]>
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" |
fec577fb | 37 | |
709c0486 AJ |
38 | #undef SCRAMBLE_DELAYED_REFS |
39 | ||
9f9b8e8d | 40 | |
5d4f98a2 | 41 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
42 | struct btrfs_delayed_ref_node *node, u64 parent, |
43 | u64 root_objectid, u64 owner_objectid, | |
44 | u64 owner_offset, int refs_to_drop, | |
45 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
46 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
47 | struct extent_buffer *leaf, | |
48 | struct btrfs_extent_item *ei); | |
49 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
50 | u64 parent, u64 root_objectid, |
51 | u64 flags, u64 owner, u64 offset, | |
52 | struct btrfs_key *ins, int ref_mod); | |
53 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 54 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 55 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
56 | static int find_next_key(struct btrfs_path *path, int level, |
57 | struct btrfs_key *key); | |
6a63209f | 58 | |
32da5386 | 59 | static int block_group_bits(struct btrfs_block_group *cache, u64 bits) |
0f9dd46c JB |
60 | { |
61 | return (cache->flags & bits) == bits; | |
62 | } | |
63 | ||
6f410d1b JB |
64 | int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info, |
65 | u64 start, u64 num_bytes) | |
817d52f8 | 66 | { |
11833d66 | 67 | u64 end = start + num_bytes - 1; |
fe119a6e NB |
68 | set_extent_bits(&fs_info->excluded_extents, start, end, |
69 | EXTENT_UPTODATE); | |
11833d66 YZ |
70 | return 0; |
71 | } | |
817d52f8 | 72 | |
32da5386 | 73 | void btrfs_free_excluded_extents(struct btrfs_block_group *cache) |
11833d66 | 74 | { |
9e715da8 | 75 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 76 | u64 start, end; |
817d52f8 | 77 | |
b3470b5d DS |
78 | start = cache->start; |
79 | end = start + cache->length - 1; | |
11833d66 | 80 | |
fe119a6e NB |
81 | clear_extent_bits(&fs_info->excluded_extents, start, end, |
82 | EXTENT_UPTODATE); | |
817d52f8 JB |
83 | } |
84 | ||
78192442 | 85 | static u64 generic_ref_to_space_flags(struct btrfs_ref *ref) |
0d9f824d | 86 | { |
ddf30cf0 QW |
87 | if (ref->type == BTRFS_REF_METADATA) { |
88 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
78192442 | 89 | return BTRFS_BLOCK_GROUP_SYSTEM; |
0d9f824d | 90 | else |
78192442 | 91 | return BTRFS_BLOCK_GROUP_METADATA; |
0d9f824d | 92 | } |
78192442 QW |
93 | return BTRFS_BLOCK_GROUP_DATA; |
94 | } | |
95 | ||
96 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, | |
97 | struct btrfs_ref *ref) | |
98 | { | |
99 | struct btrfs_space_info *space_info; | |
100 | u64 flags = generic_ref_to_space_flags(ref); | |
101 | ||
280c2908 | 102 | space_info = btrfs_find_space_info(fs_info, flags); |
78192442 QW |
103 | ASSERT(space_info); |
104 | percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len, | |
105 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
106 | } | |
107 | ||
108 | static void sub_pinned_bytes(struct btrfs_fs_info *fs_info, | |
109 | struct btrfs_ref *ref) | |
110 | { | |
111 | struct btrfs_space_info *space_info; | |
112 | u64 flags = generic_ref_to_space_flags(ref); | |
0d9f824d | 113 | |
280c2908 | 114 | space_info = btrfs_find_space_info(fs_info, flags); |
55e8196a | 115 | ASSERT(space_info); |
78192442 | 116 | percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len, |
dec59fa3 | 117 | BTRFS_TOTAL_BYTES_PINNED_BATCH); |
0d9f824d OS |
118 | } |
119 | ||
1a4ed8fd | 120 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 121 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
122 | { |
123 | int ret; | |
124 | struct btrfs_key key; | |
31840ae1 | 125 | struct btrfs_path *path; |
e02119d5 | 126 | |
31840ae1 | 127 | path = btrfs_alloc_path(); |
d8926bb3 MF |
128 | if (!path) |
129 | return -ENOMEM; | |
130 | ||
e02119d5 CM |
131 | key.objectid = start; |
132 | key.offset = len; | |
3173a18f | 133 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 134 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 135 | btrfs_free_path(path); |
7bb86316 CM |
136 | return ret; |
137 | } | |
138 | ||
a22285a6 | 139 | /* |
3173a18f | 140 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
141 | * |
142 | * the head node for delayed ref is used to store the sum of all the | |
143 | * reference count modifications queued up in the rbtree. the head | |
144 | * node may also store the extent flags to set. This way you can check | |
145 | * to see what the reference count and extent flags would be if all of | |
146 | * the delayed refs are not processed. | |
147 | */ | |
148 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 149 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 150 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
151 | { |
152 | struct btrfs_delayed_ref_head *head; | |
153 | struct btrfs_delayed_ref_root *delayed_refs; | |
154 | struct btrfs_path *path; | |
155 | struct btrfs_extent_item *ei; | |
156 | struct extent_buffer *leaf; | |
157 | struct btrfs_key key; | |
158 | u32 item_size; | |
159 | u64 num_refs; | |
160 | u64 extent_flags; | |
161 | int ret; | |
162 | ||
3173a18f JB |
163 | /* |
164 | * If we don't have skinny metadata, don't bother doing anything | |
165 | * different | |
166 | */ | |
0b246afa JM |
167 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
168 | offset = fs_info->nodesize; | |
3173a18f JB |
169 | metadata = 0; |
170 | } | |
171 | ||
a22285a6 YZ |
172 | path = btrfs_alloc_path(); |
173 | if (!path) | |
174 | return -ENOMEM; | |
175 | ||
a22285a6 YZ |
176 | if (!trans) { |
177 | path->skip_locking = 1; | |
178 | path->search_commit_root = 1; | |
179 | } | |
639eefc8 FDBM |
180 | |
181 | search_again: | |
182 | key.objectid = bytenr; | |
183 | key.offset = offset; | |
184 | if (metadata) | |
185 | key.type = BTRFS_METADATA_ITEM_KEY; | |
186 | else | |
187 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
188 | ||
0b246afa | 189 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
190 | if (ret < 0) |
191 | goto out_free; | |
192 | ||
3173a18f | 193 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
194 | if (path->slots[0]) { |
195 | path->slots[0]--; | |
196 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
197 | path->slots[0]); | |
198 | if (key.objectid == bytenr && | |
199 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 200 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
201 | ret = 0; |
202 | } | |
3173a18f JB |
203 | } |
204 | ||
a22285a6 YZ |
205 | if (ret == 0) { |
206 | leaf = path->nodes[0]; | |
207 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
208 | if (item_size >= sizeof(*ei)) { | |
209 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
210 | struct btrfs_extent_item); | |
211 | num_refs = btrfs_extent_refs(leaf, ei); | |
212 | extent_flags = btrfs_extent_flags(leaf, ei); | |
213 | } else { | |
ba3c2b19 NB |
214 | ret = -EINVAL; |
215 | btrfs_print_v0_err(fs_info); | |
216 | if (trans) | |
217 | btrfs_abort_transaction(trans, ret); | |
218 | else | |
219 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
220 | ||
221 | goto out_free; | |
a22285a6 | 222 | } |
ba3c2b19 | 223 | |
a22285a6 YZ |
224 | BUG_ON(num_refs == 0); |
225 | } else { | |
226 | num_refs = 0; | |
227 | extent_flags = 0; | |
228 | ret = 0; | |
229 | } | |
230 | ||
231 | if (!trans) | |
232 | goto out; | |
233 | ||
234 | delayed_refs = &trans->transaction->delayed_refs; | |
235 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 236 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
237 | if (head) { |
238 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 239 | refcount_inc(&head->refs); |
a22285a6 YZ |
240 | spin_unlock(&delayed_refs->lock); |
241 | ||
b3b4aa74 | 242 | btrfs_release_path(path); |
a22285a6 | 243 | |
8cc33e5c DS |
244 | /* |
245 | * Mutex was contended, block until it's released and try | |
246 | * again | |
247 | */ | |
a22285a6 YZ |
248 | mutex_lock(&head->mutex); |
249 | mutex_unlock(&head->mutex); | |
d278850e | 250 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 251 | goto search_again; |
a22285a6 | 252 | } |
d7df2c79 | 253 | spin_lock(&head->lock); |
a22285a6 YZ |
254 | if (head->extent_op && head->extent_op->update_flags) |
255 | extent_flags |= head->extent_op->flags_to_set; | |
256 | else | |
257 | BUG_ON(num_refs == 0); | |
258 | ||
d278850e | 259 | num_refs += head->ref_mod; |
d7df2c79 | 260 | spin_unlock(&head->lock); |
a22285a6 YZ |
261 | mutex_unlock(&head->mutex); |
262 | } | |
263 | spin_unlock(&delayed_refs->lock); | |
264 | out: | |
265 | WARN_ON(num_refs == 0); | |
266 | if (refs) | |
267 | *refs = num_refs; | |
268 | if (flags) | |
269 | *flags = extent_flags; | |
270 | out_free: | |
271 | btrfs_free_path(path); | |
272 | return ret; | |
273 | } | |
274 | ||
d8d5f3e1 CM |
275 | /* |
276 | * Back reference rules. Back refs have three main goals: | |
277 | * | |
278 | * 1) differentiate between all holders of references to an extent so that | |
279 | * when a reference is dropped we can make sure it was a valid reference | |
280 | * before freeing the extent. | |
281 | * | |
282 | * 2) Provide enough information to quickly find the holders of an extent | |
283 | * if we notice a given block is corrupted or bad. | |
284 | * | |
285 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
286 | * maintenance. This is actually the same as #2, but with a slightly | |
287 | * different use case. | |
288 | * | |
5d4f98a2 YZ |
289 | * There are two kinds of back refs. The implicit back refs is optimized |
290 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
291 | * back refs of this kind provide information about the block's owner tree | |
292 | * and the pointer's key. These information allow us to find the block by | |
293 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
294 | * referenced by their owner trees. The location of tree block is recorded | |
295 | * in the back refs. Actually the full back refs is generic, and can be | |
296 | * used in all cases the implicit back refs is used. The major shortcoming | |
297 | * of the full back refs is its overhead. Every time a tree block gets | |
298 | * COWed, we have to update back refs entry for all pointers in it. | |
299 | * | |
300 | * For a newly allocated tree block, we use implicit back refs for | |
301 | * pointers in it. This means most tree related operations only involve | |
302 | * implicit back refs. For a tree block created in old transaction, the | |
303 | * only way to drop a reference to it is COW it. So we can detect the | |
304 | * event that tree block loses its owner tree's reference and do the | |
305 | * back refs conversion. | |
306 | * | |
01327610 | 307 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
308 | * |
309 | * The reference count of the block is one and the tree is the block's | |
310 | * owner tree. Nothing to do in this case. | |
311 | * | |
312 | * The reference count of the block is one and the tree is not the | |
313 | * block's owner tree. In this case, full back refs is used for pointers | |
314 | * in the block. Remove these full back refs, add implicit back refs for | |
315 | * every pointers in the new block. | |
316 | * | |
317 | * The reference count of the block is greater than one and the tree is | |
318 | * the block's owner tree. In this case, implicit back refs is used for | |
319 | * pointers in the block. Add full back refs for every pointers in the | |
320 | * block, increase lower level extents' reference counts. The original | |
321 | * implicit back refs are entailed to the new block. | |
322 | * | |
323 | * The reference count of the block is greater than one and the tree is | |
324 | * not the block's owner tree. Add implicit back refs for every pointer in | |
325 | * the new block, increase lower level extents' reference count. | |
326 | * | |
327 | * Back Reference Key composing: | |
328 | * | |
329 | * The key objectid corresponds to the first byte in the extent, | |
330 | * The key type is used to differentiate between types of back refs. | |
331 | * There are different meanings of the key offset for different types | |
332 | * of back refs. | |
333 | * | |
d8d5f3e1 CM |
334 | * File extents can be referenced by: |
335 | * | |
336 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 337 | * - different files inside a single subvolume |
d8d5f3e1 CM |
338 | * - different offsets inside a file (bookend extents in file.c) |
339 | * | |
5d4f98a2 | 340 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
341 | * |
342 | * - Objectid of the subvolume root | |
d8d5f3e1 | 343 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
344 | * - original offset in the file |
345 | * - how many bookend extents | |
d8d5f3e1 | 346 | * |
5d4f98a2 YZ |
347 | * The key offset for the implicit back refs is hash of the first |
348 | * three fields. | |
d8d5f3e1 | 349 | * |
5d4f98a2 | 350 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 351 | * |
5d4f98a2 | 352 | * - number of pointers in the tree leaf |
d8d5f3e1 | 353 | * |
5d4f98a2 YZ |
354 | * The key offset for the implicit back refs is the first byte of |
355 | * the tree leaf | |
d8d5f3e1 | 356 | * |
5d4f98a2 YZ |
357 | * When a file extent is allocated, The implicit back refs is used. |
358 | * the fields are filled in: | |
d8d5f3e1 | 359 | * |
5d4f98a2 | 360 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 361 | * |
5d4f98a2 YZ |
362 | * When a file extent is removed file truncation, we find the |
363 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 364 | * |
5d4f98a2 | 365 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 366 | * |
5d4f98a2 | 367 | * Btree extents can be referenced by: |
d8d5f3e1 | 368 | * |
5d4f98a2 | 369 | * - Different subvolumes |
d8d5f3e1 | 370 | * |
5d4f98a2 YZ |
371 | * Both the implicit back refs and the full back refs for tree blocks |
372 | * only consist of key. The key offset for the implicit back refs is | |
373 | * objectid of block's owner tree. The key offset for the full back refs | |
374 | * is the first byte of parent block. | |
d8d5f3e1 | 375 | * |
5d4f98a2 YZ |
376 | * When implicit back refs is used, information about the lowest key and |
377 | * level of the tree block are required. These information are stored in | |
378 | * tree block info structure. | |
d8d5f3e1 | 379 | */ |
31840ae1 | 380 | |
167ce953 LB |
381 | /* |
382 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 383 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
384 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
385 | */ | |
386 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
387 | struct btrfs_extent_inline_ref *iref, | |
388 | enum btrfs_inline_ref_type is_data) | |
389 | { | |
390 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 391 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
392 | |
393 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
394 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
395 | type == BTRFS_SHARED_DATA_REF_KEY || | |
396 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
397 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 398 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 399 | return type; |
64ecdb64 LB |
400 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
401 | ASSERT(eb->fs_info); | |
402 | /* | |
ea57788e QW |
403 | * Every shared one has parent tree block, |
404 | * which must be aligned to sector size. | |
64ecdb64 LB |
405 | */ |
406 | if (offset && | |
ea57788e | 407 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
408 | return type; |
409 | } | |
167ce953 | 410 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 411 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 412 | return type; |
64ecdb64 LB |
413 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
414 | ASSERT(eb->fs_info); | |
415 | /* | |
ea57788e QW |
416 | * Every shared one has parent tree block, |
417 | * which must be aligned to sector size. | |
64ecdb64 LB |
418 | */ |
419 | if (offset && | |
ea57788e | 420 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
421 | return type; |
422 | } | |
167ce953 LB |
423 | } else { |
424 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
425 | return type; | |
426 | } | |
427 | } | |
428 | ||
429 | btrfs_print_leaf((struct extent_buffer *)eb); | |
ea57788e QW |
430 | btrfs_err(eb->fs_info, |
431 | "eb %llu iref 0x%lx invalid extent inline ref type %d", | |
432 | eb->start, (unsigned long)iref, type); | |
167ce953 LB |
433 | WARN_ON(1); |
434 | ||
435 | return BTRFS_REF_TYPE_INVALID; | |
436 | } | |
437 | ||
0785a9aa | 438 | u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
439 | { |
440 | u32 high_crc = ~(u32)0; | |
441 | u32 low_crc = ~(u32)0; | |
442 | __le64 lenum; | |
443 | ||
444 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 445 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 446 | lenum = cpu_to_le64(owner); |
65019df8 | 447 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 448 | lenum = cpu_to_le64(offset); |
65019df8 | 449 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
450 | |
451 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
452 | } | |
453 | ||
454 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
455 | struct btrfs_extent_data_ref *ref) | |
456 | { | |
457 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
458 | btrfs_extent_data_ref_objectid(leaf, ref), | |
459 | btrfs_extent_data_ref_offset(leaf, ref)); | |
460 | } | |
461 | ||
462 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
463 | struct btrfs_extent_data_ref *ref, | |
464 | u64 root_objectid, u64 owner, u64 offset) | |
465 | { | |
466 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
467 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
468 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
469 | return 0; | |
470 | return 1; | |
471 | } | |
472 | ||
473 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
474 | struct btrfs_path *path, |
475 | u64 bytenr, u64 parent, | |
476 | u64 root_objectid, | |
477 | u64 owner, u64 offset) | |
478 | { | |
bd1d53ef | 479 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
480 | struct btrfs_key key; |
481 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 482 | struct extent_buffer *leaf; |
5d4f98a2 | 483 | u32 nritems; |
74493f7a | 484 | int ret; |
5d4f98a2 YZ |
485 | int recow; |
486 | int err = -ENOENT; | |
74493f7a | 487 | |
31840ae1 | 488 | key.objectid = bytenr; |
5d4f98a2 YZ |
489 | if (parent) { |
490 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
491 | key.offset = parent; | |
492 | } else { | |
493 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
494 | key.offset = hash_extent_data_ref(root_objectid, | |
495 | owner, offset); | |
496 | } | |
497 | again: | |
498 | recow = 0; | |
499 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
500 | if (ret < 0) { | |
501 | err = ret; | |
502 | goto fail; | |
503 | } | |
31840ae1 | 504 | |
5d4f98a2 YZ |
505 | if (parent) { |
506 | if (!ret) | |
507 | return 0; | |
5d4f98a2 | 508 | goto fail; |
31840ae1 ZY |
509 | } |
510 | ||
511 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
512 | nritems = btrfs_header_nritems(leaf); |
513 | while (1) { | |
514 | if (path->slots[0] >= nritems) { | |
515 | ret = btrfs_next_leaf(root, path); | |
516 | if (ret < 0) | |
517 | err = ret; | |
518 | if (ret) | |
519 | goto fail; | |
520 | ||
521 | leaf = path->nodes[0]; | |
522 | nritems = btrfs_header_nritems(leaf); | |
523 | recow = 1; | |
524 | } | |
525 | ||
526 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
527 | if (key.objectid != bytenr || | |
528 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
529 | goto fail; | |
530 | ||
531 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
532 | struct btrfs_extent_data_ref); | |
533 | ||
534 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
535 | owner, offset)) { | |
536 | if (recow) { | |
b3b4aa74 | 537 | btrfs_release_path(path); |
5d4f98a2 YZ |
538 | goto again; |
539 | } | |
540 | err = 0; | |
541 | break; | |
542 | } | |
543 | path->slots[0]++; | |
31840ae1 | 544 | } |
5d4f98a2 YZ |
545 | fail: |
546 | return err; | |
31840ae1 ZY |
547 | } |
548 | ||
5d4f98a2 | 549 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
550 | struct btrfs_path *path, |
551 | u64 bytenr, u64 parent, | |
552 | u64 root_objectid, u64 owner, | |
553 | u64 offset, int refs_to_add) | |
31840ae1 | 554 | { |
62b895af | 555 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
556 | struct btrfs_key key; |
557 | struct extent_buffer *leaf; | |
5d4f98a2 | 558 | u32 size; |
31840ae1 ZY |
559 | u32 num_refs; |
560 | int ret; | |
74493f7a | 561 | |
74493f7a | 562 | key.objectid = bytenr; |
5d4f98a2 YZ |
563 | if (parent) { |
564 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
565 | key.offset = parent; | |
566 | size = sizeof(struct btrfs_shared_data_ref); | |
567 | } else { | |
568 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
569 | key.offset = hash_extent_data_ref(root_objectid, | |
570 | owner, offset); | |
571 | size = sizeof(struct btrfs_extent_data_ref); | |
572 | } | |
74493f7a | 573 | |
5d4f98a2 YZ |
574 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
575 | if (ret && ret != -EEXIST) | |
576 | goto fail; | |
577 | ||
578 | leaf = path->nodes[0]; | |
579 | if (parent) { | |
580 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 581 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
582 | struct btrfs_shared_data_ref); |
583 | if (ret == 0) { | |
584 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
585 | } else { | |
586 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
587 | num_refs += refs_to_add; | |
588 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 589 | } |
5d4f98a2 YZ |
590 | } else { |
591 | struct btrfs_extent_data_ref *ref; | |
592 | while (ret == -EEXIST) { | |
593 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
594 | struct btrfs_extent_data_ref); | |
595 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
596 | owner, offset)) | |
597 | break; | |
b3b4aa74 | 598 | btrfs_release_path(path); |
5d4f98a2 YZ |
599 | key.offset++; |
600 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
601 | size); | |
602 | if (ret && ret != -EEXIST) | |
603 | goto fail; | |
31840ae1 | 604 | |
5d4f98a2 YZ |
605 | leaf = path->nodes[0]; |
606 | } | |
607 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
608 | struct btrfs_extent_data_ref); | |
609 | if (ret == 0) { | |
610 | btrfs_set_extent_data_ref_root(leaf, ref, | |
611 | root_objectid); | |
612 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
613 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
614 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
615 | } else { | |
616 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
617 | num_refs += refs_to_add; | |
618 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 619 | } |
31840ae1 | 620 | } |
5d4f98a2 YZ |
621 | btrfs_mark_buffer_dirty(leaf); |
622 | ret = 0; | |
623 | fail: | |
b3b4aa74 | 624 | btrfs_release_path(path); |
7bb86316 | 625 | return ret; |
74493f7a CM |
626 | } |
627 | ||
5d4f98a2 | 628 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 629 | struct btrfs_path *path, |
fcebe456 | 630 | int refs_to_drop, int *last_ref) |
31840ae1 | 631 | { |
5d4f98a2 YZ |
632 | struct btrfs_key key; |
633 | struct btrfs_extent_data_ref *ref1 = NULL; | |
634 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 635 | struct extent_buffer *leaf; |
5d4f98a2 | 636 | u32 num_refs = 0; |
31840ae1 ZY |
637 | int ret = 0; |
638 | ||
639 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
640 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
641 | ||
642 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
643 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
644 | struct btrfs_extent_data_ref); | |
645 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
646 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
647 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
648 | struct btrfs_shared_data_ref); | |
649 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 650 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
651 | btrfs_print_v0_err(trans->fs_info); |
652 | btrfs_abort_transaction(trans, -EINVAL); | |
653 | return -EINVAL; | |
5d4f98a2 YZ |
654 | } else { |
655 | BUG(); | |
656 | } | |
657 | ||
56bec294 CM |
658 | BUG_ON(num_refs < refs_to_drop); |
659 | num_refs -= refs_to_drop; | |
5d4f98a2 | 660 | |
31840ae1 | 661 | if (num_refs == 0) { |
e9f6290d | 662 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 663 | *last_ref = 1; |
31840ae1 | 664 | } else { |
5d4f98a2 YZ |
665 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
666 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
667 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
668 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
669 | btrfs_mark_buffer_dirty(leaf); |
670 | } | |
31840ae1 ZY |
671 | return ret; |
672 | } | |
673 | ||
9ed0dea0 | 674 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 675 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 676 | { |
5d4f98a2 YZ |
677 | struct btrfs_key key; |
678 | struct extent_buffer *leaf; | |
679 | struct btrfs_extent_data_ref *ref1; | |
680 | struct btrfs_shared_data_ref *ref2; | |
681 | u32 num_refs = 0; | |
3de28d57 | 682 | int type; |
5d4f98a2 YZ |
683 | |
684 | leaf = path->nodes[0]; | |
685 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
686 | |
687 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 688 | if (iref) { |
3de28d57 LB |
689 | /* |
690 | * If type is invalid, we should have bailed out earlier than | |
691 | * this call. | |
692 | */ | |
693 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
694 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
695 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
696 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
697 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
698 | } else { | |
