb2f96f8 | 3531 | |
ea544149 NA |
3532 | /* Hint where to start looking for an empty space */ |
3533 | u64 hint_byte; | |
3534 | ||
cb2f96f8 NA |
3535 | /* Allocation policy */ |
3536 | enum btrfs_extent_allocation_policy policy; | |
b4bd745d QW |
3537 | }; |
3538 | ||
d06e3bb6 QW |
3539 | |
3540 | /* | |
3541 | * Helper function for find_free_extent(). | |
3542 | * | |
3543 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3544 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3545 | * Return >0 to inform caller that we find nothing | |
3546 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3547 | */ | |
32da5386 | 3548 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3549 | struct find_free_extent_ctl *ffe_ctl, |
3550 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3551 | { |
32da5386 | 3552 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3553 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3554 | u64 aligned_cluster; |
3555 | u64 offset; | |
3556 | int ret; | |
3557 | ||
3558 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3559 | if (!cluster_bg) | |
3560 | goto refill_cluster; | |
3561 | if (cluster_bg != bg && (cluster_bg->ro || | |
3562 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3563 | goto release_cluster; | |
3564 | ||
3565 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3566 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3567 | &ffe_ctl->max_extent_size); |
3568 | if (offset) { | |
3569 | /* We have a block, we're done */ | |
3570 | spin_unlock(&last_ptr->refill_lock); | |
3571 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3572 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3573 | *cluster_bg_ret = cluster_bg; | |
3574 | ffe_ctl->found_offset = offset; | |
3575 | return 0; | |
3576 | } | |
3577 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3578 | ||
3579 | release_cluster: | |
3580 | /* | |
3581 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3582 | * lets just skip it and let the allocator find whatever block it can | |
3583 | * find. If we reach this point, we will have tried the cluster | |
3584 | * allocator plenty of times and not have found anything, so we are | |
3585 | * likely way too fragmented for the clustering stuff to find anything. | |
3586 | * | |
3587 | * However, if the cluster is taken from the current block group, | |
3588 | * release the cluster first, so that we stand a better chance of | |
3589 | * succeeding in the unclustered allocation. | |
3590 | */ | |
3591 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3592 | spin_unlock(&last_ptr->refill_lock); | |
3593 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3594 | return -ENOENT; | |
3595 | } | |
3596 | ||
3597 | /* This cluster didn't work out, free it and start over */ | |
3598 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3599 | ||
3600 | if (cluster_bg != bg) | |
3601 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3602 | ||
3603 | refill_cluster: | |
3604 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3605 | spin_unlock(&last_ptr->refill_lock); | |
3606 | return -ENOENT; | |
3607 | } | |
3608 | ||
3609 | aligned_cluster = max_t(u64, | |
3610 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3611 | bg->full_stripe_len); | |
2ceeae2e DS |
3612 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3613 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3614 | if (ret == 0) { |
3615 | /* Now pull our allocation out of this cluster */ | |
3616 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3617 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3618 | &ffe_ctl->max_extent_size); | |
3619 | if (offset) { | |
3620 | /* We found one, proceed */ | |
3621 | spin_unlock(&last_ptr->refill_lock); | |
3622 | trace_btrfs_reserve_extent_cluster(bg, | |
3623 | ffe_ctl->search_start, | |
3624 | ffe_ctl->num_bytes); | |
3625 | ffe_ctl->found_offset = offset; | |
3626 | return 0; | |
3627 | } | |
3628 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3629 | !ffe_ctl->retry_clustered) { | |
3630 | spin_unlock(&last_ptr->refill_lock); | |
3631 | ||
3632 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3633 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3634 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3635 | return -EAGAIN; | |
3636 | } | |
3637 | /* | |
3638 | * At this point we either didn't find a cluster or we weren't able to | |
3639 | * allocate a block from our cluster. Free the cluster we've been | |
3640 | * trying to use, and go to the next block group. | |
3641 | */ | |
3642 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3643 | spin_unlock(&last_ptr->refill_lock); | |
3644 | return 1; | |
3645 | } | |
3646 | ||
e1a41848 QW |
3647 | /* |
3648 | * Return >0 to inform caller that we find nothing | |
3649 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3650 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3651 | */ | |
32da5386 | 3652 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3653 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3654 | { |
897cae79 | 3655 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3656 | u64 offset; |
3657 | ||
3658 | /* | |
3659 | * We are doing an unclustered allocation, set the fragmented flag so | |
3660 | * we don't bother trying to setup a cluster again until we get more | |
3661 | * space. | |
3662 | */ | |
3663 | if (unlikely(last_ptr)) { | |
3664 | spin_lock(&last_ptr->lock); | |
3665 | last_ptr->fragmented = 1; | |
3666 | spin_unlock(&last_ptr->lock); | |
3667 | } | |
3668 | if (ffe_ctl->cached) { | |
3669 | struct btrfs_free_space_ctl *free_space_ctl; | |
3670 | ||
3671 | free_space_ctl = bg->free_space_ctl; | |
3672 | spin_lock(&free_space_ctl->tree_lock); | |
3673 | if (free_space_ctl->free_space < | |
3674 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3675 | ffe_ctl->empty_size) { | |
3676 | ffe_ctl->total_free_space = max_t(u64, | |
3677 | ffe_ctl->total_free_space, | |
3678 | free_space_ctl->free_space); | |
3679 | spin_unlock(&free_space_ctl->tree_lock); | |
3680 | return 1; | |
3681 | } | |
3682 | spin_unlock(&free_space_ctl->tree_lock); | |
3683 | } | |
3684 | ||
3685 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3686 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3687 | &ffe_ctl->max_extent_size); | |
3688 | ||
3689 | /* | |
3690 | * If we didn't find a chunk, and we haven't failed on this block group | |
3691 | * before, and this block group is in the middle of caching and we are | |
3692 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3693 | * set @retry_unclustered to true. | |
3694 | * | |
3695 | * If @retry_unclustered is true then we've already waited on this | |
3696 | * block group once and should move on to the next block group. | |
3697 | */ | |
3698 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3699 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3700 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3701 | ffe_ctl->empty_size); | |
e1a41848 QW |
3702 | ffe_ctl->retry_unclustered = true; |
3703 | return -EAGAIN; | |
3704 | } else if (!offset) { | |
3705 | return 1; | |
3706 | } | |
3707 | ffe_ctl->found_offset = offset; | |
3708 | return 0; | |
3709 | } | |
3710 | ||
c668690d NA |
3711 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3712 | struct find_free_extent_ctl *ffe_ctl, | |
3713 | struct btrfs_block_group **bg_ret) | |
3714 | { | |
3715 | int ret; | |
3716 | ||
3717 | /* We want to try and use the cluster allocator, so lets look there */ | |
3718 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3719 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
c668690d NA |
3720 | if (ret >= 0 || ret == -EAGAIN) |
3721 | return ret; | |
3722 | /* ret == -ENOENT case falls through */ | |
3723 | } | |
3724 | ||
897cae79 | 3725 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3726 | } |
3727 | ||
2eda5708 NA |
3728 | /* |
3729 | * Simple allocator for sequential-only block group. It only allows sequential | |
3730 | * allocation. No need to play with trees. This function also reserves the | |
3731 | * bytes as in btrfs_add_reserved_bytes. | |
3732 | */ | |
3733 | static int do_allocation_zoned(struct btrfs_block_group *block_group, | |
3734 | struct find_free_extent_ctl *ffe_ctl, | |
3735 | struct btrfs_block_group **bg_ret) | |
3736 | { | |
3737 | struct btrfs_space_info *space_info = block_group->space_info; | |
3738 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3739 | u64 start = block_group->start; | |
3740 | u64 num_bytes = ffe_ctl->num_bytes; | |
3741 | u64 avail; | |
3742 | int ret = 0; | |
3743 | ||
3744 | ASSERT(btrfs_is_zoned(block_group->fs_info)); | |
3745 | ||
3746 | spin_lock(&space_info->lock); | |
3747 | spin_lock(&block_group->lock); | |
3748 | ||
3749 | if (block_group->ro) { | |
3750 | ret = 1; | |
3751 | goto out; | |
3752 | } | |
3753 | ||
3754 | avail = block_group->length - block_group->alloc_offset; | |
3755 | if (avail < num_bytes) { | |
3756 | if (ffe_ctl->max_extent_size < avail) { | |
3757 | /* | |
3758 | * With sequential allocator, free space is always | |
3759 | * contiguous | |
3760 | */ | |
3761 | ffe_ctl->max_extent_size = avail; | |
3762 | ffe_ctl->total_free_space = avail; | |
3763 | } | |
3764 | ret = 1; | |
3765 | goto out; | |
3766 | } | |
3767 | ||
3768 | ffe_ctl->found_offset = start + block_group->alloc_offset; | |
3769 | block_group->alloc_offset += num_bytes; | |
3770 | spin_lock(&ctl->tree_lock); | |
3771 | ctl->free_space -= num_bytes; | |
3772 | spin_unlock(&ctl->tree_lock); | |
3773 | ||
3774 | /* | |
3775 | * We do not check if found_offset is aligned to stripesize. The | |
3776 | * address is anyway rewritten when using zone append writing. | |
3777 | */ | |
3778 | ||
3779 | ffe_ctl->search_start = ffe_ctl->found_offset; | |
3780 | ||
3781 | out: | |
3782 | spin_unlock(&block_group->lock); | |
3783 | spin_unlock(&space_info->lock); | |
3784 | return ret; | |
3785 | } | |
3786 | ||
c668690d NA |
3787 | static int do_allocation(struct btrfs_block_group *block_group, |
3788 | struct find_free_extent_ctl *ffe_ctl, | |
3789 | struct btrfs_block_group **bg_ret) | |
3790 | { | |
3791 | switch (ffe_ctl->policy) { | |
3792 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3793 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
2eda5708 NA |
3794 | case BTRFS_EXTENT_ALLOC_ZONED: |
3795 | return do_allocation_zoned(block_group, ffe_ctl, bg_ret); | |
c668690d NA |
3796 | default: |
3797 | BUG(); | |
3798 | } | |
3799 | } | |
3800 | ||
baba5062 NA |
3801 | static void release_block_group(struct btrfs_block_group *block_group, |
3802 | struct find_free_extent_ctl *ffe_ctl, | |
3803 | int delalloc) | |
3804 | { | |
3805 | switch (ffe_ctl->policy) { | |
3806 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3807 | ffe_ctl->retry_clustered = false; | |
3808 | ffe_ctl->retry_unclustered = false; | |
3809 | break; | |
2eda5708 NA |
3810 | case BTRFS_EXTENT_ALLOC_ZONED: |
3811 | /* Nothing to do */ | |
3812 | break; | |
baba5062 NA |
3813 | default: |
3814 | BUG(); | |
3815 | } | |
3816 | ||
3817 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
3818 | ffe_ctl->index); | |
3819 | btrfs_release_block_group(block_group, delalloc); | |
3820 | } | |
3821 | ||
0ab9724b NA |
3822 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
3823 | struct btrfs_key *ins) | |
3824 | { | |
3825 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3826 | ||
3827 | if (!ffe_ctl->use_cluster && last_ptr) { | |
3828 | spin_lock(&last_ptr->lock); | |
3829 | last_ptr->window_start = ins->objectid; | |
3830 | spin_unlock(&last_ptr->lock); | |
3831 | } | |
3832 | } | |
3833 | ||
3834 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
3835 | struct btrfs_key *ins) | |
3836 | { | |
3837 | switch (ffe_ctl->policy) { | |
