]>
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 | ||
6534c0c9 | 156 | ret = btrfs_search_slot(NULL, 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 | |
c940d6c2 | 1269 | static int do_discard_extent(struct btrfs_io_stripe *stripe, u64 *bytes) |
6143c23c NA |
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; | |
c940d6c2 | 1316 | struct btrfs_io_context *bioc = NULL; |
e244a0ae | 1317 | |
2999241d | 1318 | /* |
c940d6c2 | 1319 | * Avoid races with device replace and make sure our bioc has devices |
2999241d FM |
1320 | * associated to its stripes that don't go away while we are discarding. |
1321 | */ | |
0b246afa | 1322 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd | 1323 | while (cur < end) { |
c940d6c2 | 1324 | struct btrfs_io_stripe *stripe; |
5d4f98a2 YZ |
1325 | int i; |
1326 | ||
6b7faadd QW |
1327 | num_bytes = end - cur; |
1328 | /* Tell the block device(s) that the sectors can be discarded */ | |
1329 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
c940d6c2 | 1330 | &num_bytes, &bioc, 0); |
6b7faadd QW |
1331 | /* |
1332 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1333 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1334 | * thus we can't continue anyway. | |
1335 | */ | |
1336 | if (ret < 0) | |
1337 | goto out; | |
5d4f98a2 | 1338 | |
c940d6c2 QW |
1339 | stripe = bioc->stripes; |
1340 | for (i = 0; i < bioc->num_stripes; i++, stripe++) { | |
d04c6b88 | 1341 | u64 bytes; |
5e753a81 | 1342 | struct btrfs_device *device = stripe->dev; |
38b5f68e | 1343 | |
5e753a81 | 1344 | if (!device->bdev) { |
627e0873 FM |
1345 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); |
1346 | continue; | |
1347 | } | |
dcba6e48 | 1348 | |
5e753a81 AJ |
1349 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
1350 | continue; | |
1351 | ||
6143c23c | 1352 | ret = do_discard_extent(stripe, &bytes); |
6b7faadd | 1353 | if (!ret) { |
d04c6b88 | 1354 | discarded_bytes += bytes; |
6b7faadd QW |
1355 | } else if (ret != -EOPNOTSUPP) { |
1356 | /* | |
1357 | * Logic errors or -ENOMEM, or -EIO, but | |
1358 | * unlikely to happen. | |
1359 | * | |
1360 | * And since there are two loops, explicitly | |
1361 | * go to out to avoid confusion. | |
1362 | */ | |
c940d6c2 | 1363 | btrfs_put_bioc(bioc); |
6b7faadd QW |
1364 | goto out; |
1365 | } | |
d5e2003c JB |
1366 | |
1367 | /* | |
1368 | * Just in case we get back EOPNOTSUPP for some reason, | |
1369 | * just ignore the return value so we don't screw up | |
1370 | * people calling discard_extent. | |
1371 | */ | |
1372 | ret = 0; | |
5d4f98a2 | 1373 | } |
c940d6c2 | 1374 | btrfs_put_bioc(bioc); |
6b7faadd | 1375 | cur += num_bytes; |
5d4f98a2 | 1376 | } |
6b7faadd | 1377 | out: |
0b246afa | 1378 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1379 | |
1380 | if (actual_bytes) | |
1381 | *actual_bytes = discarded_bytes; | |
1382 | ||
5d4f98a2 | 1383 | |
53b381b3 DW |
1384 | if (ret == -EOPNOTSUPP) |
1385 | ret = 0; | |
5d4f98a2 | 1386 | return ret; |
5d4f98a2 YZ |
1387 | } |
1388 | ||
79787eaa | 1389 | /* Can return -ENOMEM */ |
5d4f98a2 | 1390 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1391 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1392 | { |
82fa113f | 1393 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 1394 | int ret; |
66d7e7f0 | 1395 | |
82fa113f QW |
1396 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1397 | generic_ref->action); | |
1398 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1399 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1400 | |
82fa113f | 1401 | if (generic_ref->type == BTRFS_REF_METADATA) |
2187374f | 1402 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL); |
82fa113f | 1403 | else |
2187374f | 1404 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0); |
d7eae340 | 1405 | |
82fa113f | 1406 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1407 | |
5d4f98a2 YZ |
1408 | return ret; |
1409 | } | |
1410 | ||
bd3c685e NB |
1411 | /* |
1412 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1413 | * | |
07cce5cf QW |
1414 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1415 | * how it works. | |
1416 | * | |
bd3c685e NB |
1417 | * @trans: Handle of transaction |
1418 | * | |
1419 | * @node: The delayed ref node used to get the bytenr/length for | |
1420 | * extent whose references are incremented. | |
1421 | * | |
1422 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1423 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1424 | * bytenr of the parent block. Since new extents are always | |
1425 | * created with indirect references, this will only be the case | |
1426 | * when relocating a shared extent. In that case, root_objectid | |
1a9fd417 | 1427 | * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must |
bd3c685e NB |
1428 | * be 0 |
1429 | * | |
1430 | * @root_objectid: The id of the root where this modification has originated, | |
1431 | * this can be either one of the well-known metadata trees or | |
1432 | * the subvolume id which references this extent. | |
1433 | * | |
1434 | * @owner: For data extents it is the inode number of the owning file. | |
1435 | * For metadata extents this parameter holds the level in the | |
1436 | * tree of the extent. | |
1437 | * | |
1438 | * @offset: For metadata extents the offset is ignored and is currently | |
1439 | * always passed as 0. For data extents it is the fileoffset | |
1440 | * this extent belongs to. | |
1441 | * | |
1442 | * @refs_to_add Number of references to add | |
1443 | * | |
1444 | * @extent_op Pointer to a structure, holding information necessary when | |
1445 | * updating a tree block's flags | |
1446 | * | |
1447 | */ | |
5d4f98a2 | 1448 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1449 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1450 | u64 parent, u64 root_objectid, |
1451 | u64 owner, u64 offset, int refs_to_add, | |
1452 | struct btrfs_delayed_extent_op *extent_op) | |
1453 | { | |
1454 | struct btrfs_path *path; | |
1455 | struct extent_buffer *leaf; | |
1456 | struct btrfs_extent_item *item; | |
fcebe456 | 1457 | struct btrfs_key key; |
c682f9b3 QW |
1458 | u64 bytenr = node->bytenr; |
1459 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1460 | u64 refs; |
1461 | int ret; | |
5d4f98a2 YZ |
1462 | |
1463 | path = btrfs_alloc_path(); | |
1464 | if (!path) | |
1465 | return -ENOMEM; | |
1466 | ||
5d4f98a2 | 1467 | /* this will setup the path even if it fails to insert the back ref */ |
a639cdeb NB |
1468 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1469 | parent, root_objectid, owner, | |
1470 | offset, refs_to_add, extent_op); | |
0ed4792a | 1471 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1472 | goto out; |
fcebe456 JB |
1473 | |
1474 | /* | |
1475 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1476 | * inline extent ref, so just update the reference count and add a | |
1477 | * normal backref. | |
1478 | */ | |
5d4f98a2 | 1479 | leaf = path->nodes[0]; |
fcebe456 | 1480 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1481 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1482 | refs = btrfs_extent_refs(leaf, item); | |
1483 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1484 | if (extent_op) | |
1485 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1486 | |
5d4f98a2 | 1487 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1488 | btrfs_release_path(path); |
56bec294 | 1489 | |
56bec294 | 1490 | /* now insert the actual backref */ |
65cd6d9e NB |
1491 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
1492 | BUG_ON(refs_to_add != 1); | |
1493 | ret = insert_tree_block_ref(trans, path, bytenr, parent, | |
1494 | root_objectid); | |
1495 | } else { | |
1496 | ret = insert_extent_data_ref(trans, path, bytenr, parent, | |
1497 | root_objectid, owner, offset, | |
1498 | refs_to_add); | |
1499 | } | |
79787eaa | 1500 | if (ret) |
66642832 | 1501 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1502 | out: |
56bec294 | 1503 | btrfs_free_path(path); |
30d133fc | 1504 | return ret; |
56bec294 CM |
1505 | } |
1506 | ||
5d4f98a2 | 1507 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1508 | struct btrfs_delayed_ref_node *node, |
1509 | struct btrfs_delayed_extent_op *extent_op, | |
1510 | int insert_reserved) | |
56bec294 | 1511 | { |
5d4f98a2 YZ |
1512 | int ret = 0; |
1513 | struct btrfs_delayed_data_ref *ref; | |
1514 | struct btrfs_key ins; | |
1515 | u64 parent = 0; | |
1516 | u64 ref_root = 0; | |
1517 | u64 flags = 0; | |
1518 | ||
1519 | ins.objectid = node->bytenr; | |
1520 | ins.offset = node->num_bytes; | |
1521 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1522 | ||
1523 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1524 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1525 | |
5d4f98a2 YZ |
1526 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1527 | parent = ref->parent; | |
fcebe456 | 1528 | ref_root = ref->root; |
5d4f98a2 YZ |
1529 | |
1530 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1531 | if (extent_op) |
5d4f98a2 | 1532 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1533 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1534 | flags, ref->objectid, | |
1535 | ref->offset, &ins, | |
1536 | node->ref_mod); | |
5d4f98a2 | 1537 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1538 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1539 | ref->objectid, ref->offset, | |
1540 | node->ref_mod, extent_op); | |
5d4f98a2 | 1541 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1542 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1543 | ref_root, ref->objectid, |
1544 | ref->offset, node->ref_mod, | |
c682f9b3 | 1545 | extent_op); |
5d4f98a2 YZ |
1546 | } else { |
1547 | BUG(); | |
1548 | } | |
1549 | return ret; | |
1550 | } | |
1551 | ||
1552 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1553 | struct extent_buffer *leaf, | |
1554 | struct btrfs_extent_item *ei) | |
1555 | { | |
1556 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1557 | if (extent_op->update_flags) { | |
1558 | flags |= extent_op->flags_to_set; | |
1559 | btrfs_set_extent_flags(leaf, ei, flags); | |
1560 | } | |
1561 | ||
1562 | if (extent_op->update_key) { | |
1563 | struct btrfs_tree_block_info *bi; | |
1564 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1565 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1566 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1567 | } | |
1568 | } | |
1569 | ||
1570 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1571 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1572 | struct btrfs_delayed_extent_op *extent_op) |
1573 | { | |
20b9a2d6 | 1574 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1575 | struct btrfs_key key; |
1576 | struct btrfs_path *path; | |
1577 | struct btrfs_extent_item *ei; | |
1578 | struct extent_buffer *leaf; | |
1579 | u32 item_size; | |
56bec294 | 1580 | int ret; |
5d4f98a2 | 1581 | int err = 0; |
b1c79e09 | 1582 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1583 | |
bf31f87f | 1584 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1585 | return 0; |
1586 | ||
0b246afa | 1587 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1588 | metadata = 0; |
1589 | ||
5d4f98a2 YZ |
1590 | path = btrfs_alloc_path(); |
1591 | if (!path) | |
1592 | return -ENOMEM; | |
1593 | ||
d278850e | 1594 | key.objectid = head->bytenr; |
5d4f98a2 | 1595 | |
3173a18f | 1596 | if (metadata) { |
3173a18f | 1597 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1598 | key.offset = extent_op->level; |
3173a18f JB |
1599 | } else { |
1600 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1601 | key.offset = head->num_bytes; |
3173a18f JB |
1602 | } |
1603 | ||
1604 | again: | |
0b246afa | 1605 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1606 | if (ret < 0) { |
1607 | err = ret; | |
1608 | goto out; | |
1609 | } | |
1610 | if (ret > 0) { | |
3173a18f | 1611 | if (metadata) { |
55994887 FDBM |
1612 | if (path->slots[0] > 0) { |
1613 | path->slots[0]--; | |
1614 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1615 | path->slots[0]); | |
d278850e | 1616 | if (key.objectid == head->bytenr && |
55994887 | 1617 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1618 | key.offset == head->num_bytes) |
55994887 FDBM |
1619 | ret = 0; |
1620 | } | |
1621 | if (ret > 0) { | |
1622 | btrfs_release_path(path); | |
1623 | metadata = 0; | |
3173a18f | 1624 | |
d278850e JB |
1625 | key.objectid = head->bytenr; |
1626 | key.offset = head->num_bytes; | |
55994887 FDBM |
1627 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1628 | goto again; | |
1629 | } | |
1630 | } else { | |
1631 | err = -EIO; | |
1632 | goto out; | |
3173a18f | 1633 | } |
5d4f98a2 YZ |
1634 | } |
1635 | ||
1636 | leaf = path->nodes[0]; | |
1637 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1638 | |
6d8ff4e4 | 1639 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1640 | err = -EINVAL; |
1641 | btrfs_print_v0_err(fs_info); | |
1642 | btrfs_abort_transaction(trans, err); | |
1643 | goto out; | |
1644 | } | |
1645 | ||
5d4f98a2 YZ |
1646 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1647 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1648 | |
5d4f98a2 YZ |
1649 | btrfs_mark_buffer_dirty(leaf); |
1650 | out: | |
1651 | btrfs_free_path(path); | |
1652 | return err; | |
56bec294 CM |
1653 | } |
1654 | ||
5d4f98a2 | 1655 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1656 | struct btrfs_delayed_ref_node *node, |
1657 | struct btrfs_delayed_extent_op *extent_op, | |
1658 | int insert_reserved) | |
56bec294 CM |
1659 | { |
1660 | int ret = 0; | |
5d4f98a2 | 1661 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1662 | u64 parent = 0; |
1663 | u64 ref_root = 0; | |
56bec294 | 1664 | |
5d4f98a2 | 1665 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1666 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1667 | |
5d4f98a2 YZ |
1668 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1669 | parent = ref->parent; | |
fcebe456 | 1670 | ref_root = ref->root; |
5d4f98a2 | 1671 | |
02794222 | 1672 | if (node->ref_mod != 1) { |
f97806f2 | 1673 | btrfs_err(trans->fs_info, |
02794222 LB |
1674 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1675 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1676 | parent); | |
1677 | return -EIO; | |
1678 | } | |
5d4f98a2 | 1679 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1680 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1681 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1682 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1683 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1684 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1685 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1686 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1687 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1688 | } else { |
1689 | BUG(); | |
1690 | } | |
56bec294 CM |
1691 | return ret; |
1692 | } | |
1693 | ||
1694 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1695 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1696 | struct btrfs_delayed_ref_node *node, |
1697 | struct btrfs_delayed_extent_op *extent_op, | |
1698 | int insert_reserved) | |
56bec294 | 1699 | { |
79787eaa JM |
1700 | int ret = 0; |
1701 | ||
bf31f87f | 1702 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1703 | if (insert_reserved) |
b25c36f8 | 1704 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1705 | return 0; |
857cc2fc | 1706 | } |
79787eaa | 1707 | |
5d4f98a2 YZ |
1708 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1709 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1710 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1711 | insert_reserved); |
1712 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1713 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1714 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1715 | insert_reserved); |
1716 | else | |
1717 | BUG(); | |
80ee54bf | 1718 | if (ret && insert_reserved) |
b25c36f8 | 1719 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
5d4f98a2 | 1720 | return ret; |
56bec294 CM |
1721 | } |
1722 | ||
c6fc2454 | 1723 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1724 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1725 | { | |
cffc3374 FM |
1726 | struct btrfs_delayed_ref_node *ref; |
1727 | ||
e3d03965 | 1728 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1729 | return NULL; |
d7df2c79 | 1730 | |
cffc3374 FM |
1731 | /* |
1732 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1733 | * This is to prevent a ref count from going down to zero, which deletes | |
1734 | * the extent item from the extent tree, when there still are references | |
1735 | * to add, which would fail because they would not find the extent item. | |
1736 | */ | |
1d57ee94 WX |
1737 | if (!list_empty(&head->ref_add_list)) |
1738 | return list_first_entry(&head->ref_add_list, | |
1739 | struct btrfs_delayed_ref_node, add_list); | |
1740 | ||
e3d03965 | 1741 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1742 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1743 | ASSERT(list_empty(&ref->add_list)); |
1744 | return ref; | |
56bec294 CM |
1745 | } |
1746 | ||
2eadaa22 JB |
1747 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1748 | struct btrfs_delayed_ref_head *head) | |
1749 | { | |
1750 | spin_lock(&delayed_refs->lock); | |
1751 | head->processing = 0; | |
1752 | delayed_refs->num_heads_ready++; | |
1753 | spin_unlock(&delayed_refs->lock); | |
1754 | btrfs_delayed_ref_unlock(head); | |
1755 | } | |
1756 | ||
bedc6617 JB |
1757 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1758 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1759 | { |
1760 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1761 | |
1762 | if (!extent_op) | |
bedc6617 JB |
1763 | return NULL; |
1764 | ||
b00e6250 | 1765 | if (head->must_insert_reserved) { |
bedc6617 | 1766 | head->extent_op = NULL; |
b00e6250 | 1767 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1768 | return NULL; |
b00e6250 | 1769 | } |
bedc6617 JB |
1770 | return extent_op; |
1771 | } | |
1772 | ||
1773 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1774 | struct btrfs_delayed_ref_head *head) | |
1775 | { | |
1776 | struct btrfs_delayed_extent_op *extent_op; | |
1777 | int ret; | |
1778 | ||
1779 | extent_op = cleanup_extent_op(head); | |
1780 | if (!extent_op) | |
1781 | return 0; | |
1782 | head->extent_op = NULL; | |
b00e6250 | 1783 | spin_unlock(&head->lock); |
20b9a2d6 | 1784 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1785 | btrfs_free_delayed_extent_op(extent_op); |
1786 | return ret ? ret : 1; | |
1787 | } | |
1788 | ||
31890da0 JB |
1789 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1790 | struct btrfs_delayed_ref_root *delayed_refs, | |
1791 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1792 | { |
ba2c4d4e | 1793 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 | 1794 | |
81e75ac7 JB |
1795 | /* |
1796 | * We had csum deletions accounted for in our delayed refs rsv, we need | |
1797 | * to drop the csum leaves for this update from our delayed_refs_rsv. | |
1798 | */ | |
1799 | if (head->total_ref_mod < 0 && head->is_data) { | |
1800 | spin_lock(&delayed_refs->lock); | |
1801 | delayed_refs->pending_csums -= head->num_bytes; | |
1802 | spin_unlock(&delayed_refs->lock); | |
1803 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes); | |
1804 | } | |
1805 | ||
ba2c4d4e | 1806 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1807 | } |
1808 | ||
194ab0bc | 1809 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1810 | struct btrfs_delayed_ref_head *head) |
1811 | { | |
f9871edd NB |
1812 | |
1813 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1814 | struct btrfs_delayed_ref_root *delayed_refs; |
1815 | int ret; | |
1816 | ||
1817 | delayed_refs = &trans->transaction->delayed_refs; | |
1818 | ||
bedc6617 | 1819 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1820 | if (ret < 0) { |
1821 | unselect_delayed_ref_head(delayed_refs, head); | |
1822 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1823 | return ret; | |
1824 | } else if (ret) { | |
1825 | return ret; | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1830 | * and then re-check to make sure nobody got added. | |
1831 | */ | |
1832 | spin_unlock(&head->lock); | |
1833 | spin_lock(&delayed_refs->lock); | |
1834 | spin_lock(&head->lock); | |
e3d03965 | 1835 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1836 | spin_unlock(&head->lock); |
1837 | spin_unlock(&delayed_refs->lock); | |
1838 | return 1; | |
1839 | } | |
d7baffda | 1840 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1841 | spin_unlock(&head->lock); |
1e7a1421 | 1842 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1843 | |
c1103f7a | 1844 | if (head->must_insert_reserved) { |
b25c36f8 | 1845 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1846 | if (head->is_data) { |
40e046ac FM |
1847 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1848 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1849 | } |
1850 | } | |
1851 | ||
31890da0 | 1852 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1853 | |
1854 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1855 | btrfs_delayed_ref_unlock(head); |
d278850e | 1856 | btrfs_put_delayed_ref_head(head); |
856bd270 | 1857 | return ret; |
194ab0bc JB |
1858 | } |
1859 | ||
b1cdbcb5 NB |
1860 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1861 | struct btrfs_trans_handle *trans) | |
1862 | { | |
1863 | struct btrfs_delayed_ref_root *delayed_refs = | |
1864 | &trans->transaction->delayed_refs; | |
1865 | struct btrfs_delayed_ref_head *head = NULL; | |
1866 | int ret; | |
1867 | ||
1868 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1869 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1870 | if (!head) { |
1871 | spin_unlock(&delayed_refs->lock); | |
1872 | return head; | |
1873 | } | |
1874 | ||
1875 | /* | |
1876 | * Grab the lock that says we are going to process all the refs for | |
1877 | * this head | |
1878 | */ | |
9e920a6f | 1879 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1880 | spin_unlock(&delayed_refs->lock); |
1881 | ||
1882 | /* | |
1883 | * We may have dropped the spin lock to get the head mutex lock, and | |
1884 | * that might have given someone else time to free the head. If that's | |
1885 | * true, it has been removed from our list and we can move on. | |
1886 | */ | |
1887 | if (ret == -EAGAIN) | |
1888 | head = ERR_PTR(-EAGAIN); | |
1889 | ||
1890 | return head; | |
1891 | } | |
1892 | ||
e7261386 NB |
1893 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1894 | struct btrfs_delayed_ref_head *locked_ref, | |
1895 | unsigned long *run_refs) | |
1896 | { | |
1897 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1898 | struct btrfs_delayed_ref_root *delayed_refs; | |
1899 | struct btrfs_delayed_extent_op *extent_op; | |
1900 | struct btrfs_delayed_ref_node *ref; | |
1901 | int must_insert_reserved = 0; | |
1902 | int ret; | |
1903 | ||
1904 | delayed_refs = &trans->transaction->delayed_refs; | |
1905 | ||
0110a4c4 NB |
1906 | lockdep_assert_held(&locked_ref->mutex); |
1907 | lockdep_assert_held(&locked_ref->lock); | |
1908 | ||
e7261386 NB |
1909 | while ((ref = select_delayed_ref(locked_ref))) { |