699 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
700 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
701 | } | |
702 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
703 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
704 | struct btrfs_extent_data_ref); | |
705 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
706 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
707 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
708 | struct btrfs_shared_data_ref); | |
709 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
710 | } else { |
711 | WARN_ON(1); | |
712 | } | |
713 | return num_refs; | |
714 | } | |
15916de8 | 715 | |
5d4f98a2 | 716 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
717 | struct btrfs_path *path, |
718 | u64 bytenr, u64 parent, | |
719 | u64 root_objectid) | |
1f3c79a2 | 720 | { |
b8582eea | 721 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 722 | struct btrfs_key key; |
1f3c79a2 | 723 | int ret; |
1f3c79a2 | 724 | |
5d4f98a2 YZ |
725 | key.objectid = bytenr; |
726 | if (parent) { | |
727 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
728 | key.offset = parent; | |
729 | } else { | |
730 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
731 | key.offset = root_objectid; | |
1f3c79a2 LH |
732 | } |
733 | ||
5d4f98a2 YZ |
734 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
735 | if (ret > 0) | |
736 | ret = -ENOENT; | |
5d4f98a2 | 737 | return ret; |
1f3c79a2 LH |
738 | } |
739 | ||
5d4f98a2 | 740 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
741 | struct btrfs_path *path, |
742 | u64 bytenr, u64 parent, | |
743 | u64 root_objectid) | |
31840ae1 | 744 | { |
5d4f98a2 | 745 | struct btrfs_key key; |
31840ae1 | 746 | int ret; |
31840ae1 | 747 | |
5d4f98a2 YZ |
748 | key.objectid = bytenr; |
749 | if (parent) { | |
750 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
751 | key.offset = parent; | |
752 | } else { | |
753 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
754 | key.offset = root_objectid; | |
755 | } | |
756 | ||
10728404 | 757 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 758 | path, &key, 0); |
b3b4aa74 | 759 | btrfs_release_path(path); |
31840ae1 ZY |
760 | return ret; |
761 | } | |
762 | ||
5d4f98a2 | 763 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 764 | { |
5d4f98a2 YZ |
765 | int type; |
766 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
767 | if (parent > 0) | |
768 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
769 | else | |
770 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
771 | } else { | |
772 | if (parent > 0) | |
773 | type = BTRFS_SHARED_DATA_REF_KEY; | |
774 | else | |
775 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
776 | } | |
777 | return type; | |
31840ae1 | 778 | } |
56bec294 | 779 | |
2c47e605 YZ |
780 | static int find_next_key(struct btrfs_path *path, int level, |
781 | struct btrfs_key *key) | |
56bec294 | 782 | |
02217ed2 | 783 | { |
2c47e605 | 784 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
785 | if (!path->nodes[level]) |
786 | break; | |
5d4f98a2 YZ |
787 | if (path->slots[level] + 1 >= |
788 | btrfs_header_nritems(path->nodes[level])) | |
789 | continue; | |
790 | if (level == 0) | |
791 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
792 | path->slots[level] + 1); | |
793 | else | |
794 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
795 | path->slots[level] + 1); | |
796 | return 0; | |
797 | } | |
798 | return 1; | |
799 | } | |
037e6390 | 800 | |
5d4f98a2 YZ |
801 | /* |
802 | * look for inline back ref. if back ref is found, *ref_ret is set | |
803 | * to the address of inline back ref, and 0 is returned. | |
804 | * | |
805 | * if back ref isn't found, *ref_ret is set to the address where it | |
806 | * should be inserted, and -ENOENT is returned. | |
807 | * | |
808 | * if insert is true and there are too many inline back refs, the path | |
809 | * points to the extent item, and -EAGAIN is returned. | |
810 | * | |
811 | * NOTE: inline back refs are ordered in the same way that back ref | |
812 | * items in the tree are ordered. | |
813 | */ | |
814 | static noinline_for_stack | |
815 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
816 | struct btrfs_path *path, |
817 | struct btrfs_extent_inline_ref **ref_ret, | |
818 | u64 bytenr, u64 num_bytes, | |
819 | u64 parent, u64 root_objectid, | |
820 | u64 owner, u64 offset, int insert) | |
821 | { | |
867cc1fb | 822 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 823 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
824 | struct btrfs_key key; |
825 | struct extent_buffer *leaf; | |
826 | struct btrfs_extent_item *ei; | |
827 | struct btrfs_extent_inline_ref *iref; | |
828 | u64 flags; | |
829 | u64 item_size; | |
830 | unsigned long ptr; | |
831 | unsigned long end; | |
832 | int extra_size; | |
833 | int type; | |
834 | int want; | |
835 | int ret; | |
836 | int err = 0; | |
0b246afa | 837 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 838 | int needed; |
26b8003f | 839 | |
db94535d | 840 | key.objectid = bytenr; |
31840ae1 | 841 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 842 | key.offset = num_bytes; |
31840ae1 | 843 | |
5d4f98a2 YZ |
844 | want = extent_ref_type(parent, owner); |
845 | if (insert) { | |
846 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 847 | path->keep_locks = 1; |
5d4f98a2 YZ |
848 | } else |
849 | extra_size = -1; | |
3173a18f JB |
850 | |
851 | /* | |
16d1c062 NB |
852 | * Owner is our level, so we can just add one to get the level for the |
853 | * block we are interested in. | |
3173a18f JB |
854 | */ |
855 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
856 | key.type = BTRFS_METADATA_ITEM_KEY; | |
857 | key.offset = owner; | |
858 | } | |
859 | ||
860 | again: | |
5d4f98a2 | 861 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 862 | if (ret < 0) { |
5d4f98a2 YZ |
863 | err = ret; |
864 | goto out; | |
865 | } | |
3173a18f JB |
866 | |
867 | /* | |
868 | * We may be a newly converted file system which still has the old fat | |
869 | * extent entries for metadata, so try and see if we have one of those. | |
870 | */ | |
871 | if (ret > 0 && skinny_metadata) { | |
872 | skinny_metadata = false; | |
873 | if (path->slots[0]) { | |
874 | path->slots[0]--; | |
875 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
876 | path->slots[0]); | |
877 | if (key.objectid == bytenr && | |
878 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
879 | key.offset == num_bytes) | |
880 | ret = 0; | |
881 | } | |
882 | if (ret) { | |
9ce49a0b | 883 | key.objectid = bytenr; |
3173a18f JB |
884 | key.type = BTRFS_EXTENT_ITEM_KEY; |
885 | key.offset = num_bytes; | |
886 | btrfs_release_path(path); | |
887 | goto again; | |
888 | } | |
889 | } | |
890 | ||
79787eaa JM |
891 | if (ret && !insert) { |
892 | err = -ENOENT; | |
893 | goto out; | |
fae7f21c | 894 | } else if (WARN_ON(ret)) { |
492104c8 | 895 | err = -EIO; |
492104c8 | 896 | goto out; |
79787eaa | 897 | } |
5d4f98a2 YZ |
898 | |
899 | leaf = path->nodes[0]; | |
900 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 901 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
902 | err = -EINVAL; |
903 | btrfs_print_v0_err(fs_info); | |
904 | btrfs_abort_transaction(trans, err); | |
905 | goto out; | |
906 | } | |
5d4f98a2 | 907 | |
5d4f98a2 YZ |
908 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
909 | flags = btrfs_extent_flags(leaf, ei); | |
910 | ||
911 | ptr = (unsigned long)(ei + 1); | |
912 | end = (unsigned long)ei + item_size; | |
913 | ||
3173a18f | 914 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
915 | ptr += sizeof(struct btrfs_tree_block_info); |
916 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
917 | } |
918 | ||
3de28d57 LB |
919 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
920 | needed = BTRFS_REF_TYPE_DATA; | |
921 | else | |
922 | needed = BTRFS_REF_TYPE_BLOCK; | |
923 | ||
5d4f98a2 YZ |
924 | err = -ENOENT; |
925 | while (1) { | |
926 | if (ptr >= end) { | |
927 | WARN_ON(ptr > end); | |
928 | break; | |
929 | } | |
930 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
931 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
932 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 933 | err = -EUCLEAN; |
3de28d57 LB |
934 | goto out; |
935 | } | |
936 | ||
5d4f98a2 YZ |
937 | if (want < type) |
938 | break; | |
939 | if (want > type) { | |
940 | ptr += btrfs_extent_inline_ref_size(type); | |
941 | continue; | |
942 | } | |
943 | ||
944 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
945 | struct btrfs_extent_data_ref *dref; | |
946 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
947 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
948 | owner, offset)) { | |
949 | err = 0; | |
950 | break; | |
951 | } | |
952 | if (hash_extent_data_ref_item(leaf, dref) < | |
953 | hash_extent_data_ref(root_objectid, owner, offset)) | |
954 | break; | |
955 | } else { | |
956 | u64 ref_offset; | |
957 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
958 | if (parent > 0) { | |
959 | if (parent == ref_offset) { | |
960 | err = 0; | |
961 | break; | |
962 | } | |
963 | if (ref_offset < parent) | |
964 | break; | |
965 | } else { | |
966 | if (root_objectid == ref_offset) { | |
967 | err = 0; | |
968 | break; | |
969 | } | |
970 | if (ref_offset < root_objectid) | |
971 | break; | |
972 | } | |
973 | } | |
974 | ptr += btrfs_extent_inline_ref_size(type); | |
975 | } | |
976 | if (err == -ENOENT && insert) { | |
977 | if (item_size + extra_size >= | |
978 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
979 | err = -EAGAIN; | |
980 | goto out; | |
981 | } | |
982 | /* | |
983 | * To add new inline back ref, we have to make sure | |
984 | * there is no corresponding back ref item. | |
985 | * For simplicity, we just do not add new inline back | |
986 | * ref if there is any kind of item for this block | |
987 | */ | |
2c47e605 YZ |
988 | if (find_next_key(path, 0, &key) == 0 && |
989 | key.objectid == bytenr && | |
85d4198e | 990 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
991 | err = -EAGAIN; |
992 | goto out; | |
993 | } | |
994 | } | |
995 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
996 | out: | |
85d4198e | 997 | if (insert) { |
5d4f98a2 YZ |
998 | path->keep_locks = 0; |
999 | btrfs_unlock_up_safe(path, 1); | |
1000 | } | |
1001 | return err; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * helper to add new inline back ref | |
1006 | */ | |
1007 | static noinline_for_stack | |
87bde3cd | 1008 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1009 | struct btrfs_path *path, |
1010 | struct btrfs_extent_inline_ref *iref, | |
1011 | u64 parent, u64 root_objectid, | |
1012 | u64 owner, u64 offset, int refs_to_add, | |
1013 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1014 | { |
1015 | struct extent_buffer *leaf; | |
1016 | struct btrfs_extent_item *ei; | |
1017 | unsigned long ptr; | |
1018 | unsigned long end; | |
1019 | unsigned long item_offset; | |
1020 | u64 refs; | |
1021 | int size; | |
1022 | int type; | |
5d4f98a2 YZ |
1023 | |
1024 | leaf = path->nodes[0]; | |
1025 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1026 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1027 | ||
1028 | type = extent_ref_type(parent, owner); | |
1029 | size = btrfs_extent_inline_ref_size(type); | |
1030 | ||
c71dd880 | 1031 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1032 | |
1033 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1034 | refs = btrfs_extent_refs(leaf, ei); | |
1035 | refs += refs_to_add; | |
1036 | btrfs_set_extent_refs(leaf, ei, refs); | |
1037 | if (extent_op) | |
1038 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1039 | ||
1040 | ptr = (unsigned long)ei + item_offset; | |
1041 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1042 | if (ptr < end - size) | |
1043 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1044 | end - size - ptr); | |
1045 | ||
1046 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1047 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1048 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1049 | struct btrfs_extent_data_ref *dref; | |
1050 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1051 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1052 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1053 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1054 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1055 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1056 | struct btrfs_shared_data_ref *sref; | |
1057 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1058 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1059 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1060 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1061 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1062 | } else { | |
1063 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1064 | } | |
1065 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1066 | } |
1067 | ||
1068 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1069 | struct btrfs_path *path, |
1070 | struct btrfs_extent_inline_ref **ref_ret, | |
1071 | u64 bytenr, u64 num_bytes, u64 parent, | |
1072 | u64 root_objectid, u64 owner, u64 offset) | |
1073 | { | |
1074 | int ret; | |
1075 | ||
867cc1fb NB |
1076 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1077 | num_bytes, parent, root_objectid, | |
1078 | owner, offset, 0); | |
5d4f98a2 | 1079 | if (ret != -ENOENT) |
54aa1f4d | 1080 | return ret; |
5d4f98a2 | 1081 | |
b3b4aa74 | 1082 | btrfs_release_path(path); |
5d4f98a2 YZ |
1083 | *ref_ret = NULL; |
1084 | ||
1085 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1086 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1087 | root_objectid); | |
5d4f98a2 | 1088 | } else { |
bd1d53ef NB |
1089 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1090 | root_objectid, owner, offset); | |
b9473439 | 1091 | } |
5d4f98a2 YZ |
1092 | return ret; |
1093 | } | |
31840ae1 | 1094 | |
5d4f98a2 YZ |
1095 | /* |
1096 | * helper to update/remove inline back ref | |
1097 | */ | |
1098 | static noinline_for_stack | |
61a18f1c | 1099 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1100 | struct btrfs_extent_inline_ref *iref, |
1101 | int refs_to_mod, | |
fcebe456 JB |
1102 | struct btrfs_delayed_extent_op *extent_op, |
1103 | int *last_ref) | |
5d4f98a2 | 1104 | { |
61a18f1c | 1105 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1106 | struct btrfs_extent_item *ei; |
1107 | struct btrfs_extent_data_ref *dref = NULL; | |
1108 | struct btrfs_shared_data_ref *sref = NULL; | |
1109 | unsigned long ptr; | |
1110 | unsigned long end; | |
1111 | u32 item_size; | |
1112 | int size; | |
1113 | int type; | |
5d4f98a2 YZ |
1114 | u64 refs; |
1115 | ||
5d4f98a2 YZ |
1116 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1117 | refs = btrfs_extent_refs(leaf, ei); | |
1118 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1119 | refs += refs_to_mod; | |
1120 | btrfs_set_extent_refs(leaf, ei, refs); | |
1121 | if (extent_op) | |
1122 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1123 | ||
3de28d57 LB |
1124 | /* |
1125 | * If type is invalid, we should have bailed out after | |
1126 | * lookup_inline_extent_backref(). | |
1127 | */ | |
1128 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1129 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1130 | |
1131 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1132 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1133 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1134 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1135 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1136 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1137 | } else { | |
1138 | refs = 1; | |
1139 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1140 | } |
31840ae1 | 1141 | |
5d4f98a2 YZ |
1142 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1143 | refs += refs_to_mod; | |
1144 | ||
1145 | if (refs > 0) { | |
1146 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1147 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1148 | else | |
1149 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1150 | } else { | |
fcebe456 | 1151 | *last_ref = 1; |
5d4f98a2 YZ |
1152 | size = btrfs_extent_inline_ref_size(type); |
1153 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1154 | ptr = (unsigned long)iref; | |
1155 | end = (unsigned long)ei + item_size; | |
1156 | if (ptr + size < end) | |
1157 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1158 | end - ptr - size); | |
1159 | item_size -= size; | |
78ac4f9e | 1160 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1161 | } |
1162 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1163 | } |
1164 | ||
1165 | static noinline_for_stack | |
1166 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1167 | struct btrfs_path *path, |
1168 | u64 bytenr, u64 num_bytes, u64 parent, | |
1169 | u64 root_objectid, u64 owner, | |
1170 | u64 offset, int refs_to_add, | |
1171 | struct btrfs_delayed_extent_op *extent_op) | |
1172 | { | |
1173 | struct btrfs_extent_inline_ref *iref; | |
1174 | int ret; | |
1175 | ||
867cc1fb NB |
1176 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1177 | num_bytes, parent, root_objectid, | |
1178 | owner, offset, 1); | |
5d4f98a2 | 1179 | if (ret == 0) { |
07cce5cf QW |
1180 | /* |
1181 | * We're adding refs to a tree block we already own, this | |
1182 | * should not happen at all. | |
1183 | */ | |
1184 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1185 | btrfs_crit(trans->fs_info, | |
1186 | "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu", | |
1187 | bytenr, num_bytes, root_objectid); | |
1188 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
1189 | WARN_ON(1); | |
1190 | btrfs_crit(trans->fs_info, | |
1191 | "path->slots[0]=%d path->nodes[0]:", path->slots[0]); | |
1192 | btrfs_print_leaf(path->nodes[0]); | |
1193 | } | |
1194 | return -EUCLEAN; | |
1195 | } | |
61a18f1c NB |
1196 | update_inline_extent_backref(path, iref, refs_to_add, |
1197 | extent_op, NULL); | |
5d4f98a2 | 1198 | } else if (ret == -ENOENT) { |
a639cdeb | 1199 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1200 | root_objectid, owner, offset, |
1201 | refs_to_add, extent_op); | |
1202 | ret = 0; | |
771ed689 | 1203 | } |
5d4f98a2 YZ |
1204 | return ret; |
1205 | } | |
31840ae1 | 1206 | |
5d4f98a2 | 1207 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1208 | struct btrfs_path *path, |
1209 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1210 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1211 | { |
143bede5 | 1212 | int ret = 0; |
b9473439 | 1213 | |
5d4f98a2 YZ |
1214 | BUG_ON(!is_data && refs_to_drop != 1); |
1215 | if (iref) { | |
61a18f1c NB |
1216 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1217 | last_ref); | |
5d4f98a2 | 1218 | } else if (is_data) { |
e9f6290d | 1219 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1220 | last_ref); |
5d4f98a2 | 1221 | } else { |
fcebe456 | 1222 | *last_ref = 1; |
87cc7a8a | 1223 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1224 | } |
1225 | return ret; | |
1226 | } | |
1227 | ||
d04c6b88 JM |
1228 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1229 | u64 *discarded_bytes) | |
5d4f98a2 | 1230 | { |
86557861 JM |
1231 | int j, ret = 0; |
1232 | u64 bytes_left, end; | |
4d89d377 | 1233 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1234 | |
4d89d377 JM |
1235 | if (WARN_ON(start != aligned_start)) { |
1236 | len -= aligned_start - start; | |
1237 | len = round_down(len, 1 << 9); | |
1238 | start = aligned_start; | |
1239 | } | |
d04c6b88 | 1240 | |
4d89d377 | 1241 | *discarded_bytes = 0; |
86557861 JM |
1242 | |
1243 | if (!len) | |
1244 | return 0; | |
1245 | ||
1246 | end = start + len; | |
1247 | bytes_left = len; | |
1248 | ||
1249 | /* Skip any superblocks on this device. */ | |
1250 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1251 | u64 sb_start = btrfs_sb_offset(j); | |
1252 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1253 | u64 size = sb_start - start; | |
1254 | ||
1255 | if (!in_range(sb_start, start, bytes_left) && | |
1256 | !in_range(sb_end, start, bytes_left) && | |
1257 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1258 | continue; | |
1259 | ||
1260 | /* | |
1261 | * Superblock spans beginning of range. Adjust start and | |
1262 | * try again. | |
1263 | */ | |
1264 | if (sb_start <= start) { | |
1265 | start += sb_end - start; | |
1266 | if (start > end) { | |
1267 | bytes_left = 0; | |
1268 | break; | |
1269 | } | |
1270 | bytes_left = end - start; | |
1271 | continue; | |
1272 | } | |
1273 | ||
1274 | if (size) { | |
1275 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1276 | GFP_NOFS, 0); | |
1277 | if (!ret) | |
1278 | *discarded_bytes += size; | |
1279 | else if (ret != -EOPNOTSUPP) | |
1280 | return ret; | |
1281 | } | |
1282 | ||
1283 | start = sb_end; | |
1284 | if (start > end) { | |
1285 | bytes_left = 0; | |
1286 | break; | |
1287 | } | |
1288 | bytes_left = end - start; | |
1289 | } | |
1290 | ||
1291 | if (bytes_left) { | |
1292 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1293 | GFP_NOFS, 0); |
1294 | if (!ret) | |
86557861 | 1295 | *discarded_bytes += bytes_left; |
4d89d377 | 1296 | } |
d04c6b88 | 1297 | return ret; |
5d4f98a2 | 1298 | } |
5d4f98a2 | 1299 | |
2ff7e61e | 1300 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1301 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1302 | { |
6b7faadd | 1303 | int ret = 0; |
5378e607 | 1304 | u64 discarded_bytes = 0; |
6b7faadd QW |
1305 | u64 end = bytenr + num_bytes; |
1306 | u64 cur = bytenr; | |
a1d3c478 | 1307 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1308 | |
e244a0ae | 1309 | |
2999241d FM |
1310 | /* |
1311 | * Avoid races with device replace and make sure our bbio has devices | |
1312 | * associated to its stripes that don't go away while we are discarding. | |
1313 | */ | |
0b246afa | 1314 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd QW |
1315 | while (cur < end) { |
1316 | struct btrfs_bio_stripe *stripe; | |
5d4f98a2 YZ |
1317 | int i; |
1318 | ||
6b7faadd QW |
1319 | num_bytes = end - cur; |
1320 | /* Tell the block device(s) that the sectors can be discarded */ | |
1321 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
1322 | &num_bytes, &bbio, 0); | |
1323 | /* | |
1324 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1325 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1326 | * thus we can't continue anyway. | |
1327 | */ | |
1328 | if (ret < 0) | |
1329 | goto out; | |
5d4f98a2 | 1330 | |
6b7faadd | 1331 | stripe = bbio->stripes; |
a1d3c478 | 1332 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1333 | u64 bytes; |
38b5f68e AJ |
1334 | struct request_queue *req_q; |
1335 | ||
627e0873 FM |
1336 | if (!stripe->dev->bdev) { |
1337 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1338 | continue; | |
1339 | } | |
38b5f68e AJ |
1340 | req_q = bdev_get_queue(stripe->dev->bdev); |
1341 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1342 | continue; |
1343 | ||
5378e607 LD |
1344 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1345 | stripe->physical, | |
d04c6b88 JM |
1346 | stripe->length, |
1347 | &bytes); | |
6b7faadd | 1348 | if (!ret) { |
d04c6b88 | 1349 | discarded_bytes += bytes; |
6b7faadd QW |
1350 | } else if (ret != -EOPNOTSUPP) { |
1351 | /* | |
1352 | * Logic errors or -ENOMEM, or -EIO, but | |
1353 | * unlikely to happen. | |
1354 | * | |
1355 | * And since there are two loops, explicitly | |
1356 | * go to out to avoid confusion. | |
1357 | */ | |
1358 | btrfs_put_bbio(bbio); | |
1359 | goto out; | |
1360 | } | |
d5e2003c JB |
1361 | |
1362 | /* | |
1363 | * Just in case we get back EOPNOTSUPP for some reason, | |
1364 | * just ignore the return value so we don't screw up | |
1365 | * people calling discard_extent. | |
1366 | */ | |
1367 | ret = 0; | |
5d4f98a2 | 1368 | } |
6e9606d2 | 1369 | btrfs_put_bbio(bbio); |
6b7faadd | 1370 | cur += num_bytes; |
5d4f98a2 | 1371 | } |
6b7faadd | 1372 | out: |
0b246afa | 1373 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1374 | |
1375 | if (actual_bytes) | |
1376 | *actual_bytes = discarded_bytes; | |
1377 | ||
5d4f98a2 | 1378 | |
53b381b3 DW |
1379 | if (ret == -EOPNOTSUPP) |
1380 | ret = 0; | |
5d4f98a2 | 1381 | return ret; |
5d4f98a2 YZ |
1382 | } |
1383 | ||
79787eaa | 1384 | /* Can return -ENOMEM */ |
5d4f98a2 | 1385 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1386 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1387 | { |
82fa113f | 1388 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1389 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1390 | int ret; |
66d7e7f0 | 1391 | |
82fa113f QW |
1392 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1393 | generic_ref->action); | |
1394 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1395 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1396 | |
82fa113f QW |
1397 | if (generic_ref->type == BTRFS_REF_METADATA) |
1398 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1399 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1400 | else |
1401 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1402 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1403 | |
82fa113f | 1404 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1405 | |
ddf30cf0 | 1406 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1407 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 1408 | |
5d4f98a2 YZ |
1409 | return ret; |
1410 | } | |
1411 | ||
bd3c685e NB |
1412 | /* |
1413 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1414 | * | |
07cce5cf QW |
1415 | * The counterpart is in __btrfs_free_extent(), with examples and more details |
1416 | * how it works. | |
1417 | * | |
bd3c685e NB |
1418 | * @trans: Handle of transaction |
1419 | * | |