3838 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3839 | found_extent_clustered(ffe_ctl, ins); | |
3840 | break; | |
2eda5708 NA |
3841 | case BTRFS_EXTENT_ALLOC_ZONED: |
3842 | /* Nothing to do */ | |
3843 | break; | |
0ab9724b NA |
3844 | default: |
3845 | BUG(); | |
3846 | } | |
3847 | } | |
3848 | ||
c70e2139 NA |
3849 | static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl) |
3850 | { | |
3851 | switch (ffe_ctl->policy) { | |
3852 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3853 | /* | |
3854 | * If we can't allocate a new chunk we've already looped through | |
3855 | * at least once, move on to the NO_EMPTY_SIZE case. | |
3856 | */ | |
3857 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3858 | return 0; | |
2eda5708 NA |
3859 | case BTRFS_EXTENT_ALLOC_ZONED: |
3860 | /* Give up here */ | |
3861 | return -ENOSPC; | |
c70e2139 NA |
3862 | default: |
3863 | BUG(); | |
3864 | } | |
3865 | } | |
3866 | ||
e72d79d6 QW |
3867 | /* |
3868 | * Return >0 means caller needs to re-search for free extent | |
3869 | * Return 0 means we have the needed free extent. | |
3870 | * Return <0 means we failed to locate any free extent. | |
3871 | */ | |
3872 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
3873 | struct btrfs_key *ins, |
3874 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 3875 | bool full_search) |
e72d79d6 QW |
3876 | { |
3877 | struct btrfs_root *root = fs_info->extent_root; | |
3878 | int ret; | |
3879 | ||
3880 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3881 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3882 | ffe_ctl->orig_have_caching_bg = true; | |
3883 | ||
3884 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3885 | ffe_ctl->have_caching_bg) | |
3886 | return 1; | |
3887 | ||
3888 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3889 | return 1; | |
3890 | ||
3891 | if (ins->objectid) { | |
0ab9724b | 3892 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
3893 | return 0; |
3894 | } | |
3895 | ||
3896 | /* | |
3897 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3898 | * caching kthreads as we move along | |
3899 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3900 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3901 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3902 | * again | |
3903 | */ | |
3904 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3905 | ffe_ctl->index = 0; | |
3906 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3907 | /* | |
3908 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3909 | * don't have any uncached bgs and we've already done a | |
3910 | * full search through. | |
3911 | */ | |
3912 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3913 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3914 | else | |
3915 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3916 | } else { | |
3917 | ffe_ctl->loop++; | |
3918 | } | |
3919 | ||
3920 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3921 | struct btrfs_trans_handle *trans; | |
3922 | int exist = 0; | |
3923 | ||
3924 | trans = current->journal_info; | |
3925 | if (trans) | |
3926 | exist = 1; | |
3927 | else | |
3928 | trans = btrfs_join_transaction(root); | |
3929 | ||
3930 | if (IS_ERR(trans)) { | |
3931 | ret = PTR_ERR(trans); | |
3932 | return ret; | |
3933 | } | |
3934 | ||
fc471cb0 JB |
3935 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3936 | CHUNK_ALLOC_FORCE); | |
e72d79d6 | 3937 | |
e72d79d6 | 3938 | /* Do not bail out on ENOSPC since we can do more. */ |
c70e2139 NA |
3939 | if (ret == -ENOSPC) |
3940 | ret = chunk_allocation_failed(ffe_ctl); | |
3941 | else if (ret < 0) | |
e72d79d6 QW |
3942 | btrfs_abort_transaction(trans, ret); |
3943 | else | |
3944 | ret = 0; | |
3945 | if (!exist) | |
3946 | btrfs_end_transaction(trans); | |
3947 | if (ret) | |
3948 | return ret; | |
3949 | } | |
3950 | ||
3951 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
3952 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
3953 | return -ENOSPC; | |
3954 | ||
e72d79d6 QW |
3955 | /* |
3956 | * Don't loop again if we already have no empty_size and | |
3957 | * no empty_cluster. | |
3958 | */ | |
3959 | if (ffe_ctl->empty_size == 0 && | |
3960 | ffe_ctl->empty_cluster == 0) | |
3961 | return -ENOSPC; | |
3962 | ffe_ctl->empty_size = 0; | |
3963 | ffe_ctl->empty_cluster = 0; | |
3964 | } | |
3965 | return 1; | |
3966 | } | |
3967 | return -ENOSPC; | |
3968 | } | |
3969 | ||
7e895409 NA |
3970 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
3971 | struct find_free_extent_ctl *ffe_ctl, | |
3972 | struct btrfs_space_info *space_info, | |
3973 | struct btrfs_key *ins) | |
3974 | { | |
3975 | /* | |
3976 | * If our free space is heavily fragmented we may not be able to make | |
3977 | * big contiguous allocations, so instead of doing the expensive search | |
3978 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3979 | * can go ahead and search for a more manageable chunk. | |
3980 | * | |
3981 | * If our max_extent_size is large enough for our allocation simply | |
3982 | * disable clustering since we will likely not be able to find enough | |
3983 | * space to create a cluster and induce latency trying. | |
3984 | */ | |
3985 | if (space_info->max_extent_size) { | |
3986 | spin_lock(&space_info->lock); | |
3987 | if (space_info->max_extent_size && | |
3988 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
3989 | ins->offset = space_info->max_extent_size; | |
3990 | spin_unlock(&space_info->lock); | |
3991 | return -ENOSPC; | |
3992 | } else if (space_info->max_extent_size) { | |
3993 | ffe_ctl->use_cluster = false; | |
3994 | } | |
3995 | spin_unlock(&space_info->lock); | |
3996 | } | |
3997 | ||
3998 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
3999 | &ffe_ctl->empty_cluster); | |
4000 | if (ffe_ctl->last_ptr) { | |
4001 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
4002 | ||
4003 | spin_lock(&last_ptr->lock); | |
4004 | if (last_ptr->block_group) | |
4005 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4006 | if (last_ptr->fragmented) { | |
4007 | /* | |
4008 | * We still set window_start so we can keep track of the | |
4009 | * last place we found an allocation to try and save | |
4010 | * some time. | |
4011 | */ | |
4012 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4013 | ffe_ctl->use_cluster = false; | |
4014 | } | |
4015 | spin_unlock(&last_ptr->lock); | |
4016 | } | |
4017 | ||
4018 | return 0; | |
4019 | } | |
4020 | ||
4021 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
4022 | struct find_free_extent_ctl *ffe_ctl, | |
4023 | struct btrfs_space_info *space_info, | |
4024 | struct btrfs_key *ins) | |
4025 | { | |
4026 | switch (ffe_ctl->policy) { | |
4027 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4028 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
4029 | space_info, ins); | |
2eda5708 NA |
4030 | case BTRFS_EXTENT_ALLOC_ZONED: |
4031 | /* Nothing to do */ | |
4032 | return 0; | |
7e895409 NA |
4033 | default: |
4034 | BUG(); | |
4035 | } | |
4036 | } | |
4037 | ||
fec577fb CM |
4038 | /* |
4039 | * walks the btree of allocated extents and find a hole of a given size. | |
4040 | * The key ins is changed to record the hole: | |
a4820398 | 4041 | * ins->objectid == start position |
62e2749e | 4042 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4043 | * ins->offset == the size of the hole. |
fec577fb | 4044 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4045 | * |
4046 | * If there is no suitable free space, we will record the max size of | |
4047 | * the free space extent currently. | |
e72d79d6 QW |
4048 | * |
4049 | * The overall logic and call chain: | |
4050 | * | |
4051 | * find_free_extent() | |
4052 | * |- Iterate through all block groups | |
4053 | * | |- Get a valid block group | |
4054 | * | |- Try to do clustered allocation in that block group | |
4055 | * | |- Try to do unclustered allocation in that block group | |
4056 | * | |- Check if the result is valid | |
4057 | * | | |- If valid, then exit | |
4058 | * | |- Jump to next block group | |
4059 | * | | |
4060 | * |- Push harder to find free extents | |
4061 | * |- If not found, re-iterate all block groups | |
fec577fb | 4062 | */ |
437490fe | 4063 | static noinline int find_free_extent(struct btrfs_root *root, |
18513091 | 4064 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
ea544149 | 4065 | u64 hint_byte_orig, struct btrfs_key *ins, |
18513091 | 4066 | u64 flags, int delalloc) |
fec577fb | 4067 | { |
437490fe | 4068 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4069 | int ret = 0; |
db8fe64f | 4070 | int cache_block_group_error = 0; |
32da5386 | 4071 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 4072 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 4073 | struct btrfs_space_info *space_info; |
a5e681d9 | 4074 | bool full_search = false; |
fec577fb | 4075 | |
0b246afa | 4076 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 4077 | |
b4bd745d QW |
4078 | ffe_ctl.num_bytes = num_bytes; |
4079 | ffe_ctl.empty_size = empty_size; | |
4080 | ffe_ctl.flags = flags; | |
4081 | ffe_ctl.search_start = 0; | |
b4bd745d QW |
4082 | ffe_ctl.delalloc = delalloc; |
4083 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
4084 | ffe_ctl.have_caching_bg = false; | |
4085 | ffe_ctl.orig_have_caching_bg = false; | |
4086 | ffe_ctl.found_offset = 0; | |
ea544149 | 4087 | ffe_ctl.hint_byte = hint_byte_orig; |
cb2f96f8 | 4088 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED; |
b4bd745d | 4089 | |
c10859be NA |
4090 | /* For clustered allocation */ |
4091 | ffe_ctl.retry_clustered = false; | |
4092 | ffe_ctl.retry_unclustered = false; | |
4093 | ffe_ctl.last_ptr = NULL; | |
4094 | ffe_ctl.use_cluster = true; | |
4095 | ||
2eda5708 NA |
4096 | if (btrfs_is_zoned(fs_info)) |
4097 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_ZONED; | |
4098 | ||
962a298f | 4099 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4100 | ins->objectid = 0; |
4101 | ins->offset = 0; | |
b1a4d965 | 4102 | |
437490fe | 4103 | trace_find_free_extent(root, num_bytes, empty_size, flags); |
3f7de037 | 4104 | |
280c2908 | 4105 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 4106 | if (!space_info) { |
0b246afa | 4107 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
4108 | return -ENOSPC; |
4109 | } | |
2552d17e | 4110 | |
7e895409 NA |
4111 | ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins); |
4112 | if (ret < 0) | |
4113 | return ret; | |
fa9c0d79 | 4114 | |
b4bd745d QW |
4115 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
4116 | first_logical_byte(fs_info, 0)); | |
ea544149 NA |
4117 | ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte); |
4118 | if (ffe_ctl.search_start == ffe_ctl.hint_byte) { | |
b4bd745d QW |
4119 | block_group = btrfs_lookup_block_group(fs_info, |
4120 | ffe_ctl.search_start); | |
817d52f8 JB |
4121 | /* |
4122 | * we don't want to use the block group if it doesn't match our | |
4123 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4124 | * |
4125 | * However if we are re-searching with an ideal block group | |
4126 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4127 | */ |
b6919a58 | 4128 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 4129 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4130 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4131 | if (list_empty(&block_group->list) || |
4132 | block_group->ro) { | |
4133 | /* | |
4134 | * someone is removing this block group, | |
4135 | * we can't jump into the have_block_group | |
4136 | * target because our list pointers are not | |
4137 | * valid | |
4138 | */ | |
4139 | btrfs_put_block_group(block_group); | |
4140 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4141 | } else { |
b4bd745d | 4142 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 4143 | block_group->flags); |