1910 | if (ref->seq && | |
1911 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1912 | spin_unlock(&locked_ref->lock); | |
1913 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1914 | return -EAGAIN; | |
1915 | } | |
1916 | ||
1917 | (*run_refs)++; | |
1918 | ref->in_tree = 0; | |
1919 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1920 | RB_CLEAR_NODE(&ref->ref_node); | |
1921 | if (!list_empty(&ref->add_list)) | |
1922 | list_del(&ref->add_list); | |
1923 | /* | |
1924 | * When we play the delayed ref, also correct the ref_mod on | |
1925 | * head | |
1926 | */ | |
1927 | switch (ref->action) { | |
1928 | case BTRFS_ADD_DELAYED_REF: | |
1929 | case BTRFS_ADD_DELAYED_EXTENT: | |
1930 | locked_ref->ref_mod -= ref->ref_mod; | |
1931 | break; | |
1932 | case BTRFS_DROP_DELAYED_REF: | |
1933 | locked_ref->ref_mod += ref->ref_mod; | |
1934 | break; | |
1935 | default: | |
1936 | WARN_ON(1); | |
1937 | } | |
1938 | atomic_dec(&delayed_refs->num_entries); | |
1939 | ||
1940 | /* | |
1941 | * Record the must_insert_reserved flag before we drop the | |
1942 | * spin lock. | |
1943 | */ | |
1944 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1945 | locked_ref->must_insert_reserved = 0; | |
1946 | ||
1947 | extent_op = locked_ref->extent_op; | |
1948 | locked_ref->extent_op = NULL; | |
1949 | spin_unlock(&locked_ref->lock); | |
1950 | ||
1951 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1952 | must_insert_reserved); | |
1953 | ||
1954 | btrfs_free_delayed_extent_op(extent_op); | |
1955 | if (ret) { | |
1956 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1957 | btrfs_put_delayed_ref(ref); | |
1958 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1959 | ret); | |
1960 | return ret; | |
1961 | } | |
1962 | ||
1963 | btrfs_put_delayed_ref(ref); | |
1964 | cond_resched(); | |
1965 | ||
1966 | spin_lock(&locked_ref->lock); | |
1967 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1968 | } | |
1969 | ||
1970 | return 0; | |
1971 | } | |
1972 | ||
79787eaa JM |
1973 | /* |
1974 | * Returns 0 on success or if called with an already aborted transaction. | |
1975 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1976 | */ | |
d7df2c79 | 1977 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 1978 | unsigned long nr) |
56bec294 | 1979 | { |
0a1e458a | 1980 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 1981 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 1982 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 1983 | ktime_t start = ktime_get(); |
56bec294 | 1984 | int ret; |
d7df2c79 | 1985 | unsigned long count = 0; |
0a2b2a84 | 1986 | unsigned long actual_count = 0; |
56bec294 CM |
1987 | |
1988 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 1989 | do { |
56bec294 | 1990 | if (!locked_ref) { |
b1cdbcb5 | 1991 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
1992 | if (IS_ERR_OR_NULL(locked_ref)) { |
1993 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
1994 | continue; | |
1995 | } else { | |
1996 | break; | |
1997 | } | |
56bec294 | 1998 | } |
0110a4c4 | 1999 | count++; |
56bec294 | 2000 | } |
2c3cf7d5 FM |
2001 | /* |
2002 | * We need to try and merge add/drops of the same ref since we | |
2003 | * can run into issues with relocate dropping the implicit ref | |
2004 | * and then it being added back again before the drop can | |
2005 | * finish. If we merged anything we need to re-loop so we can | |
2006 | * get a good ref. | |
2007 | * Or we can get node references of the same type that weren't | |
2008 | * merged when created due to bumps in the tree mod seq, and | |
2009 | * we need to merge them to prevent adding an inline extent | |
2010 | * backref before dropping it (triggering a BUG_ON at | |
2011 | * insert_inline_extent_backref()). | |
2012 | */ | |
d7df2c79 | 2013 | spin_lock(&locked_ref->lock); |
be97f133 | 2014 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2015 | |
0110a4c4 NB |
2016 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2017 | &actual_count); | |
2018 | if (ret < 0 && ret != -EAGAIN) { | |
2019 | /* | |
2020 | * Error, btrfs_run_delayed_refs_for_head already | |
2021 | * unlocked everything so just bail out | |
2022 | */ | |
2023 | return ret; | |
2024 | } else if (!ret) { | |
2025 | /* | |
2026 | * Success, perform the usual cleanup of a processed | |
2027 | * head | |
2028 | */ | |
f9871edd | 2029 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2030 | if (ret > 0 ) { |
b00e6250 JB |
2031 | /* We dropped our lock, we need to loop. */ |
2032 | ret = 0; | |
d7df2c79 | 2033 | continue; |
194ab0bc JB |
2034 | } else if (ret) { |
2035 | return ret; | |
5d4f98a2 | 2036 | } |
22cd2e7d | 2037 | } |
1ce7a5ec | 2038 | |
b00e6250 | 2039 | /* |
0110a4c4 NB |
2040 | * Either success case or btrfs_run_delayed_refs_for_head |
2041 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2042 | */ |
b00e6250 | 2043 | |
0110a4c4 | 2044 | locked_ref = NULL; |
c3e69d58 | 2045 | cond_resched(); |
0110a4c4 | 2046 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2047 | |
2048 | /* | |
2049 | * We don't want to include ref heads since we can have empty ref heads | |
2050 | * and those will drastically skew our runtime down since we just do | |
2051 | * accounting, no actual extent tree updates. | |
2052 | */ | |
2053 | if (actual_count > 0) { | |
2054 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2055 | u64 avg; | |
2056 | ||
2057 | /* | |
2058 | * We weigh the current average higher than our current runtime | |
2059 | * to avoid large swings in the average. | |
2060 | */ | |
2061 | spin_lock(&delayed_refs->lock); | |
2062 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2063 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2064 | spin_unlock(&delayed_refs->lock); |
2065 | } | |
d7df2c79 | 2066 | return 0; |
c3e69d58 CM |
2067 | } |
2068 | ||
709c0486 AJ |
2069 | #ifdef SCRAMBLE_DELAYED_REFS |
2070 | /* | |
2071 | * Normally delayed refs get processed in ascending bytenr order. This | |
2072 | * correlates in most cases to the order added. To expose dependencies on this | |
2073 | * order, we start to process the tree in the middle instead of the beginning | |
2074 | */ | |
2075 | static u64 find_middle(struct rb_root *root) | |
2076 | { | |
2077 | struct rb_node *n = root->rb_node; | |
2078 | struct btrfs_delayed_ref_node *entry; | |
2079 | int alt = 1; | |
2080 | u64 middle; | |
2081 | u64 first = 0, last = 0; | |
2082 | ||
2083 | n = rb_first(root); | |
2084 | if (n) { | |
2085 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2086 | first = entry->bytenr; | |
2087 | } | |
2088 | n = rb_last(root); | |
2089 | if (n) { | |
2090 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2091 | last = entry->bytenr; | |
2092 | } | |
2093 | n = root->rb_node; | |
2094 | ||
2095 | while (n) { | |
2096 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2097 | WARN_ON(!entry->in_tree); | |
2098 | ||
2099 | middle = entry->bytenr; | |
2100 | ||
2101 | if (alt) | |
2102 | n = n->rb_left; | |
2103 | else | |
2104 | n = n->rb_right; | |
2105 | ||
2106 | alt = 1 - alt; | |
2107 | } | |
2108 | return middle; | |
2109 | } | |
2110 | #endif | |
2111 | ||
c3e69d58 CM |
2112 | /* |
2113 | * this starts processing the delayed reference count updates and | |
2114 | * extent insertions we have queued up so far. count can be | |
2115 | * 0, which means to process everything in the tree at the start | |
2116 | * of the run (but not newly added entries), or it can be some target | |
2117 | * number you'd like to process. | |
79787eaa JM |
2118 | * |
2119 | * Returns 0 on success or if called with an aborted transaction | |
2120 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2121 | */ |
2122 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2123 | unsigned long count) |
c3e69d58 | 2124 | { |
c79a70b1 | 2125 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2126 | struct rb_node *node; |
2127 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2128 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2129 | int ret; |
2130 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2131 | |
79787eaa | 2132 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2133 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2134 | return 0; |
2135 | ||
0b246afa | 2136 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2137 | return 0; |
2138 | ||
c3e69d58 | 2139 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2140 | if (count == 0) |
61a56a99 | 2141 | count = delayed_refs->num_heads_ready; |
bb721703 | 2142 | |
c3e69d58 | 2143 | again: |
709c0486 AJ |
2144 | #ifdef SCRAMBLE_DELAYED_REFS |
2145 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2146 | #endif | |
0a1e458a | 2147 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2148 | if (ret < 0) { |
66642832 | 2149 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2150 | return ret; |
eb099670 | 2151 | } |
c3e69d58 | 2152 | |
56bec294 | 2153 | if (run_all) { |
119e80df | 2154 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2155 | |
d7df2c79 | 2156 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2157 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2158 | if (!node) { |
2159 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2160 | goto out; |
d7df2c79 | 2161 | } |
d278850e JB |
2162 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2163 | href_node); | |
2164 | refcount_inc(&head->refs); | |
2165 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2166 | |
d278850e JB |
2167 | /* Mutex was contended, block until it's released and retry. */ |
2168 | mutex_lock(&head->mutex); | |
2169 | mutex_unlock(&head->mutex); | |
56bec294 | 2170 | |
d278850e | 2171 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2172 | cond_resched(); |
56bec294 | 2173 | goto again; |
5f39d397 | 2174 | } |
54aa1f4d | 2175 | out: |
a28ec197 CM |
2176 | return 0; |
2177 | } | |
2178 | ||
5d4f98a2 | 2179 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
42c9d0b5 | 2180 | struct extent_buffer *eb, u64 flags, |
b1c79e09 | 2181 | int level, int is_data) |
5d4f98a2 YZ |
2182 | { |
2183 | struct btrfs_delayed_extent_op *extent_op; | |
2184 | int ret; | |
2185 | ||
78a6184a | 2186 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2187 | if (!extent_op) |
2188 | return -ENOMEM; | |
2189 | ||
2190 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2191 | extent_op->update_flags = true; |
2192 | extent_op->update_key = false; | |
2193 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2194 | extent_op->level = level; |
5d4f98a2 | 2195 | |
42c9d0b5 | 2196 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2197 | if (ret) |
78a6184a | 2198 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2199 | return ret; |
2200 | } | |
2201 | ||
e4c3b2dc | 2202 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2203 | struct btrfs_path *path, |
2204 | u64 objectid, u64 offset, u64 bytenr) | |
2205 | { | |
2206 | struct btrfs_delayed_ref_head *head; | |
2207 | struct btrfs_delayed_ref_node *ref; | |
2208 | struct btrfs_delayed_data_ref *data_ref; | |
2209 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2210 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2211 | struct rb_node *node; |
5d4f98a2 YZ |
2212 | int ret = 0; |
2213 | ||
998ac6d2 | 2214 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2215 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2216 | if (cur_trans) |
2217 | refcount_inc(&cur_trans->use_count); | |
2218 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2219 | if (!cur_trans) |
2220 | return 0; | |
2221 | ||
2222 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2223 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2224 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2225 | if (!head) { |
2226 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2227 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2228 | return 0; |
2229 | } | |
5d4f98a2 YZ |
2230 | |
2231 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2232 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2233 | spin_unlock(&delayed_refs->lock); |
2234 | ||
b3b4aa74 | 2235 | btrfs_release_path(path); |
5d4f98a2 | 2236 | |
8cc33e5c DS |
2237 | /* |
2238 | * Mutex was contended, block until it's released and let | |
2239 | * caller try again | |
2240 | */ | |
5d4f98a2 YZ |
2241 | mutex_lock(&head->mutex); |
2242 | mutex_unlock(&head->mutex); | |
d278850e | 2243 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2244 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2245 | return -EAGAIN; |
2246 | } | |
d7df2c79 | 2247 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2248 | |
d7df2c79 | 2249 | spin_lock(&head->lock); |
0e0adbcf JB |
2250 | /* |
2251 | * XXX: We should replace this with a proper search function in the | |
2252 | * future. | |
2253 | */ | |
e3d03965 LB |
2254 | for (node = rb_first_cached(&head->ref_tree); node; |
2255 | node = rb_next(node)) { | |
0e0adbcf | 2256 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2257 | /* If it's a shared ref we know a cross reference exists */ |
2258 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2259 | ret = 1; | |
2260 | break; | |
2261 | } | |
5d4f98a2 | 2262 | |
d7df2c79 | 2263 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2264 | |
d7df2c79 JB |
2265 | /* |
2266 | * If our ref doesn't match the one we're currently looking at | |
2267 | * then we have a cross reference. | |
2268 | */ | |
2269 | if (data_ref->root != root->root_key.objectid || | |
2270 | data_ref->objectid != objectid || | |
2271 | data_ref->offset != offset) { | |
2272 | ret = 1; | |
2273 | break; | |
2274 | } | |
5d4f98a2 | 2275 | } |
d7df2c79 | 2276 | spin_unlock(&head->lock); |
5d4f98a2 | 2277 | mutex_unlock(&head->mutex); |
998ac6d2 | 2278 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2279 | return ret; |
2280 | } | |
2281 | ||
e4c3b2dc | 2282 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2283 | struct btrfs_path *path, |
a84d5d42 BB |
2284 | u64 objectid, u64 offset, u64 bytenr, |
2285 | bool strict) | |
be20aa9d | 2286 | { |
0b246afa JM |
2287 | struct btrfs_fs_info *fs_info = root->fs_info; |
2288 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2289 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2290 | struct btrfs_extent_data_ref *ref; |
2291 | struct btrfs_extent_inline_ref *iref; | |
2292 | struct btrfs_extent_item *ei; | |
f321e491 | 2293 | struct btrfs_key key; |
5d4f98a2 | 2294 | u32 item_size; |
3de28d57 | 2295 | int type; |
be20aa9d | 2296 | int ret; |
925baedd | 2297 | |
be20aa9d | 2298 | key.objectid = bytenr; |
31840ae1 | 2299 | key.offset = (u64)-1; |
f321e491 | 2300 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2301 | |
be20aa9d CM |
2302 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2303 | if (ret < 0) | |
2304 | goto out; | |
79787eaa | 2305 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2306 | |
2307 | ret = -ENOENT; | |
2308 | if (path->slots[0] == 0) | |
31840ae1 | 2309 | goto out; |
be20aa9d | 2310 | |
31840ae1 | 2311 | path->slots[0]--; |
f321e491 | 2312 | leaf = path->nodes[0]; |
5d4f98a2 | 2313 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2314 | |
5d4f98a2 | 2315 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2316 | goto out; |
f321e491 | 2317 | |
5d4f98a2 YZ |
2318 | ret = 1; |
2319 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2320 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2321 | |
a6bd9cd1 | 2322 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2323 | if (item_size != sizeof(*ei) + |
2324 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2325 | goto out; | |
be20aa9d | 2326 | |
a84d5d42 BB |
2327 | /* |
2328 | * If extent created before last snapshot => it's shared unless the | |
2329 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2330 | */ | |
2331 | if (!strict && | |
2332 | (btrfs_extent_generation(leaf, ei) <= | |
2333 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2334 | goto out; |
2335 | ||
2336 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2337 | |
a6bd9cd1 | 2338 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2339 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2340 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2341 | goto out; |
2342 | ||
2343 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2344 | if (btrfs_extent_refs(leaf, ei) != | |
2345 | btrfs_extent_data_ref_count(leaf, ref) || | |
2346 | btrfs_extent_data_ref_root(leaf, ref) != | |
2347 | root->root_key.objectid || | |
2348 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2349 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2350 | goto out; | |
2351 | ||
2352 | ret = 0; | |
2353 | out: | |
2354 | return ret; | |
2355 | } | |
2356 | ||
e4c3b2dc | 2357 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
a84d5d42 | 2358 | u64 bytenr, bool strict) |
5d4f98a2 YZ |
2359 | { |
2360 | struct btrfs_path *path; | |
2361 | int ret; | |
5d4f98a2 YZ |
2362 | |
2363 | path = btrfs_alloc_path(); | |
2364 | if (!path) | |
9132c4ff | 2365 | return -ENOMEM; |
5d4f98a2 YZ |
2366 | |
2367 | do { | |
e4c3b2dc | 2368 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2369 | offset, bytenr, strict); |
5d4f98a2 | 2370 | if (ret && ret != -ENOENT) |
f321e491 | 2371 | goto out; |
80ff3856 | 2372 | |
380fd066 MT |
2373 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2374 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2375 | |
be20aa9d | 2376 | out: |
80ff3856 | 2377 | btrfs_free_path(path); |
b8f762a2 | 2378 | if (btrfs_is_data_reloc_root(root)) |
f0486c68 | 2379 | WARN_ON(ret > 0); |
f321e491 | 2380 | return ret; |
be20aa9d | 2381 | } |
c5739bba | 2382 | |
5d4f98a2 | 2383 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2384 | struct btrfs_root *root, |
5d4f98a2 | 2385 | struct extent_buffer *buf, |
e339a6b0 | 2386 | int full_backref, int inc) |
31840ae1 | 2387 | { |
0b246afa | 2388 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2389 | u64 bytenr; |
5d4f98a2 YZ |
2390 | u64 num_bytes; |
2391 | u64 parent; | |
31840ae1 | 2392 | u64 ref_root; |
31840ae1 | 2393 | u32 nritems; |
31840ae1 ZY |
2394 | struct btrfs_key key; |
2395 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2396 | struct btrfs_ref generic_ref = { 0 }; |
2397 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2398 | int i; |
82fa113f | 2399 | int action; |
31840ae1 ZY |
2400 | int level; |
2401 | int ret = 0; | |
fccb84c9 | 2402 | |
0b246afa | 2403 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2404 | return 0; |
fccb84c9 | 2405 | |
31840ae1 | 2406 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2407 | nritems = btrfs_header_nritems(buf); |
2408 | level = btrfs_header_level(buf); | |
2409 | ||
92a7cc42 | 2410 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2411 | return 0; |
31840ae1 | 2412 | |
5d4f98a2 YZ |
2413 | if (full_backref) |
2414 | parent = buf->start; | |
2415 | else | |
2416 | parent = 0; | |
82fa113f QW |
2417 | if (inc) |
2418 | action = BTRFS_ADD_DELAYED_REF; | |
2419 | else | |
2420 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2421 | |
2422 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2423 | if (level == 0) { |
5d4f98a2 | 2424 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2425 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2426 | continue; |
5d4f98a2 | 2427 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2428 | struct btrfs_file_extent_item); |
2429 | if (btrfs_file_extent_type(buf, fi) == | |
2430 | BTRFS_FILE_EXTENT_INLINE) | |
2431 | continue; | |
2432 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2433 | if (bytenr == 0) | |
2434 | continue; | |
5d4f98a2 YZ |
2435 | |
2436 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2437 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2438 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2439 | num_bytes, parent); | |
2440 | generic_ref.real_root = root->root_key.objectid; | |
2441 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
4158d286 NB |
2442 | key.offset, root->root_key.objectid, |
2443 | for_reloc); | |
82fa113f | 2444 | generic_ref.skip_qgroup = for_reloc; |
dd28b6a5 | 2445 | if (inc) |
82fa113f | 2446 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2447 | else |
ffd4bb2a | 2448 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2449 | if (ret) |
2450 | goto fail; | |
2451 | } else { | |
5d4f98a2 | 2452 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2453 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2454 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2455 | num_bytes, parent); | |
2456 | generic_ref.real_root = root->root_key.objectid; | |
4158d286 NB |
2457 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root, |
2458 | root->root_key.objectid, for_reloc); | |
82fa113f | 2459 | generic_ref.skip_qgroup = for_reloc; |
dd28b6a5 | 2460 | if (inc) |
82fa113f | 2461 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2462 | else |
ffd4bb2a | 2463 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2464 | if (ret) |
2465 | goto fail; | |
2466 | } | |
2467 | } | |
2468 | return 0; | |
2469 | fail: | |
5d4f98a2 YZ |
2470 | return ret; |
2471 | } | |
2472 | ||
2473 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2474 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2475 | { |
e339a6b0 | 2476 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2477 | } |
2478 | ||
2479 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2480 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2481 | { |
e339a6b0 | 2482 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2483 | } |
2484 | ||
1b86826d | 2485 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2486 | { |
0b246afa | 2487 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2488 | u64 flags; |
53b381b3 | 2489 | u64 ret; |
9ed74f2d | 2490 | |
b742bb82 YZ |
2491 | if (data) |
2492 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2493 | else if (root == fs_info->chunk_root) |
b742bb82 | 2494 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2495 | else |
b742bb82 | 2496 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2497 | |
878d7b67 | 2498 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2499 | return ret; |
6a63209f | 2500 | } |
9ed74f2d | 2501 | |
2ff7e61e | 2502 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2503 | { |
32da5386 | 2504 | struct btrfs_block_group *cache; |
d2fb3437 | 2505 | u64 bytenr; |
0f9dd46c | 2506 | |
0b246afa JM |
2507 | spin_lock(&fs_info->block_group_cache_lock); |
2508 | bytenr = fs_info->first_logical_byte; | |
2509 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2510 | |
2511 | if (bytenr < (u64)-1) | |
2512 | return bytenr; | |
2513 | ||
0b246afa | 2514 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2515 | if (!cache) |
a061fc8d | 2516 | return 0; |
0f9dd46c | 2517 | |
b3470b5d | 2518 | bytenr = cache->start; |
fa9c0d79 | 2519 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2520 | |
2521 | return bytenr; | |
a061fc8d CM |
2522 | } |
2523 | ||
6690d071 NB |
2524 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2525 | struct btrfs_block_group *cache, | |
f0486c68 | 2526 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2527 | { |
fdf08605 DS |
2528 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2529 | ||
11833d66 YZ |