1420 | * @node: The delayed ref node used to get the bytenr/length for | |
1421 | * extent whose references are incremented. | |
1422 | * | |
1423 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1424 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1425 | * bytenr of the parent block. Since new extents are always | |
1426 | * created with indirect references, this will only be the case | |
1427 | * when relocating a shared extent. In that case, root_objectid | |
1428 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1429 | * be 0 | |
1430 | * | |
1431 | * @root_objectid: The id of the root where this modification has originated, | |
1432 | * this can be either one of the well-known metadata trees or | |
1433 | * the subvolume id which references this extent. | |
1434 | * | |
1435 | * @owner: For data extents it is the inode number of the owning file. | |
1436 | * For metadata extents this parameter holds the level in the | |
1437 | * tree of the extent. | |
1438 | * | |
1439 | * @offset: For metadata extents the offset is ignored and is currently | |
1440 | * always passed as 0. For data extents it is the fileoffset | |
1441 | * this extent belongs to. | |
1442 | * | |
1443 | * @refs_to_add Number of references to add | |
1444 | * | |
1445 | * @extent_op Pointer to a structure, holding information necessary when | |
1446 | * updating a tree block's flags | |
1447 | * | |
1448 | */ | |
5d4f98a2 | 1449 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1450 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1451 | u64 parent, u64 root_objectid, |
1452 | u64 owner, u64 offset, int refs_to_add, | |
1453 | struct btrfs_delayed_extent_op *extent_op) | |
1454 | { | |
1455 | struct btrfs_path *path; | |
1456 | struct extent_buffer *leaf; | |
1457 | struct btrfs_extent_item *item; | |
fcebe456 | 1458 | struct btrfs_key key; |
c682f9b3 QW |
1459 | u64 bytenr = node->bytenr; |
1460 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1461 | u64 refs; |
1462 | int ret; | |
5d4f98a2 YZ |
1463 | |
1464 | path = btrfs_alloc_path(); | |
1465 | if (!path) | |
1466 | return -ENOMEM; | |
1467 | ||
5d4f98a2 YZ |
1468 | path->leave_spinning = 1; |
1469 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
1470 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1471 | parent, root_objectid, owner, | |
1472 | offset, refs_to_add, extent_op); | |
0ed4792a | 1473 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1474 | goto out; |
fcebe456 JB |
1475 | |
1476 | /* | |
1477 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1478 | * inline extent ref, so just update the reference count and add a | |
1479 | * normal backref. | |
1480 | */ | |
5d4f98a2 | 1481 | leaf = path->nodes[0]; |
fcebe456 | 1482 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1483 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1484 | refs = btrfs_extent_refs(leaf, item); | |
1485 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1486 | if (extent_op) | |
1487 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1488 | |
5d4f98a2 | 1489 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1490 | btrfs_release_path(path); |
56bec294 | 1491 | |
b9473439 | 1492 | path->leave_spinning = 1; |
56bec294 | 1493 | /* now insert the actual backref */ |
65cd6d9e NB |
1494 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
1495 | BUG_ON(refs_to_add != 1); | |
1496 | ret = insert_tree_block_ref(trans, path, bytenr, parent, | |
1497 | root_objectid); | |
1498 | } else { | |
1499 | ret = insert_extent_data_ref(trans, path, bytenr, parent, | |
1500 | root_objectid, owner, offset, | |
1501 | refs_to_add); | |
1502 | } | |
79787eaa | 1503 | if (ret) |
66642832 | 1504 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1505 | out: |
56bec294 | 1506 | btrfs_free_path(path); |
30d133fc | 1507 | return ret; |
56bec294 CM |
1508 | } |
1509 | ||
5d4f98a2 | 1510 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1511 | struct btrfs_delayed_ref_node *node, |
1512 | struct btrfs_delayed_extent_op *extent_op, | |
1513 | int insert_reserved) | |
56bec294 | 1514 | { |
5d4f98a2 YZ |
1515 | int ret = 0; |
1516 | struct btrfs_delayed_data_ref *ref; | |
1517 | struct btrfs_key ins; | |
1518 | u64 parent = 0; | |
1519 | u64 ref_root = 0; | |
1520 | u64 flags = 0; | |
1521 | ||
1522 | ins.objectid = node->bytenr; | |
1523 | ins.offset = node->num_bytes; | |
1524 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1525 | ||
1526 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1527 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1528 | |
5d4f98a2 YZ |
1529 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1530 | parent = ref->parent; | |
fcebe456 | 1531 | ref_root = ref->root; |
5d4f98a2 YZ |
1532 | |
1533 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1534 | if (extent_op) |
5d4f98a2 | 1535 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1536 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1537 | flags, ref->objectid, | |
1538 | ref->offset, &ins, | |
1539 | node->ref_mod); | |
5d4f98a2 | 1540 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1541 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1542 | ref->objectid, ref->offset, | |
1543 | node->ref_mod, extent_op); | |
5d4f98a2 | 1544 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1545 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1546 | ref_root, ref->objectid, |
1547 | ref->offset, node->ref_mod, | |
c682f9b3 | 1548 | extent_op); |
5d4f98a2 YZ |
1549 | } else { |
1550 | BUG(); | |
1551 | } | |
1552 | return ret; | |
1553 | } | |
1554 | ||
1555 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1556 | struct extent_buffer *leaf, | |
1557 | struct btrfs_extent_item *ei) | |
1558 | { | |
1559 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1560 | if (extent_op->update_flags) { | |
1561 | flags |= extent_op->flags_to_set; | |
1562 | btrfs_set_extent_flags(leaf, ei, flags); | |
1563 | } | |
1564 | ||
1565 | if (extent_op->update_key) { | |
1566 | struct btrfs_tree_block_info *bi; | |
1567 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1568 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1569 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1570 | } | |
1571 | } | |
1572 | ||
1573 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1574 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1575 | struct btrfs_delayed_extent_op *extent_op) |
1576 | { | |
20b9a2d6 | 1577 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1578 | struct btrfs_key key; |
1579 | struct btrfs_path *path; | |
1580 | struct btrfs_extent_item *ei; | |
1581 | struct extent_buffer *leaf; | |
1582 | u32 item_size; | |
56bec294 | 1583 | int ret; |
5d4f98a2 | 1584 | int err = 0; |
b1c79e09 | 1585 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1586 | |
bf31f87f | 1587 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1588 | return 0; |
1589 | ||
0b246afa | 1590 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1591 | metadata = 0; |
1592 | ||
5d4f98a2 YZ |
1593 | path = btrfs_alloc_path(); |
1594 | if (!path) | |
1595 | return -ENOMEM; | |
1596 | ||
d278850e | 1597 | key.objectid = head->bytenr; |
5d4f98a2 | 1598 | |
3173a18f | 1599 | if (metadata) { |
3173a18f | 1600 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1601 | key.offset = extent_op->level; |
3173a18f JB |
1602 | } else { |
1603 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1604 | key.offset = head->num_bytes; |
3173a18f JB |
1605 | } |
1606 | ||
1607 | again: | |
5d4f98a2 | 1608 | path->leave_spinning = 1; |
0b246afa | 1609 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1610 | if (ret < 0) { |
1611 | err = ret; | |
1612 | goto out; | |
1613 | } | |
1614 | if (ret > 0) { | |
3173a18f | 1615 | if (metadata) { |
55994887 FDBM |
1616 | if (path->slots[0] > 0) { |
1617 | path->slots[0]--; | |
1618 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1619 | path->slots[0]); | |
d278850e | 1620 | if (key.objectid == head->bytenr && |
55994887 | 1621 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1622 | key.offset == head->num_bytes) |
55994887 FDBM |
1623 | ret = 0; |
1624 | } | |
1625 | if (ret > 0) { | |
1626 | btrfs_release_path(path); | |
1627 | metadata = 0; | |
3173a18f | 1628 | |
d278850e JB |
1629 | key.objectid = head->bytenr; |
1630 | key.offset = head->num_bytes; | |
55994887 FDBM |
1631 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1632 | goto again; | |
1633 | } | |
1634 | } else { | |
1635 | err = -EIO; | |
1636 | goto out; | |
3173a18f | 1637 | } |
5d4f98a2 YZ |
1638 | } |
1639 | ||
1640 | leaf = path->nodes[0]; | |
1641 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1642 | |
6d8ff4e4 | 1643 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1644 | err = -EINVAL; |
1645 | btrfs_print_v0_err(fs_info); | |
1646 | btrfs_abort_transaction(trans, err); | |
1647 | goto out; | |
1648 | } | |
1649 | ||
5d4f98a2 YZ |
1650 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1651 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1652 | |
5d4f98a2 YZ |
1653 | btrfs_mark_buffer_dirty(leaf); |
1654 | out: | |
1655 | btrfs_free_path(path); | |
1656 | return err; | |
56bec294 CM |
1657 | } |
1658 | ||
5d4f98a2 | 1659 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1660 | struct btrfs_delayed_ref_node *node, |
1661 | struct btrfs_delayed_extent_op *extent_op, | |
1662 | int insert_reserved) | |
56bec294 CM |
1663 | { |
1664 | int ret = 0; | |
5d4f98a2 | 1665 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1666 | u64 parent = 0; |
1667 | u64 ref_root = 0; | |
56bec294 | 1668 | |
5d4f98a2 | 1669 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1670 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1671 | |
5d4f98a2 YZ |
1672 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1673 | parent = ref->parent; | |
fcebe456 | 1674 | ref_root = ref->root; |
5d4f98a2 | 1675 | |
02794222 | 1676 | if (node->ref_mod != 1) { |
f97806f2 | 1677 | btrfs_err(trans->fs_info, |
02794222 LB |
1678 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1679 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1680 | parent); | |
1681 | return -EIO; | |
1682 | } | |
5d4f98a2 | 1683 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1684 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1685 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1686 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1687 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1688 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1689 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1690 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1691 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1692 | } else { |
1693 | BUG(); | |
1694 | } | |
56bec294 CM |
1695 | return ret; |
1696 | } | |
1697 | ||
1698 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1699 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1700 | struct btrfs_delayed_ref_node *node, |
1701 | struct btrfs_delayed_extent_op *extent_op, | |
1702 | int insert_reserved) | |
56bec294 | 1703 | { |
79787eaa JM |
1704 | int ret = 0; |
1705 | ||
bf31f87f | 1706 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1707 | if (insert_reserved) |
b25c36f8 | 1708 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1709 | return 0; |
857cc2fc | 1710 | } |
79787eaa | 1711 | |
5d4f98a2 YZ |
1712 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1713 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1714 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1715 | insert_reserved); |
1716 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1717 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1718 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1719 | insert_reserved); |
1720 | else | |
1721 | BUG(); | |
80ee54bf | 1722 | if (ret && insert_reserved) |
b25c36f8 | 1723 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
5d4f98a2 | 1724 | return ret; |
56bec294 CM |
1725 | } |
1726 | ||
c6fc2454 | 1727 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1728 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1729 | { | |
cffc3374 FM |
1730 | struct btrfs_delayed_ref_node *ref; |
1731 | ||
e3d03965 | 1732 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1733 | return NULL; |
d7df2c79 | 1734 | |
cffc3374 FM |
1735 | /* |
1736 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1737 | * This is to prevent a ref count from going down to zero, which deletes | |
1738 | * the extent item from the extent tree, when there still are references | |
1739 | * to add, which would fail because they would not find the extent item. | |
1740 | */ | |
1d57ee94 WX |
1741 | if (!list_empty(&head->ref_add_list)) |
1742 | return list_first_entry(&head->ref_add_list, | |
1743 | struct btrfs_delayed_ref_node, add_list); | |
1744 | ||
e3d03965 | 1745 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1746 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1747 | ASSERT(list_empty(&ref->add_list)); |
1748 | return ref; | |
56bec294 CM |
1749 | } |
1750 | ||
2eadaa22 JB |
1751 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1752 | struct btrfs_delayed_ref_head *head) | |
1753 | { | |
1754 | spin_lock(&delayed_refs->lock); | |
1755 | head->processing = 0; | |
1756 | delayed_refs->num_heads_ready++; | |
1757 | spin_unlock(&delayed_refs->lock); | |
1758 | btrfs_delayed_ref_unlock(head); | |
1759 | } | |
1760 | ||
bedc6617 JB |
1761 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1762 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1763 | { |
1764 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1765 | |
1766 | if (!extent_op) | |
bedc6617 JB |
1767 | return NULL; |
1768 | ||
b00e6250 | 1769 | if (head->must_insert_reserved) { |
bedc6617 | 1770 | head->extent_op = NULL; |
b00e6250 | 1771 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1772 | return NULL; |
b00e6250 | 1773 | } |
bedc6617 JB |
1774 | return extent_op; |
1775 | } | |
1776 | ||
1777 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1778 | struct btrfs_delayed_ref_head *head) | |
1779 | { | |
1780 | struct btrfs_delayed_extent_op *extent_op; | |
1781 | int ret; | |
1782 | ||
1783 | extent_op = cleanup_extent_op(head); | |
1784 | if (!extent_op) | |
1785 | return 0; | |
1786 | head->extent_op = NULL; | |
b00e6250 | 1787 | spin_unlock(&head->lock); |
20b9a2d6 | 1788 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1789 | btrfs_free_delayed_extent_op(extent_op); |
1790 | return ret ? ret : 1; | |
1791 | } | |
1792 | ||
31890da0 JB |
1793 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1794 | struct btrfs_delayed_ref_root *delayed_refs, | |
1795 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1796 | { |
ba2c4d4e | 1797 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
1798 | |
1799 | if (head->total_ref_mod < 0) { | |
1800 | struct btrfs_space_info *space_info; | |
1801 | u64 flags; | |
1802 | ||
1803 | if (head->is_data) | |
1804 | flags = BTRFS_BLOCK_GROUP_DATA; | |
1805 | else if (head->is_system) | |
1806 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
1807 | else | |
1808 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 1809 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
1810 | ASSERT(space_info); |
1811 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1812 | -head->num_bytes, | |
1813 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1814 | ||
ba2c4d4e JB |
1815 | /* |
1816 | * We had csum deletions accounted for in our delayed refs rsv, | |
1817 | * we need to drop the csum leaves for this update from our | |
1818 | * delayed_refs_rsv. | |
1819 | */ | |
07c47775 JB |
1820 | if (head->is_data) { |
1821 | spin_lock(&delayed_refs->lock); | |
1822 | delayed_refs->pending_csums -= head->num_bytes; | |
1823 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
1824 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
1825 | head->num_bytes); | |
07c47775 JB |
1826 | } |
1827 | } | |
1828 | ||
ba2c4d4e | 1829 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1830 | } |
1831 | ||
194ab0bc | 1832 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1833 | struct btrfs_delayed_ref_head *head) |
1834 | { | |
f9871edd NB |
1835 | |
1836 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1837 | struct btrfs_delayed_ref_root *delayed_refs; |
1838 | int ret; | |
1839 | ||
1840 | delayed_refs = &trans->transaction->delayed_refs; | |
1841 | ||
bedc6617 | 1842 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1843 | if (ret < 0) { |
1844 | unselect_delayed_ref_head(delayed_refs, head); | |
1845 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1846 | return ret; | |
1847 | } else if (ret) { | |
1848 | return ret; | |
1849 | } | |
1850 | ||
1851 | /* | |
1852 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1853 | * and then re-check to make sure nobody got added. | |
1854 | */ | |
1855 | spin_unlock(&head->lock); | |
1856 | spin_lock(&delayed_refs->lock); | |
1857 | spin_lock(&head->lock); | |
e3d03965 | 1858 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1859 | spin_unlock(&head->lock); |
1860 | spin_unlock(&delayed_refs->lock); | |
1861 | return 1; | |
1862 | } | |
d7baffda | 1863 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1864 | spin_unlock(&head->lock); |
1e7a1421 | 1865 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1866 | |
c1103f7a | 1867 | if (head->must_insert_reserved) { |
b25c36f8 | 1868 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1869 | if (head->is_data) { |
40e046ac FM |
1870 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1871 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1872 | } |
1873 | } | |
1874 | ||
31890da0 | 1875 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1876 | |
1877 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1878 | btrfs_delayed_ref_unlock(head); |
d278850e | 1879 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1880 | return 0; |
1881 | } | |
1882 | ||
b1cdbcb5 NB |
1883 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1884 | struct btrfs_trans_handle *trans) | |
1885 | { | |
1886 | struct btrfs_delayed_ref_root *delayed_refs = | |
1887 | &trans->transaction->delayed_refs; | |
1888 | struct btrfs_delayed_ref_head *head = NULL; | |
1889 | int ret; | |
1890 | ||
1891 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1892 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1893 | if (!head) { |
1894 | spin_unlock(&delayed_refs->lock); | |
1895 | return head; | |
1896 | } | |
1897 | ||
1898 | /* | |
1899 | * Grab the lock that says we are going to process all the refs for | |
1900 | * this head | |
1901 | */ | |
9e920a6f | 1902 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1903 | spin_unlock(&delayed_refs->lock); |
1904 | ||
1905 | /* | |
1906 | * We may have dropped the spin lock to get the head mutex lock, and | |
1907 | * that might have given someone else time to free the head. If that's | |
1908 | * true, it has been removed from our list and we can move on. | |
1909 | */ | |
1910 | if (ret == -EAGAIN) | |
1911 | head = ERR_PTR(-EAGAIN); | |
1912 | ||
1913 | return head; | |
1914 | } | |
1915 | ||
e7261386 NB |
1916 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1917 | struct btrfs_delayed_ref_head *locked_ref, | |
1918 | unsigned long *run_refs) | |
1919 | { | |
1920 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1921 | struct btrfs_delayed_ref_root *delayed_refs; | |
1922 | struct btrfs_delayed_extent_op *extent_op; | |
1923 | struct btrfs_delayed_ref_node *ref; | |
1924 | int must_insert_reserved = 0; | |
1925 | int ret; | |
1926 | ||
1927 | delayed_refs = &trans->transaction->delayed_refs; | |
1928 | ||
0110a4c4 NB |
1929 | lockdep_assert_held(&locked_ref->mutex); |
1930 | lockdep_assert_held(&locked_ref->lock); | |
1931 | ||
e7261386 NB |
1932 | while ((ref = select_delayed_ref(locked_ref))) { |
1933 | if (ref->seq && | |
1934 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1935 | spin_unlock(&locked_ref->lock); | |
1936 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1937 | return -EAGAIN; | |
1938 | } | |
1939 | ||
1940 | (*run_refs)++; | |
1941 | ref->in_tree = 0; | |
1942 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1943 | RB_CLEAR_NODE(&ref->ref_node); | |
1944 | if (!list_empty(&ref->add_list)) | |
1945 | list_del(&ref->add_list); | |
1946 | /* | |
1947 | * When we play the delayed ref, also correct the ref_mod on | |
1948 | * head | |
1949 | */ | |
1950 | switch (ref->action) { | |
1951 | case BTRFS_ADD_DELAYED_REF: | |
1952 | case BTRFS_ADD_DELAYED_EXTENT: | |
1953 | locked_ref->ref_mod -= ref->ref_mod; | |
1954 | break; | |
1955 | case BTRFS_DROP_DELAYED_REF: | |
1956 | locked_ref->ref_mod += ref->ref_mod; | |
1957 | break; | |
1958 | default: | |
1959 | WARN_ON(1); | |
1960 | } | |
1961 | atomic_dec(&delayed_refs->num_entries); | |
1962 | ||
1963 | /* | |
1964 | * Record the must_insert_reserved flag before we drop the | |
1965 | * spin lock. | |
1966 | */ | |
1967 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1968 | locked_ref->must_insert_reserved = 0; | |
1969 | ||
1970 | extent_op = locked_ref->extent_op; | |
1971 | locked_ref->extent_op = NULL; | |
1972 | spin_unlock(&locked_ref->lock); | |
1973 | ||
1974 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1975 | must_insert_reserved); | |
1976 | ||
1977 | btrfs_free_delayed_extent_op(extent_op); | |
1978 | if (ret) { | |
1979 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1980 | btrfs_put_delayed_ref(ref); | |
1981 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1982 | ret); | |
1983 | return ret; | |
1984 | } | |
1985 | ||
1986 | btrfs_put_delayed_ref(ref); | |
1987 | cond_resched(); | |
1988 | ||
1989 | spin_lock(&locked_ref->lock); | |
1990 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1991 | } | |
1992 | ||
1993 | return 0; | |
1994 | } | |
1995 | ||
79787eaa JM |
1996 | /* |
1997 | * Returns 0 on success or if called with an already aborted transaction. | |
1998 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1999 | */ | |
d7df2c79 | 2000 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2001 | unsigned long nr) |
56bec294 | 2002 | { |
0a1e458a | 2003 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2004 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2005 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2006 | ktime_t start = ktime_get(); |
56bec294 | 2007 | int ret; |
d7df2c79 | 2008 | unsigned long count = 0; |
0a2b2a84 | 2009 | unsigned long actual_count = 0; |
56bec294 CM |
2010 | |
2011 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2012 | do { |
56bec294 | 2013 | if (!locked_ref) { |
b1cdbcb5 | 2014 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2015 | if (IS_ERR_OR_NULL(locked_ref)) { |
2016 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2017 | continue; | |
2018 | } else { | |
2019 | break; | |
2020 | } | |
56bec294 | 2021 | } |
0110a4c4 | 2022 | count++; |
56bec294 | 2023 | } |
2c3cf7d5 FM |
2024 | /* |
2025 | * We need to try and merge add/drops of the same ref since we | |
2026 | * can run into issues with relocate dropping the implicit ref | |
2027 | * and then it being added back again before the drop can | |
2028 | * finish. If we merged anything we need to re-loop so we can | |
2029 | * get a good ref. | |
2030 | * Or we can get node references of the same type that weren't | |
2031 | * merged when created due to bumps in the tree mod seq, and | |
2032 | * we need to merge them to prevent adding an inline extent | |
2033 | * backref before dropping it (triggering a BUG_ON at | |
2034 | * insert_inline_extent_backref()). | |
2035 | */ | |
d7df2c79 | 2036 | spin_lock(&locked_ref->lock); |
be97f133 | 2037 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2038 | |
0110a4c4 NB |
2039 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2040 | &actual_count); | |
2041 | if (ret < 0 && ret != -EAGAIN) { | |
2042 | /* | |
2043 | * Error, btrfs_run_delayed_refs_for_head already | |
2044 | * unlocked everything so just bail out | |
2045 | */ | |
2046 | return ret; | |
2047 | } else if (!ret) { | |
2048 | /* | |
2049 | * Success, perform the usual cleanup of a processed | |
2050 | * head | |
2051 | */ | |
f9871edd | 2052 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2053 | if (ret > 0 ) { |
b00e6250 JB |
2054 | /* We dropped our lock, we need to loop. */ |
2055 | ret = 0; | |
d7df2c79 | 2056 | continue; |
194ab0bc JB |
2057 | } else if (ret) { |
2058 | return ret; | |
5d4f98a2 | 2059 | } |
22cd2e7d | 2060 | } |
1ce7a5ec | 2061 | |
b00e6250 | 2062 | /* |
0110a4c4 NB |
2063 | * Either success case or btrfs_run_delayed_refs_for_head |
2064 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2065 | */ |
b00e6250 | 2066 | |
0110a4c4 | 2067 | locked_ref = NULL; |
c3e69d58 | 2068 | cond_resched(); |
0110a4c4 | 2069 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2070 | |
2071 | /* | |
2072 | * We don't want to include ref heads since we can have empty ref heads | |
2073 | * and those will drastically skew our runtime down since we just do | |
2074 | * accounting, no actual extent tree updates. | |
2075 | */ | |
2076 | if (actual_count > 0) { | |
2077 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2078 | u64 avg; | |
2079 | ||
2080 | /* | |
2081 | * We weigh the current average higher than our current runtime | |
2082 | * to avoid large swings in the average. | |
2083 | */ | |
2084 | spin_lock(&delayed_refs->lock); | |
2085 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2086 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2087 | spin_unlock(&delayed_refs->lock); |
2088 | } | |
d7df2c79 | 2089 | return 0; |
c3e69d58 CM |
2090 | } |
2091 | ||
709c0486 AJ |
2092 | #ifdef SCRAMBLE_DELAYED_REFS |
2093 | /* | |
2094 | * Normally delayed refs get processed in ascending bytenr order. This | |
2095 | * correlates in most cases to the order added. To expose dependencies on this | |