e570fd27 | 4144 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 4145 | goto have_block_group; |
ccf0e725 | 4146 | } |
2552d17e | 4147 | } else if (block_group) { |
fa9c0d79 | 4148 | btrfs_put_block_group(block_group); |
2552d17e | 4149 | } |
42e70e7a | 4150 | } |
2552d17e | 4151 | search: |
b4bd745d QW |
4152 | ffe_ctl.have_caching_bg = false; |
4153 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
4154 | ffe_ctl.index == 0) | |
a5e681d9 | 4155 | full_search = true; |
80eb234a | 4156 | down_read(&space_info->groups_sem); |
b4bd745d QW |
4157 | list_for_each_entry(block_group, |
4158 | &space_info->block_groups[ffe_ctl.index], list) { | |
c668690d NA |
4159 | struct btrfs_block_group *bg_ret; |
4160 | ||
14443937 JM |
4161 | /* If the block group is read-only, we can skip it entirely. */ |
4162 | if (unlikely(block_group->ro)) | |
4163 | continue; | |
4164 | ||
e570fd27 | 4165 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 4166 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 4167 | |
83a50de9 CM |
4168 | /* |
4169 | * this can happen if we end up cycling through all the | |
4170 | * raid types, but we want to make sure we only allocate | |
4171 | * for the proper type. | |
4172 | */ | |
b6919a58 | 4173 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 4174 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4175 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4176 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4177 | BTRFS_BLOCK_GROUP_RAID10; |
4178 | ||
4179 | /* | |
4180 | * if they asked for extra copies and this block group | |
4181 | * doesn't provide them, bail. This does allow us to | |
4182 | * fill raid0 from raid1. | |
4183 | */ | |
b6919a58 | 4184 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4185 | goto loop; |
2a28468e QW |
4186 | |
4187 | /* | |
4188 | * This block group has different flags than we want. | |
4189 | * It's possible that we have MIXED_GROUP flag but no | |
4190 | * block group is mixed. Just skip such block group. | |
4191 | */ | |
4192 | btrfs_release_block_group(block_group, delalloc); | |
4193 | continue; | |
83a50de9 CM |
4194 | } |
4195 | ||
2552d17e | 4196 | have_block_group: |
32da5386 | 4197 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
4198 | if (unlikely(!ffe_ctl.cached)) { |
4199 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 4200 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
4201 | |
4202 | /* | |
4203 | * If we get ENOMEM here or something else we want to | |
4204 | * try other block groups, because it may not be fatal. | |
4205 | * However if we can't find anything else we need to | |
4206 | * save our return here so that we return the actual | |
4207 | * error that caused problems, not ENOSPC. | |
4208 | */ | |
4209 | if (ret < 0) { | |
4210 | if (!cache_block_group_error) | |
4211 | cache_block_group_error = ret; | |
4212 | ret = 0; | |
4213 | goto loop; | |
4214 | } | |
1d4284bd | 4215 | ret = 0; |
817d52f8 JB |
4216 | } |
4217 | ||
36cce922 JB |
4218 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
4219 | goto loop; | |
0f9dd46c | 4220 | |
c668690d NA |
4221 | bg_ret = NULL; |
4222 | ret = do_allocation(block_group, &ffe_ctl, &bg_ret); | |
4223 | if (ret == 0) { | |
4224 | if (bg_ret && bg_ret != block_group) { | |
4225 | btrfs_release_block_group(block_group, delalloc); | |
4226 | block_group = bg_ret; | |
fa9c0d79 | 4227 | } |
c668690d | 4228 | } else if (ret == -EAGAIN) { |
817d52f8 | 4229 | goto have_block_group; |
c668690d | 4230 | } else if (ret > 0) { |
1cdda9b8 | 4231 | goto loop; |
c668690d NA |
4232 | } |
4233 | ||
4234 | /* Checks */ | |
b4bd745d QW |
4235 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4236 | fs_info->stripesize); | |
25179201 | 4237 | |
2552d17e | 4238 | /* move on to the next group */ |
b4bd745d | 4239 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4240 | block_group->start + block_group->length) { |
2eda5708 NA |
4241 | btrfs_add_free_space_unused(block_group, |
4242 | ffe_ctl.found_offset, num_bytes); | |
2552d17e | 4243 | goto loop; |
6226cb0a | 4244 | } |
f5a31e16 | 4245 | |
b4bd745d | 4246 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
2eda5708 NA |
4247 | btrfs_add_free_space_unused(block_group, |
4248 | ffe_ctl.found_offset, | |
4249 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4250 | |
18513091 WX |
4251 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4252 | num_bytes, delalloc); | |
f0486c68 | 4253 | if (ret == -EAGAIN) { |
2eda5708 NA |
4254 | btrfs_add_free_space_unused(block_group, |
4255 | ffe_ctl.found_offset, num_bytes); | |
2552d17e | 4256 | goto loop; |
0f9dd46c | 4257 | } |
9cfa3e34 | 4258 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4259 | |
f0486c68 | 4260 | /* we are all good, lets return */ |
b4bd745d | 4261 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4262 | ins->offset = num_bytes; |
d2fb3437 | 4263 | |
b4bd745d QW |
4264 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4265 | num_bytes); | |
e570fd27 | 4266 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4267 | break; |
4268 | loop: | |
baba5062 | 4269 | release_block_group(block_group, &ffe_ctl, delalloc); |
14443937 | 4270 | cond_resched(); |
2552d17e JB |
4271 | } |
4272 | up_read(&space_info->groups_sem); | |
4273 | ||
15b7ee65 | 4274 | ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search); |
e72d79d6 | 4275 | if (ret > 0) |
b742bb82 YZ |
4276 | goto search; |
4277 | ||
db8fe64f | 4278 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4279 | /* |
4280 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4281 | * any contiguous hole. | |
4282 | */ | |
4283 | if (!ffe_ctl.max_extent_size) | |
4284 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4285 | spin_lock(&space_info->lock); |
b4bd745d | 4286 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4287 | spin_unlock(&space_info->lock); |
b4bd745d | 4288 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4289 | } else if (ret == -ENOSPC) { |
4290 | ret = cache_block_group_error; | |
4f4db217 | 4291 | } |
0f70abe2 | 4292 | return ret; |
fec577fb | 4293 | } |
ec44a35c | 4294 | |
6f47c706 NB |
4295 | /* |
4296 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4297 | * hole that is at least as big as @num_bytes. | |
4298 | * | |
4299 | * @root - The root that will contain this extent | |
4300 | * | |
4301 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4302 | * is used for accounting purposes. This value differs | |
4303 | * from @num_bytes only in the case of compressed extents. | |
4304 | * | |
4305 | * @num_bytes - Number of bytes to allocate on-disk. | |
4306 | * | |
4307 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4308 | * allocator should try to satisfy. In some cases | |
4309 | * @num_bytes may be larger than what is required and if | |
4310 | * the filesystem is fragmented then allocation fails. | |
4311 | * However, the presence of @min_alloc_size gives a | |
4312 | * chance to try and satisfy the smaller allocation. | |
4313 | * | |
4314 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4315 | * size in bytes the allocator should try to find free | |
4316 | * next to the block it returns. This is just a hint and | |
4317 | * may be ignored by the allocator. | |
4318 | * | |
4319 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4320 | * address passed. It might be ignored. | |
4321 | * | |
4322 | * @ins - This key is modified to record the found hole. It will | |
4323 | * have the following values: | |
4324 | * ins->objectid == start position | |
4325 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4326 | * ins->offset == the size of the hole. | |
4327 | * | |
4328 | * @is_data - Boolean flag indicating whether an extent is | |
4329 | * allocated for data (true) or metadata (false) | |
4330 | * | |
4331 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4332 | * delalloc or not. If 'true' data_rwsem of block groups | |
4333 | * is going to be acquired. | |
4334 | * | |
4335 | * | |
4336 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4337 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4338 | * largest available hole the allocator managed to find. | |
4339 | */ | |
18513091 | 4340 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4341 | u64 num_bytes, u64 min_alloc_size, |
4342 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4343 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4344 | { |
ab8d0fc4 | 4345 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4346 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4347 | u64 flags; |
fec577fb | 4348 | int ret; |
925baedd | 4349 | |
1b86826d | 4350 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4351 | again: |
0b246afa | 4352 | WARN_ON(num_bytes < fs_info->sectorsize); |
437490fe | 4353 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
18513091 | 4354 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4355 | if (!ret && !is_data) { |
ab8d0fc4 | 4356 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4357 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4358 | if (!final_tried && ins->offset) { |
4359 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4360 | num_bytes = round_down(num_bytes, |
0b246afa | 4361 | fs_info->sectorsize); |
9e622d6b | 4362 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4363 | ram_bytes = num_bytes; |
9e622d6b MX |
4364 | if (num_bytes == min_alloc_size) |
4365 | final_tried = true; | |
4366 | goto again; | |
ab8d0fc4 | 4367 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4368 | struct btrfs_space_info *sinfo; |
4369 | ||
280c2908 | 4370 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4371 | btrfs_err(fs_info, |
5d163e0e JM |
4372 | "allocation failed flags %llu, wanted %llu", |
4373 | flags, num_bytes); | |
53804280 | 4374 | if (sinfo) |
5da6afeb JB |
4375 | btrfs_dump_space_info(fs_info, sinfo, |
4376 | num_bytes, 1); | |
9e622d6b | 4377 | } |
925baedd | 4378 | } |
0f9dd46c JB |
4379 | |
4380 | return ret; | |
e6dcd2dc CM |
4381 | } |
4382 | ||
a0fbf736 NB |
4383 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4384 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4385 | { |
32da5386 | 4386 | struct btrfs_block_group *cache; |
0f9dd46c | 4387 | |
0b246afa | 4388 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4389 | if (!cache) { |
0b246afa JM |
4390 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4391 | start); | |
0f9dd46c JB |
4392 | return -ENOSPC; |
4393 | } | |
1f3c79a2 | 4394 | |
a0fbf736 NB |
4395 | btrfs_add_free_space(cache, start, len); |
4396 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4397 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4398 | ||
fa9c0d79 | 4399 | btrfs_put_block_group(cache); |
a0fbf736 | 4400 | return 0; |
e6dcd2dc CM |
4401 | } |
4402 | ||
7bfc1007 NB |
4403 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, |
4404 | u64 len) | |
e688b725 | 4405 | { |
7ef54d54 | 4406 | struct btrfs_block_group *cache; |
a0fbf736 | 4407 | int ret = 0; |
7ef54d54 | 4408 | |
7bfc1007 | 4409 | cache = btrfs_lookup_block_group(trans->fs_info, start); |
7ef54d54 | 4410 | if (!cache) { |
7bfc1007 NB |
4411 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
4412 | start); | |
7ef54d54 NB |
4413 | return -ENOSPC; |
4414 | } | |
4415 | ||
6690d071 | 4416 | ret = pin_down_extent(trans, cache, start, len, 1); |
7ef54d54 | 4417 | btrfs_put_block_group(cache); |
a0fbf736 | 4418 | return ret; |
e688b725 CM |
4419 | } |
4420 | ||
5d4f98a2 | 4421 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4422 | u64 parent, u64 root_objectid, |
4423 | u64 flags, u64 owner, u64 offset, | |
4424 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4425 | { |
ef89b824 | 4426 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4427 | int ret; |