2530 | spin_lock(&cache->space_info->lock); |
2531 | spin_lock(&cache->lock); | |
2532 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2533 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2534 | num_bytes); | |
11833d66 YZ |
2535 | if (reserved) { |
2536 | cache->reserved -= num_bytes; | |
2537 | cache->space_info->bytes_reserved -= num_bytes; | |
2538 | } | |
2539 | spin_unlock(&cache->lock); | |
2540 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2541 | |
fe119a6e | 2542 | set_extent_dirty(&trans->transaction->pinned_extents, bytenr, |
f0486c68 YZ |
2543 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2544 | return 0; | |
2545 | } | |
68b38550 | 2546 | |
b25c36f8 | 2547 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2548 | u64 bytenr, u64 num_bytes, int reserved) |
2549 | { | |
32da5386 | 2550 | struct btrfs_block_group *cache; |
68b38550 | 2551 | |
b25c36f8 | 2552 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2553 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2554 | |
6690d071 | 2555 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2556 | |
2557 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2558 | return 0; |
2559 | } | |
2560 | ||
f0486c68 | 2561 | /* |
e688b725 CM |
2562 | * this function must be called within transaction |
2563 | */ | |
9fce5704 | 2564 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
e688b725 CM |
2565 | u64 bytenr, u64 num_bytes) |
2566 | { | |
32da5386 | 2567 | struct btrfs_block_group *cache; |
b50c6e25 | 2568 | int ret; |
e688b725 | 2569 | |
9fce5704 | 2570 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
b50c6e25 JB |
2571 | if (!cache) |
2572 | return -EINVAL; | |
e688b725 CM |
2573 | |
2574 | /* | |
2575 | * pull in the free space cache (if any) so that our pin | |
2576 | * removes the free space from the cache. We have load_only set | |
2577 | * to one because the slow code to read in the free extents does check | |
2578 | * the pinned extents. | |
2579 | */ | |
676f1f75 | 2580 | btrfs_cache_block_group(cache, 1); |
9ad6d91f FM |
2581 | /* |
2582 | * Make sure we wait until the cache is completely built in case it is | |
2583 | * missing or is invalid and therefore needs to be rebuilt. | |
2584 | */ | |
2585 | ret = btrfs_wait_block_group_cache_done(cache); | |
2586 | if (ret) | |
2587 | goto out; | |
e688b725 | 2588 | |
6690d071 | 2589 | pin_down_extent(trans, cache, bytenr, num_bytes, 0); |
e688b725 CM |
2590 | |
2591 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2592 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
9ad6d91f | 2593 | out: |
e688b725 | 2594 | btrfs_put_block_group(cache); |
b50c6e25 | 2595 | return ret; |
e688b725 CM |
2596 | } |
2597 | ||
2ff7e61e JM |
2598 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2599 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2600 | { |
2601 | int ret; | |
32da5386 | 2602 | struct btrfs_block_group *block_group; |
8c2a1a30 | 2603 | |
0b246afa | 2604 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2605 | if (!block_group) |
2606 | return -EINVAL; | |
2607 | ||
9ad6d91f FM |
2608 | btrfs_cache_block_group(block_group, 1); |
2609 | /* | |
2610 | * Make sure we wait until the cache is completely built in case it is | |
2611 | * missing or is invalid and therefore needs to be rebuilt. | |
2612 | */ | |
2613 | ret = btrfs_wait_block_group_cache_done(block_group); | |
2614 | if (ret) | |
2615 | goto out; | |
8c2a1a30 | 2616 | |
9ad6d91f FM |
2617 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2618 | out: | |
8c2a1a30 JB |
2619 | btrfs_put_block_group(block_group); |
2620 | return ret; | |
2621 | } | |
2622 | ||
bcdc428c | 2623 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2624 | { |
bcdc428c | 2625 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2626 | struct btrfs_file_extent_item *item; |
2627 | struct btrfs_key key; | |
2628 | int found_type; | |
2629 | int i; | |
b89311ef | 2630 | int ret = 0; |
8c2a1a30 | 2631 | |
2ff7e61e | 2632 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2633 | return 0; |
2634 | ||
2635 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2636 | btrfs_item_key_to_cpu(eb, &key, i); | |
2637 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2638 | continue; | |
2639 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2640 | found_type = btrfs_file_extent_type(eb, item); | |
2641 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2642 | continue; | |
2643 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2644 | continue; | |
2645 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2646 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2647 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2648 | if (ret) | |
2649 | break; | |
8c2a1a30 JB |
2650 | } |
2651 | ||
b89311ef | 2652 | return ret; |
8c2a1a30 JB |
2653 | } |
2654 | ||
9cfa3e34 | 2655 | static void |
32da5386 | 2656 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2657 | { |
2658 | atomic_inc(&bg->reservations); | |
2659 | } | |
2660 | ||
c759c4e1 JB |
2661 | /* |
2662 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2663 | * what it should be based on the mount options. | |
2664 | */ | |
2665 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2666 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2667 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2668 | { |
2669 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2670 | |
2671 | *empty_cluster = 0; | |
2672 | if (btrfs_mixed_space_info(space_info)) | |
2673 | return ret; | |
2674 | ||
c759c4e1 | 2675 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2676 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2677 | if (btrfs_test_opt(fs_info, SSD)) |
2678 | *empty_cluster = SZ_2M; | |
2679 | else | |
ee22184b | 2680 | *empty_cluster = SZ_64K; |
583b7231 HK |
2681 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2682 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2683 | *empty_cluster = SZ_2M; | |
0b246afa | 2684 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2685 | } |
2686 | ||
2687 | return ret; | |
2688 | } | |
2689 | ||
2ff7e61e JM |
2690 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2691 | u64 start, u64 end, | |
678886bd | 2692 | const bool return_free_space) |
ccd467d6 | 2693 | { |
32da5386 | 2694 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2695 | struct btrfs_space_info *space_info; |
2696 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2697 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2698 | u64 len; |
c759c4e1 JB |
2699 | u64 total_unpinned = 0; |
2700 | u64 empty_cluster = 0; | |
7b398f8e | 2701 | bool readonly; |
ccd467d6 | 2702 | |
11833d66 | 2703 | while (start <= end) { |
7b398f8e | 2704 | readonly = false; |
11833d66 | 2705 | if (!cache || |
b3470b5d | 2706 | start >= cache->start + cache->length) { |
11833d66 YZ |
2707 | if (cache) |
2708 | btrfs_put_block_group(cache); | |
c759c4e1 | 2709 | total_unpinned = 0; |
11833d66 | 2710 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2711 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2712 | |
2ff7e61e | 2713 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2714 | cache->space_info, |
2715 | &empty_cluster); | |
2716 | empty_cluster <<= 1; | |
11833d66 YZ |
2717 | } |
2718 | ||
b3470b5d | 2719 | len = cache->start + cache->length - start; |
11833d66 YZ |
2720 | len = min(len, end + 1 - start); |
2721 | ||
2ca08c56 | 2722 | down_read(&fs_info->commit_root_sem); |
9076dbd5 JB |
2723 | if (start < cache->last_byte_to_unpin && return_free_space) { |
2724 | u64 add_len = min(len, cache->last_byte_to_unpin - start); | |
2725 | ||
2726 | btrfs_add_free_space(cache, start, add_len); | |
11833d66 | 2727 | } |
2ca08c56 | 2728 | up_read(&fs_info->commit_root_sem); |
11833d66 | 2729 | |
f0486c68 | 2730 | start += len; |
c759c4e1 | 2731 | total_unpinned += len; |
7b398f8e | 2732 | space_info = cache->space_info; |
f0486c68 | 2733 | |
c759c4e1 JB |
2734 | /* |
2735 | * If this space cluster has been marked as fragmented and we've | |
2736 | * unpinned enough in this block group to potentially allow a | |
2737 | * cluster to be created inside of it go ahead and clear the | |
2738 | * fragmented check. | |
2739 | */ | |
2740 | if (cluster && cluster->fragmented && | |
2741 | total_unpinned > empty_cluster) { | |
2742 | spin_lock(&cluster->lock); | |
2743 | cluster->fragmented = 0; | |
2744 | spin_unlock(&cluster->lock); | |
2745 | } | |
2746 | ||
7b398f8e | 2747 | spin_lock(&space_info->lock); |
11833d66 YZ |
2748 | spin_lock(&cache->lock); |
2749 | cache->pinned -= len; | |
bb96c4e5 | 2750 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2751 | space_info->max_extent_size = 0; |
7b398f8e JB |
2752 | if (cache->ro) { |
2753 | space_info->bytes_readonly += len; | |
2754 | readonly = true; | |
169e0da9 NA |
2755 | } else if (btrfs_is_zoned(fs_info)) { |
2756 | /* Need reset before reusing in a zoned block group */ | |
2757 | space_info->bytes_zone_unusable += len; | |
2758 | readonly = true; | |
7b398f8e | 2759 | } |
11833d66 | 2760 | spin_unlock(&cache->lock); |
957780eb JB |
2761 | if (!readonly && return_free_space && |
2762 | global_rsv->space_info == space_info) { | |
2763 | u64 to_add = len; | |
92ac58ec | 2764 | |
7b398f8e JB |
2765 | spin_lock(&global_rsv->lock); |
2766 | if (!global_rsv->full) { | |
957780eb JB |
2767 | to_add = min(len, global_rsv->size - |
2768 | global_rsv->reserved); | |
2769 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2770 | btrfs_space_info_update_bytes_may_use(fs_info, |
2771 | space_info, to_add); | |
7b398f8e JB |
2772 | if (global_rsv->reserved >= global_rsv->size) |
2773 | global_rsv->full = 1; | |
957780eb | 2774 | len -= to_add; |
7b398f8e JB |
2775 | } |
2776 | spin_unlock(&global_rsv->lock); | |
2777 | } | |
2732798c JB |
2778 | /* Add to any tickets we may have */ |
2779 | if (!readonly && return_free_space && len) | |
2780 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2781 | spin_unlock(&space_info->lock); |
ccd467d6 | 2782 | } |
11833d66 YZ |
2783 | |
2784 | if (cache) | |
2785 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2786 | return 0; |
2787 | } | |
2788 | ||
5ead2dd0 | 2789 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2790 | { |
5ead2dd0 | 2791 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2792 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2793 | struct list_head *deleted_bgs; |
11833d66 | 2794 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2795 | u64 start; |
2796 | u64 end; | |
a28ec197 | 2797 | int ret; |
a28ec197 | 2798 | |
fe119a6e | 2799 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2800 | |
bf31f87f | 2801 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2802 | struct extent_state *cached_state = NULL; |
2803 | ||
d4b450cd | 2804 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2805 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2806 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2807 | if (ret) { |
2808 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2809 | break; |
d4b450cd | 2810 | } |
1f3c79a2 | 2811 | |
46b27f50 | 2812 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2813 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2814 | end + 1 - start, NULL); |
1f3c79a2 | 2815 | |
0e6ec385 | 2816 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2817 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2818 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2819 | free_extent_state(cached_state); |
b9473439 | 2820 | cond_resched(); |
a28ec197 | 2821 | } |
817d52f8 | 2822 | |
a2309300 DZ |
2823 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2824 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2825 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2826 | } |
b0643e59 | 2827 | |
e33e17ee JM |
2828 | /* |
2829 | * Transaction is finished. We don't need the lock anymore. We | |
2830 | * do need to clean up the block groups in case of a transaction | |
2831 | * abort. | |
2832 | */ | |
2833 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2834 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2835 | u64 trimmed = 0; | |
2836 | ||
2837 | ret = -EROFS; | |
bf31f87f | 2838 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2839 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2840 | block_group->start, |
2841 | block_group->length, | |
e33e17ee JM |
2842 | &trimmed); |
2843 | ||
2844 | list_del_init(&block_group->bg_list); | |
6b7304af | 2845 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2846 | btrfs_put_block_group(block_group); |
2847 | ||
2848 | if (ret) { | |
2849 | const char *errstr = btrfs_decode_error(ret); | |
2850 | btrfs_warn(fs_info, | |
913e1535 | 2851 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2852 | ret, errstr); |
2853 | } | |
2854 | } | |
2855 | ||
e20d96d6 CM |
2856 | return 0; |
2857 | } | |
2858 | ||
1c2a07f5 QW |
2859 | /* |
2860 | * Drop one or more refs of @node. | |
2861 | * | |
2862 | * 1. Locate the extent refs. | |
2863 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
2864 | * Locate it, then reduce the refs number or remove the ref line completely. | |
2865 | * | |
2866 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
2867 | * | |
2868 | * Inline backref case: | |
2869 | * | |
2870 | * in extent tree we have: | |
2871 | * | |
2872 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2873 | * refs 2 gen 6 flags DATA | |
2874 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2875 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
2876 | * | |
2877 | * This function gets called with: | |
2878 | * | |
2879 | * node->bytenr = 13631488 | |
2880 | * node->num_bytes = 1048576 | |
2881 | * root_objectid = FS_TREE | |
2882 | * owner_objectid = 257 | |
2883 | * owner_offset = 0 | |
2884 | * refs_to_drop = 1 | |
2885 | * | |
2886 | * Then we should get some like: | |
2887 | * | |
2888 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2889 | * refs 1 gen 6 flags DATA | |
2890 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2891 | * | |
2892 | * Keyed backref case: | |
2893 | * | |
2894 | * in extent tree we have: | |
2895 | * | |
2896 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2897 | * refs 754 gen 6 flags DATA | |
2898 | * [...] | |
2899 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
2900 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
2901 | * | |
2902 | * This function get called with: | |
2903 | * | |
2904 | * node->bytenr = 13631488 | |
2905 | * node->num_bytes = 1048576 | |
2906 | * root_objectid = FS_TREE | |
2907 | * owner_objectid = 866 | |
2908 | * owner_offset = 0 | |
2909 | * refs_to_drop = 1 | |
2910 | * | |
2911 | * Then we should get some like: | |
2912 | * | |
2913 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2914 | * refs 753 gen 6 flags DATA | |
2915 | * | |
2916 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
2917 | */ | |
5d4f98a2 | 2918 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2919 | struct btrfs_delayed_ref_node *node, u64 parent, |
2920 | u64 root_objectid, u64 owner_objectid, | |
2921 | u64 owner_offset, int refs_to_drop, | |
2922 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2923 | { |
e72cb923 | 2924 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2925 | struct btrfs_key key; |
5d4f98a2 | 2926 | struct btrfs_path *path; |
1261ec42 | 2927 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2928 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2929 | struct btrfs_extent_item *ei; |
2930 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2931 | int ret; |
5d4f98a2 | 2932 | int is_data; |
952fccac CM |
2933 | int extent_slot = 0; |
2934 | int found_extent = 0; | |
2935 | int num_to_del = 1; | |
5d4f98a2 YZ |
2936 | u32 item_size; |
2937 | u64 refs; | |
c682f9b3 QW |
2938 | u64 bytenr = node->bytenr; |
2939 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2940 | int last_ref = 0; |
0b246afa | 2941 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 2942 | |
5caf2a00 | 2943 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2944 | if (!path) |
2945 | return -ENOMEM; | |
5f26f772 | 2946 | |
5d4f98a2 | 2947 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
1c2a07f5 QW |
2948 | |
2949 | if (!is_data && refs_to_drop != 1) { | |
2950 | btrfs_crit(info, | |
2951 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
2952 | node->bytenr, refs_to_drop); | |
2953 | ret = -EINVAL; | |
2954 | btrfs_abort_transaction(trans, ret); | |
2955 | goto out; | |
2956 | } | |
5d4f98a2 | 2957 | |
3173a18f | 2958 | if (is_data) |
897ca819 | 2959 | skinny_metadata = false; |
3173a18f | 2960 | |
fbe4801b NB |
2961 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
2962 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 2963 | owner_offset); |
7bb86316 | 2964 | if (ret == 0) { |
1c2a07f5 QW |
2965 | /* |
2966 | * Either the inline backref or the SHARED_DATA_REF/ | |
2967 | * SHARED_BLOCK_REF is found | |
2968 | * | |
2969 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
2970 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
2971 | */ | |
952fccac | 2972 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
2973 | while (extent_slot >= 0) { |
2974 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 2975 | extent_slot); |
5d4f98a2 | 2976 | if (key.objectid != bytenr) |
952fccac | 2977 | break; |
5d4f98a2 YZ |
2978 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
2979 | key.offset == num_bytes) { | |
952fccac CM |
2980 | found_extent = 1; |
2981 | break; | |
2982 | } | |
3173a18f JB |
2983 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
2984 | key.offset == owner_objectid) { | |
2985 | found_extent = 1; | |
2986 | break; | |
2987 | } | |
1c2a07f5 QW |
2988 | |
2989 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
2990 | if (path->slots[0] - extent_slot > 5) |
2991 | break; | |
5d4f98a2 | 2992 | extent_slot--; |
952fccac | 2993 | } |
a79865c6 | 2994 | |
31840ae1 | 2995 | if (!found_extent) { |
1c2a07f5 QW |
2996 | if (iref) { |
2997 | btrfs_crit(info, | |
2998 | "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref"); | |
2999 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3000 | goto err_dump; | |
3001 | } | |
3002 | /* Must be SHARED_* item, remove the backref first */ | |
87cc7a8a | 3003 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3004 | refs_to_drop, |
fcebe456 | 3005 | is_data, &last_ref); |
005d6427 | 3006 | if (ret) { |
66642832 | 3007 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3008 | goto out; |
3009 | } | |
b3b4aa74 | 3010 | btrfs_release_path(path); |
5d4f98a2 | 3011 | |
1c2a07f5 | 3012 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3013 | key.objectid = bytenr; |
3014 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3015 | key.offset = num_bytes; | |
3016 | ||
3173a18f JB |
3017 | if (!is_data && skinny_metadata) { |
3018 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3019 | key.offset = owner_objectid; | |
3020 | } | |
3021 | ||
31840ae1 ZY |
3022 | ret = btrfs_search_slot(trans, extent_root, |
3023 | &key, path, -1, 1); | |
3173a18f JB |
3024 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3025 | /* | |
3026 | * Couldn't find our skinny metadata item, | |
3027 | * see if we have ye olde extent item. | |
3028 | */ | |
3029 | path->slots[0]--; | |
3030 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3031 | path->slots[0]); | |
3032 | if (key.objectid == bytenr && | |
3033 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3034 | key.offset == num_bytes) | |
3035 | ret = 0; | |
3036 | } | |
3037 | ||
3038 | if (ret > 0 && skinny_metadata) { | |
3039 | skinny_metadata = false; | |
9ce49a0b | 3040 | key.objectid = bytenr; |
3173a18f JB |
3041 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3042 | key.offset = num_bytes; | |
3043 | btrfs_release_path(path); | |
3044 | ret = btrfs_search_slot(trans, extent_root, | |
3045 | &key, path, -1, 1); | |
3046 | } | |
3047 | ||
f3465ca4 | 3048 | if (ret) { |
5d163e0e JM |
3049 | btrfs_err(info, |
3050 | "umm, got %d back from search, was looking for %llu", | |
3051 | ret, bytenr); | |
b783e62d | 3052 | if (ret > 0) |
a4f78750 | 3053 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3054 | } |
005d6427 | 3055 | if (ret < 0) { |
66642832 | 3056 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3057 | goto out; |
3058 | } | |
31840ae1 ZY |
3059 | extent_slot = path->slots[0]; |
3060 | } | |
fae7f21c | 3061 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3062 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3063 | btrfs_err(info, |
3064 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3065 | bytenr, parent, root_objectid, owner_objectid, |
3066 | owner_offset); | |
66642832 | 3067 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3068 | goto out; |
79787eaa | 3069 | } else { |
66642832 | 3070 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3071 | goto out; |
7bb86316 | 3072 | } |
5f39d397 CM |
3073 | |
3074 | leaf = path->nodes[0]; | |
5d4f98a2 | 3075 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3076 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3077 | ret = -EINVAL; |
3078 | btrfs_print_v0_err(info); | |
3079 | btrfs_abort_transaction(trans, ret); | |
3080 | goto out; | |
3081 | } | |
952fccac | 3082 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3083 | struct btrfs_extent_item); |
3173a18f JB |
3084 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3085 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3086 | struct btrfs_tree_block_info *bi; |
1c2a07f5 QW |
3087 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
3088 | btrfs_crit(info, | |
cad69d13 | 3089 | "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu", |
1c2a07f5 QW |
3090 | key.objectid, key.type, key.offset, |
3091 | owner_objectid, item_size, | |
3092 | sizeof(*ei) + sizeof(*bi)); | |
3093 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3094 | goto err_dump; | |
3095 | } | |
5d4f98a2 YZ |
3096 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3097 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3098 | } | |
56bec294 | 3099 | |
5d4f98a2 | 3100 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3101 | if (refs < refs_to_drop) { |
1c2a07f5 QW |
3102 | btrfs_crit(info, |
3103 | "trying to drop %d refs but we only have %llu for bytenr %llu", | |
5d163e0e | 3104 | refs_to_drop, refs, bytenr); |
1c2a07f5 QW |
3105 | btrfs_abort_transaction(trans, -EUCLEAN); |
3106 | goto err_dump; | |
32b02538 | 3107 | } |
56bec294 | 3108 | refs -= refs_to_drop; |
5f39d397 | 3109 | |
5d4f98a2 YZ |
3110 | if (refs > 0) { |
3111 | if (extent_op) | |
3112 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3113 | /* | |
3114 | * In the case of inline back ref, reference count will | |
3115 | * be updated by remove_extent_backref | |
952fccac | 3116 | */ |
5d4f98a2 | 3117 | if (iref) { |
1c2a07f5 QW |
3118 | if (!found_extent) { |
3119 | btrfs_crit(info, | |
3120 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found"); | |
3121 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3122 | goto err_dump; | |
3123 | } | |
5d4f98a2 YZ |
3124 | } else { |
3125 | btrfs_set_extent_refs(leaf, ei, refs); | |
3126 | btrfs_mark_buffer_dirty(leaf); | |
3127 | } | |
3128 | if (found_extent) { | |
87cc7a8a NB |