2096 | * order, we start to process the tree in the middle instead of the beginning | |
2097 | */ | |
2098 | static u64 find_middle(struct rb_root *root) | |
2099 | { | |
2100 | struct rb_node *n = root->rb_node; | |
2101 | struct btrfs_delayed_ref_node *entry; | |
2102 | int alt = 1; | |
2103 | u64 middle; | |
2104 | u64 first = 0, last = 0; | |
2105 | ||
2106 | n = rb_first(root); | |
2107 | if (n) { | |
2108 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2109 | first = entry->bytenr; | |
2110 | } | |
2111 | n = rb_last(root); | |
2112 | if (n) { | |
2113 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2114 | last = entry->bytenr; | |
2115 | } | |
2116 | n = root->rb_node; | |
2117 | ||
2118 | while (n) { | |
2119 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2120 | WARN_ON(!entry->in_tree); | |
2121 | ||
2122 | middle = entry->bytenr; | |
2123 | ||
2124 | if (alt) | |
2125 | n = n->rb_left; | |
2126 | else | |
2127 | n = n->rb_right; | |
2128 | ||
2129 | alt = 1 - alt; | |
2130 | } | |
2131 | return middle; | |
2132 | } | |
2133 | #endif | |
2134 | ||
1262133b JB |
2135 | /* |
2136 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2137 | * would require to store the csums for that many bytes. | |
2138 | */ | |
2ff7e61e | 2139 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2140 | { |
2141 | u64 csum_size; | |
2142 | u64 num_csums_per_leaf; | |
2143 | u64 num_csums; | |
2144 | ||
0b246afa | 2145 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2146 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2147 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2148 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2149 | num_csums += num_csums_per_leaf - 1; |
2150 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2151 | return num_csums; | |
2152 | } | |
2153 | ||
c3e69d58 CM |
2154 | /* |
2155 | * this starts processing the delayed reference count updates and | |
2156 | * extent insertions we have queued up so far. count can be | |
2157 | * 0, which means to process everything in the tree at the start | |
2158 | * of the run (but not newly added entries), or it can be some target | |
2159 | * number you'd like to process. | |
79787eaa JM |
2160 | * |
2161 | * Returns 0 on success or if called with an aborted transaction | |
2162 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2163 | */ |
2164 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2165 | unsigned long count) |
c3e69d58 | 2166 | { |
c79a70b1 | 2167 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2168 | struct rb_node *node; |
2169 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2170 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2171 | int ret; |
2172 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2173 | |
79787eaa | 2174 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2175 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2176 | return 0; |
2177 | ||
0b246afa | 2178 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2179 | return 0; |
2180 | ||
c3e69d58 | 2181 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2182 | if (count == 0) |
d7df2c79 | 2183 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2184 | |
c3e69d58 | 2185 | again: |
709c0486 AJ |
2186 | #ifdef SCRAMBLE_DELAYED_REFS |
2187 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2188 | #endif | |
0a1e458a | 2189 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2190 | if (ret < 0) { |
66642832 | 2191 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2192 | return ret; |
eb099670 | 2193 | } |
c3e69d58 | 2194 | |
56bec294 | 2195 | if (run_all) { |
119e80df | 2196 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2197 | |
d7df2c79 | 2198 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2199 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2200 | if (!node) { |
2201 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2202 | goto out; |
d7df2c79 | 2203 | } |
d278850e JB |
2204 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2205 | href_node); | |
2206 | refcount_inc(&head->refs); | |
2207 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2208 | |
d278850e JB |
2209 | /* Mutex was contended, block until it's released and retry. */ |
2210 | mutex_lock(&head->mutex); | |
2211 | mutex_unlock(&head->mutex); | |
56bec294 | 2212 | |
d278850e | 2213 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2214 | cond_resched(); |
56bec294 | 2215 | goto again; |
5f39d397 | 2216 | } |
54aa1f4d | 2217 | out: |
a28ec197 CM |
2218 | return 0; |
2219 | } | |
2220 | ||
5d4f98a2 | 2221 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
42c9d0b5 | 2222 | struct extent_buffer *eb, u64 flags, |
b1c79e09 | 2223 | int level, int is_data) |
5d4f98a2 YZ |
2224 | { |
2225 | struct btrfs_delayed_extent_op *extent_op; | |
2226 | int ret; | |
2227 | ||
78a6184a | 2228 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2229 | if (!extent_op) |
2230 | return -ENOMEM; | |
2231 | ||
2232 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2233 | extent_op->update_flags = true; |
2234 | extent_op->update_key = false; | |
2235 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2236 | extent_op->level = level; |
5d4f98a2 | 2237 | |
42c9d0b5 | 2238 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2239 | if (ret) |
78a6184a | 2240 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2241 | return ret; |
2242 | } | |
2243 | ||
e4c3b2dc | 2244 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2245 | struct btrfs_path *path, |
2246 | u64 objectid, u64 offset, u64 bytenr) | |
2247 | { | |
2248 | struct btrfs_delayed_ref_head *head; | |
2249 | struct btrfs_delayed_ref_node *ref; | |
2250 | struct btrfs_delayed_data_ref *data_ref; | |
2251 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2252 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2253 | struct rb_node *node; |
5d4f98a2 YZ |
2254 | int ret = 0; |
2255 | ||
998ac6d2 | 2256 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2257 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2258 | if (cur_trans) |
2259 | refcount_inc(&cur_trans->use_count); | |
2260 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2261 | if (!cur_trans) |
2262 | return 0; | |
2263 | ||
2264 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2265 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2266 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2267 | if (!head) { |
2268 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2269 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2270 | return 0; |
2271 | } | |
5d4f98a2 YZ |
2272 | |
2273 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2274 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2275 | spin_unlock(&delayed_refs->lock); |
2276 | ||
b3b4aa74 | 2277 | btrfs_release_path(path); |
5d4f98a2 | 2278 | |
8cc33e5c DS |
2279 | /* |
2280 | * Mutex was contended, block until it's released and let | |
2281 | * caller try again | |
2282 | */ | |
5d4f98a2 YZ |
2283 | mutex_lock(&head->mutex); |
2284 | mutex_unlock(&head->mutex); | |
d278850e | 2285 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2286 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2287 | return -EAGAIN; |
2288 | } | |
d7df2c79 | 2289 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2290 | |
d7df2c79 | 2291 | spin_lock(&head->lock); |
0e0adbcf JB |
2292 | /* |
2293 | * XXX: We should replace this with a proper search function in the | |
2294 | * future. | |
2295 | */ | |
e3d03965 LB |
2296 | for (node = rb_first_cached(&head->ref_tree); node; |
2297 | node = rb_next(node)) { | |
0e0adbcf | 2298 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2299 | /* If it's a shared ref we know a cross reference exists */ |
2300 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2301 | ret = 1; | |
2302 | break; | |
2303 | } | |
5d4f98a2 | 2304 | |
d7df2c79 | 2305 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2306 | |
d7df2c79 JB |
2307 | /* |
2308 | * If our ref doesn't match the one we're currently looking at | |
2309 | * then we have a cross reference. | |
2310 | */ | |
2311 | if (data_ref->root != root->root_key.objectid || | |
2312 | data_ref->objectid != objectid || | |
2313 | data_ref->offset != offset) { | |
2314 | ret = 1; | |
2315 | break; | |
2316 | } | |
5d4f98a2 | 2317 | } |
d7df2c79 | 2318 | spin_unlock(&head->lock); |
5d4f98a2 | 2319 | mutex_unlock(&head->mutex); |
998ac6d2 | 2320 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2321 | return ret; |
2322 | } | |
2323 | ||
e4c3b2dc | 2324 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2325 | struct btrfs_path *path, |
a84d5d42 BB |
2326 | u64 objectid, u64 offset, u64 bytenr, |
2327 | bool strict) | |
be20aa9d | 2328 | { |
0b246afa JM |
2329 | struct btrfs_fs_info *fs_info = root->fs_info; |
2330 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2331 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2332 | struct btrfs_extent_data_ref *ref; |
2333 | struct btrfs_extent_inline_ref *iref; | |
2334 | struct btrfs_extent_item *ei; | |
f321e491 | 2335 | struct btrfs_key key; |
5d4f98a2 | 2336 | u32 item_size; |
3de28d57 | 2337 | int type; |
be20aa9d | 2338 | int ret; |
925baedd | 2339 | |
be20aa9d | 2340 | key.objectid = bytenr; |
31840ae1 | 2341 | key.offset = (u64)-1; |
f321e491 | 2342 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2343 | |
be20aa9d CM |
2344 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2345 | if (ret < 0) | |
2346 | goto out; | |
79787eaa | 2347 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2348 | |
2349 | ret = -ENOENT; | |
2350 | if (path->slots[0] == 0) | |
31840ae1 | 2351 | goto out; |
be20aa9d | 2352 | |
31840ae1 | 2353 | path->slots[0]--; |
f321e491 | 2354 | leaf = path->nodes[0]; |
5d4f98a2 | 2355 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2356 | |
5d4f98a2 | 2357 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2358 | goto out; |
f321e491 | 2359 | |
5d4f98a2 YZ |
2360 | ret = 1; |
2361 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2362 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2363 | |
a6bd9cd1 | 2364 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2365 | if (item_size != sizeof(*ei) + |
2366 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2367 | goto out; | |
be20aa9d | 2368 | |
a84d5d42 BB |
2369 | /* |
2370 | * If extent created before last snapshot => it's shared unless the | |
2371 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2372 | */ | |
2373 | if (!strict && | |
2374 | (btrfs_extent_generation(leaf, ei) <= | |
2375 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2376 | goto out; |
2377 | ||
2378 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2379 | |
a6bd9cd1 | 2380 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2381 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2382 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2383 | goto out; |
2384 | ||
2385 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2386 | if (btrfs_extent_refs(leaf, ei) != | |
2387 | btrfs_extent_data_ref_count(leaf, ref) || | |
2388 | btrfs_extent_data_ref_root(leaf, ref) != | |
2389 | root->root_key.objectid || | |
2390 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2391 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2392 | goto out; | |
2393 | ||
2394 | ret = 0; | |
2395 | out: | |
2396 | return ret; | |
2397 | } | |
2398 | ||
e4c3b2dc | 2399 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
a84d5d42 | 2400 | u64 bytenr, bool strict) |
5d4f98a2 YZ |
2401 | { |
2402 | struct btrfs_path *path; | |
2403 | int ret; | |
5d4f98a2 YZ |
2404 | |
2405 | path = btrfs_alloc_path(); | |
2406 | if (!path) | |
9132c4ff | 2407 | return -ENOMEM; |
5d4f98a2 YZ |
2408 | |
2409 | do { | |
e4c3b2dc | 2410 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2411 | offset, bytenr, strict); |
5d4f98a2 | 2412 | if (ret && ret != -ENOENT) |
f321e491 | 2413 | goto out; |
80ff3856 | 2414 | |
380fd066 MT |
2415 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2416 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2417 | |
be20aa9d | 2418 | out: |
80ff3856 | 2419 | btrfs_free_path(path); |
f0486c68 YZ |
2420 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2421 | WARN_ON(ret > 0); | |
f321e491 | 2422 | return ret; |
be20aa9d | 2423 | } |
c5739bba | 2424 | |
5d4f98a2 | 2425 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2426 | struct btrfs_root *root, |
5d4f98a2 | 2427 | struct extent_buffer *buf, |
e339a6b0 | 2428 | int full_backref, int inc) |
31840ae1 | 2429 | { |
0b246afa | 2430 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2431 | u64 bytenr; |
5d4f98a2 YZ |
2432 | u64 num_bytes; |
2433 | u64 parent; | |
31840ae1 | 2434 | u64 ref_root; |
31840ae1 | 2435 | u32 nritems; |
31840ae1 ZY |
2436 | struct btrfs_key key; |
2437 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2438 | struct btrfs_ref generic_ref = { 0 }; |
2439 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2440 | int i; |
82fa113f | 2441 | int action; |
31840ae1 ZY |
2442 | int level; |
2443 | int ret = 0; | |
fccb84c9 | 2444 | |
0b246afa | 2445 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2446 | return 0; |
fccb84c9 | 2447 | |
31840ae1 | 2448 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2449 | nritems = btrfs_header_nritems(buf); |
2450 | level = btrfs_header_level(buf); | |
2451 | ||
92a7cc42 | 2452 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2453 | return 0; |
31840ae1 | 2454 | |
5d4f98a2 YZ |
2455 | if (full_backref) |
2456 | parent = buf->start; | |
2457 | else | |
2458 | parent = 0; | |
82fa113f QW |
2459 | if (inc) |
2460 | action = BTRFS_ADD_DELAYED_REF; | |
2461 | else | |
2462 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2463 | |
2464 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2465 | if (level == 0) { |
5d4f98a2 | 2466 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2467 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2468 | continue; |
5d4f98a2 | 2469 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2470 | struct btrfs_file_extent_item); |
2471 | if (btrfs_file_extent_type(buf, fi) == | |
2472 | BTRFS_FILE_EXTENT_INLINE) | |
2473 | continue; | |
2474 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2475 | if (bytenr == 0) | |
2476 | continue; | |
5d4f98a2 YZ |
2477 | |
2478 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2479 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2480 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2481 | num_bytes, parent); | |
2482 | generic_ref.real_root = root->root_key.objectid; | |
2483 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2484 | key.offset); | |
2485 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2486 | if (inc) |
82fa113f | 2487 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2488 | else |
ffd4bb2a | 2489 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2490 | if (ret) |
2491 | goto fail; | |
2492 | } else { | |
5d4f98a2 | 2493 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2494 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2495 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2496 | num_bytes, parent); | |
2497 | generic_ref.real_root = root->root_key.objectid; | |
2498 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2499 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2500 | if (inc) |
82fa113f | 2501 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2502 | else |
ffd4bb2a | 2503 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2504 | if (ret) |
2505 | goto fail; | |
2506 | } | |
2507 | } | |
2508 | return 0; | |
2509 | fail: | |
5d4f98a2 YZ |
2510 | return ret; |
2511 | } | |
2512 | ||
2513 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2514 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2515 | { |
e339a6b0 | 2516 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2517 | } |
2518 | ||
2519 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2520 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2521 | { |
e339a6b0 | 2522 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2523 | } |
2524 | ||
2ff7e61e | 2525 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 | 2526 | { |
32da5386 | 2527 | struct btrfs_block_group *block_group; |
d2fb3437 YZ |
2528 | int readonly = 0; |
2529 | ||
0b246afa | 2530 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2531 | if (!block_group || block_group->ro) |
2532 | readonly = 1; | |
2533 | if (block_group) | |
fa9c0d79 | 2534 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2535 | return readonly; |
2536 | } | |
2537 | ||
1b86826d | 2538 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2539 | { |
0b246afa | 2540 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2541 | u64 flags; |
53b381b3 | 2542 | u64 ret; |
9ed74f2d | 2543 | |
b742bb82 YZ |
2544 | if (data) |
2545 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2546 | else if (root == fs_info->chunk_root) |
b742bb82 | 2547 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2548 | else |
b742bb82 | 2549 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2550 | |
878d7b67 | 2551 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2552 | return ret; |
6a63209f | 2553 | } |
9ed74f2d | 2554 | |
2ff7e61e | 2555 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2556 | { |
32da5386 | 2557 | struct btrfs_block_group *cache; |
d2fb3437 | 2558 | u64 bytenr; |
0f9dd46c | 2559 | |
0b246afa JM |
2560 | spin_lock(&fs_info->block_group_cache_lock); |
2561 | bytenr = fs_info->first_logical_byte; | |
2562 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2563 | |
2564 | if (bytenr < (u64)-1) | |
2565 | return bytenr; | |
2566 | ||
0b246afa | 2567 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2568 | if (!cache) |
a061fc8d | 2569 | return 0; |
0f9dd46c | 2570 | |
b3470b5d | 2571 | bytenr = cache->start; |
fa9c0d79 | 2572 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2573 | |
2574 | return bytenr; | |
a061fc8d CM |
2575 | } |
2576 | ||
6690d071 NB |
2577 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2578 | struct btrfs_block_group *cache, | |
f0486c68 | 2579 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2580 | { |
fdf08605 DS |
2581 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2582 | ||
11833d66 YZ |
2583 | spin_lock(&cache->space_info->lock); |
2584 | spin_lock(&cache->lock); | |
2585 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2586 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2587 | num_bytes); | |
11833d66 YZ |
2588 | if (reserved) { |
2589 | cache->reserved -= num_bytes; | |
2590 | cache->space_info->bytes_reserved -= num_bytes; | |
2591 | } | |
2592 | spin_unlock(&cache->lock); | |
2593 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2594 | |
dec59fa3 EL |
2595 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
2596 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
fe119a6e | 2597 | set_extent_dirty(&trans->transaction->pinned_extents, bytenr, |
f0486c68 YZ |
2598 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2599 | return 0; | |
2600 | } | |
68b38550 | 2601 | |
b25c36f8 | 2602 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2603 | u64 bytenr, u64 num_bytes, int reserved) |
2604 | { | |
32da5386 | 2605 | struct btrfs_block_group *cache; |
68b38550 | 2606 | |
b25c36f8 | 2607 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2608 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2609 | |
6690d071 | 2610 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2611 | |
2612 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2613 | return 0; |
2614 | } | |
2615 | ||
f0486c68 | 2616 | /* |
e688b725 CM |
2617 | * this function must be called within transaction |
2618 | */ | |
9fce5704 | 2619 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
e688b725 CM |
2620 | u64 bytenr, u64 num_bytes) |
2621 | { | |
32da5386 | 2622 | struct btrfs_block_group *cache; |
b50c6e25 | 2623 | int ret; |
e688b725 | 2624 | |
f2fb7298 NB |
2625 | btrfs_add_excluded_extent(trans->fs_info, bytenr, num_bytes); |
2626 | ||
9fce5704 | 2627 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
b50c6e25 JB |
2628 | if (!cache) |
2629 | return -EINVAL; | |
e688b725 CM |
2630 | |
2631 | /* | |
2632 | * pull in the free space cache (if any) so that our pin | |
2633 | * removes the free space from the cache. We have load_only set | |
2634 | * to one because the slow code to read in the free extents does check | |
2635 | * the pinned extents. | |
2636 | */ | |
676f1f75 | 2637 | btrfs_cache_block_group(cache, 1); |
e688b725 | 2638 | |
6690d071 | 2639 | pin_down_extent(trans, cache, bytenr, num_bytes, 0); |
e688b725 CM |
2640 | |
2641 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2642 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 2643 | btrfs_put_block_group(cache); |
b50c6e25 | 2644 | return ret; |
e688b725 CM |
2645 | } |
2646 | ||
2ff7e61e JM |
2647 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2648 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2649 | { |
2650 | int ret; | |
32da5386 | 2651 | struct btrfs_block_group *block_group; |
8c2a1a30 JB |
2652 | struct btrfs_caching_control *caching_ctl; |
2653 | ||
0b246afa | 2654 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2655 | if (!block_group) |
2656 | return -EINVAL; | |
2657 | ||
676f1f75 | 2658 | btrfs_cache_block_group(block_group, 0); |
e3cb339f | 2659 | caching_ctl = btrfs_get_caching_control(block_group); |
8c2a1a30 JB |
2660 | |
2661 | if (!caching_ctl) { | |
2662 | /* Logic error */ | |
32da5386 | 2663 | BUG_ON(!btrfs_block_group_done(block_group)); |
8c2a1a30 JB |
2664 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2665 | } else { | |
2666 | mutex_lock(&caching_ctl->mutex); | |
2667 | ||
2668 | if (start >= caching_ctl->progress) { | |
6f410d1b JB |
2669 | ret = btrfs_add_excluded_extent(fs_info, start, |
2670 | num_bytes); | |
8c2a1a30 JB |
2671 | } else if (start + num_bytes <= caching_ctl->progress) { |
2672 | ret = btrfs_remove_free_space(block_group, | |
2673 | start, num_bytes); | |
2674 | } else { | |
2675 | num_bytes = caching_ctl->progress - start; | |
2676 | ret = btrfs_remove_free_space(block_group, | |
2677 | start, num_bytes); | |
2678 | if (ret) | |
2679 | goto out_lock; | |
2680 | ||
2681 | num_bytes = (start + num_bytes) - | |
2682 | caching_ctl->progress; | |
2683 | start = caching_ctl->progress; | |
6f410d1b JB |
2684 | ret = btrfs_add_excluded_extent(fs_info, start, |
2685 | num_bytes); | |
8c2a1a30 JB |
2686 | } |
2687 | out_lock: | |
2688 | mutex_unlock(&caching_ctl->mutex); | |
e3cb339f | 2689 | btrfs_put_caching_control(caching_ctl); |
8c2a1a30 JB |
2690 | } |
2691 | btrfs_put_block_group(block_group); | |
2692 | return ret; | |
2693 | } | |
2694 | ||
bcdc428c | 2695 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2696 | { |
bcdc428c | 2697 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2698 | struct btrfs_file_extent_item *item; |
2699 | struct btrfs_key key; | |
2700 | int found_type; | |
2701 | int i; | |
b89311ef | 2702 | int ret = 0; |
8c2a1a30 | 2703 | |
2ff7e61e | 2704 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2705 | return 0; |
2706 | ||
2707 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2708 | btrfs_item_key_to_cpu(eb, &key, i); | |
2709 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2710 | continue; | |
2711 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2712 | found_type = btrfs_file_extent_type(eb, item); | |
2713 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2714 | continue; | |
2715 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2716 | continue; | |
2717 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2718 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2719 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2720 | if (ret) | |
2721 | break; | |
8c2a1a30 JB |
2722 | } |
2723 | ||
b89311ef | 2724 | return ret; |
8c2a1a30 JB |
2725 | } |
2726 | ||
9cfa3e34 | 2727 | static void |
32da5386 | 2728 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2729 | { |
2730 | atomic_inc(&bg->reservations); | |
2731 | } | |
2732 | ||
8b74c03e | 2733 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 2734 | { |
11833d66 YZ |
2735 | struct btrfs_caching_control *next; |
2736 | struct btrfs_caching_control *caching_ctl; | |
32da5386 | 2737 | struct btrfs_block_group *cache; |
e8569813 | 2738 | |
9e351cc8 | 2739 | down_write(&fs_info->commit_root_sem); |
25179201 | 2740 | |
11833d66 YZ |
2741 | list_for_each_entry_safe(caching_ctl, next, |
2742 | &fs_info->caching_block_groups, list) { | |
2743 | cache = caching_ctl->block_group; | |
32da5386 | 2744 | if (btrfs_block_group_done(cache)) { |
11833d66 YZ |
2745 | cache->last_byte_to_unpin = (u64)-1; |
2746 | list_del_init(&caching_ctl->list); | |
e3cb339f | 2747 | btrfs_put_caching_control(caching_ctl); |
e8569813 | 2748 | } else { |
11833d66 | 2749 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 2750 | } |
e8569813 | 2751 | } |
11833d66 | 2752 | |
9e351cc8 | 2753 | up_write(&fs_info->commit_root_sem); |
8929ecfa | 2754 | |
67f9c220 | 2755 | btrfs_update_global_block_rsv(fs_info); |
e8569813 ZY |
2756 | } |
2757 | ||
c759c4e1 JB |
2758 | /* |
2759 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2760 | * what it should be based on the mount options. | |
2761 | */ | |
2762 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2763 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2764 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2765 | { |
2766 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2767 | |
2768 | *empty_cluster = 0; | |
2769 | if (btrfs_mixed_space_info(space_info)) | |
2770 | return ret; | |
2771 | ||
c759c4e1 | 2772 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2773 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2774 | if (btrfs_test_opt(fs_info, SSD)) |
2775 | *empty_cluster = SZ_2M; | |
2776 | else | |