e6dcd2dc | 4428 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4429 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4430 | struct btrfs_path *path; |
5d4f98a2 YZ |
4431 | struct extent_buffer *leaf; |
4432 | int type; | |
4433 | u32 size; | |
26b8003f | 4434 | |
5d4f98a2 YZ |
4435 | if (parent > 0) |
4436 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4437 | else | |
4438 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4439 | |
5d4f98a2 | 4440 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4441 | |
4442 | path = btrfs_alloc_path(); | |
db5b493a TI |
4443 | if (!path) |
4444 | return -ENOMEM; | |
47e4bb98 | 4445 | |
5d4f98a2 YZ |
4446 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4447 | ins, size); | |
79787eaa JM |
4448 | if (ret) { |
4449 | btrfs_free_path(path); | |
4450 | return ret; | |
4451 | } | |
0f9dd46c | 4452 | |
5d4f98a2 YZ |
4453 | leaf = path->nodes[0]; |
4454 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4455 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4456 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4457 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4458 | btrfs_set_extent_flags(leaf, extent_item, | |
4459 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4460 | ||
4461 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4462 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4463 | if (parent > 0) { | |
4464 | struct btrfs_shared_data_ref *ref; | |
4465 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4466 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4467 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4468 | } else { | |
4469 | struct btrfs_extent_data_ref *ref; | |
4470 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4471 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4472 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4473 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4474 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4475 | } | |
47e4bb98 CM |
4476 | |
4477 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4478 | btrfs_free_path(path); |
f510cfec | 4479 | |
25a356d3 | 4480 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4481 | if (ret) |
4482 | return ret; | |
4483 | ||
ade4b516 | 4484 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4485 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4486 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4487 | ins->objectid, ins->offset); |
f5947066 CM |
4488 | BUG(); |
4489 | } | |
71ff6437 | 4490 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4491 | return ret; |
4492 | } | |
4493 | ||
5d4f98a2 | 4494 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4495 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4496 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4497 | { |
9dcdbe01 | 4498 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4499 | int ret; |
5d4f98a2 | 4500 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4501 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4502 | struct btrfs_tree_block_info *block_info; |
4503 | struct btrfs_extent_inline_ref *iref; | |
4504 | struct btrfs_path *path; | |
4505 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4506 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4507 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4508 | u64 num_bytes; |
21ebfbe7 | 4509 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4510 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4511 | |
4e6bd4e0 NB |
4512 | ref = btrfs_delayed_node_to_tree_ref(node); |
4513 | ||
4e6bd4e0 NB |
4514 | extent_key.objectid = node->bytenr; |
4515 | if (skinny_metadata) { | |
4516 | extent_key.offset = ref->level; | |
4517 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4518 | num_bytes = fs_info->nodesize; | |
4519 | } else { | |
4520 | extent_key.offset = node->num_bytes; | |
4521 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4522 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4523 | num_bytes = node->num_bytes; |
4524 | } | |
1c2308f8 | 4525 | |
5d4f98a2 | 4526 | path = btrfs_alloc_path(); |
80ee54bf | 4527 | if (!path) |
d8926bb3 | 4528 | return -ENOMEM; |
56bec294 | 4529 | |
5d4f98a2 | 4530 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4e6bd4e0 | 4531 | &extent_key, size); |
79787eaa | 4532 | if (ret) { |
dd825259 | 4533 | btrfs_free_path(path); |
79787eaa JM |
4534 | return ret; |
4535 | } | |
5d4f98a2 YZ |
4536 | |
4537 | leaf = path->nodes[0]; | |
4538 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4539 | struct btrfs_extent_item); | |
4540 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4541 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4542 | btrfs_set_extent_flags(leaf, extent_item, | |
4543 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4544 | |
3173a18f JB |
4545 | if (skinny_metadata) { |
4546 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4547 | } else { | |
4548 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4549 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4550 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4551 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4552 | } | |
5d4f98a2 | 4553 | |
d4b20733 | 4554 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4555 | btrfs_set_extent_inline_ref_type(leaf, iref, |
4556 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4557 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4558 | } else { |
4559 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4560 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4561 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4562 | } |
4563 | ||
4564 | btrfs_mark_buffer_dirty(leaf); | |
4565 | btrfs_free_path(path); | |
4566 | ||
4e6bd4e0 NB |
4567 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4568 | num_bytes); | |
1e144fb8 OS |
4569 | if (ret) |
4570 | return ret; | |
4571 | ||
ade4b516 JB |
4572 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4573 | fs_info->nodesize, 1); | |
79787eaa | 4574 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4575 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4576 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4577 | BUG(); |
4578 | } | |
0be5dc67 | 4579 | |
4e6bd4e0 | 4580 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4581 | fs_info->nodesize); |
5d4f98a2 YZ |
4582 | return ret; |
4583 | } | |
4584 | ||
4585 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4586 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4587 | u64 offset, u64 ram_bytes, |
4588 | struct btrfs_key *ins) | |
5d4f98a2 | 4589 | { |
76675593 | 4590 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 | 4591 | |
84f7d8e6 | 4592 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4593 | |
76675593 QW |
4594 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4595 | ins->objectid, ins->offset, 0); | |
4596 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4597 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
2187374f JB |
4598 | |
4599 | return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes); | |
e6dcd2dc | 4600 | } |
e02119d5 CM |
4601 | |
4602 | /* | |
4603 | * this is used by the tree logging recovery code. It records that | |
4604 | * an extent has been allocated and makes sure to clear the free | |
4605 | * space cache bits as well | |
4606 | */ | |
5d4f98a2 | 4607 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4608 | u64 root_objectid, u64 owner, u64 offset, |
4609 | struct btrfs_key *ins) | |
e02119d5 | 4610 | { |
61da2abf | 4611 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4612 | int ret; |
32da5386 | 4613 | struct btrfs_block_group *block_group; |
ed7a6948 | 4614 | struct btrfs_space_info *space_info; |
11833d66 | 4615 | |
8c2a1a30 JB |
4616 | /* |
4617 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4618 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4619 | */ |
0b246afa | 4620 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4621 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4622 | ins->offset); | |
b50c6e25 | 4623 | if (ret) |
8c2a1a30 | 4624 | return ret; |
11833d66 YZ |
4625 | } |
4626 | ||
0b246afa | 4627 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4628 | if (!block_group) |
4629 | return -EINVAL; | |
4630 | ||
ed7a6948 WX |
4631 | space_info = block_group->space_info; |
4632 | spin_lock(&space_info->lock); | |
4633 | spin_lock(&block_group->lock); | |
4634 | space_info->bytes_reserved += ins->offset; | |
4635 | block_group->reserved += ins->offset; | |
4636 | spin_unlock(&block_group->lock); | |
4637 | spin_unlock(&space_info->lock); | |
4638 | ||
ef89b824 NB |
4639 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4640 | offset, ins, 1); | |
bd727173 | 4641 | if (ret) |
ab9b2c7b | 4642 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
b50c6e25 | 4643 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4644 | return ret; |
4645 | } | |
4646 | ||
48a3b636 ES |
4647 | static struct extent_buffer * |
4648 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
9631e4cc JB |
4649 | u64 bytenr, int level, u64 owner, |
4650 | enum btrfs_lock_nesting nest) | |
65b51a00 | 4651 | { |
0b246afa | 4652 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4653 | struct extent_buffer *buf; |
4654 | ||
3fbaf258 | 4655 | buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level); |
c871b0f2 LB |
4656 | if (IS_ERR(buf)) |
4657 | return buf; | |
4658 | ||
b72c3aba QW |
4659 | /* |
4660 | * Extra safety check in case the extent tree is corrupted and extent | |
4661 | * allocator chooses to use a tree block which is already used and | |
4662 | * locked. | |
4663 | */ | |
4664 | if (buf->lock_owner == current->pid) { | |
4665 | btrfs_err_rl(fs_info, | |
4666 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4667 | buf->start, btrfs_header_owner(buf), current->pid); | |
4668 | free_extent_buffer(buf); | |
4669 | return ERR_PTR(-EUCLEAN); | |
4670 | } | |
4671 | ||
e114c545 JB |
4672 | /* |
4673 | * This needs to stay, because we could allocate a freed block from an | |
4674 | * old tree into a new tree, so we need to make sure this new block is | |
4675 | * set to the appropriate level and owner. | |
4676 | */ | |
ad244665 | 4677 | btrfs_set_buffer_lockdep_class(owner, buf, level); |
9631e4cc | 4678 | __btrfs_tree_lock(buf, nest); |
6a884d7d | 4679 | btrfs_clean_tree_block(buf); |
3083ee2e | 4680 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
d3575156 | 4681 | clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags); |
b4ce94de | 4682 | |
4db8c528 | 4683 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4684 | |
bc877d28 NB |
4685 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4686 | btrfs_set_header_level(buf, level); | |
4687 | btrfs_set_header_bytenr(buf, buf->start); | |
4688 | btrfs_set_header_generation(buf, trans->transid); | |
4689 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4690 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4691 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4692 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4693 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4694 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4695 | /* |
4696 | * we allow two log transactions at a time, use different | |
52042d8e | 4697 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4698 | */ |
656f30db | 4699 | if (buf->log_index == 0) |
8cef4e16 YZ |
4700 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4701 | buf->start + buf->len - 1, GFP_NOFS); | |
4702 | else | |
4703 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4704 | buf->start + buf->len - 1); |
d0c803c4 | 4705 | } else { |
656f30db | 4706 | buf->log_index = -1; |
d0c803c4 | 4707 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4708 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4709 | } |