3129 | ret = remove_extent_backref(trans, path, iref, |
3130 | refs_to_drop, is_data, | |
3131 | &last_ref); | |
005d6427 | 3132 | if (ret) { |
66642832 | 3133 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3134 | goto out; |
3135 | } | |
952fccac | 3136 | } |
5d4f98a2 | 3137 | } else { |
1c2a07f5 | 3138 | /* In this branch refs == 1 */ |
5d4f98a2 | 3139 | if (found_extent) { |
1c2a07f5 QW |
3140 | if (is_data && refs_to_drop != |
3141 | extent_data_ref_count(path, iref)) { | |
3142 | btrfs_crit(info, | |
3143 | "invalid refs_to_drop, current refs %u refs_to_drop %u", | |
3144 | extent_data_ref_count(path, iref), | |
3145 | refs_to_drop); | |
3146 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3147 | goto err_dump; | |
3148 | } | |
5d4f98a2 | 3149 | if (iref) { |
1c2a07f5 QW |
3150 | if (path->slots[0] != extent_slot) { |
3151 | btrfs_crit(info, | |
3152 | "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref", | |
3153 | key.objectid, key.type, | |
3154 | key.offset); | |
3155 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3156 | goto err_dump; | |
3157 | } | |
5d4f98a2 | 3158 | } else { |
1c2a07f5 QW |
3159 | /* |
3160 | * No inline ref, we must be at SHARED_* item, | |
3161 | * And it's single ref, it must be: | |
3162 | * | extent_slot ||extent_slot + 1| | |
3163 | * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ] | |
3164 | */ | |
3165 | if (path->slots[0] != extent_slot + 1) { | |
3166 | btrfs_crit(info, | |
3167 | "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM"); | |
3168 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3169 | goto err_dump; | |
3170 | } | |
5d4f98a2 YZ |
3171 | path->slots[0] = extent_slot; |
3172 | num_to_del = 2; | |
3173 | } | |
78fae27e | 3174 | } |
b9473439 | 3175 | |
fcebe456 | 3176 | last_ref = 1; |
952fccac CM |
3177 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3178 | num_to_del); | |
005d6427 | 3179 | if (ret) { |
66642832 | 3180 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3181 | goto out; |
3182 | } | |
b3b4aa74 | 3183 | btrfs_release_path(path); |
21af804c | 3184 | |
5d4f98a2 | 3185 | if (is_data) { |
40e046ac FM |
3186 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3187 | num_bytes); | |
005d6427 | 3188 | if (ret) { |
66642832 | 3189 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3190 | goto out; |
3191 | } | |
459931ec CM |
3192 | } |
3193 | ||
e7355e50 | 3194 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3195 | if (ret) { |
66642832 | 3196 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3197 | goto out; |
3198 | } | |
3199 | ||
ade4b516 | 3200 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3201 | if (ret) { |
66642832 | 3202 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3203 | goto out; |
3204 | } | |
a28ec197 | 3205 | } |
fcebe456 JB |
3206 | btrfs_release_path(path); |
3207 | ||
79787eaa | 3208 | out: |
5caf2a00 | 3209 | btrfs_free_path(path); |
a28ec197 | 3210 | return ret; |
1c2a07f5 QW |
3211 | err_dump: |
3212 | /* | |
3213 | * Leaf dump can take up a lot of log buffer, so we only do full leaf | |
3214 | * dump for debug build. | |
3215 | */ | |
3216 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
3217 | btrfs_crit(info, "path->slots[0]=%d extent_slot=%d", | |
3218 | path->slots[0], extent_slot); | |
3219 | btrfs_print_leaf(path->nodes[0]); | |
3220 | } | |
3221 | ||
3222 | btrfs_free_path(path); | |
3223 | return -EUCLEAN; | |
a28ec197 CM |
3224 | } |
3225 | ||
1887be66 | 3226 | /* |
f0486c68 | 3227 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3228 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3229 | * a given extent, and if there are no other delayed refs to be processed, it | |
3230 | * removes it from the tree. | |
3231 | */ | |
3232 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3233 | u64 bytenr) |
1887be66 CM |
3234 | { |
3235 | struct btrfs_delayed_ref_head *head; | |
3236 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3237 | int ret = 0; |
1887be66 CM |
3238 | |
3239 | delayed_refs = &trans->transaction->delayed_refs; | |
3240 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3241 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3242 | if (!head) |
cf93da7b | 3243 | goto out_delayed_unlock; |
1887be66 | 3244 | |
d7df2c79 | 3245 | spin_lock(&head->lock); |
e3d03965 | 3246 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3247 | goto out; |
3248 | ||
bedc6617 JB |
3249 | if (cleanup_extent_op(head) != NULL) |
3250 | goto out; | |
5d4f98a2 | 3251 | |
1887be66 CM |
3252 | /* |
3253 | * waiting for the lock here would deadlock. If someone else has it | |
3254 | * locked they are already in the process of dropping it anyway | |
3255 | */ | |
3256 | if (!mutex_trylock(&head->mutex)) | |
3257 | goto out; | |
3258 | ||
d7baffda | 3259 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3260 | head->processing = 0; |
d7baffda | 3261 | |
d7df2c79 | 3262 | spin_unlock(&head->lock); |
1887be66 CM |
3263 | spin_unlock(&delayed_refs->lock); |
3264 | ||
f0486c68 YZ |
3265 | BUG_ON(head->extent_op); |
3266 | if (head->must_insert_reserved) | |
3267 | ret = 1; | |
3268 | ||
31890da0 | 3269 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3270 | mutex_unlock(&head->mutex); |
d278850e | 3271 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3272 | return ret; |
1887be66 | 3273 | out: |
d7df2c79 | 3274 | spin_unlock(&head->lock); |
cf93da7b CM |
3275 | |
3276 | out_delayed_unlock: | |
1887be66 CM |
3277 | spin_unlock(&delayed_refs->lock); |
3278 | return 0; | |
3279 | } | |
3280 | ||
f0486c68 | 3281 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4d6fc4ce | 3282 | u64 root_id, |
f0486c68 | 3283 | struct extent_buffer *buf, |
5581a51a | 3284 | u64 parent, int last_ref) |
f0486c68 | 3285 | { |
4d6fc4ce | 3286 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ed4f255b | 3287 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
3288 | int ret; |
3289 | ||
ed4f255b QW |
3290 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3291 | buf->start, buf->len, parent); | |
3292 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
4d6fc4ce | 3293 | root_id, 0, false); |
ed4f255b | 3294 | |
4d6fc4ce | 3295 | if (root_id != BTRFS_TREE_LOG_OBJECTID) { |
8a5040f7 | 3296 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 3297 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL); |
79787eaa | 3298 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
3299 | } |
3300 | ||
0a16c7d7 | 3301 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3302 | struct btrfs_block_group *cache; |
485df755 | 3303 | bool must_pin = false; |
6219872d | 3304 | |
4d6fc4ce | 3305 | if (root_id != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3306 | ret = check_ref_cleanup(trans, buf->start); |
d3575156 NA |
3307 | if (!ret) { |
3308 | btrfs_redirty_list_add(trans->transaction, buf); | |
37be25bc | 3309 | goto out; |
d3575156 | 3310 | } |
f0486c68 YZ |
3311 | } |
3312 | ||
0b246afa | 3313 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3314 | |
f0486c68 | 3315 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3316 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3317 | btrfs_put_block_group(cache); |
37be25bc | 3318 | goto out; |
f0486c68 YZ |
3319 | } |
3320 | ||
485df755 FM |
3321 | /* |
3322 | * If this is a leaf and there are tree mod log users, we may | |
3323 | * have recorded mod log operations that point to this leaf. | |
3324 | * So we must make sure no one reuses this leaf's extent before | |
3325 | * mod log operations are applied to a node, otherwise after | |
3326 | * rewinding a node using the mod log operations we get an | |
3327 | * inconsistent btree, as the leaf's extent may now be used as | |
3328 | * a node or leaf for another different btree. | |
3329 | * We are safe from races here because at this point no other | |
3330 | * node or root points to this extent buffer, so if after this | |
3331 | * check a new tree mod log user joins, it will not be able to | |
3332 | * find a node pointing to this leaf and record operations that | |
3333 | * point to this leaf. | |
3334 | */ | |
888dd183 FM |
3335 | if (btrfs_header_level(buf) == 0 && |
3336 | test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags)) | |
3337 | must_pin = true; | |
485df755 FM |
3338 | |
3339 | if (must_pin || btrfs_is_zoned(fs_info)) { | |
d3575156 NA |
3340 | btrfs_redirty_list_add(trans->transaction, buf); |
3341 | pin_down_extent(trans, cache, buf->start, buf->len, 1); | |
3342 | btrfs_put_block_group(cache); | |
3343 | goto out; | |
3344 | } | |
3345 | ||
f0486c68 YZ |
3346 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); |
3347 | ||
3348 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3349 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3350 | btrfs_put_block_group(cache); |
71ff6437 | 3351 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3352 | } |
3353 | out: | |
0a16c7d7 OS |
3354 | if (last_ref) { |
3355 | /* | |
3356 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3357 | * matter anymore. | |
3358 | */ | |
3359 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3360 | } | |
f0486c68 YZ |
3361 | } |
3362 | ||
79787eaa | 3363 | /* Can return -ENOMEM */ |
ffd4bb2a | 3364 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3365 | { |
ffd4bb2a | 3366 | struct btrfs_fs_info *fs_info = trans->fs_info; |
925baedd CM |
3367 | int ret; |
3368 | ||
f5ee5c9a | 3369 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3370 | return 0; |
fccb84c9 | 3371 | |
56bec294 CM |
3372 | /* |
3373 | * tree log blocks never actually go into the extent allocation | |
3374 | * tree, just update pinning info and exit early. | |
56bec294 | 3375 | */ |
ffd4bb2a QW |
3376 | if ((ref->type == BTRFS_REF_METADATA && |
3377 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3378 | (ref->type == BTRFS_REF_DATA && | |
3379 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3380 | /* unlocks the pinned mutex */ |
b25c36f8 | 3381 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
56bec294 | 3382 | ret = 0; |
ffd4bb2a | 3383 | } else if (ref->type == BTRFS_REF_METADATA) { |
2187374f | 3384 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL); |
5d4f98a2 | 3385 | } else { |
2187374f | 3386 | ret = btrfs_add_delayed_data_ref(trans, ref, 0); |
56bec294 | 3387 | } |
d7eae340 | 3388 | |
ffd4bb2a QW |
3389 | if (!((ref->type == BTRFS_REF_METADATA && |
3390 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3391 | (ref->type == BTRFS_REF_DATA && | |
3392 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3393 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3394 | |
925baedd CM |
3395 | return ret; |
3396 | } | |
3397 | ||
817d52f8 | 3398 | enum btrfs_loop_type { |
f262fa8d DS |
3399 | LOOP_CACHING_NOWAIT, |
3400 | LOOP_CACHING_WAIT, | |
3401 | LOOP_ALLOC_CHUNK, | |
3402 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3403 | }; |
3404 | ||
e570fd27 | 3405 | static inline void |
32da5386 | 3406 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3407 | int delalloc) |
3408 | { | |
3409 | if (delalloc) | |
3410 | down_read(&cache->data_rwsem); | |
3411 | } | |
3412 | ||
32da5386 | 3413 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3414 | int delalloc) |
3415 | { | |
3416 | btrfs_get_block_group(cache); | |
3417 | if (delalloc) | |
3418 | down_read(&cache->data_rwsem); | |
3419 | } | |
3420 | ||
32da5386 DS |
3421 | static struct btrfs_block_group *btrfs_lock_cluster( |
3422 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3423 | struct btrfs_free_cluster *cluster, |
3424 | int delalloc) | |
c142c6a4 | 3425 | __acquires(&cluster->refill_lock) |
e570fd27 | 3426 | { |
32da5386 | 3427 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3428 | |
e570fd27 | 3429 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3430 | while (1) { |
3431 | used_bg = cluster->block_group; | |
3432 | if (!used_bg) | |
3433 | return NULL; | |
3434 | ||
3435 | if (used_bg == block_group) | |
e570fd27 MX |
3436 | return used_bg; |
3437 | ||
6719afdc | 3438 | btrfs_get_block_group(used_bg); |
e570fd27 | 3439 | |
6719afdc GU |
3440 | if (!delalloc) |
3441 | return used_bg; | |
e570fd27 | 3442 | |
6719afdc GU |
3443 | if (down_read_trylock(&used_bg->data_rwsem)) |
3444 | return used_bg; | |
e570fd27 | 3445 | |
6719afdc | 3446 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3447 | |
e321f8a8 LB |
3448 | /* We should only have one-level nested. */ |
3449 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3450 | |
6719afdc GU |
3451 | spin_lock(&cluster->refill_lock); |
3452 | if (used_bg == cluster->block_group) | |
3453 | return used_bg; | |
e570fd27 | 3454 | |
6719afdc GU |
3455 | up_read(&used_bg->data_rwsem); |
3456 | btrfs_put_block_group(used_bg); | |
3457 | } | |
e570fd27 MX |
3458 | } |
3459 | ||
3460 | static inline void | |
32da5386 | 3461 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3462 | int delalloc) |
3463 | { | |
3464 | if (delalloc) | |
3465 | up_read(&cache->data_rwsem); | |
3466 | btrfs_put_block_group(cache); | |
3467 | } | |
3468 | ||
cb2f96f8 NA |
3469 | enum btrfs_extent_allocation_policy { |
3470 | BTRFS_EXTENT_ALLOC_CLUSTERED, | |
2eda5708 | 3471 | BTRFS_EXTENT_ALLOC_ZONED, |
cb2f96f8 NA |
3472 | }; |
3473 | ||
b4bd745d QW |
3474 | /* |
3475 | * Structure used internally for find_free_extent() function. Wraps needed | |
3476 | * parameters. | |
3477 | */ | |
3478 | struct find_free_extent_ctl { | |
3479 | /* Basic allocation info */ | |
b4bd745d QW |
3480 | u64 num_bytes; |
3481 | u64 empty_size; | |
3482 | u64 flags; | |
3483 | int delalloc; | |
3484 | ||
3485 | /* Where to start the search inside the bg */ | |
3486 | u64 search_start; | |
3487 | ||
3488 | /* For clustered allocation */ | |
3489 | u64 empty_cluster; | |
c10859be NA |
3490 | struct btrfs_free_cluster *last_ptr; |
3491 | bool use_cluster; | |
b4bd745d QW |
3492 | |
3493 | bool have_caching_bg; | |
3494 | bool orig_have_caching_bg; | |
3495 | ||
40ab3be1 NA |
3496 | /* Allocation is called for tree-log */ |
3497 | bool for_treelog; | |
3498 | ||
763e2672 JT |
3499 | /* Allocation is called for data relocation */ |
3500 | bool for_data_reloc; | |
3501 | ||
b4bd745d QW |
3502 | /* RAID index, converted from flags */ |
3503 | int index; | |
3504 | ||
e72d79d6 QW |
3505 | /* |
3506 | * Current loop number, check find_free_extent_update_loop() for details | |
3507 | */ | |
b4bd745d QW |
3508 | int loop; |
3509 | ||
3510 | /* | |
3511 | * Whether we're refilling a cluster, if true we need to re-search | |
3512 | * current block group but don't try to refill the cluster again. | |
3513 | */ | |
3514 | bool retry_clustered; | |
3515 | ||
3516 | /* | |
3517 | * Whether we're updating free space cache, if true we need to re-search | |
3518 | * current block group but don't try updating free space cache again. | |
3519 | */ | |
3520 | bool retry_unclustered; | |
3521 | ||
3522 | /* If current block group is cached */ | |
3523 | int cached; | |
3524 | ||
3525 | /* Max contiguous hole found */ | |
3526 | u64 max_extent_size; | |
3527 | ||
3528 | /* Total free space from free space cache, not always contiguous */ | |
3529 | u64 total_free_space; | |
3530 | ||
3531 | /* Found result */ | |
3532 | u64 found_offset; | |
cb2f96f8 | 3533 | |
ea544149 NA |
3534 | /* Hint where to start looking for an empty space */ |
3535 | u64 hint_byte; | |
3536 | ||
cb2f96f8 NA |
3537 | /* Allocation policy */ |
3538 | enum btrfs_extent_allocation_policy policy; | |
b4bd745d QW |
3539 | }; |
3540 | ||
d06e3bb6 QW |
3541 | |
3542 | /* | |
3543 | * Helper function for find_free_extent(). | |
3544 | * | |
3545 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3546 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3547 | * Return >0 to inform caller that we find nothing | |
3548 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3549 | */ | |
32da5386 | 3550 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3551 | struct find_free_extent_ctl *ffe_ctl, |
3552 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3553 | { |
32da5386 | 3554 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3555 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3556 | u64 aligned_cluster; |
3557 | u64 offset; | |
3558 | int ret; | |
3559 | ||
3560 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3561 | if (!cluster_bg) | |
3562 | goto refill_cluster; | |
3563 | if (cluster_bg != bg && (cluster_bg->ro || | |
3564 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3565 | goto release_cluster; | |
3566 | ||
3567 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3568 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3569 | &ffe_ctl->max_extent_size); |
3570 | if (offset) { | |
3571 | /* We have a block, we're done */ | |
3572 | spin_unlock(&last_ptr->refill_lock); | |
3573 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3574 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3575 | *cluster_bg_ret = cluster_bg; | |
3576 | ffe_ctl->found_offset = offset; | |
3577 | return 0; | |
3578 | } | |
3579 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3580 | ||
3581 | release_cluster: | |
3582 | /* | |
3583 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3584 | * lets just skip it and let the allocator find whatever block it can | |
3585 | * find. If we reach this point, we will have tried the cluster | |
3586 | * allocator plenty of times and not have found anything, so we are | |
3587 | * likely way too fragmented for the clustering stuff to find anything. | |
3588 | * | |
3589 | * However, if the cluster is taken from the current block group, | |
3590 | * release the cluster first, so that we stand a better chance of | |
3591 | * succeeding in the unclustered allocation. | |
3592 | */ | |
3593 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3594 | spin_unlock(&last_ptr->refill_lock); | |
3595 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3596 | return -ENOENT; | |
3597 | } | |
3598 | ||
3599 | /* This cluster didn't work out, free it and start over */ | |
3600 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3601 | ||
3602 | if (cluster_bg != bg) | |
3603 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3604 | ||
3605 | refill_cluster: | |
3606 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3607 | spin_unlock(&last_ptr->refill_lock); | |
3608 | return -ENOENT; | |
3609 | } | |
3610 | ||
3611 | aligned_cluster = max_t(u64, | |
3612 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3613 | bg->full_stripe_len); | |
2ceeae2e DS |
3614 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3615 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3616 | if (ret == 0) { |
3617 | /* Now pull our allocation out of this cluster */ | |
3618 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3619 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3620 | &ffe_ctl->max_extent_size); | |
3621 | if (offset) { | |
3622 | /* We found one, proceed */ | |
3623 | spin_unlock(&last_ptr->refill_lock); | |
3624 | trace_btrfs_reserve_extent_cluster(bg, | |
3625 | ffe_ctl->search_start, | |
3626 | ffe_ctl->num_bytes); | |
3627 | ffe_ctl->found_offset = offset; | |
3628 | return 0; | |
3629 | } | |
3630 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3631 | !ffe_ctl->retry_clustered) { | |
3632 | spin_unlock(&last_ptr->refill_lock); | |
3633 | ||
3634 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3635 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3636 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3637 | return -EAGAIN; | |
3638 | } | |
3639 | /* | |
3640 | * At this point we either didn't find a cluster or we weren't able to | |
3641 | * allocate a block from our cluster. Free the cluster we've been | |
3642 | * trying to use, and go to the next block group. | |
3643 | */ | |
3644 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3645 | spin_unlock(&last_ptr->refill_lock); | |
3646 | return 1; | |
3647 | } | |
3648 | ||
e1a41848 QW |
3649 | /* |
3650 | * Return >0 to inform caller that we find nothing | |
3651 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3652 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3653 | */ | |
32da5386 | 3654 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3655 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3656 | { |
897cae79 | 3657 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3658 | u64 offset; |
3659 | ||
3660 | /* | |
3661 | * We are doing an unclustered allocation, set the fragmented flag so | |
3662 | * we don't bother trying to setup a cluster again until we get more | |
3663 | * space. | |
3664 | */ | |
3665 | if (unlikely(last_ptr)) { | |
3666 | spin_lock(&last_ptr->lock); | |
3667 | last_ptr->fragmented = 1; | |
3668 | spin_unlock(&last_ptr->lock); | |
3669 | } | |
3670 | if (ffe_ctl->cached) { | |
3671 | struct btrfs_free_space_ctl *free_space_ctl; | |
3672 | ||
3673 | free_space_ctl = bg->free_space_ctl; | |
3674 | spin_lock(&free_space_ctl->tree_lock); | |
3675 | if (free_space_ctl->free_space < | |
3676 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3677 | ffe_ctl->empty_size) { | |
3678 | ffe_ctl->total_free_space = max_t(u64, | |
3679 | ffe_ctl->total_free_space, | |
3680 | free_space_ctl->free_space); | |
3681 | spin_unlock(&free_space_ctl->tree_lock); | |
3682 | return 1; | |
3683 | } | |
3684 | spin_unlock(&free_space_ctl->tree_lock); | |
3685 | } | |
3686 | ||
3687 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3688 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3689 | &ffe_ctl->max_extent_size); | |
3690 | ||
3691 | /* | |
3692 | * If we didn't find a chunk, and we haven't failed on this block group | |
3693 | * before, and this block group is in the middle of caching and we are | |
3694 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3695 | * set @retry_unclustered to true. | |
3696 | * | |
3697 | * If @retry_unclustered is true then we've already waited on this | |
3698 | * block group once and should move on to the next block group. | |
3699 | */ | |
3700 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3701 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3702 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3703 | ffe_ctl->empty_size); | |
e1a41848 QW |
3704 | ffe_ctl->retry_unclustered = true; |
3705 | return -EAGAIN; | |
3706 | } else if (!offset) { | |
3707 | return 1; | |
3708 | } | |
3709 | ffe_ctl->found_offset = offset; | |
3710 | return 0; | |
3711 | } | |
3712 | ||
c668690d NA |
3713 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3714 | struct find_free_extent_ctl *ffe_ctl, | |
3715 | struct btrfs_block_group **bg_ret) | |
3716 | { | |
3717 | int ret; | |
3718 | ||
3719 | /* We want to try and use the cluster allocator, so lets look there */ | |
3720 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3721 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
c668690d NA |
3722 | if (ret >= 0 || ret == -EAGAIN) |
3723 | return ret; | |
3724 | /* ret == -ENOENT case falls through */ | |
3725 | } | |
3726 | ||
897cae79 | 3727 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3728 | } |
3729 | ||
40ab3be1 NA |
3730 | /* |
3731 | * Tree-log block group locking | |
3732 | * ============================ | |
3733 | * | |
3734 | * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which | |