ee22184b | 2777 | *empty_cluster = SZ_64K; |
583b7231 HK |
2778 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2779 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2780 | *empty_cluster = SZ_2M; | |
0b246afa | 2781 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2782 | } |
2783 | ||
2784 | return ret; | |
2785 | } | |
2786 | ||
2ff7e61e JM |
2787 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2788 | u64 start, u64 end, | |
678886bd | 2789 | const bool return_free_space) |
ccd467d6 | 2790 | { |
32da5386 | 2791 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2792 | struct btrfs_space_info *space_info; |
2793 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2794 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2795 | u64 len; |
c759c4e1 JB |
2796 | u64 total_unpinned = 0; |
2797 | u64 empty_cluster = 0; | |
7b398f8e | 2798 | bool readonly; |
ccd467d6 | 2799 | |
11833d66 | 2800 | while (start <= end) { |
7b398f8e | 2801 | readonly = false; |
11833d66 | 2802 | if (!cache || |
b3470b5d | 2803 | start >= cache->start + cache->length) { |
11833d66 YZ |
2804 | if (cache) |
2805 | btrfs_put_block_group(cache); | |
c759c4e1 | 2806 | total_unpinned = 0; |
11833d66 | 2807 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2808 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2809 | |
2ff7e61e | 2810 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2811 | cache->space_info, |
2812 | &empty_cluster); | |
2813 | empty_cluster <<= 1; | |
11833d66 YZ |
2814 | } |
2815 | ||
b3470b5d | 2816 | len = cache->start + cache->length - start; |
11833d66 YZ |
2817 | len = min(len, end + 1 - start); |
2818 | ||
2819 | if (start < cache->last_byte_to_unpin) { | |
2820 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
2821 | if (return_free_space) |
2822 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
2823 | } |
2824 | ||
f0486c68 | 2825 | start += len; |
c759c4e1 | 2826 | total_unpinned += len; |
7b398f8e | 2827 | space_info = cache->space_info; |
f0486c68 | 2828 | |
c759c4e1 JB |
2829 | /* |
2830 | * If this space cluster has been marked as fragmented and we've | |
2831 | * unpinned enough in this block group to potentially allow a | |
2832 | * cluster to be created inside of it go ahead and clear the | |
2833 | * fragmented check. | |
2834 | */ | |
2835 | if (cluster && cluster->fragmented && | |
2836 | total_unpinned > empty_cluster) { | |
2837 | spin_lock(&cluster->lock); | |
2838 | cluster->fragmented = 0; | |
2839 | spin_unlock(&cluster->lock); | |
2840 | } | |
2841 | ||
7b398f8e | 2842 | spin_lock(&space_info->lock); |
11833d66 YZ |
2843 | spin_lock(&cache->lock); |
2844 | cache->pinned -= len; | |
bb96c4e5 | 2845 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2846 | space_info->max_extent_size = 0; |
dec59fa3 EL |
2847 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2848 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
2849 | if (cache->ro) { |
2850 | space_info->bytes_readonly += len; | |
2851 | readonly = true; | |
2852 | } | |
11833d66 | 2853 | spin_unlock(&cache->lock); |
957780eb JB |
2854 | if (!readonly && return_free_space && |
2855 | global_rsv->space_info == space_info) { | |
2856 | u64 to_add = len; | |
92ac58ec | 2857 | |
7b398f8e JB |
2858 | spin_lock(&global_rsv->lock); |
2859 | if (!global_rsv->full) { | |
957780eb JB |
2860 | to_add = min(len, global_rsv->size - |
2861 | global_rsv->reserved); | |
2862 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2863 | btrfs_space_info_update_bytes_may_use(fs_info, |
2864 | space_info, to_add); | |
7b398f8e JB |
2865 | if (global_rsv->reserved >= global_rsv->size) |
2866 | global_rsv->full = 1; | |
957780eb | 2867 | len -= to_add; |
7b398f8e JB |
2868 | } |
2869 | spin_unlock(&global_rsv->lock); | |
2870 | } | |
2732798c JB |
2871 | /* Add to any tickets we may have */ |
2872 | if (!readonly && return_free_space && len) | |
2873 | btrfs_try_granting_tickets(fs_info, space_info); | |
7b398f8e | 2874 | spin_unlock(&space_info->lock); |
ccd467d6 | 2875 | } |
11833d66 YZ |
2876 | |
2877 | if (cache) | |
2878 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2879 | return 0; |
2880 | } | |
2881 | ||
5ead2dd0 | 2882 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2883 | { |
5ead2dd0 | 2884 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2885 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2886 | struct list_head *deleted_bgs; |
11833d66 | 2887 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2888 | u64 start; |
2889 | u64 end; | |
a28ec197 | 2890 | int ret; |
a28ec197 | 2891 | |
fe119a6e | 2892 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2893 | |
bf31f87f | 2894 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2895 | struct extent_state *cached_state = NULL; |
2896 | ||
d4b450cd | 2897 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2898 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2899 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2900 | if (ret) { |
2901 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2902 | break; |
d4b450cd | 2903 | } |
fe119a6e NB |
2904 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
2905 | clear_extent_bits(&fs_info->excluded_extents, start, | |
f2fb7298 | 2906 | end, EXTENT_UPTODATE); |
1f3c79a2 | 2907 | |
46b27f50 | 2908 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2909 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2910 | end + 1 - start, NULL); |
1f3c79a2 | 2911 | |
0e6ec385 | 2912 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2913 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2914 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2915 | free_extent_state(cached_state); |
b9473439 | 2916 | cond_resched(); |
a28ec197 | 2917 | } |
817d52f8 | 2918 | |
a2309300 DZ |
2919 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2920 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2921 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2922 | } |
b0643e59 | 2923 | |
e33e17ee JM |
2924 | /* |
2925 | * Transaction is finished. We don't need the lock anymore. We | |
2926 | * do need to clean up the block groups in case of a transaction | |
2927 | * abort. | |
2928 | */ | |
2929 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2930 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2931 | u64 trimmed = 0; | |
2932 | ||
2933 | ret = -EROFS; | |
bf31f87f | 2934 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2935 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2936 | block_group->start, |
2937 | block_group->length, | |
e33e17ee JM |
2938 | &trimmed); |
2939 | ||
2940 | list_del_init(&block_group->bg_list); | |
6b7304af | 2941 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2942 | btrfs_put_block_group(block_group); |
2943 | ||
2944 | if (ret) { | |
2945 | const char *errstr = btrfs_decode_error(ret); | |
2946 | btrfs_warn(fs_info, | |
913e1535 | 2947 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2948 | ret, errstr); |
2949 | } | |
2950 | } | |
2951 | ||
e20d96d6 CM |
2952 | return 0; |
2953 | } | |
2954 | ||
1c2a07f5 QW |
2955 | /* |
2956 | * Drop one or more refs of @node. | |
2957 | * | |
2958 | * 1. Locate the extent refs. | |
2959 | * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item. | |
2960 | * Locate it, then reduce the refs number or remove the ref line completely. | |
2961 | * | |
2962 | * 2. Update the refs count in EXTENT/METADATA_ITEM | |
2963 | * | |
2964 | * Inline backref case: | |
2965 | * | |
2966 | * in extent tree we have: | |
2967 | * | |
2968 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2969 | * refs 2 gen 6 flags DATA | |
2970 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2971 | * extent data backref root FS_TREE objectid 257 offset 0 count 1 | |
2972 | * | |
2973 | * This function gets called with: | |
2974 | * | |
2975 | * node->bytenr = 13631488 | |
2976 | * node->num_bytes = 1048576 | |
2977 | * root_objectid = FS_TREE | |
2978 | * owner_objectid = 257 | |
2979 | * owner_offset = 0 | |
2980 | * refs_to_drop = 1 | |
2981 | * | |
2982 | * Then we should get some like: | |
2983 | * | |
2984 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82 | |
2985 | * refs 1 gen 6 flags DATA | |
2986 | * extent data backref root FS_TREE objectid 258 offset 0 count 1 | |
2987 | * | |
2988 | * Keyed backref case: | |
2989 | * | |
2990 | * in extent tree we have: | |
2991 | * | |
2992 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
2993 | * refs 754 gen 6 flags DATA | |
2994 | * [...] | |
2995 | * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28 | |
2996 | * extent data backref root FS_TREE objectid 866 offset 0 count 1 | |
2997 | * | |
2998 | * This function get called with: | |
2999 | * | |
3000 | * node->bytenr = 13631488 | |
3001 | * node->num_bytes = 1048576 | |
3002 | * root_objectid = FS_TREE | |
3003 | * owner_objectid = 866 | |
3004 | * owner_offset = 0 | |
3005 | * refs_to_drop = 1 | |
3006 | * | |
3007 | * Then we should get some like: | |
3008 | * | |
3009 | * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24 | |
3010 | * refs 753 gen 6 flags DATA | |
3011 | * | |
3012 | * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed. | |
3013 | */ | |
5d4f98a2 | 3014 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
3015 | struct btrfs_delayed_ref_node *node, u64 parent, |
3016 | u64 root_objectid, u64 owner_objectid, | |
3017 | u64 owner_offset, int refs_to_drop, | |
3018 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 3019 | { |
e72cb923 | 3020 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 3021 | struct btrfs_key key; |
5d4f98a2 | 3022 | struct btrfs_path *path; |
1261ec42 | 3023 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 3024 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3025 | struct btrfs_extent_item *ei; |
3026 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 3027 | int ret; |
5d4f98a2 | 3028 | int is_data; |
952fccac CM |
3029 | int extent_slot = 0; |
3030 | int found_extent = 0; | |
3031 | int num_to_del = 1; | |
5d4f98a2 YZ |
3032 | u32 item_size; |
3033 | u64 refs; | |
c682f9b3 QW |
3034 | u64 bytenr = node->bytenr; |
3035 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 3036 | int last_ref = 0; |
0b246afa | 3037 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 3038 | |
5caf2a00 | 3039 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3040 | if (!path) |
3041 | return -ENOMEM; | |
5f26f772 | 3042 | |
b9473439 | 3043 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3044 | |
3045 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
1c2a07f5 QW |
3046 | |
3047 | if (!is_data && refs_to_drop != 1) { | |
3048 | btrfs_crit(info, | |
3049 | "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u", | |
3050 | node->bytenr, refs_to_drop); | |
3051 | ret = -EINVAL; | |
3052 | btrfs_abort_transaction(trans, ret); | |
3053 | goto out; | |
3054 | } | |
5d4f98a2 | 3055 | |
3173a18f | 3056 | if (is_data) |
897ca819 | 3057 | skinny_metadata = false; |
3173a18f | 3058 | |
fbe4801b NB |
3059 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
3060 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 3061 | owner_offset); |
7bb86316 | 3062 | if (ret == 0) { |
1c2a07f5 QW |
3063 | /* |
3064 | * Either the inline backref or the SHARED_DATA_REF/ | |
3065 | * SHARED_BLOCK_REF is found | |
3066 | * | |
3067 | * Here is a quick path to locate EXTENT/METADATA_ITEM. | |
3068 | * It's possible the EXTENT/METADATA_ITEM is near current slot. | |
3069 | */ | |
952fccac | 3070 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3071 | while (extent_slot >= 0) { |
3072 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3073 | extent_slot); |
5d4f98a2 | 3074 | if (key.objectid != bytenr) |
952fccac | 3075 | break; |
5d4f98a2 YZ |
3076 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3077 | key.offset == num_bytes) { | |
952fccac CM |
3078 | found_extent = 1; |
3079 | break; | |
3080 | } | |
3173a18f JB |
3081 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
3082 | key.offset == owner_objectid) { | |
3083 | found_extent = 1; | |
3084 | break; | |
3085 | } | |
1c2a07f5 QW |
3086 | |
3087 | /* Quick path didn't find the EXTEMT/METADATA_ITEM */ | |
952fccac CM |
3088 | if (path->slots[0] - extent_slot > 5) |
3089 | break; | |
5d4f98a2 | 3090 | extent_slot--; |
952fccac | 3091 | } |
a79865c6 | 3092 | |
31840ae1 | 3093 | if (!found_extent) { |
1c2a07f5 QW |
3094 | if (iref) { |
3095 | btrfs_crit(info, | |
3096 | "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref"); | |
3097 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3098 | goto err_dump; | |
3099 | } | |
3100 | /* Must be SHARED_* item, remove the backref first */ | |
87cc7a8a | 3101 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3102 | refs_to_drop, |
fcebe456 | 3103 | is_data, &last_ref); |
005d6427 | 3104 | if (ret) { |
66642832 | 3105 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3106 | goto out; |
3107 | } | |
b3b4aa74 | 3108 | btrfs_release_path(path); |
b9473439 | 3109 | path->leave_spinning = 1; |
5d4f98a2 | 3110 | |
1c2a07f5 | 3111 | /* Slow path to locate EXTENT/METADATA_ITEM */ |
5d4f98a2 YZ |
3112 | key.objectid = bytenr; |
3113 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3114 | key.offset = num_bytes; | |
3115 | ||
3173a18f JB |
3116 | if (!is_data && skinny_metadata) { |
3117 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3118 | key.offset = owner_objectid; | |
3119 | } | |
3120 | ||
31840ae1 ZY |
3121 | ret = btrfs_search_slot(trans, extent_root, |
3122 | &key, path, -1, 1); | |
3173a18f JB |
3123 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3124 | /* | |
3125 | * Couldn't find our skinny metadata item, | |
3126 | * see if we have ye olde extent item. | |
3127 | */ | |
3128 | path->slots[0]--; | |
3129 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3130 | path->slots[0]); | |
3131 | if (key.objectid == bytenr && | |
3132 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3133 | key.offset == num_bytes) | |
3134 | ret = 0; | |
3135 | } | |
3136 | ||
3137 | if (ret > 0 && skinny_metadata) { | |
3138 | skinny_metadata = false; | |
9ce49a0b | 3139 | key.objectid = bytenr; |
3173a18f JB |
3140 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3141 | key.offset = num_bytes; | |
3142 | btrfs_release_path(path); | |
3143 | ret = btrfs_search_slot(trans, extent_root, | |
3144 | &key, path, -1, 1); | |
3145 | } | |
3146 | ||
f3465ca4 | 3147 | if (ret) { |
5d163e0e JM |
3148 | btrfs_err(info, |
3149 | "umm, got %d back from search, was looking for %llu", | |
3150 | ret, bytenr); | |
b783e62d | 3151 | if (ret > 0) |
a4f78750 | 3152 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3153 | } |
005d6427 | 3154 | if (ret < 0) { |
66642832 | 3155 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3156 | goto out; |
3157 | } | |
31840ae1 ZY |
3158 | extent_slot = path->slots[0]; |
3159 | } | |
fae7f21c | 3160 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3161 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3162 | btrfs_err(info, |
3163 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3164 | bytenr, parent, root_objectid, owner_objectid, |
3165 | owner_offset); | |
66642832 | 3166 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3167 | goto out; |
79787eaa | 3168 | } else { |
66642832 | 3169 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3170 | goto out; |
7bb86316 | 3171 | } |
5f39d397 CM |
3172 | |
3173 | leaf = path->nodes[0]; | |
5d4f98a2 | 3174 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3175 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3176 | ret = -EINVAL; |
3177 | btrfs_print_v0_err(info); | |
3178 | btrfs_abort_transaction(trans, ret); | |
3179 | goto out; | |
3180 | } | |
952fccac | 3181 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3182 | struct btrfs_extent_item); |
3173a18f JB |
3183 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3184 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 | 3185 | struct btrfs_tree_block_info *bi; |
1c2a07f5 QW |
3186 | if (item_size < sizeof(*ei) + sizeof(*bi)) { |
3187 | btrfs_crit(info, | |
3188 | "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %lu", | |
3189 | key.objectid, key.type, key.offset, | |
3190 | owner_objectid, item_size, | |
3191 | sizeof(*ei) + sizeof(*bi)); | |
3192 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3193 | goto err_dump; | |
3194 | } | |
5d4f98a2 YZ |
3195 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
3196 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3197 | } | |
56bec294 | 3198 | |
5d4f98a2 | 3199 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3200 | if (refs < refs_to_drop) { |
1c2a07f5 QW |
3201 | btrfs_crit(info, |
3202 | "trying to drop %d refs but we only have %llu for bytenr %llu", | |
5d163e0e | 3203 | refs_to_drop, refs, bytenr); |
1c2a07f5 QW |
3204 | btrfs_abort_transaction(trans, -EUCLEAN); |
3205 | goto err_dump; | |
32b02538 | 3206 | } |
56bec294 | 3207 | refs -= refs_to_drop; |
5f39d397 | 3208 | |
5d4f98a2 YZ |
3209 | if (refs > 0) { |
3210 | if (extent_op) | |
3211 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3212 | /* | |
3213 | * In the case of inline back ref, reference count will | |
3214 | * be updated by remove_extent_backref | |
952fccac | 3215 | */ |
5d4f98a2 | 3216 | if (iref) { |
1c2a07f5 QW |
3217 | if (!found_extent) { |
3218 | btrfs_crit(info, | |
3219 | "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found"); | |
3220 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3221 | goto err_dump; | |
3222 | } | |
5d4f98a2 YZ |
3223 | } else { |
3224 | btrfs_set_extent_refs(leaf, ei, refs); | |
3225 | btrfs_mark_buffer_dirty(leaf); | |
3226 | } | |
3227 | if (found_extent) { | |
87cc7a8a NB |
3228 | ret = remove_extent_backref(trans, path, iref, |
3229 | refs_to_drop, is_data, | |
3230 | &last_ref); | |
005d6427 | 3231 | if (ret) { |
66642832 | 3232 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3233 | goto out; |
3234 | } | |
952fccac | 3235 | } |
5d4f98a2 | 3236 | } else { |
1c2a07f5 | 3237 | /* In this branch refs == 1 */ |
5d4f98a2 | 3238 | if (found_extent) { |
1c2a07f5 QW |
3239 | if (is_data && refs_to_drop != |
3240 | extent_data_ref_count(path, iref)) { | |
3241 | btrfs_crit(info, | |
3242 | "invalid refs_to_drop, current refs %u refs_to_drop %u", | |
3243 | extent_data_ref_count(path, iref), | |
3244 | refs_to_drop); | |
3245 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3246 | goto err_dump; | |
3247 | } | |
5d4f98a2 | 3248 | if (iref) { |
1c2a07f5 QW |
3249 | if (path->slots[0] != extent_slot) { |
3250 | btrfs_crit(info, | |
3251 | "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref", | |
3252 | key.objectid, key.type, | |
3253 | key.offset); | |
3254 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3255 | goto err_dump; | |
3256 | } | |
5d4f98a2 | 3257 | } else { |
1c2a07f5 QW |
3258 | /* |
3259 | * No inline ref, we must be at SHARED_* item, | |
3260 | * And it's single ref, it must be: | |
3261 | * | extent_slot ||extent_slot + 1| | |
3262 | * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ] | |
3263 | */ | |
3264 | if (path->slots[0] != extent_slot + 1) { | |
3265 | btrfs_crit(info, | |
3266 | "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM"); | |
3267 | btrfs_abort_transaction(trans, -EUCLEAN); | |
3268 | goto err_dump; | |
3269 | } | |
5d4f98a2 YZ |
3270 | path->slots[0] = extent_slot; |
3271 | num_to_del = 2; | |
3272 | } | |
78fae27e | 3273 | } |
b9473439 | 3274 | |
fcebe456 | 3275 | last_ref = 1; |
952fccac CM |
3276 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3277 | num_to_del); | |
005d6427 | 3278 | if (ret) { |
66642832 | 3279 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3280 | goto out; |
3281 | } | |
b3b4aa74 | 3282 | btrfs_release_path(path); |
21af804c | 3283 | |
5d4f98a2 | 3284 | if (is_data) { |
40e046ac FM |
3285 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3286 | num_bytes); | |
005d6427 | 3287 | if (ret) { |
66642832 | 3288 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3289 | goto out; |
3290 | } | |
459931ec CM |
3291 | } |
3292 | ||
e7355e50 | 3293 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3294 | if (ret) { |
66642832 | 3295 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3296 | goto out; |
3297 | } | |
3298 | ||
ade4b516 | 3299 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3300 | if (ret) { |
66642832 | 3301 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3302 | goto out; |
3303 | } | |
a28ec197 | 3304 | } |
fcebe456 JB |
3305 | btrfs_release_path(path); |
3306 | ||
79787eaa | 3307 | out: |
5caf2a00 | 3308 | btrfs_free_path(path); |
a28ec197 | 3309 | return ret; |
1c2a07f5 QW |
3310 | err_dump: |
3311 | /* | |
3312 | * Leaf dump can take up a lot of log buffer, so we only do full leaf | |
3313 | * dump for debug build. | |
3314 | */ | |
3315 | if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { | |
3316 | btrfs_crit(info, "path->slots[0]=%d extent_slot=%d", | |
3317 | path->slots[0], extent_slot); | |
3318 | btrfs_print_leaf(path->nodes[0]); | |
3319 | } | |
3320 | ||
3321 | btrfs_free_path(path); | |
3322 | return -EUCLEAN; | |
a28ec197 CM |
3323 | } |
3324 | ||
1887be66 | 3325 | /* |
f0486c68 | 3326 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3327 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3328 | * a given extent, and if there are no other delayed refs to be processed, it | |
3329 | * removes it from the tree. | |
3330 | */ | |
3331 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3332 | u64 bytenr) |
1887be66 CM |
3333 | { |
3334 | struct btrfs_delayed_ref_head *head; | |
3335 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3336 | int ret = 0; |
1887be66 CM |
3337 | |
3338 | delayed_refs = &trans->transaction->delayed_refs; | |
3339 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3340 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3341 | if (!head) |
cf93da7b | 3342 | goto out_delayed_unlock; |
1887be66 | 3343 | |
d7df2c79 | 3344 | spin_lock(&head->lock); |
e3d03965 | 3345 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3346 | goto out; |
3347 | ||
bedc6617 JB |
3348 | if (cleanup_extent_op(head) != NULL) |
3349 | goto out; | |
5d4f98a2 | 3350 | |
1887be66 CM |
3351 | /* |
3352 | * waiting for the lock here would deadlock. If someone else has it | |
3353 | * locked they are already in the process of dropping it anyway | |
3354 | */ | |
3355 | if (!mutex_trylock(&head->mutex)) | |
3356 | goto out; | |
3357 | ||
d7baffda | 3358 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3359 | head->processing = 0; |
d7baffda | 3360 | |
d7df2c79 | 3361 | spin_unlock(&head->lock); |
1887be66 CM |
3362 | spin_unlock(&delayed_refs->lock); |
3363 | ||
f0486c68 YZ |
3364 | BUG_ON(head->extent_op); |
3365 | if (head->must_insert_reserved) | |
3366 | ret = 1; | |
3367 | ||
31890da0 | 3368 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3369 | mutex_unlock(&head->mutex); |
d278850e | 3370 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3371 | return ret; |
1887be66 | 3372 | out: |
d7df2c79 | 3373 | spin_unlock(&head->lock); |
cf93da7b CM |
3374 | |
3375 | out_delayed_unlock: | |
1887be66 CM |
3376 | spin_unlock(&delayed_refs->lock); |
3377 | return 0; | |
3378 | } | |
3379 | ||
f0486c68 YZ |
3380 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3381 | struct btrfs_root *root, | |
3382 | struct extent_buffer *buf, | |
5581a51a | 3383 | u64 parent, int last_ref) |
f0486c68 | 3384 | { |
0b246afa | 3385 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3386 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 3387 | int pin = 1; |
f0486c68 YZ |
3388 | int ret; |
3389 | ||
ed4f255b QW |
3390 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3391 | buf->start, buf->len, parent); | |
3392 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3393 | root->root_key.objectid); | |
3394 | ||
f0486c68 | 3395 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
3396 | int old_ref_mod, new_ref_mod; |
3397 | ||
8a5040f7 | 3398 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 3399 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 3400 | &old_ref_mod, &new_ref_mod); |
79787eaa | 3401 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 3402 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
3403 | } |
3404 | ||
0a16c7d7 | 3405 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3406 | struct btrfs_block_group *cache; |
6219872d | 3407 | |
f0486c68 | 3408 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3409 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 3410 | if (!ret) |
37be25bc | 3411 | goto out; |
f0486c68 YZ |
3412 | } |
3413 | ||
4da8b76d | 3414 | pin = 0; |
0b246afa | 3415 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3416 | |
f0486c68 | 3417 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3418 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3419 | btrfs_put_block_group(cache); |
37be25bc | 3420 | goto out; |
f0486c68 YZ |
3421 | } |
3422 | ||
3423 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3424 | ||
3425 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3426 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3427 | btrfs_put_block_group(cache); |
71ff6437 | 3428 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3429 | } |
3430 | out: | |
b150a4f1 | 3431 | if (pin) |
78192442 | 3432 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 3433 | |
0a16c7d7 OS |
3434 | if (last_ref) { |
3435 | /* | |
3436 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3437 | * matter anymore. | |
3438 | */ | |
3439 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3440 | } | |
f0486c68 YZ |
3441 | } |
3442 | ||
79787eaa | 3443 | /* Can return -ENOMEM */ |
ffd4bb2a | 3444 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3445 | { |
ffd4bb2a | 3446 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 3447 | int old_ref_mod, new_ref_mod; |