64c12921 | 4710 | trans->dirty = true; |
b4ce94de | 4711 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4712 | return buf; |
4713 | } | |
4714 | ||
fec577fb | 4715 | /* |
f0486c68 | 4716 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4717 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4718 | */ |
4d75f8a9 | 4719 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4720 | struct btrfs_root *root, |
4721 | u64 parent, u64 root_objectid, | |
4722 | const struct btrfs_disk_key *key, | |
4723 | int level, u64 hint, | |
9631e4cc JB |
4724 | u64 empty_size, |
4725 | enum btrfs_lock_nesting nest) | |
fec577fb | 4726 | { |
0b246afa | 4727 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4728 | struct btrfs_key ins; |
f0486c68 | 4729 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4730 | struct extent_buffer *buf; |
67b7859e | 4731 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4732 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4733 | u64 flags = 0; |
4734 | int ret; | |
0b246afa JM |
4735 | u32 blocksize = fs_info->nodesize; |
4736 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4737 | |
05653ef3 | 4738 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4739 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4740 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
9631e4cc | 4741 | level, root_objectid, nest); |
faa2dbf0 JB |
4742 | if (!IS_ERR(buf)) |
4743 | root->alloc_bytenr += blocksize; | |
4744 | return buf; | |
4745 | } | |
05653ef3 | 4746 | #endif |
fccb84c9 | 4747 | |
67f9c220 | 4748 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4749 | if (IS_ERR(block_rsv)) |
4750 | return ERR_CAST(block_rsv); | |
4751 | ||
18513091 | 4752 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4753 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4754 | if (ret) |
4755 | goto out_unuse; | |
55c69072 | 4756 | |
bc877d28 | 4757 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
9631e4cc | 4758 | root_objectid, nest); |
67b7859e OS |
4759 | if (IS_ERR(buf)) { |
4760 | ret = PTR_ERR(buf); | |
4761 | goto out_free_reserved; | |
4762 | } | |
f0486c68 YZ |
4763 | |
4764 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4765 | if (parent == 0) | |
4766 | parent = ins.objectid; | |
4767 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4768 | } else | |
4769 | BUG_ON(parent > 0); | |
4770 | ||
4771 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4772 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4773 | if (!extent_op) { |
4774 | ret = -ENOMEM; | |
4775 | goto out_free_buf; | |
4776 | } | |
f0486c68 YZ |
4777 | if (key) |
4778 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4779 | else | |
4780 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4781 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4782 | extent_op->update_key = skinny_metadata ? false : true; |
4783 | extent_op->update_flags = true; | |
4784 | extent_op->is_data = false; | |
b1c79e09 | 4785 | extent_op->level = level; |
f0486c68 | 4786 | |
ed4f255b QW |
4787 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4788 | ins.objectid, ins.offset, parent); | |
4789 | generic_ref.real_root = root->root_key.objectid; | |
4790 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4791 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 4792 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op); |
67b7859e OS |
4793 | if (ret) |
4794 | goto out_free_delayed; | |
f0486c68 | 4795 | } |
fec577fb | 4796 | return buf; |
67b7859e OS |
4797 | |
4798 | out_free_delayed: | |
4799 | btrfs_free_delayed_extent_op(extent_op); | |
4800 | out_free_buf: | |
4801 | free_extent_buffer(buf); | |
4802 | out_free_reserved: | |
2ff7e61e | 4803 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4804 | out_unuse: |
67f9c220 | 4805 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4806 | return ERR_PTR(ret); |
fec577fb | 4807 | } |
a28ec197 | 4808 | |
2c47e605 YZ |
4809 | struct walk_control { |
4810 | u64 refs[BTRFS_MAX_LEVEL]; | |
4811 | u64 flags[BTRFS_MAX_LEVEL]; | |
4812 | struct btrfs_key update_progress; | |
aea6f028 JB |
4813 | struct btrfs_key drop_progress; |
4814 | int drop_level; | |
2c47e605 YZ |
4815 | int stage; |
4816 | int level; | |
4817 | int shared_level; | |
4818 | int update_ref; | |
4819 | int keep_locks; | |
1c4850e2 YZ |
4820 | int reada_slot; |
4821 | int reada_count; | |
78c52d9e | 4822 | int restarted; |
2c47e605 YZ |
4823 | }; |
4824 | ||
4825 | #define DROP_REFERENCE 1 | |
4826 | #define UPDATE_BACKREF 2 | |
4827 | ||
1c4850e2 YZ |
4828 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4829 | struct btrfs_root *root, | |
4830 | struct walk_control *wc, | |
4831 | struct btrfs_path *path) | |
6407bf6d | 4832 | { |
0b246afa | 4833 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4834 | u64 bytenr; |
4835 | u64 generation; | |
4836 | u64 refs; | |
94fcca9f | 4837 | u64 flags; |
5d4f98a2 | 4838 | u32 nritems; |
1c4850e2 YZ |
4839 | struct btrfs_key key; |
4840 | struct extent_buffer *eb; | |
6407bf6d | 4841 | int ret; |
1c4850e2 YZ |
4842 | int slot; |
4843 | int nread = 0; | |
6407bf6d | 4844 | |
1c4850e2 YZ |
4845 | if (path->slots[wc->level] < wc->reada_slot) { |
4846 | wc->reada_count = wc->reada_count * 2 / 3; | |
4847 | wc->reada_count = max(wc->reada_count, 2); | |
4848 | } else { | |
4849 | wc->reada_count = wc->reada_count * 3 / 2; | |
4850 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4851 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4852 | } |
7bb86316 | 4853 | |
1c4850e2 YZ |
4854 | eb = path->nodes[wc->level]; |
4855 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4856 | |
1c4850e2 YZ |
4857 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4858 | if (nread >= wc->reada_count) | |
4859 | break; | |
bd56b302 | 4860 | |
2dd3e67b | 4861 | cond_resched(); |
1c4850e2 YZ |
4862 | bytenr = btrfs_node_blockptr(eb, slot); |
4863 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4864 | |
1c4850e2 YZ |
4865 | if (slot == path->slots[wc->level]) |
4866 | goto reada; | |
5d4f98a2 | 4867 | |
1c4850e2 YZ |
4868 | if (wc->stage == UPDATE_BACKREF && |
4869 | generation <= root->root_key.offset) | |
bd56b302 CM |
4870 | continue; |
4871 | ||
94fcca9f | 4872 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4873 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4874 | wc->level - 1, 1, &refs, |
4875 | &flags); | |
79787eaa JM |
4876 | /* We don't care about errors in readahead. */ |
4877 | if (ret < 0) | |
4878 | continue; | |
94fcca9f YZ |
4879 | BUG_ON(refs == 0); |
4880 | ||
1c4850e2 | 4881 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4882 | if (refs == 1) |
4883 | goto reada; | |
bd56b302 | 4884 | |
94fcca9f YZ |
4885 | if (wc->level == 1 && |
4886 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4887 | continue; | |
1c4850e2 YZ |
4888 | if (!wc->update_ref || |
4889 | generation <= root->root_key.offset) | |
4890 | continue; | |
4891 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4892 | ret = btrfs_comp_cpu_keys(&key, | |
4893 | &wc->update_progress); | |
4894 | if (ret < 0) | |
4895 | continue; | |
94fcca9f YZ |
4896 | } else { |
4897 | if (wc->level == 1 && | |
4898 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4899 | continue; | |
6407bf6d | 4900 | } |
1c4850e2 | 4901 | reada: |
bfb484d9 | 4902 | btrfs_readahead_node_child(eb, slot); |
1c4850e2 | 4903 | nread++; |
20524f02 | 4904 | } |
1c4850e2 | 4905 | wc->reada_slot = slot; |
20524f02 | 4906 | } |
2c47e605 | 4907 | |
f82d02d9 | 4908 | /* |
2c016dc2 | 4909 | * helper to process tree block while walking down the tree. |
2c47e605 | 4910 | * |
2c47e605 YZ |
4911 | * when wc->stage == UPDATE_BACKREF, this function updates |
4912 | * back refs for pointers in the block. | |
4913 | * | |
4914 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4915 | */ |
2c47e605 | 4916 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4917 | struct btrfs_root *root, |
2c47e605 | 4918 | struct btrfs_path *path, |
94fcca9f | 4919 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4920 | { |
2ff7e61e | 4921 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4922 | int level = wc->level; |
4923 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4924 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4925 | int ret; |
4926 | ||
2c47e605 YZ |
4927 | if (wc->stage == UPDATE_BACKREF && |
4928 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4929 | return 1; | |
f82d02d9 | 4930 | |
2c47e605 YZ |
4931 | /* |
4932 | * when reference count of tree block is 1, it won't increase | |
4933 | * again. once full backref flag is set, we never clear it. | |
4934 | */ | |
94fcca9f YZ |
4935 | if (lookup_info && |
4936 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4937 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4938 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4939 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4940 | eb->start, level, 1, |
2c47e605 YZ |
4941 | &wc->refs[level], |
4942 | &wc->flags[level]); | |
79787eaa JM |
4943 | BUG_ON(ret == -ENOMEM); |
4944 | if (ret) | |
4945 | return ret; | |
2c47e605 YZ |
4946 | BUG_ON(wc->refs[level] == 0); |
4947 | } | |
5d4f98a2 | 4948 | |
2c47e605 YZ |
4949 | if (wc->stage == DROP_REFERENCE) { |
4950 | if (wc->refs[level] > 1) | |
4951 | return 1; | |
f82d02d9 | 4952 | |
2c47e605 | 4953 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4954 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4955 | path->locks[level] = 0; |
4956 | } | |
4957 | return 0; | |
4958 | } | |
f82d02d9 | 4959 | |
2c47e605 YZ |
4960 | /* wc->stage == UPDATE_BACKREF */ |
4961 | if (!(wc->flags[level] & flag)) { | |
4962 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4963 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4964 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4965 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4966 | BUG_ON(ret); /* -ENOMEM */ |
42c9d0b5 | 4967 | ret = btrfs_set_disk_extent_flags(trans, eb, flag, |
b1c79e09 | 4968 | btrfs_header_level(eb), 0); |
79787eaa | 4969 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4970 | wc->flags[level] |= flag; |
4971 | } | |
4972 | ||
4973 | /* | |
4974 | * the block is shared by multiple trees, so it's not good to | |
4975 | * keep the tree lock | |
4976 | */ | |
4977 | if (path->locks[level] && level > 0) { | |
bd681513 | 4978 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4979 | path->locks[level] = 0; |
4980 | } | |
4981 | return 0; | |
4982 | } | |
4983 | ||
78c52d9e JB |
4984 | /* |
4985 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4986 | * would have a drop_progress key that hadn't been updated properly. | |
4987 | */ | |
4988 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4989 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4990 | int level) | |
4991 | { | |
4992 | struct btrfs_path *path; | |
4993 | struct btrfs_extent_inline_ref *iref; | |
4994 | int ret; | |
4995 | ||
4996 | path = btrfs_alloc_path(); | |
4997 | if (!path) | |
4998 | return -ENOMEM; | |
4999 | ||
5000 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
5001 | root->fs_info->nodesize, parent, | |
5002 | root->root_key.objectid, level, 0); | |
5003 | btrfs_free_path(path); | |
5004 | if (ret == -ENOENT) | |
5005 | return 0; | |
5006 | if (ret < 0) | |
5007 | return ret; | |
5008 | return 1; | |
5009 | } | |
5010 | ||
1c4850e2 | 5011 | /* |
2c016dc2 | 5012 | * helper to process tree block pointer. |
1c4850e2 YZ |
5013 | * |
5014 | * when wc->stage == DROP_REFERENCE, this function checks | |
5015 | * reference count of the block pointed to. if the block | |
5016 | * is shared and we need update back refs for the subtree | |
5017 | * rooted at the block, this function changes wc->stage to | |
5018 | * UPDATE_BACKREF. if the block is shared and there is no | |