3735 | * indicates the starting address of a block group, which is reserved only | |
3736 | * for tree-log metadata. | |
3737 | * | |
3738 | * Lock nesting | |
3739 | * ============ | |
3740 | * | |
3741 | * space_info::lock | |
3742 | * block_group::lock | |
3743 | * fs_info::treelog_bg_lock | |
3744 | */ | |
3745 | ||
2eda5708 NA |
3746 | /* |
3747 | * Simple allocator for sequential-only block group. It only allows sequential | |
3748 | * allocation. No need to play with trees. This function also reserves the | |
3749 | * bytes as in btrfs_add_reserved_bytes. | |
3750 | */ | |
3751 | static int do_allocation_zoned(struct btrfs_block_group *block_group, | |
3752 | struct find_free_extent_ctl *ffe_ctl, | |
3753 | struct btrfs_block_group **bg_ret) | |
3754 | { | |
40ab3be1 | 3755 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
2eda5708 NA |
3756 | struct btrfs_space_info *space_info = block_group->space_info; |
3757 | struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl; | |
3758 | u64 start = block_group->start; | |
3759 | u64 num_bytes = ffe_ctl->num_bytes; | |
3760 | u64 avail; | |
40ab3be1 NA |
3761 | u64 bytenr = block_group->start; |
3762 | u64 log_bytenr; | |
763e2672 | 3763 | u64 data_reloc_bytenr; |
2eda5708 | 3764 | int ret = 0; |
40ab3be1 | 3765 | bool skip; |
2eda5708 NA |
3766 | |
3767 | ASSERT(btrfs_is_zoned(block_group->fs_info)); | |
3768 | ||
40ab3be1 NA |
3769 | /* |
3770 | * Do not allow non-tree-log blocks in the dedicated tree-log block | |
3771 | * group, and vice versa. | |
3772 | */ | |
3773 | spin_lock(&fs_info->treelog_bg_lock); | |
3774 | log_bytenr = fs_info->treelog_bg; | |
3775 | skip = log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) || | |
3776 | (!ffe_ctl->for_treelog && bytenr == log_bytenr)); | |
3777 | spin_unlock(&fs_info->treelog_bg_lock); | |
3778 | if (skip) | |
3779 | return 1; | |
3780 | ||
763e2672 JT |
3781 | /* |
3782 | * Do not allow non-relocation blocks in the dedicated relocation block | |
3783 | * group, and vice versa. | |
3784 | */ | |
3785 | spin_lock(&fs_info->relocation_bg_lock); | |
3786 | data_reloc_bytenr = fs_info->data_reloc_bg; | |
3787 | if (data_reloc_bytenr && | |
3788 | ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) || | |
3789 | (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr))) | |
3790 | skip = true; | |
3791 | spin_unlock(&fs_info->relocation_bg_lock); | |
3792 | if (skip) | |
3793 | return 1; | |
3794 | ||
2eda5708 NA |
3795 | spin_lock(&space_info->lock); |
3796 | spin_lock(&block_group->lock); | |
40ab3be1 | 3797 | spin_lock(&fs_info->treelog_bg_lock); |
763e2672 | 3798 | spin_lock(&fs_info->relocation_bg_lock); |
40ab3be1 NA |
3799 | |
3800 | ASSERT(!ffe_ctl->for_treelog || | |
3801 | block_group->start == fs_info->treelog_bg || | |
3802 | fs_info->treelog_bg == 0); | |
763e2672 JT |
3803 | ASSERT(!ffe_ctl->for_data_reloc || |
3804 | block_group->start == fs_info->data_reloc_bg || | |
3805 | fs_info->data_reloc_bg == 0); | |
2eda5708 | 3806 | |
bf2e8c79 | 3807 | if (block_group->ro || block_group->zoned_data_reloc_ongoing) { |
2eda5708 NA |
3808 | ret = 1; |
3809 | goto out; | |
3810 | } | |
3811 | ||
40ab3be1 NA |
3812 | /* |
3813 | * Do not allow currently using block group to be tree-log dedicated | |
3814 | * block group. | |
3815 | */ | |
3816 | if (ffe_ctl->for_treelog && !fs_info->treelog_bg && | |
3817 | (block_group->used || block_group->reserved)) { | |
3818 | ret = 1; | |
3819 | goto out; | |
3820 | } | |
3821 | ||
763e2672 JT |
3822 | /* |
3823 | * Do not allow currently used block group to be the data relocation | |
3824 | * dedicated block group. | |
3825 | */ | |
3826 | if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg && | |
3827 | (block_group->used || block_group->reserved)) { | |
3828 | ret = 1; | |
3829 | goto out; | |
3830 | } | |
3831 | ||
2eda5708 NA |
3832 | avail = block_group->length - block_group->alloc_offset; |
3833 | if (avail < num_bytes) { | |
3834 | if (ffe_ctl->max_extent_size < avail) { | |
3835 | /* | |
3836 | * With sequential allocator, free space is always | |
3837 | * contiguous | |
3838 | */ | |
3839 | ffe_ctl->max_extent_size = avail; | |
3840 | ffe_ctl->total_free_space = avail; | |
3841 | } | |
3842 | ret = 1; | |
3843 | goto out; | |
3844 | } | |
3845 | ||
40ab3be1 NA |
3846 | if (ffe_ctl->for_treelog && !fs_info->treelog_bg) |
3847 | fs_info->treelog_bg = block_group->start; | |
3848 | ||
763e2672 JT |
3849 | if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg) |
3850 | fs_info->data_reloc_bg = block_group->start; | |
3851 | ||
2eda5708 NA |
3852 | ffe_ctl->found_offset = start + block_group->alloc_offset; |
3853 | block_group->alloc_offset += num_bytes; | |
3854 | spin_lock(&ctl->tree_lock); | |
3855 | ctl->free_space -= num_bytes; | |
3856 | spin_unlock(&ctl->tree_lock); | |
3857 | ||
3858 | /* | |
3859 | * We do not check if found_offset is aligned to stripesize. The | |
3860 | * address is anyway rewritten when using zone append writing. | |
3861 | */ | |
3862 | ||
3863 | ffe_ctl->search_start = ffe_ctl->found_offset; | |
3864 | ||
3865 | out: | |
40ab3be1 NA |
3866 | if (ret && ffe_ctl->for_treelog) |
3867 | fs_info->treelog_bg = 0; | |
bf2e8c79 NA |
3868 | if (ret && ffe_ctl->for_data_reloc && |
3869 | fs_info->data_reloc_bg == block_group->start) { | |
3870 | /* | |
3871 | * Do not allow further allocations from this block group. | |
3872 | * Compared to increasing the ->ro, setting the | |
3873 | * ->zoned_data_reloc_ongoing flag still allows nocow | |
3874 | * writers to come in. See btrfs_inc_nocow_writers(). | |
3875 | * | |
3876 | * We need to disable an allocation to avoid an allocation of | |
3877 | * regular (non-relocation data) extent. With mix of relocation | |
3878 | * extents and regular extents, we can dispatch WRITE commands | |
3879 | * (for relocation extents) and ZONE APPEND commands (for | |
3880 | * regular extents) at the same time to the same zone, which | |
3881 | * easily break the write pointer. | |
3882 | */ | |
3883 | block_group->zoned_data_reloc_ongoing = 1; | |
763e2672 | 3884 | fs_info->data_reloc_bg = 0; |
bf2e8c79 | 3885 | } |
763e2672 | 3886 | spin_unlock(&fs_info->relocation_bg_lock); |
40ab3be1 | 3887 | spin_unlock(&fs_info->treelog_bg_lock); |
2eda5708 NA |
3888 | spin_unlock(&block_group->lock); |
3889 | spin_unlock(&space_info->lock); | |
3890 | return ret; | |
3891 | } | |
3892 | ||
c668690d NA |
3893 | static int do_allocation(struct btrfs_block_group *block_group, |
3894 | struct find_free_extent_ctl *ffe_ctl, | |
3895 | struct btrfs_block_group **bg_ret) | |
3896 | { | |
3897 | switch (ffe_ctl->policy) { | |
3898 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3899 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
2eda5708 NA |
3900 | case BTRFS_EXTENT_ALLOC_ZONED: |
3901 | return do_allocation_zoned(block_group, ffe_ctl, bg_ret); | |
c668690d NA |
3902 | default: |
3903 | BUG(); | |
3904 | } | |
3905 | } | |
3906 | ||
baba5062 NA |
3907 | static void release_block_group(struct btrfs_block_group *block_group, |
3908 | struct find_free_extent_ctl *ffe_ctl, | |
3909 | int delalloc) | |
3910 | { | |
3911 | switch (ffe_ctl->policy) { | |
3912 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3913 | ffe_ctl->retry_clustered = false; | |
3914 | ffe_ctl->retry_unclustered = false; | |
3915 | break; | |
2eda5708 NA |
3916 | case BTRFS_EXTENT_ALLOC_ZONED: |
3917 | /* Nothing to do */ | |
3918 | break; | |
baba5062 NA |
3919 | default: |
3920 | BUG(); | |
3921 | } | |
3922 | ||
3923 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
3924 | ffe_ctl->index); | |
3925 | btrfs_release_block_group(block_group, delalloc); | |
3926 | } | |
3927 | ||
0ab9724b NA |
3928 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
3929 | struct btrfs_key *ins) | |
3930 | { | |
3931 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3932 | ||
3933 | if (!ffe_ctl->use_cluster && last_ptr) { | |
3934 | spin_lock(&last_ptr->lock); | |
3935 | last_ptr->window_start = ins->objectid; | |
3936 | spin_unlock(&last_ptr->lock); | |
3937 | } | |
3938 | } | |
3939 | ||
3940 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
3941 | struct btrfs_key *ins) | |
3942 | { | |
3943 | switch (ffe_ctl->policy) { | |
3944 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3945 | found_extent_clustered(ffe_ctl, ins); | |
3946 | break; | |
2eda5708 NA |
3947 | case BTRFS_EXTENT_ALLOC_ZONED: |
3948 | /* Nothing to do */ | |
3949 | break; | |
0ab9724b NA |
3950 | default: |
3951 | BUG(); | |
3952 | } | |
3953 | } | |
3954 | ||
c70e2139 NA |
3955 | static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl) |
3956 | { | |
3957 | switch (ffe_ctl->policy) { | |
3958 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3959 | /* | |
3960 | * If we can't allocate a new chunk we've already looped through | |
3961 | * at least once, move on to the NO_EMPTY_SIZE case. | |
3962 | */ | |
3963 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3964 | return 0; | |
2eda5708 NA |
3965 | case BTRFS_EXTENT_ALLOC_ZONED: |
3966 | /* Give up here */ | |
3967 | return -ENOSPC; | |
c70e2139 NA |
3968 | default: |
3969 | BUG(); | |
3970 | } | |
3971 | } | |
3972 | ||
e72d79d6 QW |
3973 | /* |
3974 | * Return >0 means caller needs to re-search for free extent | |
3975 | * Return 0 means we have the needed free extent. | |
3976 | * Return <0 means we failed to locate any free extent. | |
3977 | */ | |
3978 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
3979 | struct btrfs_key *ins, |
3980 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 3981 | bool full_search) |
e72d79d6 QW |
3982 | { |
3983 | struct btrfs_root *root = fs_info->extent_root; | |
3984 | int ret; | |
3985 | ||
3986 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3987 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3988 | ffe_ctl->orig_have_caching_bg = true; | |
3989 | ||
3990 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3991 | ffe_ctl->have_caching_bg) | |
3992 | return 1; | |
3993 | ||
3994 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3995 | return 1; | |
3996 | ||
3997 | if (ins->objectid) { | |
0ab9724b | 3998 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
3999 | return 0; |
4000 | } | |
4001 | ||
4002 | /* | |
4003 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
4004 | * caching kthreads as we move along | |
4005 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
4006 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
4007 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
4008 | * again | |
4009 | */ | |
4010 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
4011 | ffe_ctl->index = 0; | |
4012 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
4013 | /* | |
4014 | * We want to skip the LOOP_CACHING_WAIT step if we | |
4015 | * don't have any uncached bgs and we've already done a | |
4016 | * full search through. | |
4017 | */ | |
4018 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
4019 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
4020 | else | |
4021 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
4022 | } else { | |
4023 | ffe_ctl->loop++; | |
4024 | } | |
4025 | ||
4026 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
4027 | struct btrfs_trans_handle *trans; | |
4028 | int exist = 0; | |
4029 | ||
4030 | trans = current->journal_info; | |
4031 | if (trans) | |
4032 | exist = 1; | |
4033 | else | |
4034 | trans = btrfs_join_transaction(root); | |
4035 | ||
4036 | if (IS_ERR(trans)) { | |
4037 | ret = PTR_ERR(trans); | |
4038 | return ret; | |
4039 | } | |
4040 | ||
fc471cb0 JB |
4041 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
4042 | CHUNK_ALLOC_FORCE); | |
e72d79d6 | 4043 | |
e72d79d6 | 4044 | /* Do not bail out on ENOSPC since we can do more. */ |
c70e2139 NA |
4045 | if (ret == -ENOSPC) |
4046 | ret = chunk_allocation_failed(ffe_ctl); | |
4047 | else if (ret < 0) | |
e72d79d6 QW |
4048 | btrfs_abort_transaction(trans, ret); |
4049 | else | |
4050 | ret = 0; | |
4051 | if (!exist) | |
4052 | btrfs_end_transaction(trans); | |
4053 | if (ret) | |
4054 | return ret; | |
4055 | } | |
4056 | ||
4057 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
4058 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
4059 | return -ENOSPC; | |
4060 | ||
e72d79d6 QW |
4061 | /* |
4062 | * Don't loop again if we already have no empty_size and | |
4063 | * no empty_cluster. | |
4064 | */ | |
4065 | if (ffe_ctl->empty_size == 0 && | |
4066 | ffe_ctl->empty_cluster == 0) | |
4067 | return -ENOSPC; | |
4068 | ffe_ctl->empty_size = 0; | |
4069 | ffe_ctl->empty_cluster = 0; | |
4070 | } | |
4071 | return 1; | |
4072 | } | |
4073 | return -ENOSPC; | |
4074 | } | |
4075 | ||
7e895409 NA |
4076 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
4077 | struct find_free_extent_ctl *ffe_ctl, | |
4078 | struct btrfs_space_info *space_info, | |
4079 | struct btrfs_key *ins) | |
4080 | { | |
4081 | /* | |
4082 | * If our free space is heavily fragmented we may not be able to make | |
4083 | * big contiguous allocations, so instead of doing the expensive search | |
4084 | * for free space, simply return ENOSPC with our max_extent_size so we | |
4085 | * can go ahead and search for a more manageable chunk. | |
4086 | * | |
4087 | * If our max_extent_size is large enough for our allocation simply | |
4088 | * disable clustering since we will likely not be able to find enough | |
4089 | * space to create a cluster and induce latency trying. | |
4090 | */ | |
4091 | if (space_info->max_extent_size) { | |
4092 | spin_lock(&space_info->lock); | |
4093 | if (space_info->max_extent_size && | |
4094 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
4095 | ins->offset = space_info->max_extent_size; | |
4096 | spin_unlock(&space_info->lock); | |
4097 | return -ENOSPC; | |
4098 | } else if (space_info->max_extent_size) { | |
4099 | ffe_ctl->use_cluster = false; | |
4100 | } | |
4101 | spin_unlock(&space_info->lock); | |
4102 | } | |
4103 | ||
4104 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
4105 | &ffe_ctl->empty_cluster); | |
4106 | if (ffe_ctl->last_ptr) { | |
4107 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
4108 | ||
4109 | spin_lock(&last_ptr->lock); | |
4110 | if (last_ptr->block_group) | |
4111 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4112 | if (last_ptr->fragmented) { | |
4113 | /* | |
4114 | * We still set window_start so we can keep track of the | |
4115 | * last place we found an allocation to try and save | |
4116 | * some time. | |
4117 | */ | |
4118 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4119 | ffe_ctl->use_cluster = false; | |
4120 | } | |
4121 | spin_unlock(&last_ptr->lock); | |
4122 | } | |
4123 | ||
4124 | return 0; | |
4125 | } | |
4126 | ||
4127 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
4128 | struct find_free_extent_ctl *ffe_ctl, | |
4129 | struct btrfs_space_info *space_info, | |
4130 | struct btrfs_key *ins) | |
4131 | { | |
4132 | switch (ffe_ctl->policy) { | |
4133 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4134 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
4135 | space_info, ins); | |
2eda5708 | 4136 | case BTRFS_EXTENT_ALLOC_ZONED: |
40ab3be1 NA |
4137 | if (ffe_ctl->for_treelog) { |
4138 | spin_lock(&fs_info->treelog_bg_lock); | |
4139 | if (fs_info->treelog_bg) | |
4140 | ffe_ctl->hint_byte = fs_info->treelog_bg; | |
4141 | spin_unlock(&fs_info->treelog_bg_lock); | |
4142 | } | |
763e2672 JT |
4143 | if (ffe_ctl->for_data_reloc) { |
4144 | spin_lock(&fs_info->relocation_bg_lock); | |
4145 | if (fs_info->data_reloc_bg) | |
4146 | ffe_ctl->hint_byte = fs_info->data_reloc_bg; | |
4147 | spin_unlock(&fs_info->relocation_bg_lock); | |
4148 | } | |
2eda5708 | 4149 | return 0; |
7e895409 NA |
4150 | default: |
4151 | BUG(); | |
4152 | } | |
4153 | } | |
4154 | ||
fec577fb CM |
4155 | /* |
4156 | * walks the btree of allocated extents and find a hole of a given size. | |
4157 | * The key ins is changed to record the hole: | |
a4820398 | 4158 | * ins->objectid == start position |
62e2749e | 4159 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4160 | * ins->offset == the size of the hole. |
fec577fb | 4161 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4162 | * |
4163 | * If there is no suitable free space, we will record the max size of | |
4164 | * the free space extent currently. | |
e72d79d6 QW |
4165 | * |
4166 | * The overall logic and call chain: | |
4167 | * | |
4168 | * find_free_extent() | |
4169 | * |- Iterate through all block groups | |
4170 | * | |- Get a valid block group | |
4171 | * | |- Try to do clustered allocation in that block group | |
4172 | * | |- Try to do unclustered allocation in that block group | |
4173 | * | |- Check if the result is valid | |
4174 | * | | |- If valid, then exit | |
4175 | * | |- Jump to next block group | |
4176 | * | | |
4177 | * |- Push harder to find free extents | |
4178 | * |- If not found, re-iterate all block groups | |
fec577fb | 4179 | */ |
437490fe | 4180 | static noinline int find_free_extent(struct btrfs_root *root, |
18513091 | 4181 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
ea544149 | 4182 | u64 hint_byte_orig, struct btrfs_key *ins, |
18513091 | 4183 | u64 flags, int delalloc) |
fec577fb | 4184 | { |
437490fe | 4185 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4186 | int ret = 0; |
db8fe64f | 4187 | int cache_block_group_error = 0; |
32da5386 | 4188 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 4189 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 4190 | struct btrfs_space_info *space_info; |
a5e681d9 | 4191 | bool full_search = false; |
40ab3be1 | 4192 | bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
763e2672 JT |
4193 | bool for_data_reloc = (btrfs_is_data_reloc_root(root) && |
4194 | flags & BTRFS_BLOCK_GROUP_DATA); | |
fec577fb | 4195 | |
0b246afa | 4196 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 4197 | |
b4bd745d QW |
4198 | ffe_ctl.num_bytes = num_bytes; |
4199 | ffe_ctl.empty_size = empty_size; | |
4200 | ffe_ctl.flags = flags; | |
4201 | ffe_ctl.search_start = 0; | |
b4bd745d QW |
4202 | ffe_ctl.delalloc = delalloc; |
4203 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
4204 | ffe_ctl.have_caching_bg = false; | |
4205 | ffe_ctl.orig_have_caching_bg = false; | |
4206 | ffe_ctl.found_offset = 0; | |
ea544149 | 4207 | ffe_ctl.hint_byte = hint_byte_orig; |
40ab3be1 | 4208 | ffe_ctl.for_treelog = for_treelog; |
763e2672 | 4209 | ffe_ctl.for_data_reloc = for_data_reloc; |
cb2f96f8 | 4210 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED; |
b4bd745d | 4211 | |
c10859be NA |
4212 | /* For clustered allocation */ |
4213 | ffe_ctl.retry_clustered = false; | |
4214 | ffe_ctl.retry_unclustered = false; | |
4215 | ffe_ctl.last_ptr = NULL; | |
4216 | ffe_ctl.use_cluster = true; | |
4217 | ||
2eda5708 NA |
4218 | if (btrfs_is_zoned(fs_info)) |
4219 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_ZONED; | |
4220 | ||
962a298f | 4221 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4222 | ins->objectid = 0; |
4223 | ins->offset = 0; | |
b1a4d965 | 4224 | |
437490fe | 4225 | trace_find_free_extent(root, num_bytes, empty_size, flags); |
3f7de037 | 4226 | |
280c2908 | 4227 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 4228 | if (!space_info) { |
0b246afa | 4229 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
4230 | return -ENOSPC; |
4231 | } | |
2552d17e | 4232 | |
7e895409 NA |
4233 | ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins); |
4234 | if (ret < 0) | |
4235 | return ret; | |
fa9c0d79 | 4236 | |
b4bd745d QW |
4237 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
4238 | first_logical_byte(fs_info, 0)); | |
ea544149 NA |
4239 | ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte); |
4240 | if (ffe_ctl.search_start == ffe_ctl.hint_byte) { | |
b4bd745d QW |
4241 | block_group = btrfs_lookup_block_group(fs_info, |
4242 | ffe_ctl.search_start); | |
817d52f8 JB |
4243 | /* |
4244 | * we don't want to use the block group if it doesn't match our | |
4245 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4246 | * |
4247 | * However if we are re-searching with an ideal block group | |
4248 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4249 | */ |
b6919a58 | 4250 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 4251 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4252 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4253 | if (list_empty(&block_group->list) || |
4254 | block_group->ro) { | |
4255 | /* | |
4256 | * someone is removing this block group, | |
4257 | * we can't jump into the have_block_group | |
4258 | * target because our list pointers are not | |
4259 | * valid | |
4260 | */ | |
4261 | btrfs_put_block_group(block_group); | |
4262 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4263 | } else { |
b4bd745d | 4264 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 4265 | block_group->flags); |
e570fd27 | 4266 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 4267 | goto have_block_group; |
ccf0e725 | 4268 | } |
2552d17e | 4269 | } else if (block_group) { |
fa9c0d79 | 4270 | btrfs_put_block_group(block_group); |
2552d17e | 4271 | } |
42e70e7a | 4272 | } |
2552d17e | 4273 | search: |
b4bd745d QW |
4274 | ffe_ctl.have_caching_bg = false; |
4275 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
4276 | ffe_ctl.index == 0) | |
a5e681d9 | 4277 | full_search = true; |
80eb234a | 4278 | down_read(&space_info->groups_sem); |
b4bd745d QW |
4279 | list_for_each_entry(block_group, |
4280 | &space_info->block_groups[ffe_ctl.index], list) { | |
c668690d NA |
4281 | struct btrfs_block_group *bg_ret; |
4282 | ||
14443937 | 4283 | /* If the block group is read-only, we can skip it entirely. */ |
40ab3be1 NA |
4284 | if (unlikely(block_group->ro)) { |
4285 | if (for_treelog) | |
4286 | btrfs_clear_treelog_bg(block_group); | |
763e2672 JT |
4287 | if (ffe_ctl.for_data_reloc) |
4288 | btrfs_clear_data_reloc_bg(block_group); | |
14443937 | 4289 | continue; |
40ab3be1 | 4290 | } |
14443937 | 4291 | |
e570fd27 | 4292 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 4293 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 4294 | |
83a50de9 CM |
4295 | /* |
4296 | * this can happen if we end up cycling through all the | |
4297 | * raid types, but we want to make sure we only allocate | |
4298 | * for the proper type. | |
4299 | */ | |
b6919a58 | 4300 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 4301 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4302 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4303 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4304 | BTRFS_BLOCK_GROUP_RAID10; |
4305 | ||
4306 | /* | |
4307 | * if they asked for extra copies and this block group | |
4308 | * doesn't provide them, bail. This does allow us to | |
4309 | * fill raid0 from raid1. | |
4310 | */ | |
b6919a58 | 4311 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4312 | goto loop; |
2a28468e QW |
4313 | |
4314 | /* | |
4315 | * This block group has different flags than we want. | |
4316 | * It's possible that we have MIXED_GROUP flag but no | |
4317 | * block group is mixed. Just skip such block group. | |
4318 | */ | |
4319 | btrfs_release_block_group(block_group, delalloc); | |
4320 | continue; | |
83a50de9 CM |
4321 | } |