925baedd CM |
3448 | int ret; |
3449 | ||
f5ee5c9a | 3450 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3451 | return 0; |
fccb84c9 | 3452 | |
56bec294 CM |
3453 | /* |
3454 | * tree log blocks never actually go into the extent allocation | |
3455 | * tree, just update pinning info and exit early. | |
56bec294 | 3456 | */ |
ffd4bb2a QW |
3457 | if ((ref->type == BTRFS_REF_METADATA && |
3458 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3459 | (ref->type == BTRFS_REF_DATA && | |
3460 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3461 | /* unlocks the pinned mutex */ |
b25c36f8 | 3462 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
d7eae340 | 3463 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 3464 | ret = 0; |
ffd4bb2a QW |
3465 | } else if (ref->type == BTRFS_REF_METADATA) { |
3466 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 3467 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 3468 | } else { |
ffd4bb2a | 3469 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 3470 | &old_ref_mod, &new_ref_mod); |
56bec294 | 3471 | } |
d7eae340 | 3472 | |
ffd4bb2a QW |
3473 | if (!((ref->type == BTRFS_REF_METADATA && |
3474 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3475 | (ref->type == BTRFS_REF_DATA && | |
3476 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3477 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3478 | |
ddf30cf0 | 3479 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 3480 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 3481 | |
925baedd CM |
3482 | return ret; |
3483 | } | |
3484 | ||
817d52f8 | 3485 | enum btrfs_loop_type { |
f262fa8d DS |
3486 | LOOP_CACHING_NOWAIT, |
3487 | LOOP_CACHING_WAIT, | |
3488 | LOOP_ALLOC_CHUNK, | |
3489 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3490 | }; |
3491 | ||
e570fd27 | 3492 | static inline void |
32da5386 | 3493 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3494 | int delalloc) |
3495 | { | |
3496 | if (delalloc) | |
3497 | down_read(&cache->data_rwsem); | |
3498 | } | |
3499 | ||
32da5386 | 3500 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3501 | int delalloc) |
3502 | { | |
3503 | btrfs_get_block_group(cache); | |
3504 | if (delalloc) | |
3505 | down_read(&cache->data_rwsem); | |
3506 | } | |
3507 | ||
32da5386 DS |
3508 | static struct btrfs_block_group *btrfs_lock_cluster( |
3509 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3510 | struct btrfs_free_cluster *cluster, |
3511 | int delalloc) | |
c142c6a4 | 3512 | __acquires(&cluster->refill_lock) |
e570fd27 | 3513 | { |
32da5386 | 3514 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3515 | |
e570fd27 | 3516 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3517 | while (1) { |
3518 | used_bg = cluster->block_group; | |
3519 | if (!used_bg) | |
3520 | return NULL; | |
3521 | ||
3522 | if (used_bg == block_group) | |
e570fd27 MX |
3523 | return used_bg; |
3524 | ||
6719afdc | 3525 | btrfs_get_block_group(used_bg); |
e570fd27 | 3526 | |
6719afdc GU |
3527 | if (!delalloc) |
3528 | return used_bg; | |
e570fd27 | 3529 | |
6719afdc GU |
3530 | if (down_read_trylock(&used_bg->data_rwsem)) |
3531 | return used_bg; | |
e570fd27 | 3532 | |
6719afdc | 3533 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3534 | |
e321f8a8 LB |
3535 | /* We should only have one-level nested. */ |
3536 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3537 | |
6719afdc GU |
3538 | spin_lock(&cluster->refill_lock); |
3539 | if (used_bg == cluster->block_group) | |
3540 | return used_bg; | |
e570fd27 | 3541 | |
6719afdc GU |
3542 | up_read(&used_bg->data_rwsem); |
3543 | btrfs_put_block_group(used_bg); | |
3544 | } | |
e570fd27 MX |
3545 | } |
3546 | ||
3547 | static inline void | |
32da5386 | 3548 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3549 | int delalloc) |
3550 | { | |
3551 | if (delalloc) | |
3552 | up_read(&cache->data_rwsem); | |
3553 | btrfs_put_block_group(cache); | |
3554 | } | |
3555 | ||
cb2f96f8 NA |
3556 | enum btrfs_extent_allocation_policy { |
3557 | BTRFS_EXTENT_ALLOC_CLUSTERED, | |
3558 | }; | |
3559 | ||
b4bd745d QW |
3560 | /* |
3561 | * Structure used internally for find_free_extent() function. Wraps needed | |
3562 | * parameters. | |
3563 | */ | |
3564 | struct find_free_extent_ctl { | |
3565 | /* Basic allocation info */ | |
b4bd745d QW |
3566 | u64 num_bytes; |
3567 | u64 empty_size; | |
3568 | u64 flags; | |
3569 | int delalloc; | |
3570 | ||
3571 | /* Where to start the search inside the bg */ | |
3572 | u64 search_start; | |
3573 | ||
3574 | /* For clustered allocation */ | |
3575 | u64 empty_cluster; | |
c10859be NA |
3576 | struct btrfs_free_cluster *last_ptr; |
3577 | bool use_cluster; | |
b4bd745d QW |
3578 | |
3579 | bool have_caching_bg; | |
3580 | bool orig_have_caching_bg; | |
3581 | ||
3582 | /* RAID index, converted from flags */ | |
3583 | int index; | |
3584 | ||
e72d79d6 QW |
3585 | /* |
3586 | * Current loop number, check find_free_extent_update_loop() for details | |
3587 | */ | |
b4bd745d QW |
3588 | int loop; |
3589 | ||
3590 | /* | |
3591 | * Whether we're refilling a cluster, if true we need to re-search | |
3592 | * current block group but don't try to refill the cluster again. | |
3593 | */ | |
3594 | bool retry_clustered; | |
3595 | ||
3596 | /* | |
3597 | * Whether we're updating free space cache, if true we need to re-search | |
3598 | * current block group but don't try updating free space cache again. | |
3599 | */ | |
3600 | bool retry_unclustered; | |
3601 | ||
3602 | /* If current block group is cached */ | |
3603 | int cached; | |
3604 | ||
3605 | /* Max contiguous hole found */ | |
3606 | u64 max_extent_size; | |
3607 | ||
3608 | /* Total free space from free space cache, not always contiguous */ | |
3609 | u64 total_free_space; | |
3610 | ||
3611 | /* Found result */ | |
3612 | u64 found_offset; | |
cb2f96f8 | 3613 | |
ea544149 NA |
3614 | /* Hint where to start looking for an empty space */ |
3615 | u64 hint_byte; | |
3616 | ||
cb2f96f8 NA |
3617 | /* Allocation policy */ |
3618 | enum btrfs_extent_allocation_policy policy; | |
b4bd745d QW |
3619 | }; |
3620 | ||
d06e3bb6 QW |
3621 | |
3622 | /* | |
3623 | * Helper function for find_free_extent(). | |
3624 | * | |
3625 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3626 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3627 | * Return >0 to inform caller that we find nothing | |
3628 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3629 | */ | |
32da5386 | 3630 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3631 | struct find_free_extent_ctl *ffe_ctl, |
3632 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3633 | { |
32da5386 | 3634 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3635 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3636 | u64 aligned_cluster; |
3637 | u64 offset; | |
3638 | int ret; | |
3639 | ||
3640 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3641 | if (!cluster_bg) | |
3642 | goto refill_cluster; | |
3643 | if (cluster_bg != bg && (cluster_bg->ro || | |
3644 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3645 | goto release_cluster; | |
3646 | ||
3647 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3648 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3649 | &ffe_ctl->max_extent_size); |
3650 | if (offset) { | |
3651 | /* We have a block, we're done */ | |
3652 | spin_unlock(&last_ptr->refill_lock); | |
3653 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3654 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3655 | *cluster_bg_ret = cluster_bg; | |
3656 | ffe_ctl->found_offset = offset; | |
3657 | return 0; | |
3658 | } | |
3659 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3660 | ||
3661 | release_cluster: | |
3662 | /* | |
3663 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3664 | * lets just skip it and let the allocator find whatever block it can | |
3665 | * find. If we reach this point, we will have tried the cluster | |
3666 | * allocator plenty of times and not have found anything, so we are | |
3667 | * likely way too fragmented for the clustering stuff to find anything. | |
3668 | * | |
3669 | * However, if the cluster is taken from the current block group, | |
3670 | * release the cluster first, so that we stand a better chance of | |
3671 | * succeeding in the unclustered allocation. | |
3672 | */ | |
3673 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3674 | spin_unlock(&last_ptr->refill_lock); | |
3675 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3676 | return -ENOENT; | |
3677 | } | |
3678 | ||
3679 | /* This cluster didn't work out, free it and start over */ | |
3680 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3681 | ||
3682 | if (cluster_bg != bg) | |
3683 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3684 | ||
3685 | refill_cluster: | |
3686 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3687 | spin_unlock(&last_ptr->refill_lock); | |
3688 | return -ENOENT; | |
3689 | } | |
3690 | ||
3691 | aligned_cluster = max_t(u64, | |
3692 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3693 | bg->full_stripe_len); | |
2ceeae2e DS |
3694 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3695 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3696 | if (ret == 0) { |
3697 | /* Now pull our allocation out of this cluster */ | |
3698 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3699 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3700 | &ffe_ctl->max_extent_size); | |
3701 | if (offset) { | |
3702 | /* We found one, proceed */ | |
3703 | spin_unlock(&last_ptr->refill_lock); | |
3704 | trace_btrfs_reserve_extent_cluster(bg, | |
3705 | ffe_ctl->search_start, | |
3706 | ffe_ctl->num_bytes); | |
3707 | ffe_ctl->found_offset = offset; | |
3708 | return 0; | |
3709 | } | |
3710 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3711 | !ffe_ctl->retry_clustered) { | |
3712 | spin_unlock(&last_ptr->refill_lock); | |
3713 | ||
3714 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3715 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3716 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3717 | return -EAGAIN; | |
3718 | } | |
3719 | /* | |
3720 | * At this point we either didn't find a cluster or we weren't able to | |
3721 | * allocate a block from our cluster. Free the cluster we've been | |
3722 | * trying to use, and go to the next block group. | |
3723 | */ | |
3724 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3725 | spin_unlock(&last_ptr->refill_lock); | |
3726 | return 1; | |
3727 | } | |
3728 | ||
e1a41848 QW |
3729 | /* |
3730 | * Return >0 to inform caller that we find nothing | |
3731 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3732 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3733 | */ | |
32da5386 | 3734 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3735 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3736 | { |
897cae79 | 3737 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3738 | u64 offset; |
3739 | ||
3740 | /* | |
3741 | * We are doing an unclustered allocation, set the fragmented flag so | |
3742 | * we don't bother trying to setup a cluster again until we get more | |
3743 | * space. | |
3744 | */ | |
3745 | if (unlikely(last_ptr)) { | |
3746 | spin_lock(&last_ptr->lock); | |
3747 | last_ptr->fragmented = 1; | |
3748 | spin_unlock(&last_ptr->lock); | |
3749 | } | |
3750 | if (ffe_ctl->cached) { | |
3751 | struct btrfs_free_space_ctl *free_space_ctl; | |
3752 | ||
3753 | free_space_ctl = bg->free_space_ctl; | |
3754 | spin_lock(&free_space_ctl->tree_lock); | |
3755 | if (free_space_ctl->free_space < | |
3756 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3757 | ffe_ctl->empty_size) { | |
3758 | ffe_ctl->total_free_space = max_t(u64, | |
3759 | ffe_ctl->total_free_space, | |
3760 | free_space_ctl->free_space); | |
3761 | spin_unlock(&free_space_ctl->tree_lock); | |
3762 | return 1; | |
3763 | } | |
3764 | spin_unlock(&free_space_ctl->tree_lock); | |
3765 | } | |
3766 | ||
3767 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3768 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3769 | &ffe_ctl->max_extent_size); | |
3770 | ||
3771 | /* | |
3772 | * If we didn't find a chunk, and we haven't failed on this block group | |
3773 | * before, and this block group is in the middle of caching and we are | |
3774 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3775 | * set @retry_unclustered to true. | |
3776 | * | |
3777 | * If @retry_unclustered is true then we've already waited on this | |
3778 | * block group once and should move on to the next block group. | |
3779 | */ | |
3780 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3781 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3782 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3783 | ffe_ctl->empty_size); | |
e1a41848 QW |
3784 | ffe_ctl->retry_unclustered = true; |
3785 | return -EAGAIN; | |
3786 | } else if (!offset) { | |
3787 | return 1; | |
3788 | } | |
3789 | ffe_ctl->found_offset = offset; | |
3790 | return 0; | |
3791 | } | |
3792 | ||
c668690d NA |
3793 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3794 | struct find_free_extent_ctl *ffe_ctl, | |
3795 | struct btrfs_block_group **bg_ret) | |
3796 | { | |
3797 | int ret; | |
3798 | ||
3799 | /* We want to try and use the cluster allocator, so lets look there */ | |
3800 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3801 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
c668690d NA |
3802 | if (ret >= 0 || ret == -EAGAIN) |
3803 | return ret; | |
3804 | /* ret == -ENOENT case falls through */ | |
3805 | } | |
3806 | ||
897cae79 | 3807 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3808 | } |
3809 | ||
3810 | static int do_allocation(struct btrfs_block_group *block_group, | |
3811 | struct find_free_extent_ctl *ffe_ctl, | |
3812 | struct btrfs_block_group **bg_ret) | |
3813 | { | |
3814 | switch (ffe_ctl->policy) { | |
3815 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3816 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
3817 | default: | |
3818 | BUG(); | |
3819 | } | |
3820 | } | |
3821 | ||
baba5062 NA |
3822 | static void release_block_group(struct btrfs_block_group *block_group, |
3823 | struct find_free_extent_ctl *ffe_ctl, | |
3824 | int delalloc) | |
3825 | { | |
3826 | switch (ffe_ctl->policy) { | |
3827 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3828 | ffe_ctl->retry_clustered = false; | |
3829 | ffe_ctl->retry_unclustered = false; | |
3830 | break; | |
3831 | default: | |
3832 | BUG(); | |
3833 | } | |
3834 | ||
3835 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
3836 | ffe_ctl->index); | |
3837 | btrfs_release_block_group(block_group, delalloc); | |
3838 | } | |
3839 | ||
0ab9724b NA |
3840 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
3841 | struct btrfs_key *ins) | |
3842 | { | |
3843 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3844 | ||
3845 | if (!ffe_ctl->use_cluster && last_ptr) { | |
3846 | spin_lock(&last_ptr->lock); | |
3847 | last_ptr->window_start = ins->objectid; | |
3848 | spin_unlock(&last_ptr->lock); | |
3849 | } | |
3850 | } | |
3851 | ||
3852 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
3853 | struct btrfs_key *ins) | |
3854 | { | |
3855 | switch (ffe_ctl->policy) { | |
3856 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3857 | found_extent_clustered(ffe_ctl, ins); | |
3858 | break; | |
3859 | default: | |
3860 | BUG(); | |
3861 | } | |
3862 | } | |
3863 | ||
c70e2139 NA |
3864 | static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl) |
3865 | { | |
3866 | switch (ffe_ctl->policy) { | |
3867 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3868 | /* | |
3869 | * If we can't allocate a new chunk we've already looped through | |
3870 | * at least once, move on to the NO_EMPTY_SIZE case. | |
3871 | */ | |
3872 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3873 | return 0; | |
3874 | default: | |
3875 | BUG(); | |
3876 | } | |
3877 | } | |
3878 | ||
e72d79d6 QW |
3879 | /* |
3880 | * Return >0 means caller needs to re-search for free extent | |
3881 | * Return 0 means we have the needed free extent. | |
3882 | * Return <0 means we failed to locate any free extent. | |
3883 | */ | |
3884 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
3885 | struct btrfs_key *ins, |
3886 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 3887 | bool full_search) |
e72d79d6 QW |
3888 | { |
3889 | struct btrfs_root *root = fs_info->extent_root; | |
3890 | int ret; | |
3891 | ||
3892 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3893 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3894 | ffe_ctl->orig_have_caching_bg = true; | |
3895 | ||
3896 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3897 | ffe_ctl->have_caching_bg) | |
3898 | return 1; | |
3899 | ||
3900 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3901 | return 1; | |
3902 | ||
3903 | if (ins->objectid) { | |
0ab9724b | 3904 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
3905 | return 0; |
3906 | } | |
3907 | ||
3908 | /* | |
3909 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3910 | * caching kthreads as we move along | |
3911 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3912 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3913 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3914 | * again | |
3915 | */ | |
3916 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3917 | ffe_ctl->index = 0; | |
3918 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3919 | /* | |
3920 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3921 | * don't have any uncached bgs and we've already done a | |
3922 | * full search through. | |
3923 | */ | |
3924 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3925 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3926 | else | |
3927 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3928 | } else { | |
3929 | ffe_ctl->loop++; | |
3930 | } | |
3931 | ||
3932 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3933 | struct btrfs_trans_handle *trans; | |
3934 | int exist = 0; | |
3935 | ||
3936 | trans = current->journal_info; | |
3937 | if (trans) | |
3938 | exist = 1; | |
3939 | else | |
3940 | trans = btrfs_join_transaction(root); | |
3941 | ||
3942 | if (IS_ERR(trans)) { | |
3943 | ret = PTR_ERR(trans); | |
3944 | return ret; | |
3945 | } | |
3946 | ||
fc471cb0 JB |
3947 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3948 | CHUNK_ALLOC_FORCE); | |
e72d79d6 | 3949 | |
e72d79d6 | 3950 | /* Do not bail out on ENOSPC since we can do more. */ |
c70e2139 NA |
3951 | if (ret == -ENOSPC) |
3952 | ret = chunk_allocation_failed(ffe_ctl); | |
3953 | else if (ret < 0) | |
e72d79d6 QW |
3954 | btrfs_abort_transaction(trans, ret); |
3955 | else | |
3956 | ret = 0; | |
3957 | if (!exist) | |
3958 | btrfs_end_transaction(trans); | |
3959 | if (ret) | |
3960 | return ret; | |
3961 | } | |
3962 | ||
3963 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
3964 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
3965 | return -ENOSPC; | |
3966 | ||
e72d79d6 QW |
3967 | /* |
3968 | * Don't loop again if we already have no empty_size and | |
3969 | * no empty_cluster. | |
3970 | */ | |
3971 | if (ffe_ctl->empty_size == 0 && | |
3972 | ffe_ctl->empty_cluster == 0) | |
3973 | return -ENOSPC; | |
3974 | ffe_ctl->empty_size = 0; | |
3975 | ffe_ctl->empty_cluster = 0; | |
3976 | } | |
3977 | return 1; | |
3978 | } | |
3979 | return -ENOSPC; | |
3980 | } | |
3981 | ||
7e895409 NA |
3982 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
3983 | struct find_free_extent_ctl *ffe_ctl, | |
3984 | struct btrfs_space_info *space_info, | |
3985 | struct btrfs_key *ins) | |
3986 | { | |
3987 | /* | |
3988 | * If our free space is heavily fragmented we may not be able to make | |
3989 | * big contiguous allocations, so instead of doing the expensive search | |
3990 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3991 | * can go ahead and search for a more manageable chunk. | |
3992 | * | |
3993 | * If our max_extent_size is large enough for our allocation simply | |
3994 | * disable clustering since we will likely not be able to find enough | |
3995 | * space to create a cluster and induce latency trying. | |
3996 | */ | |
3997 | if (space_info->max_extent_size) { | |
3998 | spin_lock(&space_info->lock); | |
3999 | if (space_info->max_extent_size && | |
4000 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
4001 | ins->offset = space_info->max_extent_size; | |
4002 | spin_unlock(&space_info->lock); | |
4003 | return -ENOSPC; | |
4004 | } else if (space_info->max_extent_size) { | |
4005 | ffe_ctl->use_cluster = false; | |
4006 | } | |
4007 | spin_unlock(&space_info->lock); | |
4008 | } | |
4009 | ||
4010 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
4011 | &ffe_ctl->empty_cluster); | |
4012 | if (ffe_ctl->last_ptr) { | |
4013 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
4014 | ||
4015 | spin_lock(&last_ptr->lock); | |
4016 | if (last_ptr->block_group) | |
4017 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4018 | if (last_ptr->fragmented) { | |
4019 | /* | |
4020 | * We still set window_start so we can keep track of the | |
4021 | * last place we found an allocation to try and save | |
4022 | * some time. | |
4023 | */ | |
4024 | ffe_ctl->hint_byte = last_ptr->window_start; | |
4025 | ffe_ctl->use_cluster = false; | |
4026 | } | |
4027 | spin_unlock(&last_ptr->lock); | |
4028 | } | |
4029 | ||
4030 | return 0; | |
4031 | } | |
4032 | ||
4033 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
4034 | struct find_free_extent_ctl *ffe_ctl, | |
4035 | struct btrfs_space_info *space_info, | |
4036 | struct btrfs_key *ins) | |
4037 | { | |
4038 | switch (ffe_ctl->policy) { | |
4039 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
4040 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
4041 | space_info, ins); | |
4042 | default: | |
4043 | BUG(); | |
4044 | } | |
4045 | } | |
4046 | ||
fec577fb CM |
4047 | /* |
4048 | * walks the btree of allocated extents and find a hole of a given size. | |
4049 | * The key ins is changed to record the hole: | |
a4820398 | 4050 | * ins->objectid == start position |
62e2749e | 4051 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 4052 | * ins->offset == the size of the hole. |
fec577fb | 4053 | * Any available blocks before search_start are skipped. |
a4820398 MX |
4054 | * |
4055 | * If there is no suitable free space, we will record the max size of | |
4056 | * the free space extent currently. | |
e72d79d6 QW |
4057 | * |
4058 | * The overall logic and call chain: | |
4059 | * | |
4060 | * find_free_extent() | |
4061 | * |- Iterate through all block groups | |
4062 | * | |- Get a valid block group | |
4063 | * | |- Try to do clustered allocation in that block group | |
4064 | * | |- Try to do unclustered allocation in that block group | |
4065 | * | |- Check if the result is valid | |
4066 | * | | |- If valid, then exit | |
4067 | * | |- Jump to next block group | |
4068 | * | | |
4069 | * |- Push harder to find free extents | |
4070 | * |- If not found, re-iterate all block groups | |
fec577fb | 4071 | */ |
437490fe | 4072 | static noinline int find_free_extent(struct btrfs_root *root, |
18513091 | 4073 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
ea544149 | 4074 | u64 hint_byte_orig, struct btrfs_key *ins, |
18513091 | 4075 | u64 flags, int delalloc) |
fec577fb | 4076 | { |
437490fe | 4077 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 4078 | int ret = 0; |
db8fe64f | 4079 | int cache_block_group_error = 0; |
32da5386 | 4080 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 4081 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 4082 | struct btrfs_space_info *space_info; |
a5e681d9 | 4083 | bool full_search = false; |
fec577fb | 4084 | |
0b246afa | 4085 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 4086 | |
b4bd745d QW |
4087 | ffe_ctl.num_bytes = num_bytes; |
4088 | ffe_ctl.empty_size = empty_size; | |
4089 | ffe_ctl.flags = flags; | |
4090 | ffe_ctl.search_start = 0; | |
b4bd745d QW |
4091 | ffe_ctl.delalloc = delalloc; |
4092 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
4093 | ffe_ctl.have_caching_bg = false; | |
4094 | ffe_ctl.orig_have_caching_bg = false; | |
4095 | ffe_ctl.found_offset = 0; | |
ea544149 | 4096 | ffe_ctl.hint_byte = hint_byte_orig; |
cb2f96f8 | 4097 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED; |
b4bd745d | 4098 | |
c10859be NA |
4099 | /* For clustered allocation */ |
4100 | ffe_ctl.retry_clustered = false; | |
4101 | ffe_ctl.retry_unclustered = false; | |
4102 | ffe_ctl.last_ptr = NULL; | |
4103 | ffe_ctl.use_cluster = true; | |
4104 | ||
962a298f | 4105 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
4106 | ins->objectid = 0; |
4107 | ins->offset = 0; | |
b1a4d965 | 4108 | |
437490fe | 4109 | trace_find_free_extent(root, num_bytes, empty_size, flags); |
3f7de037 | 4110 | |
280c2908 | 4111 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 4112 | if (!space_info) { |
0b246afa | 4113 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
4114 | return -ENOSPC; |
4115 | } | |
2552d17e | 4116 | |
7e895409 NA |
4117 | ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins); |
4118 | if (ret < 0) | |
4119 | return ret; | |
fa9c0d79 | 4120 | |
b4bd745d QW |
4121 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
4122 | first_logical_byte(fs_info, 0)); | |
ea544149 NA |
4123 | ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte); |
4124 | if (ffe_ctl.search_start == ffe_ctl.hint_byte) { | |
b4bd745d QW |
4125 | block_group = btrfs_lookup_block_group(fs_info, |
4126 | ffe_ctl.search_start); | |
817d52f8 JB |
4127 | /* |
4128 | * we don't want to use the block group if it doesn't match our | |
4129 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4130 | * |