5019 | * need to update back, this function drops the reference | |
5020 | * to the block. | |
5021 | * | |
5022 | * NOTE: return value 1 means we should stop walking down. | |
5023 | */ | |
5024 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5025 | struct btrfs_root *root, | |
5026 | struct btrfs_path *path, | |
94fcca9f | 5027 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 5028 | { |
0b246afa | 5029 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
5030 | u64 bytenr; |
5031 | u64 generation; | |
5032 | u64 parent; | |
1c4850e2 | 5033 | struct btrfs_key key; |
581c1760 | 5034 | struct btrfs_key first_key; |
ffd4bb2a | 5035 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
5036 | struct extent_buffer *next; |
5037 | int level = wc->level; | |
5038 | int reada = 0; | |
5039 | int ret = 0; | |
1152651a | 5040 | bool need_account = false; |
1c4850e2 YZ |
5041 | |
5042 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5043 | path->slots[level]); | |
5044 | /* | |
5045 | * if the lower level block was created before the snapshot | |
5046 | * was created, we know there is no need to update back refs | |
5047 | * for the subtree | |
5048 | */ | |
5049 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5050 | generation <= root->root_key.offset) { |
5051 | *lookup_info = 1; | |
1c4850e2 | 5052 | return 1; |
94fcca9f | 5053 | } |
1c4850e2 YZ |
5054 | |
5055 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
5056 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
5057 | path->slots[level]); | |
1c4850e2 | 5058 | |
0b246afa | 5059 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5060 | if (!next) { |
3fbaf258 JB |
5061 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
5062 | root->root_key.objectid, level - 1); | |
c871b0f2 LB |
5063 | if (IS_ERR(next)) |
5064 | return PTR_ERR(next); | |
1c4850e2 YZ |
5065 | reada = 1; |
5066 | } | |
5067 | btrfs_tree_lock(next); | |
1c4850e2 | 5068 | |
2ff7e61e | 5069 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
5070 | &wc->refs[level - 1], |
5071 | &wc->flags[level - 1]); | |
4867268c JB |
5072 | if (ret < 0) |
5073 | goto out_unlock; | |
79787eaa | 5074 | |
c2cf52eb | 5075 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5076 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5077 | ret = -EIO; |
5078 | goto out_unlock; | |
c2cf52eb | 5079 | } |
94fcca9f | 5080 | *lookup_info = 0; |
1c4850e2 | 5081 | |
94fcca9f | 5082 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5083 | if (wc->refs[level - 1] > 1) { |
1152651a | 5084 | need_account = true; |
94fcca9f YZ |
5085 | if (level == 1 && |
5086 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5087 | goto skip; | |
5088 | ||
1c4850e2 YZ |
5089 | if (!wc->update_ref || |
5090 | generation <= root->root_key.offset) | |
5091 | goto skip; | |
5092 | ||
5093 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5094 | path->slots[level]); | |
5095 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5096 | if (ret < 0) | |
5097 | goto skip; | |
5098 | ||
5099 | wc->stage = UPDATE_BACKREF; | |
5100 | wc->shared_level = level - 1; | |
5101 | } | |
94fcca9f YZ |
5102 | } else { |
5103 | if (level == 1 && | |
5104 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5105 | goto skip; | |
1c4850e2 YZ |
5106 | } |
5107 | ||
b9fab919 | 5108 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5109 | btrfs_tree_unlock(next); |
5110 | free_extent_buffer(next); | |
5111 | next = NULL; | |
94fcca9f | 5112 | *lookup_info = 1; |
1c4850e2 YZ |
5113 | } |
5114 | ||
5115 | if (!next) { | |
5116 | if (reada && level == 1) | |
5117 | reada_walk_down(trans, root, wc, path); | |
1b7ec85e JB |
5118 | next = read_tree_block(fs_info, bytenr, root->root_key.objectid, |
5119 | generation, level - 1, &first_key); | |
64c043de LB |
5120 | if (IS_ERR(next)) { |
5121 | return PTR_ERR(next); | |
5122 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5123 | free_extent_buffer(next); |
97d9a8a4 | 5124 | return -EIO; |
416bc658 | 5125 | } |
1c4850e2 | 5126 | btrfs_tree_lock(next); |
1c4850e2 YZ |
5127 | } |
5128 | ||
5129 | level--; | |
4867268c JB |
5130 | ASSERT(level == btrfs_header_level(next)); |
5131 | if (level != btrfs_header_level(next)) { | |
5132 | btrfs_err(root->fs_info, "mismatched level"); | |
5133 | ret = -EIO; | |
5134 | goto out_unlock; | |
5135 | } | |
1c4850e2 YZ |
5136 | path->nodes[level] = next; |
5137 | path->slots[level] = 0; | |
ac5887c8 | 5138 | path->locks[level] = BTRFS_WRITE_LOCK; |
1c4850e2 YZ |
5139 | wc->level = level; |
5140 | if (wc->level == 1) | |
5141 | wc->reada_slot = 0; | |
5142 | return 0; | |
5143 | skip: | |
5144 | wc->refs[level - 1] = 0; | |
5145 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5146 | if (wc->stage == DROP_REFERENCE) { |
5147 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5148 | parent = path->nodes[level]->start; | |
5149 | } else { | |
4867268c | 5150 | ASSERT(root->root_key.objectid == |
94fcca9f | 5151 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5152 | if (root->root_key.objectid != |
5153 | btrfs_header_owner(path->nodes[level])) { | |
5154 | btrfs_err(root->fs_info, | |
5155 | "mismatched block owner"); | |
5156 | ret = -EIO; | |
5157 | goto out_unlock; | |
5158 | } | |
94fcca9f YZ |
5159 | parent = 0; |
5160 | } | |
1c4850e2 | 5161 | |
78c52d9e JB |
5162 | /* |
5163 | * If we had a drop_progress we need to verify the refs are set | |
5164 | * as expected. If we find our ref then we know that from here | |
5165 | * on out everything should be correct, and we can clear the | |
5166 | * ->restarted flag. | |
5167 | */ | |
5168 | if (wc->restarted) { | |
5169 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5170 | level - 1); | |
5171 | if (ret < 0) | |
5172 | goto out_unlock; | |
5173 | if (ret == 0) | |
5174 | goto no_delete; | |
5175 | ret = 0; | |
5176 | wc->restarted = 0; | |
5177 | } | |
5178 | ||
2cd86d30 QW |
5179 | /* |
5180 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5181 | * already accounted them at merge time (replace_path), | |
5182 | * thus we could skip expensive subtree trace here. | |
5183 | */ | |
5184 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5185 | need_account) { | |
deb40627 | 5186 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5187 | generation, level - 1); |
1152651a | 5188 | if (ret) { |
0b246afa | 5189 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5190 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5191 | ret); | |
1152651a MF |
5192 | } |
5193 | } | |
aea6f028 JB |
5194 | |
5195 | /* | |
5196 | * We need to update the next key in our walk control so we can | |
5197 | * update the drop_progress key accordingly. We don't care if | |
5198 | * find_next_key doesn't find a key because that means we're at | |
5199 | * the end and are going to clean up now. | |
5200 | */ | |
5201 | wc->drop_level = level; | |
5202 | find_next_key(path, level, &wc->drop_progress); | |
5203 | ||
ffd4bb2a QW |
5204 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
5205 | fs_info->nodesize, parent); | |
5206 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
5207 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
5208 | if (ret) |
5209 | goto out_unlock; | |
1c4850e2 | 5210 | } |
78c52d9e | 5211 | no_delete: |
4867268c JB |
5212 | *lookup_info = 1; |
5213 | ret = 1; | |
5214 | ||
5215 | out_unlock: | |
1c4850e2 YZ |
5216 | btrfs_tree_unlock(next); |
5217 | free_extent_buffer(next); | |
4867268c JB |
5218 | |
5219 | return ret; | |
1c4850e2 YZ |
5220 | } |
5221 | ||
2c47e605 | 5222 | /* |
2c016dc2 | 5223 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5224 | * |
5225 | * when wc->stage == DROP_REFERENCE, this function drops | |
5226 | * reference count on the block. | |
5227 | * | |
5228 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5229 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5230 | * to UPDATE_BACKREF previously while processing the block. | |
5231 | * | |
5232 | * NOTE: return value 1 means we should stop walking up. | |
5233 | */ | |
5234 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5235 | struct btrfs_root *root, | |
5236 | struct btrfs_path *path, | |
5237 | struct walk_control *wc) | |
5238 | { | |
0b246afa | 5239 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5240 | int ret; |
2c47e605 YZ |
5241 | int level = wc->level; |
5242 | struct extent_buffer *eb = path->nodes[level]; | |
5243 | u64 parent = 0; | |
5244 | ||
5245 | if (wc->stage == UPDATE_BACKREF) { | |
5246 | BUG_ON(wc->shared_level < level); | |
5247 | if (level < wc->shared_level) | |
5248 | goto out; | |
5249 | ||
2c47e605 YZ |
5250 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5251 | if (ret > 0) | |
5252 | wc->update_ref = 0; | |
5253 | ||
5254 | wc->stage = DROP_REFERENCE; | |
5255 | wc->shared_level = -1; | |
5256 | path->slots[level] = 0; | |
5257 | ||
5258 | /* | |
5259 | * check reference count again if the block isn't locked. | |
5260 | * we should start walking down the tree again if reference | |
5261 | * count is one. | |
5262 | */ | |
5263 | if (!path->locks[level]) { | |
5264 | BUG_ON(level == 0); | |
5265 | btrfs_tree_lock(eb); | |
ac5887c8 | 5266 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5267 | |
2ff7e61e | 5268 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5269 | eb->start, level, 1, |
2c47e605 YZ |
5270 | &wc->refs[level], |
5271 | &wc->flags[level]); | |
79787eaa JM |
5272 | if (ret < 0) { |
5273 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5274 | path->locks[level] = 0; |
79787eaa JM |
5275 | return ret; |
5276 | } | |
2c47e605 YZ |
5277 | BUG_ON(wc->refs[level] == 0); |
5278 | if (wc->refs[level] == 1) { | |
bd681513 | 5279 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5280 | path->locks[level] = 0; |
2c47e605 YZ |
5281 | return 1; |
5282 | } | |
f82d02d9 | 5283 | } |
2c47e605 | 5284 | } |
f82d02d9 | 5285 | |
2c47e605 YZ |
5286 | /* wc->stage == DROP_REFERENCE */ |
5287 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5288 | |
2c47e605 YZ |
5289 | if (wc->refs[level] == 1) { |
5290 | if (level == 0) { | |
5291 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5292 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5293 | else |
e339a6b0 | 5294 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5295 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5296 | if (is_fstree(root->root_key.objectid)) { |
5297 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5298 | if (ret) { | |
5299 | btrfs_err_rl(fs_info, | |
5300 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5301 | ret); |
c4140cbf | 5302 | } |
1152651a | 5303 | } |
2c47e605 | 5304 | } |
6a884d7d | 5305 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5306 | if (!path->locks[level] && |
5307 | btrfs_header_generation(eb) == trans->transid) { | |
5308 | btrfs_tree_lock(eb); | |
ac5887c8 | 5309 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5310 | } |
6a884d7d | 5311 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5312 | } |
5313 | ||
5314 | if (eb == root->node) { | |
5315 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5316 | parent = eb->start; | |
65c6e82b QW |
5317 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5318 | goto owner_mismatch; | |
2c47e605 YZ |
5319 | } else { |
5320 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5321 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5322 | else if (root->root_key.objectid != |
5323 | btrfs_header_owner(path->nodes[level + 1])) | |
5324 | goto owner_mismatch; | |
f82d02d9 | 5325 | } |
f82d02d9 | 5326 | |
5581a51a | 5327 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5328 | out: |
5329 | wc->refs[level] = 0; | |
5330 | wc->flags[level] = 0; | |
f0486c68 | 5331 | return 0; |
65c6e82b QW |
5332 | |
5333 | owner_mismatch: | |
5334 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5335 | btrfs_header_owner(eb), root->root_key.objectid); | |