4322 | ||
2552d17e | 4323 | have_block_group: |
32da5386 | 4324 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
4325 | if (unlikely(!ffe_ctl.cached)) { |
4326 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 4327 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
4328 | |
4329 | /* | |
4330 | * If we get ENOMEM here or something else we want to | |
4331 | * try other block groups, because it may not be fatal. | |
4332 | * However if we can't find anything else we need to | |
4333 | * save our return here so that we return the actual | |
4334 | * error that caused problems, not ENOSPC. | |
4335 | */ | |
4336 | if (ret < 0) { | |
4337 | if (!cache_block_group_error) | |
4338 | cache_block_group_error = ret; | |
4339 | ret = 0; | |
4340 | goto loop; | |
4341 | } | |
1d4284bd | 4342 | ret = 0; |
817d52f8 JB |
4343 | } |
4344 | ||
36cce922 JB |
4345 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
4346 | goto loop; | |
0f9dd46c | 4347 | |
c668690d NA |
4348 | bg_ret = NULL; |
4349 | ret = do_allocation(block_group, &ffe_ctl, &bg_ret); | |
4350 | if (ret == 0) { | |
4351 | if (bg_ret && bg_ret != block_group) { | |
4352 | btrfs_release_block_group(block_group, delalloc); | |
4353 | block_group = bg_ret; | |
fa9c0d79 | 4354 | } |
c668690d | 4355 | } else if (ret == -EAGAIN) { |
817d52f8 | 4356 | goto have_block_group; |
c668690d | 4357 | } else if (ret > 0) { |
1cdda9b8 | 4358 | goto loop; |
c668690d NA |
4359 | } |
4360 | ||
4361 | /* Checks */ | |
b4bd745d QW |
4362 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4363 | fs_info->stripesize); | |
25179201 | 4364 | |
2552d17e | 4365 | /* move on to the next group */ |
b4bd745d | 4366 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4367 | block_group->start + block_group->length) { |
2eda5708 NA |
4368 | btrfs_add_free_space_unused(block_group, |
4369 | ffe_ctl.found_offset, num_bytes); | |
2552d17e | 4370 | goto loop; |
6226cb0a | 4371 | } |
f5a31e16 | 4372 | |
b4bd745d | 4373 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
2eda5708 NA |
4374 | btrfs_add_free_space_unused(block_group, |
4375 | ffe_ctl.found_offset, | |
4376 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4377 | |
18513091 WX |
4378 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4379 | num_bytes, delalloc); | |
f0486c68 | 4380 | if (ret == -EAGAIN) { |
2eda5708 NA |
4381 | btrfs_add_free_space_unused(block_group, |
4382 | ffe_ctl.found_offset, num_bytes); | |
2552d17e | 4383 | goto loop; |
0f9dd46c | 4384 | } |
9cfa3e34 | 4385 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4386 | |
f0486c68 | 4387 | /* we are all good, lets return */ |
b4bd745d | 4388 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4389 | ins->offset = num_bytes; |
d2fb3437 | 4390 | |
b4bd745d QW |
4391 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4392 | num_bytes); | |
e570fd27 | 4393 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4394 | break; |
4395 | loop: | |
baba5062 | 4396 | release_block_group(block_group, &ffe_ctl, delalloc); |
14443937 | 4397 | cond_resched(); |
2552d17e JB |
4398 | } |
4399 | up_read(&space_info->groups_sem); | |
4400 | ||
15b7ee65 | 4401 | ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search); |
e72d79d6 | 4402 | if (ret > 0) |
b742bb82 YZ |
4403 | goto search; |
4404 | ||
db8fe64f | 4405 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4406 | /* |
4407 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4408 | * any contiguous hole. | |
4409 | */ | |
4410 | if (!ffe_ctl.max_extent_size) | |
4411 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4412 | spin_lock(&space_info->lock); |
b4bd745d | 4413 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4414 | spin_unlock(&space_info->lock); |
b4bd745d | 4415 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4416 | } else if (ret == -ENOSPC) { |
4417 | ret = cache_block_group_error; | |
4f4db217 | 4418 | } |
0f70abe2 | 4419 | return ret; |
fec577fb | 4420 | } |
ec44a35c | 4421 | |
6f47c706 NB |
4422 | /* |
4423 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4424 | * hole that is at least as big as @num_bytes. | |
4425 | * | |
4426 | * @root - The root that will contain this extent | |
4427 | * | |
4428 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4429 | * is used for accounting purposes. This value differs | |
4430 | * from @num_bytes only in the case of compressed extents. | |
4431 | * | |
4432 | * @num_bytes - Number of bytes to allocate on-disk. | |
4433 | * | |
4434 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4435 | * allocator should try to satisfy. In some cases | |
4436 | * @num_bytes may be larger than what is required and if | |
4437 | * the filesystem is fragmented then allocation fails. | |
4438 | * However, the presence of @min_alloc_size gives a | |
4439 | * chance to try and satisfy the smaller allocation. | |
4440 | * | |
4441 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4442 | * size in bytes the allocator should try to find free | |
4443 | * next to the block it returns. This is just a hint and | |
4444 | * may be ignored by the allocator. | |
4445 | * | |
4446 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4447 | * address passed. It might be ignored. | |
4448 | * | |
4449 | * @ins - This key is modified to record the found hole. It will | |
4450 | * have the following values: | |
4451 | * ins->objectid == start position | |
4452 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4453 | * ins->offset == the size of the hole. | |
4454 | * | |
4455 | * @is_data - Boolean flag indicating whether an extent is | |
4456 | * allocated for data (true) or metadata (false) | |
4457 | * | |
4458 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4459 | * delalloc or not. If 'true' data_rwsem of block groups | |
4460 | * is going to be acquired. | |
4461 | * | |
4462 | * | |
4463 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4464 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4465 | * largest available hole the allocator managed to find. | |
4466 | */ | |
18513091 | 4467 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4468 | u64 num_bytes, u64 min_alloc_size, |
4469 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4470 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4471 | { |
ab8d0fc4 | 4472 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4473 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4474 | u64 flags; |
fec577fb | 4475 | int ret; |
40ab3be1 | 4476 | bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
763e2672 | 4477 | bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data); |
925baedd | 4478 | |
1b86826d | 4479 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4480 | again: |
0b246afa | 4481 | WARN_ON(num_bytes < fs_info->sectorsize); |
437490fe | 4482 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
18513091 | 4483 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4484 | if (!ret && !is_data) { |
ab8d0fc4 | 4485 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4486 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4487 | if (!final_tried && ins->offset) { |
4488 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4489 | num_bytes = round_down(num_bytes, |
0b246afa | 4490 | fs_info->sectorsize); |
9e622d6b | 4491 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4492 | ram_bytes = num_bytes; |
9e622d6b MX |
4493 | if (num_bytes == min_alloc_size) |
4494 | final_tried = true; | |
4495 | goto again; | |
ab8d0fc4 | 4496 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4497 | struct btrfs_space_info *sinfo; |
4498 | ||
280c2908 | 4499 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4500 | btrfs_err(fs_info, |
763e2672 JT |
4501 | "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d", |
4502 | flags, num_bytes, for_treelog, for_data_reloc); | |
53804280 | 4503 | if (sinfo) |
5da6afeb JB |
4504 | btrfs_dump_space_info(fs_info, sinfo, |
4505 | num_bytes, 1); | |
9e622d6b | 4506 | } |
925baedd | 4507 | } |
0f9dd46c JB |
4508 | |
4509 | return ret; | |
e6dcd2dc CM |
4510 | } |
4511 | ||
a0fbf736 NB |
4512 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4513 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4514 | { |
32da5386 | 4515 | struct btrfs_block_group *cache; |
0f9dd46c | 4516 | |
0b246afa | 4517 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4518 | if (!cache) { |
0b246afa JM |
4519 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4520 | start); | |
0f9dd46c JB |
4521 | return -ENOSPC; |
4522 | } | |
1f3c79a2 | 4523 | |
a0fbf736 NB |
4524 | btrfs_add_free_space(cache, start, len); |
4525 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4526 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4527 | ||
fa9c0d79 | 4528 | btrfs_put_block_group(cache); |
a0fbf736 | 4529 | return 0; |
e6dcd2dc CM |
4530 | } |
4531 | ||
7bfc1007 NB |
4532 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, |
4533 | u64 len) | |
e688b725 | 4534 | { |
7ef54d54 | 4535 | struct btrfs_block_group *cache; |
a0fbf736 | 4536 | int ret = 0; |
7ef54d54 | 4537 | |
7bfc1007 | 4538 | cache = btrfs_lookup_block_group(trans->fs_info, start); |
7ef54d54 | 4539 | if (!cache) { |
7bfc1007 NB |
4540 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
4541 | start); | |
7ef54d54 NB |
4542 | return -ENOSPC; |
4543 | } | |
4544 | ||
6690d071 | 4545 | ret = pin_down_extent(trans, cache, start, len, 1); |
7ef54d54 | 4546 | btrfs_put_block_group(cache); |
a0fbf736 | 4547 | return ret; |
e688b725 CM |
4548 | } |
4549 | ||
5d4f98a2 | 4550 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4551 | u64 parent, u64 root_objectid, |
4552 | u64 flags, u64 owner, u64 offset, | |
4553 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4554 | { |
ef89b824 | 4555 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4556 | int ret; |
e6dcd2dc | 4557 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4558 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4559 | struct btrfs_path *path; |
5d4f98a2 YZ |
4560 | struct extent_buffer *leaf; |
4561 | int type; | |
4562 | u32 size; | |
26b8003f | 4563 | |
5d4f98a2 YZ |
4564 | if (parent > 0) |
4565 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4566 | else | |
4567 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4568 | |
5d4f98a2 | 4569 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4570 | |
4571 | path = btrfs_alloc_path(); | |
db5b493a TI |
4572 | if (!path) |
4573 | return -ENOMEM; | |
47e4bb98 | 4574 | |
5d4f98a2 YZ |
4575 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4576 | ins, size); | |
79787eaa JM |
4577 | if (ret) { |
4578 | btrfs_free_path(path); | |
4579 | return ret; | |
4580 | } | |
0f9dd46c | 4581 | |
5d4f98a2 YZ |
4582 | leaf = path->nodes[0]; |
4583 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4584 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4585 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4586 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4587 | btrfs_set_extent_flags(leaf, extent_item, | |
4588 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4589 | ||
4590 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4591 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4592 | if (parent > 0) { | |
4593 | struct btrfs_shared_data_ref *ref; | |
4594 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4595 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4596 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4597 | } else { | |
4598 | struct btrfs_extent_data_ref *ref; | |
4599 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4600 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4601 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4602 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4603 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4604 | } | |
47e4bb98 CM |
4605 | |
4606 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4607 | btrfs_free_path(path); |
f510cfec | 4608 | |
25a356d3 | 4609 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4610 | if (ret) |
4611 | return ret; | |
4612 | ||
ade4b516 | 4613 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4614 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4615 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4616 | ins->objectid, ins->offset); |
f5947066 CM |
4617 | BUG(); |
4618 | } | |
71ff6437 | 4619 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4620 | return ret; |
4621 | } | |
4622 | ||
5d4f98a2 | 4623 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4624 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4625 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4626 | { |
9dcdbe01 | 4627 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4628 | int ret; |
5d4f98a2 | 4629 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4630 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4631 | struct btrfs_tree_block_info *block_info; |
4632 | struct btrfs_extent_inline_ref *iref; | |
4633 | struct btrfs_path *path; | |
4634 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4635 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4636 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4637 | u64 num_bytes; |
21ebfbe7 | 4638 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4639 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4640 | |
4e6bd4e0 NB |
4641 | ref = btrfs_delayed_node_to_tree_ref(node); |
4642 | ||
4e6bd4e0 NB |
4643 | extent_key.objectid = node->bytenr; |
4644 | if (skinny_metadata) { | |
4645 | extent_key.offset = ref->level; | |
4646 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4647 | num_bytes = fs_info->nodesize; | |
4648 | } else { | |
4649 | extent_key.offset = node->num_bytes; | |
4650 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4651 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4652 | num_bytes = node->num_bytes; |
4653 | } | |
1c2308f8 | 4654 | |
5d4f98a2 | 4655 | path = btrfs_alloc_path(); |
80ee54bf | 4656 | if (!path) |
d8926bb3 | 4657 | return -ENOMEM; |
56bec294 | 4658 | |
5d4f98a2 | 4659 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4e6bd4e0 | 4660 | &extent_key, size); |
79787eaa | 4661 | if (ret) { |
dd825259 | 4662 | btrfs_free_path(path); |
79787eaa JM |
4663 | return ret; |
4664 | } | |
5d4f98a2 YZ |
4665 | |
4666 | leaf = path->nodes[0]; | |
4667 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4668 | struct btrfs_extent_item); | |
4669 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4670 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4671 | btrfs_set_extent_flags(leaf, extent_item, | |
4672 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4673 | |
3173a18f JB |
4674 | if (skinny_metadata) { |
4675 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4676 | } else { | |
4677 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4678 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4679 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4680 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4681 | } | |
5d4f98a2 | 4682 | |
d4b20733 | 4683 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4684 | btrfs_set_extent_inline_ref_type(leaf, iref, |
4685 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4686 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4687 | } else { |
4688 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4689 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4690 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4691 | } |
4692 | ||
4693 | btrfs_mark_buffer_dirty(leaf); | |
4694 | btrfs_free_path(path); | |
4695 | ||
4e6bd4e0 NB |
4696 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4697 | num_bytes); | |
1e144fb8 OS |
4698 | if (ret) |
4699 | return ret; | |
4700 | ||
ade4b516 JB |
4701 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4702 | fs_info->nodesize, 1); | |
79787eaa | 4703 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4704 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4705 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4706 | BUG(); |
4707 | } | |
0be5dc67 | 4708 | |
4e6bd4e0 | 4709 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4710 | fs_info->nodesize); |
5d4f98a2 YZ |
4711 | return ret; |
4712 | } | |
4713 | ||
4714 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4715 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4716 | u64 offset, u64 ram_bytes, |
4717 | struct btrfs_key *ins) | |
5d4f98a2 | 4718 | { |
76675593 | 4719 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 | 4720 | |
84f7d8e6 | 4721 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4722 | |
76675593 QW |
4723 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4724 | ins->objectid, ins->offset, 0); | |
4158d286 NB |
4725 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, |
4726 | offset, 0, false); | |
8a5040f7 | 4727 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
2187374f JB |
4728 | |
4729 | return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes); | |
e6dcd2dc | 4730 | } |
e02119d5 CM |
4731 | |
4732 | /* | |
4733 | * this is used by the tree logging recovery code. It records that | |
4734 | * an extent has been allocated and makes sure to clear the free | |
4735 | * space cache bits as well | |
4736 | */ | |
5d4f98a2 | 4737 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4738 | u64 root_objectid, u64 owner, u64 offset, |
4739 | struct btrfs_key *ins) | |
e02119d5 | 4740 | { |
61da2abf | 4741 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4742 | int ret; |
32da5386 | 4743 | struct btrfs_block_group *block_group; |
ed7a6948 | 4744 | struct btrfs_space_info *space_info; |
11833d66 | 4745 | |
8c2a1a30 JB |
4746 | /* |
4747 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4748 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4749 | */ |
0b246afa | 4750 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4751 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4752 | ins->offset); | |
b50c6e25 | 4753 | if (ret) |
8c2a1a30 | 4754 | return ret; |
11833d66 YZ |
4755 | } |
4756 | ||
0b246afa | 4757 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4758 | if (!block_group) |
4759 | return -EINVAL; | |
4760 | ||
ed7a6948 WX |
4761 | space_info = block_group->space_info; |
4762 | spin_lock(&space_info->lock); | |
4763 | spin_lock(&block_group->lock); | |
4764 | space_info->bytes_reserved += ins->offset; | |
4765 | block_group->reserved += ins->offset; | |
4766 | spin_unlock(&block_group->lock); | |
4767 | spin_unlock(&space_info->lock); | |
4768 | ||
ef89b824 NB |
4769 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4770 | offset, ins, 1); | |
bd727173 | 4771 | if (ret) |
ab9b2c7b | 4772 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
b50c6e25 | 4773 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4774 | return ret; |
4775 | } | |
4776 | ||
48a3b636 ES |
4777 | static struct extent_buffer * |
4778 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
9631e4cc JB |
4779 | u64 bytenr, int level, u64 owner, |
4780 | enum btrfs_lock_nesting nest) | |
65b51a00 | 4781 | { |
0b246afa | 4782 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4783 | struct extent_buffer *buf; |
4784 | ||
3fbaf258 | 4785 | buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level); |
c871b0f2 LB |
4786 | if (IS_ERR(buf)) |
4787 | return buf; | |
4788 | ||
b72c3aba QW |
4789 | /* |
4790 | * Extra safety check in case the extent tree is corrupted and extent | |
4791 | * allocator chooses to use a tree block which is already used and | |
4792 | * locked. | |
4793 | */ | |
4794 | if (buf->lock_owner == current->pid) { | |
4795 | btrfs_err_rl(fs_info, | |
4796 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4797 | buf->start, btrfs_header_owner(buf), current->pid); | |
4798 | free_extent_buffer(buf); | |
4799 | return ERR_PTR(-EUCLEAN); | |
4800 | } | |
4801 | ||
e114c545 JB |
4802 | /* |
4803 | * This needs to stay, because we could allocate a freed block from an | |
4804 | * old tree into a new tree, so we need to make sure this new block is | |
4805 | * set to the appropriate level and owner. | |
4806 | */ | |
ad244665 | 4807 | btrfs_set_buffer_lockdep_class(owner, buf, level); |
9631e4cc | 4808 | __btrfs_tree_lock(buf, nest); |
6a884d7d | 4809 | btrfs_clean_tree_block(buf); |
3083ee2e | 4810 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
d3575156 | 4811 | clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags); |
b4ce94de | 4812 | |
4db8c528 | 4813 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4814 | |
bc877d28 NB |
4815 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4816 | btrfs_set_header_level(buf, level); | |
4817 | btrfs_set_header_bytenr(buf, buf->start); | |
4818 | btrfs_set_header_generation(buf, trans->transid); | |
4819 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4820 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4821 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4822 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4823 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4824 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4825 | /* |
4826 | * we allow two log transactions at a time, use different | |
52042d8e | 4827 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4828 | */ |
656f30db | 4829 | if (buf->log_index == 0) |
8cef4e16 YZ |
4830 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4831 | buf->start + buf->len - 1, GFP_NOFS); | |
4832 | else | |
4833 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4834 | buf->start + buf->len - 1); |
d0c803c4 | 4835 | } else { |
656f30db | 4836 | buf->log_index = -1; |
d0c803c4 | 4837 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4838 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4839 | } |
b4ce94de | 4840 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4841 | return buf; |
4842 | } | |
4843 | ||
fec577fb | 4844 | /* |
f0486c68 | 4845 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4846 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4847 | */ |
4d75f8a9 | 4848 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4849 | struct btrfs_root *root, |
4850 | u64 parent, u64 root_objectid, | |
4851 | const struct btrfs_disk_key *key, | |
4852 | int level, u64 hint, | |
9631e4cc JB |
4853 | u64 empty_size, |
4854 | enum btrfs_lock_nesting nest) | |
fec577fb | 4855 | { |
0b246afa | 4856 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4857 | struct btrfs_key ins; |
f0486c68 | 4858 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4859 | struct extent_buffer *buf; |
67b7859e | 4860 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4861 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4862 | u64 flags = 0; |
4863 | int ret; | |
0b246afa JM |
4864 | u32 blocksize = fs_info->nodesize; |
4865 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4866 | |
05653ef3 | 4867 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4868 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4869 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
9631e4cc | 4870 | level, root_objectid, nest); |
faa2dbf0 JB |
4871 | if (!IS_ERR(buf)) |
4872 | root->alloc_bytenr += blocksize; | |
4873 | return buf; | |
4874 | } | |
05653ef3 | 4875 | #endif |
fccb84c9 | 4876 | |