4131 | * However if we are re-searching with an ideal block group | |
4132 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 4133 | */ |
b6919a58 | 4134 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 4135 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 4136 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4137 | if (list_empty(&block_group->list) || |
4138 | block_group->ro) { | |
4139 | /* | |
4140 | * someone is removing this block group, | |
4141 | * we can't jump into the have_block_group | |
4142 | * target because our list pointers are not | |
4143 | * valid | |
4144 | */ | |
4145 | btrfs_put_block_group(block_group); | |
4146 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4147 | } else { |
b4bd745d | 4148 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 4149 | block_group->flags); |
e570fd27 | 4150 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 4151 | goto have_block_group; |
ccf0e725 | 4152 | } |
2552d17e | 4153 | } else if (block_group) { |
fa9c0d79 | 4154 | btrfs_put_block_group(block_group); |
2552d17e | 4155 | } |
42e70e7a | 4156 | } |
2552d17e | 4157 | search: |
b4bd745d QW |
4158 | ffe_ctl.have_caching_bg = false; |
4159 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
4160 | ffe_ctl.index == 0) | |
a5e681d9 | 4161 | full_search = true; |
80eb234a | 4162 | down_read(&space_info->groups_sem); |
b4bd745d QW |
4163 | list_for_each_entry(block_group, |
4164 | &space_info->block_groups[ffe_ctl.index], list) { | |
c668690d NA |
4165 | struct btrfs_block_group *bg_ret; |
4166 | ||
14443937 JM |
4167 | /* If the block group is read-only, we can skip it entirely. */ |
4168 | if (unlikely(block_group->ro)) | |
4169 | continue; | |
4170 | ||
e570fd27 | 4171 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 4172 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 4173 | |
83a50de9 CM |
4174 | /* |
4175 | * this can happen if we end up cycling through all the | |
4176 | * raid types, but we want to make sure we only allocate | |
4177 | * for the proper type. | |
4178 | */ | |
b6919a58 | 4179 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 4180 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4181 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4182 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4183 | BTRFS_BLOCK_GROUP_RAID10; |
4184 | ||
4185 | /* | |
4186 | * if they asked for extra copies and this block group | |
4187 | * doesn't provide them, bail. This does allow us to | |
4188 | * fill raid0 from raid1. | |
4189 | */ | |
b6919a58 | 4190 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4191 | goto loop; |
2a28468e QW |
4192 | |
4193 | /* | |
4194 | * This block group has different flags than we want. | |
4195 | * It's possible that we have MIXED_GROUP flag but no | |
4196 | * block group is mixed. Just skip such block group. | |
4197 | */ | |
4198 | btrfs_release_block_group(block_group, delalloc); | |
4199 | continue; | |
83a50de9 CM |
4200 | } |
4201 | ||
2552d17e | 4202 | have_block_group: |
32da5386 | 4203 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
4204 | if (unlikely(!ffe_ctl.cached)) { |
4205 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 4206 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
4207 | |
4208 | /* | |
4209 | * If we get ENOMEM here or something else we want to | |
4210 | * try other block groups, because it may not be fatal. | |
4211 | * However if we can't find anything else we need to | |
4212 | * save our return here so that we return the actual | |
4213 | * error that caused problems, not ENOSPC. | |
4214 | */ | |
4215 | if (ret < 0) { | |
4216 | if (!cache_block_group_error) | |
4217 | cache_block_group_error = ret; | |
4218 | ret = 0; | |
4219 | goto loop; | |
4220 | } | |
1d4284bd | 4221 | ret = 0; |
817d52f8 JB |
4222 | } |
4223 | ||
36cce922 JB |
4224 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
4225 | goto loop; | |
0f9dd46c | 4226 | |
c668690d NA |
4227 | bg_ret = NULL; |
4228 | ret = do_allocation(block_group, &ffe_ctl, &bg_ret); | |
4229 | if (ret == 0) { | |
4230 | if (bg_ret && bg_ret != block_group) { | |
4231 | btrfs_release_block_group(block_group, delalloc); | |
4232 | block_group = bg_ret; | |
fa9c0d79 | 4233 | } |
c668690d | 4234 | } else if (ret == -EAGAIN) { |
817d52f8 | 4235 | goto have_block_group; |
c668690d | 4236 | } else if (ret > 0) { |
1cdda9b8 | 4237 | goto loop; |
c668690d NA |
4238 | } |
4239 | ||
4240 | /* Checks */ | |
b4bd745d QW |
4241 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4242 | fs_info->stripesize); | |
25179201 | 4243 | |
2552d17e | 4244 | /* move on to the next group */ |
b4bd745d | 4245 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4246 | block_group->start + block_group->length) { |
b4bd745d QW |
4247 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4248 | num_bytes); | |
2552d17e | 4249 | goto loop; |
6226cb0a | 4250 | } |
f5a31e16 | 4251 | |
b4bd745d QW |
4252 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
4253 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
4254 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4255 | |
18513091 WX |
4256 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4257 | num_bytes, delalloc); | |
f0486c68 | 4258 | if (ret == -EAGAIN) { |
b4bd745d QW |
4259 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4260 | num_bytes); | |
2552d17e | 4261 | goto loop; |
0f9dd46c | 4262 | } |
9cfa3e34 | 4263 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4264 | |
f0486c68 | 4265 | /* we are all good, lets return */ |
b4bd745d | 4266 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4267 | ins->offset = num_bytes; |
d2fb3437 | 4268 | |
b4bd745d QW |
4269 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4270 | num_bytes); | |
e570fd27 | 4271 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4272 | break; |
4273 | loop: | |
baba5062 | 4274 | release_block_group(block_group, &ffe_ctl, delalloc); |
14443937 | 4275 | cond_resched(); |
2552d17e JB |
4276 | } |
4277 | up_read(&space_info->groups_sem); | |
4278 | ||
15b7ee65 | 4279 | ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search); |
e72d79d6 | 4280 | if (ret > 0) |
b742bb82 YZ |
4281 | goto search; |
4282 | ||
db8fe64f | 4283 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4284 | /* |
4285 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4286 | * any contiguous hole. | |
4287 | */ | |
4288 | if (!ffe_ctl.max_extent_size) | |
4289 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4290 | spin_lock(&space_info->lock); |
b4bd745d | 4291 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4292 | spin_unlock(&space_info->lock); |
b4bd745d | 4293 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4294 | } else if (ret == -ENOSPC) { |
4295 | ret = cache_block_group_error; | |
4f4db217 | 4296 | } |
0f70abe2 | 4297 | return ret; |
fec577fb | 4298 | } |
ec44a35c | 4299 | |
6f47c706 NB |
4300 | /* |
4301 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4302 | * hole that is at least as big as @num_bytes. | |
4303 | * | |
4304 | * @root - The root that will contain this extent | |
4305 | * | |
4306 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4307 | * is used for accounting purposes. This value differs | |
4308 | * from @num_bytes only in the case of compressed extents. | |
4309 | * | |
4310 | * @num_bytes - Number of bytes to allocate on-disk. | |
4311 | * | |
4312 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4313 | * allocator should try to satisfy. In some cases | |
4314 | * @num_bytes may be larger than what is required and if | |
4315 | * the filesystem is fragmented then allocation fails. | |
4316 | * However, the presence of @min_alloc_size gives a | |
4317 | * chance to try and satisfy the smaller allocation. | |
4318 | * | |
4319 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4320 | * size in bytes the allocator should try to find free | |
4321 | * next to the block it returns. This is just a hint and | |
4322 | * may be ignored by the allocator. | |
4323 | * | |
4324 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4325 | * address passed. It might be ignored. | |
4326 | * | |
4327 | * @ins - This key is modified to record the found hole. It will | |
4328 | * have the following values: | |
4329 | * ins->objectid == start position | |
4330 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4331 | * ins->offset == the size of the hole. | |
4332 | * | |
4333 | * @is_data - Boolean flag indicating whether an extent is | |
4334 | * allocated for data (true) or metadata (false) | |
4335 | * | |
4336 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4337 | * delalloc or not. If 'true' data_rwsem of block groups | |
4338 | * is going to be acquired. | |
4339 | * | |
4340 | * | |
4341 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4342 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4343 | * largest available hole the allocator managed to find. | |
4344 | */ | |
18513091 | 4345 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4346 | u64 num_bytes, u64 min_alloc_size, |
4347 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4348 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4349 | { |
ab8d0fc4 | 4350 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4351 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4352 | u64 flags; |
fec577fb | 4353 | int ret; |
925baedd | 4354 | |
1b86826d | 4355 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4356 | again: |
0b246afa | 4357 | WARN_ON(num_bytes < fs_info->sectorsize); |
437490fe | 4358 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
18513091 | 4359 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4360 | if (!ret && !is_data) { |
ab8d0fc4 | 4361 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4362 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4363 | if (!final_tried && ins->offset) { |
4364 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4365 | num_bytes = round_down(num_bytes, |
0b246afa | 4366 | fs_info->sectorsize); |
9e622d6b | 4367 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4368 | ram_bytes = num_bytes; |
9e622d6b MX |
4369 | if (num_bytes == min_alloc_size) |
4370 | final_tried = true; | |
4371 | goto again; | |
ab8d0fc4 | 4372 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4373 | struct btrfs_space_info *sinfo; |
4374 | ||
280c2908 | 4375 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4376 | btrfs_err(fs_info, |
5d163e0e JM |
4377 | "allocation failed flags %llu, wanted %llu", |
4378 | flags, num_bytes); | |
53804280 | 4379 | if (sinfo) |
5da6afeb JB |
4380 | btrfs_dump_space_info(fs_info, sinfo, |
4381 | num_bytes, 1); | |
9e622d6b | 4382 | } |
925baedd | 4383 | } |
0f9dd46c JB |
4384 | |
4385 | return ret; | |
e6dcd2dc CM |
4386 | } |
4387 | ||
a0fbf736 NB |
4388 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4389 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4390 | { |
32da5386 | 4391 | struct btrfs_block_group *cache; |
0f9dd46c | 4392 | |
0b246afa | 4393 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4394 | if (!cache) { |
0b246afa JM |
4395 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4396 | start); | |
0f9dd46c JB |
4397 | return -ENOSPC; |
4398 | } | |
1f3c79a2 | 4399 | |
a0fbf736 NB |
4400 | btrfs_add_free_space(cache, start, len); |
4401 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4402 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4403 | ||
fa9c0d79 | 4404 | btrfs_put_block_group(cache); |
a0fbf736 | 4405 | return 0; |
e6dcd2dc CM |
4406 | } |
4407 | ||
7bfc1007 NB |
4408 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, |
4409 | u64 len) | |
e688b725 | 4410 | { |
7ef54d54 | 4411 | struct btrfs_block_group *cache; |
a0fbf736 | 4412 | int ret = 0; |
7ef54d54 | 4413 | |
7bfc1007 | 4414 | cache = btrfs_lookup_block_group(trans->fs_info, start); |
7ef54d54 | 4415 | if (!cache) { |
7bfc1007 NB |
4416 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
4417 | start); | |
7ef54d54 NB |
4418 | return -ENOSPC; |
4419 | } | |
4420 | ||
6690d071 | 4421 | ret = pin_down_extent(trans, cache, start, len, 1); |
7ef54d54 | 4422 | btrfs_put_block_group(cache); |
a0fbf736 | 4423 | return ret; |
e688b725 CM |
4424 | } |
4425 | ||
5d4f98a2 | 4426 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4427 | u64 parent, u64 root_objectid, |
4428 | u64 flags, u64 owner, u64 offset, | |
4429 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4430 | { |
ef89b824 | 4431 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4432 | int ret; |
e6dcd2dc | 4433 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4434 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4435 | struct btrfs_path *path; |
5d4f98a2 YZ |
4436 | struct extent_buffer *leaf; |
4437 | int type; | |
4438 | u32 size; | |
26b8003f | 4439 | |
5d4f98a2 YZ |
4440 | if (parent > 0) |
4441 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4442 | else | |
4443 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4444 | |
5d4f98a2 | 4445 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4446 | |
4447 | path = btrfs_alloc_path(); | |
db5b493a TI |
4448 | if (!path) |
4449 | return -ENOMEM; | |
47e4bb98 | 4450 | |
b9473439 | 4451 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4452 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4453 | ins, size); | |
79787eaa JM |
4454 | if (ret) { |
4455 | btrfs_free_path(path); | |
4456 | return ret; | |
4457 | } | |
0f9dd46c | 4458 | |
5d4f98a2 YZ |
4459 | leaf = path->nodes[0]; |
4460 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4461 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4462 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4463 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4464 | btrfs_set_extent_flags(leaf, extent_item, | |
4465 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4466 | ||
4467 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4468 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4469 | if (parent > 0) { | |
4470 | struct btrfs_shared_data_ref *ref; | |
4471 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4472 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4473 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4474 | } else { | |
4475 | struct btrfs_extent_data_ref *ref; | |
4476 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4477 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4478 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4479 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4480 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4481 | } | |
47e4bb98 CM |
4482 | |
4483 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4484 | btrfs_free_path(path); |
f510cfec | 4485 | |
25a356d3 | 4486 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4487 | if (ret) |
4488 | return ret; | |
4489 | ||
ade4b516 | 4490 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4491 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4492 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4493 | ins->objectid, ins->offset); |
f5947066 CM |
4494 | BUG(); |
4495 | } | |
71ff6437 | 4496 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4497 | return ret; |
4498 | } | |
4499 | ||
5d4f98a2 | 4500 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4501 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4502 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4503 | { |
9dcdbe01 | 4504 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4505 | int ret; |
5d4f98a2 | 4506 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4507 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4508 | struct btrfs_tree_block_info *block_info; |
4509 | struct btrfs_extent_inline_ref *iref; | |
4510 | struct btrfs_path *path; | |
4511 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4512 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4513 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4514 | u64 num_bytes; |
21ebfbe7 | 4515 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4516 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4517 | |
4e6bd4e0 NB |
4518 | ref = btrfs_delayed_node_to_tree_ref(node); |
4519 | ||
4e6bd4e0 NB |
4520 | extent_key.objectid = node->bytenr; |
4521 | if (skinny_metadata) { | |
4522 | extent_key.offset = ref->level; | |
4523 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4524 | num_bytes = fs_info->nodesize; | |
4525 | } else { | |
4526 | extent_key.offset = node->num_bytes; | |
4527 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4528 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4529 | num_bytes = node->num_bytes; |
4530 | } | |
1c2308f8 | 4531 | |
5d4f98a2 | 4532 | path = btrfs_alloc_path(); |
80ee54bf | 4533 | if (!path) |
d8926bb3 | 4534 | return -ENOMEM; |
56bec294 | 4535 | |
5d4f98a2 YZ |
4536 | path->leave_spinning = 1; |
4537 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 4538 | &extent_key, size); |
79787eaa | 4539 | if (ret) { |
dd825259 | 4540 | btrfs_free_path(path); |
79787eaa JM |
4541 | return ret; |
4542 | } | |
5d4f98a2 YZ |
4543 | |
4544 | leaf = path->nodes[0]; | |
4545 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4546 | struct btrfs_extent_item); | |
4547 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4548 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4549 | btrfs_set_extent_flags(leaf, extent_item, | |
4550 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4551 | |
3173a18f JB |
4552 | if (skinny_metadata) { |
4553 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4554 | } else { | |
4555 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4556 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4557 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4558 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4559 | } | |
5d4f98a2 | 4560 | |
d4b20733 | 4561 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4562 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4563 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4564 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4565 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4566 | } else { |
4567 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4568 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4569 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4570 | } |
4571 | ||
4572 | btrfs_mark_buffer_dirty(leaf); | |
4573 | btrfs_free_path(path); | |
4574 | ||
4e6bd4e0 NB |
4575 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4576 | num_bytes); | |
1e144fb8 OS |
4577 | if (ret) |
4578 | return ret; | |
4579 | ||
ade4b516 JB |
4580 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4581 | fs_info->nodesize, 1); | |
79787eaa | 4582 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4583 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4584 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4585 | BUG(); |
4586 | } | |
0be5dc67 | 4587 | |
4e6bd4e0 | 4588 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4589 | fs_info->nodesize); |
5d4f98a2 YZ |
4590 | return ret; |
4591 | } | |
4592 | ||
4593 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4594 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4595 | u64 offset, u64 ram_bytes, |
4596 | struct btrfs_key *ins) | |
5d4f98a2 | 4597 | { |
76675593 | 4598 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
4599 | int ret; |
4600 | ||
84f7d8e6 | 4601 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4602 | |
76675593 QW |
4603 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4604 | ins->objectid, ins->offset, 0); | |
4605 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4606 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
4607 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
4608 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
4609 | return ret; |
4610 | } | |
e02119d5 CM |
4611 | |
4612 | /* | |
4613 | * this is used by the tree logging recovery code. It records that | |
4614 | * an extent has been allocated and makes sure to clear the free | |
4615 | * space cache bits as well | |
4616 | */ | |
5d4f98a2 | 4617 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4618 | u64 root_objectid, u64 owner, u64 offset, |
4619 | struct btrfs_key *ins) | |
e02119d5 | 4620 | { |
61da2abf | 4621 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4622 | int ret; |
32da5386 | 4623 | struct btrfs_block_group *block_group; |
ed7a6948 | 4624 | struct btrfs_space_info *space_info; |
11833d66 | 4625 | |
8c2a1a30 JB |
4626 | /* |
4627 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4628 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4629 | */ |
0b246afa | 4630 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4631 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4632 | ins->offset); | |
b50c6e25 | 4633 | if (ret) |
8c2a1a30 | 4634 | return ret; |
11833d66 YZ |
4635 | } |
4636 | ||
0b246afa | 4637 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4638 | if (!block_group) |
4639 | return -EINVAL; | |
4640 | ||
ed7a6948 WX |
4641 | space_info = block_group->space_info; |
4642 | spin_lock(&space_info->lock); | |
4643 | spin_lock(&block_group->lock); | |
4644 | space_info->bytes_reserved += ins->offset; | |
4645 | block_group->reserved += ins->offset; | |
4646 | spin_unlock(&block_group->lock); | |
4647 | spin_unlock(&space_info->lock); | |
4648 | ||
ef89b824 NB |
4649 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4650 | offset, ins, 1); | |
bd727173 | 4651 | if (ret) |
ab9b2c7b | 4652 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
b50c6e25 | 4653 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4654 | return ret; |
4655 | } | |
4656 | ||
48a3b636 ES |
4657 | static struct extent_buffer * |
4658 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 4659 | u64 bytenr, int level, u64 owner) |
65b51a00 | 4660 | { |
0b246afa | 4661 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4662 | struct extent_buffer *buf; |
4663 | ||
2ff7e61e | 4664 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4665 | if (IS_ERR(buf)) |
4666 | return buf; | |
4667 | ||
b72c3aba QW |
4668 | /* |
4669 | * Extra safety check in case the extent tree is corrupted and extent | |
4670 | * allocator chooses to use a tree block which is already used and | |
4671 | * locked. | |
4672 | */ | |
4673 | if (buf->lock_owner == current->pid) { | |
4674 | btrfs_err_rl(fs_info, | |
4675 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4676 | buf->start, btrfs_header_owner(buf), current->pid); | |
4677 | free_extent_buffer(buf); | |
4678 | return ERR_PTR(-EUCLEAN); | |
4679 | } | |
4680 | ||
ad244665 | 4681 | btrfs_set_buffer_lockdep_class(owner, buf, level); |
65b51a00 | 4682 | btrfs_tree_lock(buf); |
6a884d7d | 4683 | btrfs_clean_tree_block(buf); |
3083ee2e | 4684 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 4685 | |
8bead258 | 4686 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 4687 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4688 | |
bc877d28 NB |
4689 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4690 | btrfs_set_header_level(buf, level); | |
4691 | btrfs_set_header_bytenr(buf, buf->start); | |
4692 | btrfs_set_header_generation(buf, trans->transid); | |
4693 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4694 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4695 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4696 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4697 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4698 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4699 | /* |
4700 | * we allow two log transactions at a time, use different | |
52042d8e | 4701 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4702 | */ |
656f30db | 4703 | if (buf->log_index == 0) |
8cef4e16 YZ |
4704 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4705 | buf->start + buf->len - 1, GFP_NOFS); | |
4706 | else | |
4707 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4708 | buf->start + buf->len - 1); |
d0c803c4 | 4709 | } else { |
656f30db | 4710 | buf->log_index = -1; |
d0c803c4 | 4711 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4712 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4713 | } |
64c12921 | 4714 | trans->dirty = true; |
b4ce94de | 4715 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4716 | return buf; |
4717 | } | |
4718 | ||
fec577fb | 4719 | /* |
f0486c68 | 4720 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4721 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4722 | */ |
4d75f8a9 | 4723 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4724 | struct btrfs_root *root, |
4725 | u64 parent, u64 root_objectid, | |
4726 | const struct btrfs_disk_key *key, | |
4727 | int level, u64 hint, | |
4728 | u64 empty_size) | |
fec577fb | 4729 | { |
0b246afa | 4730 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4731 | struct btrfs_key ins; |
f0486c68 | 4732 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4733 | struct extent_buffer *buf; |
67b7859e | 4734 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4735 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4736 | u64 flags = 0; |
4737 | int ret; | |
0b246afa JM |
4738 | u32 blocksize = fs_info->nodesize; |
4739 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4740 | |
05653ef3 | 4741 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4742 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4743 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 4744 | level, root_objectid); |
faa2dbf0 JB |
4745 | if (!IS_ERR(buf)) |
4746 | root->alloc_bytenr += blocksize; | |
4747 | return buf; | |
4748 | } | |
05653ef3 | 4749 | #endif |
fccb84c9 | 4750 | |
67f9c220 | 4751 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4752 | if (IS_ERR(block_rsv)) |
4753 | return ERR_CAST(block_rsv); | |
4754 | ||
18513091 | 4755 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4756 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4757 | if (ret) |