5336 | return -EUCLEAN; | |
2c47e605 YZ |
5337 | } |
5338 | ||
5339 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5340 | struct btrfs_root *root, | |
5341 | struct btrfs_path *path, | |
5342 | struct walk_control *wc) | |
5343 | { | |
2c47e605 | 5344 | int level = wc->level; |
94fcca9f | 5345 | int lookup_info = 1; |
2c47e605 YZ |
5346 | int ret; |
5347 | ||
5348 | while (level >= 0) { | |
94fcca9f | 5349 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5350 | if (ret > 0) |
5351 | break; | |
5352 | ||
5353 | if (level == 0) | |
5354 | break; | |
5355 | ||
7a7965f8 YZ |
5356 | if (path->slots[level] >= |
5357 | btrfs_header_nritems(path->nodes[level])) | |
5358 | break; | |
5359 | ||
94fcca9f | 5360 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5361 | if (ret > 0) { |
5362 | path->slots[level]++; | |
5363 | continue; | |
90d2c51d MX |
5364 | } else if (ret < 0) |
5365 | return ret; | |
1c4850e2 | 5366 | level = wc->level; |
f82d02d9 | 5367 | } |
f82d02d9 YZ |
5368 | return 0; |
5369 | } | |
5370 | ||
d397712b | 5371 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5372 | struct btrfs_root *root, |
f82d02d9 | 5373 | struct btrfs_path *path, |
2c47e605 | 5374 | struct walk_control *wc, int max_level) |
20524f02 | 5375 | { |
2c47e605 | 5376 | int level = wc->level; |
20524f02 | 5377 | int ret; |
9f3a7427 | 5378 | |
2c47e605 YZ |
5379 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5380 | while (level < max_level && path->nodes[level]) { | |
5381 | wc->level = level; | |
5382 | if (path->slots[level] + 1 < | |
5383 | btrfs_header_nritems(path->nodes[level])) { | |
5384 | path->slots[level]++; | |
20524f02 CM |
5385 | return 0; |
5386 | } else { | |
2c47e605 YZ |
5387 | ret = walk_up_proc(trans, root, path, wc); |
5388 | if (ret > 0) | |
5389 | return 0; | |
65c6e82b QW |
5390 | if (ret < 0) |
5391 | return ret; | |
bd56b302 | 5392 | |
2c47e605 | 5393 | if (path->locks[level]) { |
bd681513 CM |
5394 | btrfs_tree_unlock_rw(path->nodes[level], |
5395 | path->locks[level]); | |
2c47e605 | 5396 | path->locks[level] = 0; |
f82d02d9 | 5397 | } |
2c47e605 YZ |
5398 | free_extent_buffer(path->nodes[level]); |
5399 | path->nodes[level] = NULL; | |
5400 | level++; | |
20524f02 CM |
5401 | } |
5402 | } | |
5403 | return 1; | |
5404 | } | |
5405 | ||
9aca1d51 | 5406 | /* |
2c47e605 YZ |
5407 | * drop a subvolume tree. |
5408 | * | |
5409 | * this function traverses the tree freeing any blocks that only | |
5410 | * referenced by the tree. | |
5411 | * | |
5412 | * when a shared tree block is found. this function decreases its | |
5413 | * reference count by one. if update_ref is true, this function | |
5414 | * also make sure backrefs for the shared block and all lower level | |
5415 | * blocks are properly updated. | |
9d1a2a3a DS |
5416 | * |
5417 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5418 | */ |
0078a9f9 | 5419 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5420 | { |
ab8d0fc4 | 5421 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5422 | struct btrfs_path *path; |
2c47e605 | 5423 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5424 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5425 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5426 | struct walk_control *wc; |
5427 | struct btrfs_key key; | |
5428 | int err = 0; | |
5429 | int ret; | |
5430 | int level; | |
d29a9f62 | 5431 | bool root_dropped = false; |
20524f02 | 5432 | |
4fd786e6 | 5433 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5434 | |
5caf2a00 | 5435 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5436 | if (!path) { |
5437 | err = -ENOMEM; | |
5438 | goto out; | |
5439 | } | |
20524f02 | 5440 | |
2c47e605 | 5441 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5442 | if (!wc) { |
5443 | btrfs_free_path(path); | |
cb1b69f4 TI |
5444 | err = -ENOMEM; |
5445 | goto out; | |
38a1a919 | 5446 | } |
2c47e605 | 5447 | |
f3e3d9cc QW |
5448 | /* |
5449 | * Use join to avoid potential EINTR from transaction start. See | |
5450 | * wait_reserve_ticket and the whole reservation callchain. | |
5451 | */ | |
5452 | if (for_reloc) | |
5453 | trans = btrfs_join_transaction(tree_root); | |
5454 | else | |
5455 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5456 | if (IS_ERR(trans)) { |
5457 | err = PTR_ERR(trans); | |
5458 | goto out_free; | |
5459 | } | |
98d5dc13 | 5460 | |
0568e82d JB |
5461 | err = btrfs_run_delayed_items(trans); |
5462 | if (err) | |
5463 | goto out_end_trans; | |
5464 | ||
83354f07 JB |
5465 | /* |
5466 | * This will help us catch people modifying the fs tree while we're | |
5467 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5468 | * dropped as we unlock the root node and parent nodes as we walk down | |
5469 | * the tree, assuming nothing will change. If something does change | |
5470 | * then we'll have stale information and drop references to blocks we've | |
5471 | * already dropped. | |
5472 | */ | |
5473 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5474 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5475 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5476 | path->nodes[level] = btrfs_lock_root_node(root); |
9f3a7427 | 5477 | path->slots[level] = 0; |
ac5887c8 | 5478 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5479 | memset(&wc->update_progress, 0, |
5480 | sizeof(wc->update_progress)); | |
9f3a7427 | 5481 | } else { |
9f3a7427 | 5482 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5483 | memcpy(&wc->update_progress, &key, |
5484 | sizeof(wc->update_progress)); | |
5485 | ||
c8422684 | 5486 | level = btrfs_root_drop_level(root_item); |
2c47e605 | 5487 | BUG_ON(level == 0); |
6702ed49 | 5488 | path->lowest_level = level; |
2c47e605 YZ |
5489 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5490 | path->lowest_level = 0; | |
5491 | if (ret < 0) { | |
5492 | err = ret; | |
79787eaa | 5493 | goto out_end_trans; |
9f3a7427 | 5494 | } |
1c4850e2 | 5495 | WARN_ON(ret > 0); |
2c47e605 | 5496 | |
7d9eb12c CM |
5497 | /* |
5498 | * unlock our path, this is safe because only this | |
5499 | * function is allowed to delete this snapshot | |
5500 | */ | |
5d4f98a2 | 5501 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5502 | |
5503 | level = btrfs_header_level(root->node); | |
5504 | while (1) { | |
5505 | btrfs_tree_lock(path->nodes[level]); | |
ac5887c8 | 5506 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5507 | |
2ff7e61e | 5508 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5509 | path->nodes[level]->start, |
3173a18f | 5510 | level, 1, &wc->refs[level], |
2c47e605 | 5511 | &wc->flags[level]); |
79787eaa JM |
5512 | if (ret < 0) { |
5513 | err = ret; | |
5514 | goto out_end_trans; | |
5515 | } | |
2c47e605 YZ |
5516 | BUG_ON(wc->refs[level] == 0); |
5517 | ||
c8422684 | 5518 | if (level == btrfs_root_drop_level(root_item)) |
2c47e605 YZ |
5519 | break; |
5520 | ||
5521 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5522 | path->locks[level] = 0; |
2c47e605 YZ |
5523 | WARN_ON(wc->refs[level] != 1); |
5524 | level--; | |
5525 | } | |
9f3a7427 | 5526 | } |
2c47e605 | 5527 | |
78c52d9e | 5528 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5529 | wc->level = level; |
5530 | wc->shared_level = -1; | |
5531 | wc->stage = DROP_REFERENCE; | |
5532 | wc->update_ref = update_ref; | |
5533 | wc->keep_locks = 0; | |
0b246afa | 5534 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5535 | |
d397712b | 5536 | while (1) { |
9d1a2a3a | 5537 | |
2c47e605 YZ |
5538 | ret = walk_down_tree(trans, root, path, wc); |
5539 | if (ret < 0) { | |
5540 | err = ret; | |
20524f02 | 5541 | break; |
2c47e605 | 5542 | } |
9aca1d51 | 5543 | |
2c47e605 YZ |
5544 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5545 | if (ret < 0) { | |
5546 | err = ret; | |
20524f02 | 5547 | break; |
2c47e605 YZ |
5548 | } |
5549 | ||
5550 | if (ret > 0) { | |
5551 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5552 | break; |
5553 | } | |
2c47e605 YZ |
5554 | |
5555 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5556 | wc->drop_level = wc->level; |
5557 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5558 | &wc->drop_progress, | |
5559 | path->slots[wc->drop_level]); | |
5560 | } | |
5561 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5562 | &wc->drop_progress); | |
c8422684 | 5563 | btrfs_set_root_drop_level(root_item, wc->drop_level); |
2c47e605 YZ |
5564 | |
5565 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5566 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5567 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5568 | ret = btrfs_update_root(trans, tree_root, |
5569 | &root->root_key, | |
5570 | root_item); | |
79787eaa | 5571 | if (ret) { |
66642832 | 5572 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5573 | err = ret; |
5574 | goto out_end_trans; | |
5575 | } | |
2c47e605 | 5576 | |
3a45bb20 | 5577 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5578 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5579 | btrfs_debug(fs_info, |
5580 | "drop snapshot early exit"); | |
3c8f2422 JB |
5581 | err = -EAGAIN; |
5582 | goto out_free; | |
5583 | } | |
5584 | ||
18d3bff4 JB |
5585 | /* |
5586 | * Use join to avoid potential EINTR from transaction | |
5587 | * start. See wait_reserve_ticket and the whole | |
5588 | * reservation callchain. | |
5589 | */ | |
5590 | if (for_reloc) | |
5591 | trans = btrfs_join_transaction(tree_root); | |
5592 | else | |
5593 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5594 | if (IS_ERR(trans)) { |
5595 | err = PTR_ERR(trans); | |
5596 | goto out_free; | |
5597 | } | |
c3e69d58 | 5598 | } |
20524f02 | 5599 | } |
b3b4aa74 | 5600 | btrfs_release_path(path); |
79787eaa JM |
5601 | if (err) |
5602 | goto out_end_trans; | |
2c47e605 | 5603 | |
ab9ce7d4 | 5604 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5605 | if (ret) { |
66642832 | 5606 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5607 | err = ret; |
79787eaa JM |
5608 | goto out_end_trans; |
5609 | } | |
2c47e605 | 5610 | |
76dda93c | 5611 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5612 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5613 | NULL, NULL); | |
79787eaa | 5614 | if (ret < 0) { |
66642832 | 5615 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5616 | err = ret; |
5617 | goto out_end_trans; | |
5618 | } else if (ret > 0) { | |
84cd948c JB |
5619 | /* if we fail to delete the orphan item this time |
5620 | * around, it'll get picked up the next time. | |
5621 | * | |
5622 | * The most common failure here is just -ENOENT. | |
5623 | */ | |
5624 | btrfs_del_orphan_item(trans, tree_root, | |
5625 | root->root_key.objectid); | |
76dda93c YZ |
5626 | } |
5627 | } | |
5628 | ||
a3cf0e43 QW |
5629 | /* |
5630 | * This subvolume is going to be completely dropped, and won't be | |
5631 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
5632 | * commit transaction time. So free it here manually. | |
5633 | */ | |
5634 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
5635 | btrfs_qgroup_free_meta_all_pertrans(root); | |
5636 | ||
8c38938c | 5637 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 5638 | btrfs_add_dropped_root(trans, root); |
8c38938c | 5639 | else |
00246528 | 5640 | btrfs_put_root(root); |
d29a9f62 | 5641 | root_dropped = true; |
79787eaa | 5642 | out_end_trans: |
3a45bb20 | 5643 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5644 | out_free: |
2c47e605 | 5645 | kfree(wc); |
5caf2a00 | 5646 | btrfs_free_path(path); |
cb1b69f4 | 5647 | out: |
d29a9f62 JB |
5648 | /* |
5649 | * So if we need to stop dropping the snapshot for whatever reason we | |
5650 | * need to make sure to add it back to the dead root list so that we | |
5651 | * keep trying to do the work later. This also cleans up roots if we | |