67f9c220 | 4877 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4878 | if (IS_ERR(block_rsv)) |
4879 | return ERR_CAST(block_rsv); | |
4880 | ||
18513091 | 4881 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4882 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4883 | if (ret) |
4884 | goto out_unuse; | |
55c69072 | 4885 | |
bc877d28 | 4886 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
9631e4cc | 4887 | root_objectid, nest); |
67b7859e OS |
4888 | if (IS_ERR(buf)) { |
4889 | ret = PTR_ERR(buf); | |
4890 | goto out_free_reserved; | |
4891 | } | |
f0486c68 YZ |
4892 | |
4893 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4894 | if (parent == 0) | |
4895 | parent = ins.objectid; | |
4896 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4897 | } else | |
4898 | BUG_ON(parent > 0); | |
4899 | ||
4900 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4901 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4902 | if (!extent_op) { |
4903 | ret = -ENOMEM; | |
4904 | goto out_free_buf; | |
4905 | } | |
f0486c68 YZ |
4906 | if (key) |
4907 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4908 | else | |
4909 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4910 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4911 | extent_op->update_key = skinny_metadata ? false : true; |
4912 | extent_op->update_flags = true; | |
4913 | extent_op->is_data = false; | |
b1c79e09 | 4914 | extent_op->level = level; |
f0486c68 | 4915 | |
ed4f255b QW |
4916 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4917 | ins.objectid, ins.offset, parent); | |
4918 | generic_ref.real_root = root->root_key.objectid; | |
4158d286 NB |
4919 | btrfs_init_tree_ref(&generic_ref, level, root_objectid, |
4920 | root->root_key.objectid, false); | |
8a5040f7 | 4921 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
2187374f | 4922 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op); |
67b7859e OS |
4923 | if (ret) |
4924 | goto out_free_delayed; | |
f0486c68 | 4925 | } |
fec577fb | 4926 | return buf; |
67b7859e OS |
4927 | |
4928 | out_free_delayed: | |
4929 | btrfs_free_delayed_extent_op(extent_op); | |
4930 | out_free_buf: | |
19ea40dd | 4931 | btrfs_tree_unlock(buf); |
67b7859e OS |
4932 | free_extent_buffer(buf); |
4933 | out_free_reserved: | |
2ff7e61e | 4934 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4935 | out_unuse: |
67f9c220 | 4936 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4937 | return ERR_PTR(ret); |
fec577fb | 4938 | } |
a28ec197 | 4939 | |
2c47e605 YZ |
4940 | struct walk_control { |
4941 | u64 refs[BTRFS_MAX_LEVEL]; | |
4942 | u64 flags[BTRFS_MAX_LEVEL]; | |
4943 | struct btrfs_key update_progress; | |
aea6f028 JB |
4944 | struct btrfs_key drop_progress; |
4945 | int drop_level; | |
2c47e605 YZ |
4946 | int stage; |
4947 | int level; | |
4948 | int shared_level; | |
4949 | int update_ref; | |
4950 | int keep_locks; | |
1c4850e2 YZ |
4951 | int reada_slot; |
4952 | int reada_count; | |
78c52d9e | 4953 | int restarted; |
2c47e605 YZ |
4954 | }; |
4955 | ||
4956 | #define DROP_REFERENCE 1 | |
4957 | #define UPDATE_BACKREF 2 | |
4958 | ||
1c4850e2 YZ |
4959 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4960 | struct btrfs_root *root, | |
4961 | struct walk_control *wc, | |
4962 | struct btrfs_path *path) | |
6407bf6d | 4963 | { |
0b246afa | 4964 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4965 | u64 bytenr; |
4966 | u64 generation; | |
4967 | u64 refs; | |
94fcca9f | 4968 | u64 flags; |
5d4f98a2 | 4969 | u32 nritems; |
1c4850e2 YZ |
4970 | struct btrfs_key key; |
4971 | struct extent_buffer *eb; | |
6407bf6d | 4972 | int ret; |
1c4850e2 YZ |
4973 | int slot; |
4974 | int nread = 0; | |
6407bf6d | 4975 | |
1c4850e2 YZ |
4976 | if (path->slots[wc->level] < wc->reada_slot) { |
4977 | wc->reada_count = wc->reada_count * 2 / 3; | |
4978 | wc->reada_count = max(wc->reada_count, 2); | |
4979 | } else { | |
4980 | wc->reada_count = wc->reada_count * 3 / 2; | |
4981 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4982 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4983 | } |
7bb86316 | 4984 | |
1c4850e2 YZ |
4985 | eb = path->nodes[wc->level]; |
4986 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4987 | |
1c4850e2 YZ |
4988 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4989 | if (nread >= wc->reada_count) | |
4990 | break; | |
bd56b302 | 4991 | |
2dd3e67b | 4992 | cond_resched(); |
1c4850e2 YZ |
4993 | bytenr = btrfs_node_blockptr(eb, slot); |
4994 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4995 | |
1c4850e2 YZ |
4996 | if (slot == path->slots[wc->level]) |
4997 | goto reada; | |
5d4f98a2 | 4998 | |
1c4850e2 YZ |
4999 | if (wc->stage == UPDATE_BACKREF && |
5000 | generation <= root->root_key.offset) | |
bd56b302 CM |
5001 | continue; |
5002 | ||
94fcca9f | 5003 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 5004 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
5005 | wc->level - 1, 1, &refs, |
5006 | &flags); | |
79787eaa JM |
5007 | /* We don't care about errors in readahead. */ |
5008 | if (ret < 0) | |
5009 | continue; | |
94fcca9f YZ |
5010 | BUG_ON(refs == 0); |
5011 | ||
1c4850e2 | 5012 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5013 | if (refs == 1) |
5014 | goto reada; | |
bd56b302 | 5015 | |
94fcca9f YZ |
5016 | if (wc->level == 1 && |
5017 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5018 | continue; | |
1c4850e2 YZ |
5019 | if (!wc->update_ref || |
5020 | generation <= root->root_key.offset) | |
5021 | continue; | |
5022 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5023 | ret = btrfs_comp_cpu_keys(&key, | |
5024 | &wc->update_progress); | |
5025 | if (ret < 0) | |
5026 | continue; | |
94fcca9f YZ |
5027 | } else { |
5028 | if (wc->level == 1 && | |
5029 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5030 | continue; | |
6407bf6d | 5031 | } |
1c4850e2 | 5032 | reada: |
bfb484d9 | 5033 | btrfs_readahead_node_child(eb, slot); |
1c4850e2 | 5034 | nread++; |
20524f02 | 5035 | } |
1c4850e2 | 5036 | wc->reada_slot = slot; |
20524f02 | 5037 | } |
2c47e605 | 5038 | |
f82d02d9 | 5039 | /* |
2c016dc2 | 5040 | * helper to process tree block while walking down the tree. |
2c47e605 | 5041 | * |
2c47e605 YZ |
5042 | * when wc->stage == UPDATE_BACKREF, this function updates |
5043 | * back refs for pointers in the block. | |
5044 | * | |
5045 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5046 | */ |
2c47e605 | 5047 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5048 | struct btrfs_root *root, |
2c47e605 | 5049 | struct btrfs_path *path, |
94fcca9f | 5050 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5051 | { |
2ff7e61e | 5052 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
5053 | int level = wc->level; |
5054 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5055 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5056 | int ret; |
5057 | ||
2c47e605 YZ |
5058 | if (wc->stage == UPDATE_BACKREF && |
5059 | btrfs_header_owner(eb) != root->root_key.objectid) | |
5060 | return 1; | |
f82d02d9 | 5061 | |
2c47e605 YZ |
5062 | /* |
5063 | * when reference count of tree block is 1, it won't increase | |
5064 | * again. once full backref flag is set, we never clear it. | |
5065 | */ | |
94fcca9f YZ |
5066 | if (lookup_info && |
5067 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5068 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 5069 | BUG_ON(!path->locks[level]); |
2ff7e61e | 5070 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5071 | eb->start, level, 1, |
2c47e605 YZ |
5072 | &wc->refs[level], |
5073 | &wc->flags[level]); | |
79787eaa JM |
5074 | BUG_ON(ret == -ENOMEM); |
5075 | if (ret) | |
5076 | return ret; | |
2c47e605 YZ |
5077 | BUG_ON(wc->refs[level] == 0); |
5078 | } | |
5d4f98a2 | 5079 | |
2c47e605 YZ |
5080 | if (wc->stage == DROP_REFERENCE) { |
5081 | if (wc->refs[level] > 1) | |
5082 | return 1; | |
f82d02d9 | 5083 | |
2c47e605 | 5084 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 5085 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5086 | path->locks[level] = 0; |
5087 | } | |
5088 | return 0; | |
5089 | } | |
f82d02d9 | 5090 | |
2c47e605 YZ |
5091 | /* wc->stage == UPDATE_BACKREF */ |
5092 | if (!(wc->flags[level] & flag)) { | |
5093 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 5094 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 5095 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 5096 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5097 | BUG_ON(ret); /* -ENOMEM */ |
42c9d0b5 | 5098 | ret = btrfs_set_disk_extent_flags(trans, eb, flag, |
b1c79e09 | 5099 | btrfs_header_level(eb), 0); |
79787eaa | 5100 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
5101 | wc->flags[level] |= flag; |
5102 | } | |
5103 | ||
5104 | /* | |
5105 | * the block is shared by multiple trees, so it's not good to | |
5106 | * keep the tree lock | |
5107 | */ | |
5108 | if (path->locks[level] && level > 0) { | |
bd681513 | 5109 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5110 | path->locks[level] = 0; |
5111 | } | |
5112 | return 0; | |
5113 | } | |
5114 | ||
78c52d9e JB |
5115 | /* |
5116 | * This is used to verify a ref exists for this root to deal with a bug where we | |
5117 | * would have a drop_progress key that hadn't been updated properly. | |
5118 | */ | |
5119 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
5120 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
5121 | int level) | |
5122 | { | |
5123 | struct btrfs_path *path; | |
5124 | struct btrfs_extent_inline_ref *iref; | |
5125 | int ret; | |
5126 | ||
5127 | path = btrfs_alloc_path(); | |
5128 | if (!path) | |
5129 | return -ENOMEM; | |
5130 | ||
5131 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
5132 | root->fs_info->nodesize, parent, | |
5133 | root->root_key.objectid, level, 0); | |
5134 | btrfs_free_path(path); | |
5135 | if (ret == -ENOENT) | |
5136 | return 0; | |
5137 | if (ret < 0) | |
5138 | return ret; | |
5139 | return 1; | |
5140 | } | |
5141 | ||
1c4850e2 | 5142 | /* |
2c016dc2 | 5143 | * helper to process tree block pointer. |
1c4850e2 YZ |
5144 | * |
5145 | * when wc->stage == DROP_REFERENCE, this function checks | |
5146 | * reference count of the block pointed to. if the block | |
5147 | * is shared and we need update back refs for the subtree | |
5148 | * rooted at the block, this function changes wc->stage to | |
5149 | * UPDATE_BACKREF. if the block is shared and there is no | |
5150 | * need to update back, this function drops the reference | |
5151 | * to the block. | |
5152 | * | |
5153 | * NOTE: return value 1 means we should stop walking down. | |
5154 | */ | |
5155 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5156 | struct btrfs_root *root, | |
5157 | struct btrfs_path *path, | |
94fcca9f | 5158 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 5159 | { |
0b246afa | 5160 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
5161 | u64 bytenr; |
5162 | u64 generation; | |
5163 | u64 parent; | |
1c4850e2 | 5164 | struct btrfs_key key; |
581c1760 | 5165 | struct btrfs_key first_key; |
ffd4bb2a | 5166 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
5167 | struct extent_buffer *next; |
5168 | int level = wc->level; | |
5169 | int reada = 0; | |
5170 | int ret = 0; | |
1152651a | 5171 | bool need_account = false; |
1c4850e2 YZ |
5172 | |
5173 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5174 | path->slots[level]); | |
5175 | /* | |
5176 | * if the lower level block was created before the snapshot | |
5177 | * was created, we know there is no need to update back refs | |
5178 | * for the subtree | |
5179 | */ | |
5180 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5181 | generation <= root->root_key.offset) { |
5182 | *lookup_info = 1; | |
1c4850e2 | 5183 | return 1; |
94fcca9f | 5184 | } |
1c4850e2 YZ |
5185 | |
5186 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
5187 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
5188 | path->slots[level]); | |
1c4850e2 | 5189 | |
0b246afa | 5190 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5191 | if (!next) { |
3fbaf258 JB |
5192 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
5193 | root->root_key.objectid, level - 1); | |
c871b0f2 LB |
5194 | if (IS_ERR(next)) |
5195 | return PTR_ERR(next); | |
1c4850e2 YZ |
5196 | reada = 1; |
5197 | } | |
5198 | btrfs_tree_lock(next); | |
1c4850e2 | 5199 | |
2ff7e61e | 5200 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
5201 | &wc->refs[level - 1], |
5202 | &wc->flags[level - 1]); | |
4867268c JB |
5203 | if (ret < 0) |
5204 | goto out_unlock; | |
79787eaa | 5205 | |
c2cf52eb | 5206 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5207 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5208 | ret = -EIO; |
5209 | goto out_unlock; | |
c2cf52eb | 5210 | } |
94fcca9f | 5211 | *lookup_info = 0; |
1c4850e2 | 5212 | |
94fcca9f | 5213 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5214 | if (wc->refs[level - 1] > 1) { |
1152651a | 5215 | need_account = true; |
94fcca9f YZ |
5216 | if (level == 1 && |
5217 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5218 | goto skip; | |
5219 | ||
1c4850e2 YZ |
5220 | if (!wc->update_ref || |
5221 | generation <= root->root_key.offset) | |
5222 | goto skip; | |
5223 | ||
5224 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5225 | path->slots[level]); | |
5226 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5227 | if (ret < 0) | |
5228 | goto skip; | |
5229 | ||
5230 | wc->stage = UPDATE_BACKREF; | |
5231 | wc->shared_level = level - 1; | |
5232 | } | |
94fcca9f YZ |
5233 | } else { |
5234 | if (level == 1 && | |
5235 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5236 | goto skip; | |
1c4850e2 YZ |
5237 | } |
5238 | ||
b9fab919 | 5239 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5240 | btrfs_tree_unlock(next); |
5241 | free_extent_buffer(next); | |
5242 | next = NULL; | |
94fcca9f | 5243 | *lookup_info = 1; |
1c4850e2 YZ |
5244 | } |
5245 | ||
5246 | if (!next) { | |
5247 | if (reada && level == 1) | |
5248 | reada_walk_down(trans, root, wc, path); | |
1b7ec85e JB |
5249 | next = read_tree_block(fs_info, bytenr, root->root_key.objectid, |
5250 | generation, level - 1, &first_key); | |
64c043de LB |
5251 | if (IS_ERR(next)) { |
5252 | return PTR_ERR(next); | |
5253 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5254 | free_extent_buffer(next); |
97d9a8a4 | 5255 | return -EIO; |
416bc658 | 5256 | } |
1c4850e2 | 5257 | btrfs_tree_lock(next); |
1c4850e2 YZ |
5258 | } |
5259 | ||
5260 | level--; | |
4867268c JB |
5261 | ASSERT(level == btrfs_header_level(next)); |
5262 | if (level != btrfs_header_level(next)) { | |
5263 | btrfs_err(root->fs_info, "mismatched level"); | |
5264 | ret = -EIO; | |
5265 | goto out_unlock; | |
5266 | } | |
1c4850e2 YZ |
5267 | path->nodes[level] = next; |
5268 | path->slots[level] = 0; | |
ac5887c8 | 5269 | path->locks[level] = BTRFS_WRITE_LOCK; |
1c4850e2 YZ |
5270 | wc->level = level; |
5271 | if (wc->level == 1) | |
5272 | wc->reada_slot = 0; | |
5273 | return 0; | |
5274 | skip: | |
5275 | wc->refs[level - 1] = 0; | |
5276 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5277 | if (wc->stage == DROP_REFERENCE) { |
5278 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5279 | parent = path->nodes[level]->start; | |
5280 | } else { | |
4867268c | 5281 | ASSERT(root->root_key.objectid == |
94fcca9f | 5282 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5283 | if (root->root_key.objectid != |
5284 | btrfs_header_owner(path->nodes[level])) { | |
5285 | btrfs_err(root->fs_info, | |
5286 | "mismatched block owner"); | |
5287 | ret = -EIO; | |
5288 | goto out_unlock; | |
5289 | } | |
94fcca9f YZ |
5290 | parent = 0; |
5291 | } | |
1c4850e2 | 5292 | |
78c52d9e JB |
5293 | /* |
5294 | * If we had a drop_progress we need to verify the refs are set | |
5295 | * as expected. If we find our ref then we know that from here | |
5296 | * on out everything should be correct, and we can clear the | |
5297 | * ->restarted flag. | |
5298 | */ | |
5299 | if (wc->restarted) { | |
5300 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5301 | level - 1); | |
5302 | if (ret < 0) | |
5303 | goto out_unlock; | |
5304 | if (ret == 0) | |
5305 | goto no_delete; | |
5306 | ret = 0; | |
5307 | wc->restarted = 0; | |
5308 | } | |
5309 | ||
2cd86d30 QW |
5310 | /* |
5311 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5312 | * already accounted them at merge time (replace_path), | |
5313 | * thus we could skip expensive subtree trace here. | |
5314 | */ | |
5315 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5316 | need_account) { | |
deb40627 | 5317 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5318 | generation, level - 1); |
1152651a | 5319 | if (ret) { |
0b246afa | 5320 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5321 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5322 | ret); | |
1152651a MF |
5323 | } |
5324 | } | |
aea6f028 JB |
5325 | |
5326 | /* | |
5327 | * We need to update the next key in our walk control so we can | |
5328 | * update the drop_progress key accordingly. We don't care if | |
5329 | * find_next_key doesn't find a key because that means we're at | |
5330 | * the end and are going to clean up now. | |
5331 | */ | |
5332 | wc->drop_level = level; | |
5333 | find_next_key(path, level, &wc->drop_progress); | |
5334 | ||
ffd4bb2a QW |
5335 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
5336 | fs_info->nodesize, parent); | |
4158d286 NB |
5337 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid, |
5338 | 0, false); | |
ffd4bb2a | 5339 | ret = btrfs_free_extent(trans, &ref); |
4867268c JB |
5340 | if (ret) |
5341 | goto out_unlock; | |
1c4850e2 | 5342 | } |
78c52d9e | 5343 | no_delete: |
4867268c JB |
5344 | *lookup_info = 1; |
5345 | ret = 1; | |
5346 | ||
5347 | out_unlock: | |
1c4850e2 YZ |
5348 | btrfs_tree_unlock(next); |
5349 | free_extent_buffer(next); | |
4867268c JB |
5350 | |
5351 | return ret; | |
1c4850e2 YZ |
5352 | } |
5353 | ||
2c47e605 | 5354 | /* |
2c016dc2 | 5355 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5356 | * |
5357 | * when wc->stage == DROP_REFERENCE, this function drops | |
5358 | * reference count on the block. | |
5359 | * | |
5360 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5361 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5362 | * to UPDATE_BACKREF previously while processing the block. | |
5363 | * | |
5364 | * NOTE: return value 1 means we should stop walking up. | |
5365 | */ | |
5366 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5367 | struct btrfs_root *root, | |
5368 | struct btrfs_path *path, | |
5369 | struct walk_control *wc) | |
5370 | { | |
0b246afa | 5371 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5372 | int ret; |
2c47e605 YZ |
5373 | int level = wc->level; |
5374 | struct extent_buffer *eb = path->nodes[level]; | |
5375 | u64 parent = 0; | |
5376 | ||
5377 | if (wc->stage == UPDATE_BACKREF) { | |
5378 | BUG_ON(wc->shared_level < level); | |
5379 | if (level < wc->shared_level) | |
5380 | goto out; | |
5381 | ||
2c47e605 YZ |
5382 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5383 | if (ret > 0) | |
5384 | wc->update_ref = 0; | |
5385 | ||
5386 | wc->stage = DROP_REFERENCE; | |
5387 | wc->shared_level = -1; | |
5388 | path->slots[level] = 0; | |
5389 | ||
5390 | /* | |
5391 | * check reference count again if the block isn't locked. | |
5392 | * we should start walking down the tree again if reference | |
5393 | * count is one. | |
5394 | */ | |
5395 | if (!path->locks[level]) { | |
5396 | BUG_ON(level == 0); | |
5397 | btrfs_tree_lock(eb); | |
ac5887c8 | 5398 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5399 | |
2ff7e61e | 5400 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5401 | eb->start, level, 1, |
2c47e605 YZ |
5402 | &wc->refs[level], |
5403 | &wc->flags[level]); | |
79787eaa JM |
5404 | if (ret < 0) { |
5405 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5406 | path->locks[level] = 0; |
79787eaa JM |
5407 | return ret; |
5408 | } | |
2c47e605 YZ |
5409 | BUG_ON(wc->refs[level] == 0); |
5410 | if (wc->refs[level] == 1) { | |
bd681513 | 5411 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5412 | path->locks[level] = 0; |
2c47e605 YZ |
5413 | return 1; |
5414 | } | |
f82d02d9 | 5415 | } |
2c47e605 | 5416 | } |
f82d02d9 | 5417 | |
2c47e605 YZ |
5418 | /* wc->stage == DROP_REFERENCE */ |
5419 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5420 | |
2c47e605 YZ |
5421 | if (wc->refs[level] == 1) { |
5422 | if (level == 0) { | |
5423 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5424 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5425 | else |
e339a6b0 | 5426 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5427 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5428 | if (is_fstree(root->root_key.objectid)) { |
5429 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5430 | if (ret) { | |
5431 | btrfs_err_rl(fs_info, | |
5432 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5433 | ret); |
c4140cbf | 5434 | } |
1152651a | 5435 | } |
2c47e605 | 5436 | } |
6a884d7d | 5437 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5438 | if (!path->locks[level] && |
5439 | btrfs_header_generation(eb) == trans->transid) { | |
5440 | btrfs_tree_lock(eb); | |
ac5887c8 | 5441 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5442 | } |
6a884d7d | 5443 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5444 | } |
5445 | ||
5446 | if (eb == root->node) { | |
5447 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5448 | parent = eb->start; | |
65c6e82b QW |
5449 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5450 | goto owner_mismatch; | |
2c47e605 YZ |
5451 | } else { |
5452 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5453 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5454 | else if (root->root_key.objectid != |
5455 | btrfs_header_owner(path->nodes[level + 1])) | |
5456 | goto owner_mismatch; | |
f82d02d9 | 5457 | } |
f82d02d9 | 5458 | |
4d6fc4ce FM |
5459 | btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent, |
5460 | wc->refs[level] == 1); | |
2c47e605 YZ |
5461 | out: |
5462 | wc->refs[level] = 0; | |
5463 | wc->flags[level] = 0; | |
f0486c68 | 5464 | return 0; |
65c6e82b QW |
5465 | |
5466 | owner_mismatch: | |
5467 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5468 | btrfs_header_owner(eb), root->root_key.objectid); | |
5469 | return -EUCLEAN; | |
2c47e605 YZ |
5470 | } |
5471 | ||
5472 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5473 | struct btrfs_root *root, | |
5474 | struct btrfs_path *path, | |
5475 | struct walk_control *wc) | |
5476 | { | |
2c47e605 | 5477 | int level = wc->level; |
94fcca9f | 5478 | int lookup_info = 1; |
2c47e605 YZ |
5479 | int ret; |
5480 | ||
5481 | while (level >= 0) { | |
94fcca9f | 5482 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5483 | if (ret > 0) |
5484 | break; | |
5485 | ||
5486 | if (level == 0) | |
5487 | break; | |
5488 | ||
7a7965f8 YZ |
5489 | if (path->slots[level] >= |
5490 | btrfs_header_nritems(path->nodes[level])) | |
5491 | break; | |
5492 | ||
94fcca9f | 5493 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5494 | if (ret > 0) { |
5495 | path->slots[level]++; | |
5496 | continue; | |
90d2c51d MX |
5497 | } else if (ret < 0) |
5498 | return ret; | |
1c4850e2 | 5499 | level = wc->level; |
f82d02d9 | 5500 | } |
f82d02d9 YZ |
5501 | return 0; |
5502 | } | |
5503 | ||
d397712b | 5504 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5505 | struct btrfs_root *root, |
f82d02d9 | 5506 | struct btrfs_path *path, |
2c47e605 | 5507 | struct walk_control *wc, int max_level) |
20524f02 | 5508 | { |
2c47e605 | 5509 | int level = wc->level; |
20524f02 | 5510 | int ret; |
9f3a7427 | 5511 | |
2c47e605 YZ |
5512 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5513 | while (level < max_level && path->nodes[level]) { | |
5514 | wc->level = level; | |
5515 | if (path->slots[level] + 1 < | |
5516 | btrfs_header_nritems(path->nodes[level])) { | |
5517 | path->slots[level]++; | |
20524f02 CM |
5518 | return 0; |
5519 | } else { | |
2c47e605 YZ |
5520 | ret = walk_up_proc(trans, root, path, wc); |
5521 | if (ret > 0) | |
5522 | return 0; | |
65c6e82b QW |
5523 | if (ret < 0) |
5524 | return ret; | |
bd56b302 | 5525 | |
2c47e605 | 5526 | if (path->locks[level]) { |
bd681513 CM |
5527 | btrfs_tree_unlock_rw(path->nodes[level], |
5528 | path->locks[level]); | |
2c47e605 | 5529 | path->locks[level] = 0; |
f82d02d9 | 5530 | } |
2c47e605 YZ |
5531 | free_extent_buffer(path->nodes[level]); |
5532 | path->nodes[level] = NULL; | |
5533 | level++; | |
20524f02 CM |
5534 | } |
5535 | } | |
5536 | return 1; | |
5537 | } | |
5538 | ||
9aca1d51 | 5539 | /* |
2c47e605 YZ |
5540 | * drop a subvolume tree. |
5541 | * | |
5542 | * this function traverses the tree freeing any blocks that only | |
5543 | * referenced by the tree. | |
5544 | * | |
5545 | * when a shared tree block is found. this function decreases its | |
5546 | * reference count by one. if update_ref is true, this function | |
5547 | * also make sure backrefs for the shared block and all lower level | |
5548 | * blocks are properly updated. | |
9d1a2a3a DS |
5549 | * |
5550 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5551 | */ |
0078a9f9 | 5552 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5553 | { |
ab8d0fc4 | 5554 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5555 | struct btrfs_path *path; |
2c47e605 | 5556 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5557 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5558 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5559 | struct walk_control *wc; |
5560 | struct btrfs_key key; | |
5561 | int err = 0; | |
5562 | int ret; | |
5563 | int level; | |
d29a9f62 | 5564 | bool root_dropped = false; |
78db1921 | 5565 | bool unfinished_drop = false; |
20524f02 | 5566 | |
4fd786e6 | 5567 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5568 | |
5caf2a00 | 5569 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5570 | if (!path) { |
5571 | err = -ENOMEM; | |
5572 | goto out; | |
5573 | } | |
20524f02 | 5574 | |
2c47e605 | 5575 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5576 | if (!wc) { |
5577 | btrfs_free_path(path); | |
cb1b69f4 TI |
5578 | err = -ENOMEM; |
5579 | goto out; | |
38a1a919 | 5580 | } |
2c47e605 | 5581 | |
f3e3d9cc QW |
5582 | /* |
5583 | * Use join to avoid potential EINTR from transaction start. See | |
5584 | * wait_reserve_ticket and the whole reservation callchain. | |
5585 | */ | |
5586 | if (for_reloc) | |
5587 | trans = btrfs_join_transaction(tree_root); | |
5588 | else | |
5589 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5590 | if (IS_ERR(trans)) { |
5591 | err = PTR_ERR(trans); | |
5592 | goto out_free; | |
5593 | } | |
98d5dc13 | 5594 | |
0568e82d JB |
5595 | err = btrfs_run_delayed_items(trans); |
5596 | if (err) | |
5597 | goto out_end_trans; | |
5598 | ||
83354f07 JB |
5599 | /* |
5600 | * This will help us catch people modifying the fs tree while we're | |
5601 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5602 | * dropped as we unlock the root node and parent nodes as we walk down | |
5603 | * the tree, assuming nothing will change. If something does change | |
5604 | * then we'll have stale information and drop references to blocks we've | |
5605 | * already dropped. | |
5606 | */ | |
5607 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
78db1921 JB |
5608 | unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state); |
5609 | ||
9f3a7427 | 5610 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5611 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5612 | path->nodes[level] = btrfs_lock_root_node(root); |
9f3a7427 | 5613 | path->slots[level] = 0; |
ac5887c8 | 5614 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5615 | memset(&wc->update_progress, 0, |
5616 | sizeof(wc->update_progress)); | |
9f3a7427 | 5617 | } else { |
9f3a7427 | 5618 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5619 | memcpy(&wc->update_progress, &key, |
5620 | sizeof(wc->update_progress)); | |
5621 | ||
c8422684 | 5622 | level = btrfs_root_drop_level(root_item); |
2c47e605 | 5623 | BUG_ON(level == 0); |
6702ed49 | 5624 | path->lowest_level = level; |
2c47e605 YZ |
5625 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5626 | path->lowest_level = 0; | |
5627 | if (ret < 0) { | |
5628 | err = ret; | |
79787eaa | 5629 | goto out_end_trans; |
9f3a7427 | 5630 | } |
1c4850e2 | 5631 | WARN_ON(ret > 0); |
2c47e605 | 5632 | |
7d9eb12c CM |
5633 | /* |
5634 | * unlock our path, this is safe because only this | |
5635 | * function is allowed to delete this snapshot | |
5636 | */ | |
5d4f98a2 | 5637 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5638 | |
5639 | level = btrfs_header_level(root->node); | |
5640 | while (1) { | |
5641 | btrfs_tree_lock(path->nodes[level]); | |
ac5887c8 | 5642 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 | 5643 | |
2ff7e61e | 5644 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5645 | path->nodes[level]->start, |
3173a18f | 5646 | level, 1, &wc->refs[level], |
2c47e605 | 5647 | &wc->flags[level]); |
79787eaa JM |
5648 | if (ret < 0) { |
5649 | err = ret; | |
5650 | goto out_end_trans; | |
5651 | } | |
2c47e605 YZ |
5652 | BUG_ON(wc->refs[level] == 0); |
5653 | ||
c8422684 | 5654 | if (level == btrfs_root_drop_level(root_item)) |
2c47e605 YZ |
5655 | break; |
5656 | ||
5657 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5658 | path->locks[level] = 0; |
2c47e605 YZ |
5659 | WARN_ON(wc->refs[level] != 1); |
5660 | level--; | |
5661 | } | |
9f3a7427 | 5662 | } |
2c47e605 | 5663 | |
78c52d9e | 5664 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5665 | wc->level = level; |
5666 | wc->shared_level = -1; | |
5667 | wc->stage = DROP_REFERENCE; | |
5668 | wc->update_ref = update_ref; | |
5669 | wc->keep_locks = 0; | |
0b246afa | 5670 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5671 | |
d397712b | 5672 | while (1) { |
9d1a2a3a | 5673 | |
2c47e605 YZ |
5674 | ret = walk_down_tree(trans, root, path, wc); |
5675 | if (ret < 0) { | |
5676 | err = ret; | |
20524f02 | 5677 | break; |
2c47e605 | 5678 | } |
9aca1d51 | 5679 | |
2c47e605 YZ |
5680 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5681 | if (ret < 0) { | |
5682 | err = ret; | |
20524f02 | 5683 | break; |
2c47e605 YZ |
5684 | } |
5685 | ||
5686 | if (ret > 0) { | |
5687 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5688 | break; |
5689 | } | |
2c47e605 YZ |
5690 | |
5691 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5692 | wc->drop_level = wc->level; |
5693 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5694 | &wc->drop_progress, | |
5695 | path->slots[wc->drop_level]); | |
5696 | } | |
5697 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5698 | &wc->drop_progress); | |
c8422684 | 5699 | btrfs_set_root_drop_level(root_item, wc->drop_level); |
2c47e605 YZ |
5700 | |
5701 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5702 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5703 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5704 | ret = btrfs_update_root(trans, tree_root, |
5705 | &root->root_key, | |
5706 | root_item); | |
79787eaa | 5707 | if (ret) { |
66642832 | 5708 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5709 | err = ret; |
5710 | goto out_end_trans; | |
5711 | } | |
2c47e605 | 5712 | |
3a45bb20 | 5713 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5714 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5715 | btrfs_debug(fs_info, |
5716 | "drop snapshot early exit"); | |
3c8f2422 JB |
5717 | err = -EAGAIN; |
5718 | goto out_free; | |
5719 | } | |
5720 | ||
18d3bff4 JB |
5721 | /* |
5722 | * Use join to avoid potential EINTR from transaction | |
5723 | * start. See wait_reserve_ticket and the whole | |
5724 | * reservation callchain. | |
5725 | */ | |
5726 | if (for_reloc) | |
5727 | trans = btrfs_join_transaction(tree_root); | |
5728 | else | |
5729 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5730 | if (IS_ERR(trans)) { |
5731 | err = PTR_ERR(trans); | |
5732 | goto out_free; | |
5733 | } | |
c3e69d58 | 5734 | } |
20524f02 | 5735 | } |
b3b4aa74 | 5736 | btrfs_release_path(path); |
79787eaa JM |
5737 | if (err) |
5738 | goto out_end_trans; | |
2c47e605 | 5739 | |
ab9ce7d4 | 5740 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5741 | if (ret) { |
66642832 | 5742 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5743 | err = ret; |
79787eaa JM |
5744 | goto out_end_trans; |
5745 | } | |
2c47e605 | 5746 | |
76dda93c | 5747 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5748 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5749 | NULL, NULL); | |
79787eaa | 5750 | if (ret < 0) { |
66642832 | 5751 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5752 | err = ret; |
5753 | goto out_end_trans; | |
5754 | } else if (ret > 0) { | |
84cd948c JB |
5755 | /* if we fail to delete the orphan item this time |
5756 | * around, it'll get picked up the next time. | |
5757 | * | |
5758 | * The most common failure here is just -ENOENT. | |
5759 | */ | |
5760 | btrfs_del_orphan_item(trans, tree_root, | |
5761 | root->root_key.objectid); | |
76dda93c YZ |
5762 | } |
5763 | } | |
5764 | ||
a3cf0e43 QW |
5765 | /* |
5766 | * This subvolume is going to be completely dropped, and won't be | |
5767 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
5768 | * commit transaction time. So free it here manually. | |
5769 | */ | |
5770 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
5771 | btrfs_qgroup_free_meta_all_pertrans(root); | |
5772 | ||
8c38938c | 5773 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 5774 | btrfs_add_dropped_root(trans, root); |
8c38938c | 5775 | else |
00246528 | 5776 | btrfs_put_root(root); |
d29a9f62 | 5777 | root_dropped = true; |
79787eaa | 5778 | out_end_trans: |
3a45bb20 | 5779 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5780 | out_free: |
2c47e605 | 5781 | kfree(wc); |
5caf2a00 | 5782 | btrfs_free_path(path); |
cb1b69f4 | 5783 | out: |
78db1921 JB |
5784 | /* |
5785 | * We were an unfinished drop root, check to see if there are any | |
5786 | * pending, and if not clear and wake up any waiters. | |
5787 | */ | |
5788 | if (!err && unfinished_drop) | |
5789 | btrfs_maybe_wake_unfinished_drop(fs_info); | |
5790 | ||
d29a9f62 JB |
5791 | /* |
5792 | * So if we need to stop dropping the snapshot for whatever reason we | |
5793 | * need to make sure to add it back to the dead root list so that we | |
5794 | * keep trying to do the work later. This also cleans up roots if we | |
5795 | * don't have it in the radix (like when we recover after a power fail | |
5796 | * or unmount) so we don't leak memory. | |
5797 | */ | |
897ca819 | 5798 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5799 | btrfs_add_dead_root(root); |
2c536799 | 5800 | return err; |
20524f02 | 5801 | } |
9078a3e1 | 5802 | |
2c47e605 YZ |
5803 | /* |
5804 | * drop subtree rooted at tree block 'node'. | |
5805 | * | |
5806 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5807 | * only used by relocation code |
2c47e605 | 5808 | */ |
f82d02d9 YZ |
5809 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5810 | struct btrfs_root *root, | |
5811 | struct extent_buffer *node, | |
5812 | struct extent_buffer *parent) | |
5813 | { | |
0b246afa | 5814 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5815 | struct btrfs_path *path; |
2c47e605 | 5816 | struct walk_control *wc; |
f82d02d9 YZ |
5817 | int level; |
5818 | int parent_level; | |
5819 | int ret = 0; | |
5820 | int wret; | |
5821 | ||
2c47e605 YZ |
5822 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5823 | ||
f82d02d9 | 5824 | path = btrfs_alloc_path(); |
db5b493a TI |
5825 | if (!path) |
5826 | return -ENOMEM; | |
f82d02d9 | 5827 | |
2c47e605 | 5828 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5829 | if (!wc) { |
5830 | btrfs_free_path(path); | |
5831 | return -ENOMEM; | |
5832 | } | |
2c47e605 | 5833 | |
b9447ef8 | 5834 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5835 | parent_level = btrfs_header_level(parent); |
67439dad | 5836 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5837 | path->nodes[parent_level] = parent; |
5838 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5839 | ||
b9447ef8 | 5840 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5841 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5842 | path->nodes[level] = node; |
5843 | path->slots[level] = 0; | |
ac5887c8 | 5844 | path->locks[level] = BTRFS_WRITE_LOCK; |
2c47e605 YZ |
5845 | |
5846 | wc->refs[parent_level] = 1; | |
5847 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5848 | wc->level = level; | |
5849 | wc->shared_level = -1; | |
5850 | wc->stage = DROP_REFERENCE; | |
5851 | wc->update_ref = 0; | |
5852 | wc->keep_locks = 1; | |
0b246afa | 5853 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5854 | |
5855 | while (1) { | |
2c47e605 YZ |
5856 | wret = walk_down_tree(trans, root, path, wc); |
5857 | if (wret < 0) { | |
f82d02d9 | 5858 | ret = wret; |
f82d02d9 | 5859 | break; |
2c47e605 | 5860 | } |
f82d02d9 | 5861 | |
2c47e605 | 5862 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5863 | if (wret < 0) |
5864 | ret = wret; | |
5865 | if (wret != 0) | |
5866 | break; | |
5867 | } | |
5868 | ||
2c47e605 | 5869 | kfree(wc); |
f82d02d9 YZ |
5870 | btrfs_free_path(path); |
5871 | return ret; | |
5872 | } | |
5873 | ||
6d07bcec MX |
5874 | /* |
5875 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5876 | * space_info. takes mirrors into account. |
6d07bcec | 5877 | */ |
633c0aad | 5878 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5879 | { |
32da5386 | 5880 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5881 | u64 free_bytes = 0; |
5882 | int factor; | |
5883 | ||
01327610 | 5884 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5885 | if (list_empty(&sinfo->ro_bgs)) |
5886 | return 0; | |
5887 | ||
5888 | spin_lock(&sinfo->lock); | |
5889 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5890 | spin_lock(&block_group->lock); |
5891 | ||
5892 | if (!block_group->ro) { | |
5893 | spin_unlock(&block_group->lock); | |
5894 | continue; | |
5895 | } | |
5896 | ||
46df06b8 | 5897 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5898 | free_bytes += (block_group->length - |
bf38be65 | 5899 | block_group->used) * factor; |
6d07bcec MX |
5900 | |
5901 | spin_unlock(&block_group->lock); | |
5902 | } | |
6d07bcec MX |
5903 | spin_unlock(&sinfo->lock); |
5904 | ||
5905 | return free_bytes; | |
5906 | } | |
5907 | ||
2ff7e61e JM |
5908 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5909 | u64 start, u64 end) | |
acce952b | 5910 | { |
2ff7e61e | 5911 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5912 | } |
5913 | ||
499f377f JM |
5914 | /* |
5915 | * It used to be that old block groups would be left around forever. | |
5916 | * Iterating over them would be enough to trim unused space. Since we | |
5917 | * now automatically remove them, we also need to iterate over unallocated | |
5918 | * space. | |
5919 | * | |
5920 | * We don't want a transaction for this since the discard may take a | |
5921 | * substantial amount of time. We don't require that a transaction be | |
5922 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5923 | * to ensure that we're not discarding chunks that were released or |
5924 | * allocated in the current transaction. | |
499f377f JM |
5925 | * |
5926 | * Holding the chunks lock will prevent other threads from allocating | |
5927 | * or releasing chunks, but it won't prevent a running transaction | |
5928 | * from committing and releasing the memory that the pending chunks | |
5929 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5930 | * transaction and hold the commit root sem. We only need to hold |
5931 | * it while performing the free space search since we have already | |
5932 | * held back allocations. | |
499f377f | 5933 | */ |
8103d10b | 5934 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5935 | { |
8103d10b | 5936 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5937 | int ret; |
5938 | ||
5939 | *trimmed = 0; | |
5940 | ||
0be88e36 JM |
5941 | /* Discard not supported = nothing to do. */ |
5942 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5943 | return 0; | |
5944 | ||
52042d8e | 5945 | /* Not writable = nothing to do. */ |
ebbede42 | 5946 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5947 | return 0; |
5948 | ||
5949 | /* No free space = nothing to do. */ | |
5950 | if (device->total_bytes <= device->bytes_used) | |
5951 | return 0; | |
5952 | ||
5953 | ret = 0; | |
5954 | ||
5955 | while (1) { | |
fb456252 | 5956 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5957 | u64 bytes; |
5958 | ||
5959 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5960 | if (ret) | |
fee7acc3 | 5961 | break; |
499f377f | 5962 | |
929be17a NB |
5963 | find_first_clear_extent_bit(&device->alloc_state, start, |
5964 | &start, &end, | |
5965 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 5966 | |
c57dd1f2 QW |
5967 | /* Check if there are any CHUNK_* bits left */ |
5968 | if (start > device->total_bytes) { | |
5969 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5970 | btrfs_warn_in_rcu(fs_info, | |
5971 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
5972 | start, end - start + 1, | |
5973 | rcu_str_deref(device->name), | |
5974 | device->total_bytes); | |
5975 | mutex_unlock(&fs_info->chunk_mutex); | |
5976 | ret = 0; | |
5977 | break; | |
5978 | } | |
5979 | ||
53460a45 NB |
5980 | /* Ensure we skip the reserved area in the first 1M */ |
5981 | start = max_t(u64, start, SZ_1M); | |
5982 | ||
929be17a NB |
5983 | /* |
5984 | * If find_first_clear_extent_bit find a range that spans the | |
5985 | * end of the device it will set end to -1, in this case it's up | |
5986 | * to the caller to trim the value to the size of the device. | |
5987 | */ | |
5988 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5989 | |
929be17a | 5990 | len = end - start + 1; |
499f377f | 5991 | |
929be17a NB |
5992 | /* We didn't find any extents */ |
5993 | if (!len) { | |
499f377f | 5994 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5995 | ret = 0; |
499f377f JM |
5996 | break; |
5997 | } | |
5998 | ||
929be17a NB |
5999 | ret = btrfs_issue_discard(device->bdev, start, len, |
6000 | &bytes); | |
6001 | if (!ret) | |
6002 | set_extent_bits(&device->alloc_state, start, | |
6003 | start + bytes - 1, | |
6004 | CHUNK_TRIMMED); | |
499f377f JM |
6005 | mutex_unlock(&fs_info->chunk_mutex); |
6006 | ||
6007 | if (ret) | |
6008 | break; | |
6009 | ||
6010 | start += len; | |
6011 | *trimmed += bytes; | |
6012 | ||
6013 | if (fatal_signal_pending(current)) { | |
6014 | ret = -ERESTARTSYS; | |
6015 | break; | |
6016 | } | |
6017 | ||
6018 | cond_resched(); | |
6019 | } | |
6020 | ||
6021 | return ret; | |
6022 | } | |
6023 | ||
93bba24d QW |
6024 | /* |
6025 | * Trim the whole filesystem by: | |
6026 | * 1) trimming the free space in each block group | |
6027 | * 2) trimming the unallocated space on each device | |
6028 | * | |
6029 | * This will also continue trimming even if a block group or device encounters | |
6030 | * an error. The return value will be the last error, or 0 if nothing bad | |
6031 | * happens. | |
6032 | */ | |
2ff7e61e | 6033 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 6034 | { |
23608d51 | 6035 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
32da5386 | 6036 | struct btrfs_block_group *cache = NULL; |
499f377f | 6037 | struct btrfs_device *device; |
f7039b1d | 6038 | u64 group_trimmed; |
07301df7 | 6039 | u64 range_end = U64_MAX; |
f7039b1d LD |
6040 | u64 start; |
6041 | u64 end; | |
6042 | u64 trimmed = 0; | |
93bba24d QW |
6043 | u64 bg_failed = 0; |
6044 | u64 dev_failed = 0; | |
6045 | int bg_ret = 0; | |
6046 | int dev_ret = 0; | |
f7039b1d LD |
6047 | int ret = 0; |
6048 | ||
07301df7 QW |
6049 | /* |
6050 | * Check range overflow if range->len is set. | |
6051 | * The default range->len is U64_MAX. | |
6052 | */ | |
6053 | if (range->len != U64_MAX && | |
6054 | check_add_overflow(range->start, range->len, &range_end)) | |
6055 | return -EINVAL; | |
6056 | ||
6ba9fc8e | 6057 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 6058 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 6059 | if (cache->start >= range_end) { |
f7039b1d LD |
6060 | btrfs_put_block_group(cache); |
6061 | break; | |
6062 | } | |
6063 | ||
b3470b5d DS |
6064 | start = max(range->start, cache->start); |
6065 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
6066 | |
6067 | if (end - start >= range->minlen) { | |
32da5386 | 6068 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 6069 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 6070 | if (ret) { |
93bba24d QW |
6071 | bg_failed++; |
6072 | bg_ret = ret; | |
6073 | continue; | |
1be41b78 | 6074 | } |
676f1f75 | 6075 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 6076 | if (ret) { |
93bba24d QW |
6077 | bg_failed++; |
6078 | bg_ret = ret; | |
6079 | continue; | |
1be41b78 | 6080 | } |
f7039b1d LD |
6081 | } |
6082 | ret = btrfs_trim_block_group(cache, | |
6083 | &group_trimmed, | |
6084 | start, | |
6085 | end, | |
6086 | range->minlen); | |
6087 | ||
6088 | trimmed += group_trimmed; | |
6089 | if (ret) { | |
93bba24d QW |
6090 | bg_failed++; |
6091 | bg_ret = ret; | |
6092 | continue; | |
f7039b1d LD |
6093 | } |
6094 | } | |
f7039b1d LD |
6095 | } |
6096 | ||
93bba24d QW |
6097 | if (bg_failed) |
6098 | btrfs_warn(fs_info, | |
6099 | "failed to trim %llu block group(s), last error %d", | |
6100 | bg_failed, bg_ret); | |
23608d51 AJ |
6101 | |
6102 | mutex_lock(&fs_devices->device_list_mutex); | |
6103 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
16a200f6 AJ |
6104 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
6105 | continue; | |
6106 | ||
8103d10b | 6107 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
6108 | if (ret) { |
6109 | dev_failed++; | |
6110 | dev_ret = ret; | |
499f377f | 6111 | break; |
93bba24d | 6112 | } |
499f377f JM |
6113 | |
6114 | trimmed += group_trimmed; | |
6115 | } | |
23608d51 | 6116 | mutex_unlock(&fs_devices->device_list_mutex); |
499f377f | 6117 | |
93bba24d QW |
6118 | if (dev_failed) |
6119 | btrfs_warn(fs_info, | |
6120 | "failed to trim %llu device(s), last error %d", | |
6121 | dev_failed, dev_ret); | |
f7039b1d | 6122 | range->len = trimmed; |
93bba24d QW |
6123 | if (bg_ret) |
6124 | return bg_ret; | |
6125 | return dev_ret; | |
f7039b1d | 6126 | } |