5652 | * don't have it in the radix (like when we recover after a power fail | |
5653 | * or unmount) so we don't leak memory. | |
5654 | */ | |
897ca819 | 5655 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5656 | btrfs_add_dead_root(root); |
2c536799 | 5657 | return err; |
20524f02 | 5658 | } |
9078a3e1 | 5659 | |
2c47e605 YZ |
5660 | /* |
5661 | * drop subtree rooted at tree block 'node'. | |
5662 | * | |
5663 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5664 | * only used by relocation code |
2c47e605 | 5665 | */ |
f82d02d9 YZ |
5666 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5667 | struct btrfs_root *root, | |
5668 | struct extent_buffer *node, | |
5669 | struct extent_buffer *parent) | |
5670 | { | |
0b246afa | 5671 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5672 | struct btrfs_path *path; |
2c47e605 | 5673 | struct walk_control *wc; |
f82d02d9 YZ |
5674 | int level; |
5675 | int parent_level; | |
5676 | int ret = 0; | |
5677 | int wret; | |
5678 | ||
2c47e605 YZ |
5679 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5680 | ||
f82d02d9 | 5681 | path = btrfs_alloc_path(); |
db5b493a TI |
5682 | if (!path) |
5683 | return -ENOMEM; | |
f82d02d9 | 5684 | |
2c47e605 | 5685 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5686 | if (!wc) { |
5687 | btrfs_free_path(path); | |
5688 | return -ENOMEM; | |
5689 | } | |
2c47e605 | 5690 | |
b9447ef8 | 5691 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5692 | parent_level = btrfs_header_level(parent); |
67439dad | 5693 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5694 | path->nodes[parent_level] = parent; |
5695 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5696 | ||
b9447ef8 | 5697 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5698 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5699 | path->nodes[level] = node; |
5700 | path->slots[level] = 0; | |
ac5887c8 | 5701 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5702 | |
5703 | wc->refs[parent_level] = 1; | |
5704 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5705 | wc->level = level; | |
5706 | wc->shared_level = -1; | |
5707 | wc->stage = DROP_REFERENCE; | |
5708 | wc->update_ref = 0; | |
5709 | wc->keep_locks = 1; | |
0b246afa | 5710 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5711 | |
5712 | while (1) { | |
2c47e605 YZ |
5713 | wret = walk_down_tree(trans, root, path, wc); |
5714 | if (wret < 0) { | |
f82d02d9 | 5715 | ret = wret; |
f82d02d9 | 5716 | break; |
2c47e605 | 5717 | } |
f82d02d9 | 5718 | |
2c47e605 | 5719 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5720 | if (wret < 0) |
5721 | ret = wret; | |
5722 | if (wret != 0) | |
5723 | break; | |
5724 | } | |
5725 | ||
2c47e605 | 5726 | kfree(wc); |
f82d02d9 YZ |
5727 | btrfs_free_path(path); |
5728 | return ret; | |
5729 | } | |
5730 | ||
6d07bcec MX |
5731 | /* |
5732 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5733 | * space_info. takes mirrors into account. |
6d07bcec | 5734 | */ |
633c0aad | 5735 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5736 | { |
32da5386 | 5737 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5738 | u64 free_bytes = 0; |
5739 | int factor; | |
5740 | ||
01327610 | 5741 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5742 | if (list_empty(&sinfo->ro_bgs)) |
5743 | return 0; | |
5744 | ||
5745 | spin_lock(&sinfo->lock); | |
5746 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5747 | spin_lock(&block_group->lock); |
5748 | ||
5749 | if (!block_group->ro) { | |
5750 | spin_unlock(&block_group->lock); | |
5751 | continue; | |
5752 | } | |
5753 | ||
46df06b8 | 5754 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5755 | free_bytes += (block_group->length - |
bf38be65 | 5756 | block_group->used) * factor; |
6d07bcec MX |
5757 | |
5758 | spin_unlock(&block_group->lock); | |
5759 | } | |
6d07bcec MX |
5760 | spin_unlock(&sinfo->lock); |
5761 | ||
5762 | return free_bytes; | |
5763 | } | |
5764 | ||
2ff7e61e JM |
5765 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5766 | u64 start, u64 end) | |
acce952b | 5767 | { |
2ff7e61e | 5768 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5769 | } |
5770 | ||
499f377f JM |
5771 | /* |
5772 | * It used to be that old block groups would be left around forever. | |
5773 | * Iterating over them would be enough to trim unused space. Since we | |
5774 | * now automatically remove them, we also need to iterate over unallocated | |
5775 | * space. | |
5776 | * | |
5777 | * We don't want a transaction for this since the discard may take a | |
5778 | * substantial amount of time. We don't require that a transaction be | |
5779 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5780 | * to ensure that we're not discarding chunks that were released or |
5781 | * allocated in the current transaction. | |
499f377f JM |
5782 | * |
5783 | * Holding the chunks lock will prevent other threads from allocating | |
5784 | * or releasing chunks, but it won't prevent a running transaction | |
5785 | * from committing and releasing the memory that the pending chunks | |
5786 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5787 | * transaction and hold the commit root sem. We only need to hold |
5788 | * it while performing the free space search since we have already | |
5789 | * held back allocations. | |
499f377f | 5790 | */ |
8103d10b | 5791 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5792 | { |
8103d10b | 5793 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5794 | int ret; |
5795 | ||
5796 | *trimmed = 0; | |
5797 | ||
0be88e36 JM |
5798 | /* Discard not supported = nothing to do. */ |
5799 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5800 | return 0; | |
5801 | ||
52042d8e | 5802 | /* Not writable = nothing to do. */ |
ebbede42 | 5803 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5804 | return 0; |
5805 | ||
5806 | /* No free space = nothing to do. */ | |
5807 | if (device->total_bytes <= device->bytes_used) | |
5808 | return 0; | |
5809 | ||
5810 | ret = 0; | |
5811 | ||
5812 | while (1) { | |
fb456252 | 5813 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5814 | u64 bytes; |
5815 | ||
5816 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5817 | if (ret) | |
fee7acc3 | 5818 | break; |
499f377f | 5819 | |
929be17a NB |
5820 | find_first_clear_extent_bit(&device->alloc_state, start, |
5821 | &start, &end, | |
5822 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 5823 | |
c57dd1f2 QW |
5824 | /* Check if there are any CHUNK_* bits left */ |
5825 | if (start > device->total_bytes) { | |
5826 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5827 | btrfs_warn_in_rcu(fs_info, | |
5828 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
5829 | start, end - start + 1, | |
5830 | rcu_str_deref(device->name), | |
5831 | device->total_bytes); | |
5832 | mutex_unlock(&fs_info->chunk_mutex); | |
5833 | ret = 0; | |
5834 | break; | |
5835 | } | |
5836 | ||
53460a45 NB |
5837 | /* Ensure we skip the reserved area in the first 1M */ |
5838 | start = max_t(u64, start, SZ_1M); | |
5839 | ||
929be17a NB |
5840 | /* |
5841 | * If find_first_clear_extent_bit find a range that spans the | |
5842 | * end of the device it will set end to -1, in this case it's up | |
5843 | * to the caller to trim the value to the size of the device. | |
5844 | */ | |
5845 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5846 | |
929be17a | 5847 | len = end - start + 1; |
499f377f | 5848 | |
929be17a NB |
5849 | /* We didn't find any extents */ |
5850 | if (!len) { | |
499f377f | 5851 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5852 | ret = 0; |
499f377f JM |
5853 | break; |
5854 | } | |
5855 | ||
929be17a NB |
5856 | ret = btrfs_issue_discard(device->bdev, start, len, |
5857 | &bytes); | |
5858 | if (!ret) | |
5859 | set_extent_bits(&device->alloc_state, start, | |
5860 | start + bytes - 1, | |
5861 | CHUNK_TRIMMED); | |
499f377f JM |
5862 | mutex_unlock(&fs_info->chunk_mutex); |
5863 | ||
5864 | if (ret) | |
5865 | break; | |
5866 | ||
5867 | start += len; | |
5868 | *trimmed += bytes; | |
5869 | ||
5870 | if (fatal_signal_pending(current)) { | |
5871 | ret = -ERESTARTSYS; | |
5872 | break; | |
5873 | } | |
5874 | ||
5875 | cond_resched(); | |
5876 | } | |
5877 | ||
5878 | return ret; | |
5879 | } | |
5880 | ||
93bba24d QW |
5881 | /* |
5882 | * Trim the whole filesystem by: | |
5883 | * 1) trimming the free space in each block group | |
5884 | * 2) trimming the unallocated space on each device | |
5885 | * | |
5886 | * This will also continue trimming even if a block group or device encounters | |
5887 | * an error. The return value will be the last error, or 0 if nothing bad | |
5888 | * happens. | |
5889 | */ | |
2ff7e61e | 5890 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5891 | { |
32da5386 | 5892 | struct btrfs_block_group *cache = NULL; |
499f377f JM |
5893 | struct btrfs_device *device; |
5894 | struct list_head *devices; | |
f7039b1d | 5895 | u64 group_trimmed; |
07301df7 | 5896 | u64 range_end = U64_MAX; |
f7039b1d LD |
5897 | u64 start; |
5898 | u64 end; | |
5899 | u64 trimmed = 0; | |
93bba24d QW |
5900 | u64 bg_failed = 0; |
5901 | u64 dev_failed = 0; | |
5902 | int bg_ret = 0; | |
5903 | int dev_ret = 0; | |
f7039b1d LD |
5904 | int ret = 0; |
5905 | ||
07301df7 QW |
5906 | /* |
5907 | * Check range overflow if range->len is set. | |
5908 | * The default range->len is U64_MAX. | |
5909 | */ | |
5910 | if (range->len != U64_MAX && | |
5911 | check_add_overflow(range->start, range->len, &range_end)) | |
5912 | return -EINVAL; | |
5913 | ||
6ba9fc8e | 5914 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5915 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 5916 | if (cache->start >= range_end) { |
f7039b1d LD |
5917 | btrfs_put_block_group(cache); |
5918 | break; | |
5919 | } | |
5920 | ||
b3470b5d DS |
5921 | start = max(range->start, cache->start); |
5922 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
5923 | |
5924 | if (end - start >= range->minlen) { | |
32da5386 | 5925 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 5926 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 5927 | if (ret) { |
93bba24d QW |
5928 | bg_failed++; |
5929 | bg_ret = ret; | |
5930 | continue; | |
1be41b78 | 5931 | } |
676f1f75 | 5932 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5933 | if (ret) { |
93bba24d QW |
5934 | bg_failed++; |
5935 | bg_ret = ret; | |
5936 | continue; | |
1be41b78 | 5937 | } |
f7039b1d LD |
5938 | } |
5939 | ret = btrfs_trim_block_group(cache, | |
5940 | &group_trimmed, | |
5941 | start, | |
5942 | end, | |
5943 | range->minlen); | |
5944 | ||
5945 | trimmed += group_trimmed; | |
5946 | if (ret) { | |
93bba24d QW |
5947 | bg_failed++; |
5948 | bg_ret = ret; | |
5949 | continue; | |
f7039b1d LD |
5950 | } |
5951 | } | |
f7039b1d LD |
5952 | } |
5953 | ||
93bba24d QW |
5954 | if (bg_failed) |
5955 | btrfs_warn(fs_info, | |
5956 | "failed to trim %llu block group(s), last error %d", | |
5957 | bg_failed, bg_ret); | |
0b246afa | 5958 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5959 | devices = &fs_info->fs_devices->devices; |
5960 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5961 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5962 | if (ret) { |
5963 | dev_failed++; | |
5964 | dev_ret = ret; | |
499f377f | 5965 | break; |
93bba24d | 5966 | } |
499f377f JM |
5967 | |
5968 | trimmed += group_trimmed; | |
5969 | } | |
0b246afa | 5970 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5971 | |
93bba24d QW |
5972 | if (dev_failed) |
5973 | btrfs_warn(fs_info, | |
5974 | "failed to trim %llu device(s), last error %d", | |
5975 | dev_failed, dev_ret); | |
f7039b1d | 5976 | range->len = trimmed; |
93bba24d QW |
5977 | if (bg_ret) |
5978 | return bg_ret; | |
5979 | return dev_ret; | |
f7039b1d | 5980 | } |