4758 | goto out_unuse; | |
55c69072 | 4759 | |
bc877d28 NB |
4760 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
4761 | root_objectid); | |
67b7859e OS |
4762 | if (IS_ERR(buf)) { |
4763 | ret = PTR_ERR(buf); | |
4764 | goto out_free_reserved; | |
4765 | } | |
f0486c68 YZ |
4766 | |
4767 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4768 | if (parent == 0) | |
4769 | parent = ins.objectid; | |
4770 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4771 | } else | |
4772 | BUG_ON(parent > 0); | |
4773 | ||
4774 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4775 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4776 | if (!extent_op) { |
4777 | ret = -ENOMEM; | |
4778 | goto out_free_buf; | |
4779 | } | |
f0486c68 YZ |
4780 | if (key) |
4781 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4782 | else | |
4783 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4784 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4785 | extent_op->update_key = skinny_metadata ? false : true; |
4786 | extent_op->update_flags = true; | |
4787 | extent_op->is_data = false; | |
b1c79e09 | 4788 | extent_op->level = level; |
f0486c68 | 4789 | |
ed4f255b QW |
4790 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4791 | ins.objectid, ins.offset, parent); | |
4792 | generic_ref.real_root = root->root_key.objectid; | |
4793 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4794 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 4795 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 4796 | extent_op, NULL, NULL); |
67b7859e OS |
4797 | if (ret) |
4798 | goto out_free_delayed; | |
f0486c68 | 4799 | } |
fec577fb | 4800 | return buf; |
67b7859e OS |
4801 | |
4802 | out_free_delayed: | |
4803 | btrfs_free_delayed_extent_op(extent_op); | |
4804 | out_free_buf: | |
4805 | free_extent_buffer(buf); | |
4806 | out_free_reserved: | |
2ff7e61e | 4807 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4808 | out_unuse: |
67f9c220 | 4809 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4810 | return ERR_PTR(ret); |
fec577fb | 4811 | } |
a28ec197 | 4812 | |
2c47e605 YZ |
4813 | struct walk_control { |
4814 | u64 refs[BTRFS_MAX_LEVEL]; | |
4815 | u64 flags[BTRFS_MAX_LEVEL]; | |
4816 | struct btrfs_key update_progress; | |
aea6f028 JB |
4817 | struct btrfs_key drop_progress; |
4818 | int drop_level; | |
2c47e605 YZ |
4819 | int stage; |
4820 | int level; | |
4821 | int shared_level; | |
4822 | int update_ref; | |
4823 | int keep_locks; | |
1c4850e2 YZ |
4824 | int reada_slot; |
4825 | int reada_count; | |
78c52d9e | 4826 | int restarted; |
2c47e605 YZ |
4827 | }; |
4828 | ||
4829 | #define DROP_REFERENCE 1 | |
4830 | #define UPDATE_BACKREF 2 | |
4831 | ||
1c4850e2 YZ |
4832 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4833 | struct btrfs_root *root, | |
4834 | struct walk_control *wc, | |
4835 | struct btrfs_path *path) | |
6407bf6d | 4836 | { |
0b246afa | 4837 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4838 | u64 bytenr; |
4839 | u64 generation; | |
4840 | u64 refs; | |
94fcca9f | 4841 | u64 flags; |
5d4f98a2 | 4842 | u32 nritems; |
1c4850e2 YZ |
4843 | struct btrfs_key key; |
4844 | struct extent_buffer *eb; | |
6407bf6d | 4845 | int ret; |
1c4850e2 YZ |
4846 | int slot; |
4847 | int nread = 0; | |
6407bf6d | 4848 | |
1c4850e2 YZ |
4849 | if (path->slots[wc->level] < wc->reada_slot) { |
4850 | wc->reada_count = wc->reada_count * 2 / 3; | |
4851 | wc->reada_count = max(wc->reada_count, 2); | |
4852 | } else { | |
4853 | wc->reada_count = wc->reada_count * 3 / 2; | |
4854 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4855 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4856 | } |
7bb86316 | 4857 | |
1c4850e2 YZ |
4858 | eb = path->nodes[wc->level]; |
4859 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4860 | |
1c4850e2 YZ |
4861 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4862 | if (nread >= wc->reada_count) | |
4863 | break; | |
bd56b302 | 4864 | |
2dd3e67b | 4865 | cond_resched(); |
1c4850e2 YZ |
4866 | bytenr = btrfs_node_blockptr(eb, slot); |
4867 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4868 | |
1c4850e2 YZ |
4869 | if (slot == path->slots[wc->level]) |
4870 | goto reada; | |
5d4f98a2 | 4871 | |
1c4850e2 YZ |
4872 | if (wc->stage == UPDATE_BACKREF && |
4873 | generation <= root->root_key.offset) | |
bd56b302 CM |
4874 | continue; |
4875 | ||
94fcca9f | 4876 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4877 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4878 | wc->level - 1, 1, &refs, |
4879 | &flags); | |
79787eaa JM |
4880 | /* We don't care about errors in readahead. */ |
4881 | if (ret < 0) | |
4882 | continue; | |
94fcca9f YZ |
4883 | BUG_ON(refs == 0); |
4884 | ||
1c4850e2 | 4885 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4886 | if (refs == 1) |
4887 | goto reada; | |
bd56b302 | 4888 | |
94fcca9f YZ |
4889 | if (wc->level == 1 && |
4890 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4891 | continue; | |
1c4850e2 YZ |
4892 | if (!wc->update_ref || |
4893 | generation <= root->root_key.offset) | |
4894 | continue; | |
4895 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4896 | ret = btrfs_comp_cpu_keys(&key, | |
4897 | &wc->update_progress); | |
4898 | if (ret < 0) | |
4899 | continue; | |
94fcca9f YZ |
4900 | } else { |
4901 | if (wc->level == 1 && | |
4902 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4903 | continue; | |
6407bf6d | 4904 | } |
1c4850e2 | 4905 | reada: |
2ff7e61e | 4906 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 4907 | nread++; |
20524f02 | 4908 | } |
1c4850e2 | 4909 | wc->reada_slot = slot; |
20524f02 | 4910 | } |
2c47e605 | 4911 | |
f82d02d9 | 4912 | /* |
2c016dc2 | 4913 | * helper to process tree block while walking down the tree. |
2c47e605 | 4914 | * |
2c47e605 YZ |
4915 | * when wc->stage == UPDATE_BACKREF, this function updates |
4916 | * back refs for pointers in the block. | |
4917 | * | |
4918 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4919 | */ |
2c47e605 | 4920 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4921 | struct btrfs_root *root, |
2c47e605 | 4922 | struct btrfs_path *path, |
94fcca9f | 4923 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4924 | { |
2ff7e61e | 4925 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4926 | int level = wc->level; |
4927 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4928 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4929 | int ret; |
4930 | ||
2c47e605 YZ |
4931 | if (wc->stage == UPDATE_BACKREF && |
4932 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4933 | return 1; | |
f82d02d9 | 4934 | |
2c47e605 YZ |
4935 | /* |
4936 | * when reference count of tree block is 1, it won't increase | |
4937 | * again. once full backref flag is set, we never clear it. | |
4938 | */ | |
94fcca9f YZ |
4939 | if (lookup_info && |
4940 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4941 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4942 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4943 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4944 | eb->start, level, 1, |
2c47e605 YZ |
4945 | &wc->refs[level], |
4946 | &wc->flags[level]); | |
79787eaa JM |
4947 | BUG_ON(ret == -ENOMEM); |
4948 | if (ret) | |
4949 | return ret; | |
2c47e605 YZ |
4950 | BUG_ON(wc->refs[level] == 0); |
4951 | } | |
5d4f98a2 | 4952 | |
2c47e605 YZ |
4953 | if (wc->stage == DROP_REFERENCE) { |
4954 | if (wc->refs[level] > 1) | |
4955 | return 1; | |
f82d02d9 | 4956 | |
2c47e605 | 4957 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4958 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4959 | path->locks[level] = 0; |
4960 | } | |
4961 | return 0; | |
4962 | } | |
f82d02d9 | 4963 | |
2c47e605 YZ |
4964 | /* wc->stage == UPDATE_BACKREF */ |
4965 | if (!(wc->flags[level] & flag)) { | |
4966 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4967 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4968 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4969 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4970 | BUG_ON(ret); /* -ENOMEM */ |
42c9d0b5 | 4971 | ret = btrfs_set_disk_extent_flags(trans, eb, flag, |
b1c79e09 | 4972 | btrfs_header_level(eb), 0); |
79787eaa | 4973 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4974 | wc->flags[level] |= flag; |
4975 | } | |
4976 | ||
4977 | /* | |
4978 | * the block is shared by multiple trees, so it's not good to | |
4979 | * keep the tree lock | |
4980 | */ | |
4981 | if (path->locks[level] && level > 0) { | |
bd681513 | 4982 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4983 | path->locks[level] = 0; |
4984 | } | |
4985 | return 0; | |
4986 | } | |
4987 | ||
78c52d9e JB |
4988 | /* |
4989 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4990 | * would have a drop_progress key that hadn't been updated properly. | |
4991 | */ | |
4992 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4993 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4994 | int level) | |
4995 | { | |
4996 | struct btrfs_path *path; | |
4997 | struct btrfs_extent_inline_ref *iref; | |
4998 | int ret; | |
4999 | ||
5000 | path = btrfs_alloc_path(); | |
5001 | if (!path) | |
5002 | return -ENOMEM; | |
5003 | ||
5004 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
5005 | root->fs_info->nodesize, parent, | |
5006 | root->root_key.objectid, level, 0); | |
5007 | btrfs_free_path(path); | |
5008 | if (ret == -ENOENT) | |
5009 | return 0; | |
5010 | if (ret < 0) | |
5011 | return ret; | |
5012 | return 1; | |
5013 | } | |
5014 | ||
1c4850e2 | 5015 | /* |
2c016dc2 | 5016 | * helper to process tree block pointer. |
1c4850e2 YZ |
5017 | * |
5018 | * when wc->stage == DROP_REFERENCE, this function checks | |
5019 | * reference count of the block pointed to. if the block | |
5020 | * is shared and we need update back refs for the subtree | |
5021 | * rooted at the block, this function changes wc->stage to | |
5022 | * UPDATE_BACKREF. if the block is shared and there is no | |
5023 | * need to update back, this function drops the reference | |
5024 | * to the block. | |
5025 | * | |
5026 | * NOTE: return value 1 means we should stop walking down. | |
5027 | */ | |
5028 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5029 | struct btrfs_root *root, | |
5030 | struct btrfs_path *path, | |
94fcca9f | 5031 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 5032 | { |
0b246afa | 5033 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
5034 | u64 bytenr; |
5035 | u64 generation; | |
5036 | u64 parent; | |
1c4850e2 | 5037 | struct btrfs_key key; |
581c1760 | 5038 | struct btrfs_key first_key; |
ffd4bb2a | 5039 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
5040 | struct extent_buffer *next; |
5041 | int level = wc->level; | |
5042 | int reada = 0; | |
5043 | int ret = 0; | |
1152651a | 5044 | bool need_account = false; |
1c4850e2 YZ |
5045 | |
5046 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5047 | path->slots[level]); | |
5048 | /* | |
5049 | * if the lower level block was created before the snapshot | |
5050 | * was created, we know there is no need to update back refs | |
5051 | * for the subtree | |
5052 | */ | |
5053 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5054 | generation <= root->root_key.offset) { |
5055 | *lookup_info = 1; | |
1c4850e2 | 5056 | return 1; |
94fcca9f | 5057 | } |
1c4850e2 YZ |
5058 | |
5059 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
5060 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
5061 | path->slots[level]); | |
1c4850e2 | 5062 | |
0b246afa | 5063 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 5064 | if (!next) { |
2ff7e61e | 5065 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
5066 | if (IS_ERR(next)) |
5067 | return PTR_ERR(next); | |
5068 | ||
b2aaaa3b JB |
5069 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
5070 | level - 1); | |
1c4850e2 YZ |
5071 | reada = 1; |
5072 | } | |
5073 | btrfs_tree_lock(next); | |
8bead258 | 5074 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 5075 | |
2ff7e61e | 5076 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
5077 | &wc->refs[level - 1], |
5078 | &wc->flags[level - 1]); | |
4867268c JB |
5079 | if (ret < 0) |
5080 | goto out_unlock; | |
79787eaa | 5081 | |
c2cf52eb | 5082 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 5083 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
5084 | ret = -EIO; |
5085 | goto out_unlock; | |
c2cf52eb | 5086 | } |
94fcca9f | 5087 | *lookup_info = 0; |
1c4850e2 | 5088 | |
94fcca9f | 5089 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5090 | if (wc->refs[level - 1] > 1) { |
1152651a | 5091 | need_account = true; |
94fcca9f YZ |
5092 | if (level == 1 && |
5093 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5094 | goto skip; | |
5095 | ||
1c4850e2 YZ |
5096 | if (!wc->update_ref || |
5097 | generation <= root->root_key.offset) | |
5098 | goto skip; | |
5099 | ||
5100 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5101 | path->slots[level]); | |
5102 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5103 | if (ret < 0) | |
5104 | goto skip; | |
5105 | ||
5106 | wc->stage = UPDATE_BACKREF; | |
5107 | wc->shared_level = level - 1; | |
5108 | } | |
94fcca9f YZ |
5109 | } else { |
5110 | if (level == 1 && | |
5111 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5112 | goto skip; | |
1c4850e2 YZ |
5113 | } |
5114 | ||
b9fab919 | 5115 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
5116 | btrfs_tree_unlock(next); |
5117 | free_extent_buffer(next); | |
5118 | next = NULL; | |
94fcca9f | 5119 | *lookup_info = 1; |
1c4850e2 YZ |
5120 | } |
5121 | ||
5122 | if (!next) { | |
5123 | if (reada && level == 1) | |
5124 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
5125 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
5126 | &first_key); | |
64c043de LB |
5127 | if (IS_ERR(next)) { |
5128 | return PTR_ERR(next); | |
5129 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 5130 | free_extent_buffer(next); |
97d9a8a4 | 5131 | return -EIO; |
416bc658 | 5132 | } |
1c4850e2 | 5133 | btrfs_tree_lock(next); |
8bead258 | 5134 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
5135 | } |
5136 | ||
5137 | level--; | |
4867268c JB |
5138 | ASSERT(level == btrfs_header_level(next)); |
5139 | if (level != btrfs_header_level(next)) { | |
5140 | btrfs_err(root->fs_info, "mismatched level"); | |
5141 | ret = -EIO; | |
5142 | goto out_unlock; | |
5143 | } | |
1c4850e2 YZ |
5144 | path->nodes[level] = next; |
5145 | path->slots[level] = 0; | |
bd681513 | 5146 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
5147 | wc->level = level; |
5148 | if (wc->level == 1) | |
5149 | wc->reada_slot = 0; | |
5150 | return 0; | |
5151 | skip: | |
5152 | wc->refs[level - 1] = 0; | |
5153 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5154 | if (wc->stage == DROP_REFERENCE) { |
5155 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5156 | parent = path->nodes[level]->start; | |
5157 | } else { | |
4867268c | 5158 | ASSERT(root->root_key.objectid == |
94fcca9f | 5159 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5160 | if (root->root_key.objectid != |
5161 | btrfs_header_owner(path->nodes[level])) { | |
5162 | btrfs_err(root->fs_info, | |
5163 | "mismatched block owner"); | |
5164 | ret = -EIO; | |
5165 | goto out_unlock; | |
5166 | } | |
94fcca9f YZ |
5167 | parent = 0; |
5168 | } | |
1c4850e2 | 5169 | |
78c52d9e JB |
5170 | /* |
5171 | * If we had a drop_progress we need to verify the refs are set | |
5172 | * as expected. If we find our ref then we know that from here | |
5173 | * on out everything should be correct, and we can clear the | |
5174 | * ->restarted flag. | |
5175 | */ | |
5176 | if (wc->restarted) { | |
5177 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5178 | level - 1); | |
5179 | if (ret < 0) | |
5180 | goto out_unlock; | |
5181 | if (ret == 0) | |
5182 | goto no_delete; | |
5183 | ret = 0; | |
5184 | wc->restarted = 0; | |
5185 | } | |
5186 | ||
2cd86d30 QW |
5187 | /* |
5188 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5189 | * already accounted them at merge time (replace_path), | |
5190 | * thus we could skip expensive subtree trace here. | |
5191 | */ | |
5192 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5193 | need_account) { | |
deb40627 | 5194 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5195 | generation, level - 1); |
1152651a | 5196 | if (ret) { |
0b246afa | 5197 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5198 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5199 | ret); | |
1152651a MF |
5200 | } |
5201 | } | |
aea6f028 JB |
5202 | |
5203 | /* | |
5204 | * We need to update the next key in our walk control so we can | |
5205 | * update the drop_progress key accordingly. We don't care if | |
5206 | * find_next_key doesn't find a key because that means we're at | |
5207 | * the end and are going to clean up now. | |
5208 | */ | |
5209 | wc->drop_level = level; | |
5210 | find_next_key(path, level, &wc->drop_progress); | |
5211 | ||
ffd4bb2a QW |
5212 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
5213 | fs_info->nodesize, parent); | |
5214 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
5215 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
5216 | if (ret) |
5217 | goto out_unlock; | |
1c4850e2 | 5218 | } |
78c52d9e | 5219 | no_delete: |
4867268c JB |
5220 | *lookup_info = 1; |
5221 | ret = 1; | |
5222 | ||
5223 | out_unlock: | |
1c4850e2 YZ |
5224 | btrfs_tree_unlock(next); |
5225 | free_extent_buffer(next); | |
4867268c JB |
5226 | |
5227 | return ret; | |
1c4850e2 YZ |
5228 | } |
5229 | ||
2c47e605 | 5230 | /* |
2c016dc2 | 5231 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5232 | * |
5233 | * when wc->stage == DROP_REFERENCE, this function drops | |
5234 | * reference count on the block. | |
5235 | * | |
5236 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5237 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5238 | * to UPDATE_BACKREF previously while processing the block. | |
5239 | * | |
5240 | * NOTE: return value 1 means we should stop walking up. | |
5241 | */ | |
5242 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5243 | struct btrfs_root *root, | |
5244 | struct btrfs_path *path, | |
5245 | struct walk_control *wc) | |
5246 | { | |
0b246afa | 5247 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5248 | int ret; |
2c47e605 YZ |
5249 | int level = wc->level; |
5250 | struct extent_buffer *eb = path->nodes[level]; | |
5251 | u64 parent = 0; | |
5252 | ||
5253 | if (wc->stage == UPDATE_BACKREF) { | |
5254 | BUG_ON(wc->shared_level < level); | |
5255 | if (level < wc->shared_level) | |
5256 | goto out; | |
5257 | ||
2c47e605 YZ |
5258 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5259 | if (ret > 0) | |
5260 | wc->update_ref = 0; | |
5261 | ||
5262 | wc->stage = DROP_REFERENCE; | |
5263 | wc->shared_level = -1; | |
5264 | path->slots[level] = 0; | |
5265 | ||
5266 | /* | |
5267 | * check reference count again if the block isn't locked. | |
5268 | * we should start walking down the tree again if reference | |
5269 | * count is one. | |
5270 | */ | |
5271 | if (!path->locks[level]) { | |
5272 | BUG_ON(level == 0); | |
5273 | btrfs_tree_lock(eb); | |
8bead258 | 5274 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5275 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5276 | |
2ff7e61e | 5277 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5278 | eb->start, level, 1, |
2c47e605 YZ |
5279 | &wc->refs[level], |
5280 | &wc->flags[level]); | |
79787eaa JM |
5281 | if (ret < 0) { |
5282 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5283 | path->locks[level] = 0; |
79787eaa JM |
5284 | return ret; |
5285 | } | |
2c47e605 YZ |
5286 | BUG_ON(wc->refs[level] == 0); |
5287 | if (wc->refs[level] == 1) { | |
bd681513 | 5288 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5289 | path->locks[level] = 0; |
2c47e605 YZ |
5290 | return 1; |
5291 | } | |
f82d02d9 | 5292 | } |
2c47e605 | 5293 | } |
f82d02d9 | 5294 | |
2c47e605 YZ |
5295 | /* wc->stage == DROP_REFERENCE */ |
5296 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5297 | |
2c47e605 YZ |
5298 | if (wc->refs[level] == 1) { |
5299 | if (level == 0) { | |
5300 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5301 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5302 | else |
e339a6b0 | 5303 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5304 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5305 | if (is_fstree(root->root_key.objectid)) { |
5306 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5307 | if (ret) { | |
5308 | btrfs_err_rl(fs_info, | |
5309 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5310 | ret); |
c4140cbf | 5311 | } |
1152651a | 5312 | } |
2c47e605 | 5313 | } |
6a884d7d | 5314 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5315 | if (!path->locks[level] && |
5316 | btrfs_header_generation(eb) == trans->transid) { | |
5317 | btrfs_tree_lock(eb); | |
8bead258 | 5318 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5319 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5320 | } |
6a884d7d | 5321 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5322 | } |
5323 | ||
5324 | if (eb == root->node) { | |
5325 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5326 | parent = eb->start; | |
65c6e82b QW |
5327 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5328 | goto owner_mismatch; | |
2c47e605 YZ |
5329 | } else { |
5330 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5331 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5332 | else if (root->root_key.objectid != |
5333 | btrfs_header_owner(path->nodes[level + 1])) | |
5334 | goto owner_mismatch; | |
f82d02d9 | 5335 | } |
f82d02d9 | 5336 | |
5581a51a | 5337 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5338 | out: |
5339 | wc->refs[level] = 0; | |
5340 | wc->flags[level] = 0; | |
f0486c68 | 5341 | return 0; |
65c6e82b QW |
5342 | |
5343 | owner_mismatch: | |
5344 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5345 | btrfs_header_owner(eb), root->root_key.objectid); | |
5346 | return -EUCLEAN; | |
2c47e605 YZ |
5347 | } |
5348 | ||
5349 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5350 | struct btrfs_root *root, | |
5351 | struct btrfs_path *path, | |
5352 | struct walk_control *wc) | |
5353 | { | |
2c47e605 | 5354 | int level = wc->level; |
94fcca9f | 5355 | int lookup_info = 1; |
2c47e605 YZ |
5356 | int ret; |
5357 | ||
5358 | while (level >= 0) { | |
94fcca9f | 5359 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5360 | if (ret > 0) |
5361 | break; | |
5362 | ||
5363 | if (level == 0) | |
5364 | break; | |
5365 | ||
7a7965f8 YZ |
5366 | if (path->slots[level] >= |
5367 | btrfs_header_nritems(path->nodes[level])) | |
5368 | break; | |
5369 | ||
94fcca9f | 5370 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5371 | if (ret > 0) { |
5372 | path->slots[level]++; | |
5373 | continue; | |
90d2c51d MX |
5374 | } else if (ret < 0) |
5375 | return ret; | |
1c4850e2 | 5376 | level = wc->level; |
f82d02d9 | 5377 | } |
f82d02d9 YZ |
5378 | return 0; |
5379 | } | |
5380 | ||
d397712b | 5381 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5382 | struct btrfs_root *root, |
f82d02d9 | 5383 | struct btrfs_path *path, |
2c47e605 | 5384 | struct walk_control *wc, int max_level) |
20524f02 | 5385 | { |
2c47e605 | 5386 | int level = wc->level; |
20524f02 | 5387 | int ret; |
9f3a7427 | 5388 | |
2c47e605 YZ |
5389 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5390 | while (level < max_level && path->nodes[level]) { | |
5391 | wc->level = level; | |
5392 | if (path->slots[level] + 1 < | |
5393 | btrfs_header_nritems(path->nodes[level])) { | |
5394 | path->slots[level]++; | |
20524f02 CM |
5395 | return 0; |
5396 | } else { | |
2c47e605 YZ |
5397 | ret = walk_up_proc(trans, root, path, wc); |
5398 | if (ret > 0) | |
5399 | return 0; | |
65c6e82b QW |
5400 | if (ret < 0) |
5401 | return ret; | |
bd56b302 | 5402 | |
2c47e605 | 5403 | if (path->locks[level]) { |
bd681513 CM |
5404 | btrfs_tree_unlock_rw(path->nodes[level], |
5405 | path->locks[level]); | |
2c47e605 | 5406 | path->locks[level] = 0; |
f82d02d9 | 5407 | } |
2c47e605 YZ |
5408 | free_extent_buffer(path->nodes[level]); |
5409 | path->nodes[level] = NULL; | |
5410 | level++; | |
20524f02 CM |
5411 | } |
5412 | } | |
5413 | return 1; | |
5414 | } | |
5415 | ||
9aca1d51 | 5416 | /* |
2c47e605 YZ |
5417 | * drop a subvolume tree. |
5418 | * | |
5419 | * this function traverses the tree freeing any blocks that only | |
5420 | * referenced by the tree. | |
5421 | * | |
5422 | * when a shared tree block is found. this function decreases its | |
5423 | * reference count by one. if update_ref is true, this function | |
5424 | * also make sure backrefs for the shared block and all lower level | |
5425 | * blocks are properly updated. | |
9d1a2a3a DS |
5426 | * |
5427 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5428 | */ |
0078a9f9 | 5429 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5430 | { |
ab8d0fc4 | 5431 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5432 | struct btrfs_path *path; |
2c47e605 | 5433 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5434 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5435 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5436 | struct walk_control *wc; |
5437 | struct btrfs_key key; | |
5438 | int err = 0; | |
5439 | int ret; | |
5440 | int level; | |
d29a9f62 | 5441 | bool root_dropped = false; |
20524f02 | 5442 | |
4fd786e6 | 5443 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5444 | |
5caf2a00 | 5445 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5446 | if (!path) { |
5447 | err = -ENOMEM; | |
5448 | goto out; | |
5449 | } | |
20524f02 | 5450 | |
2c47e605 | 5451 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5452 | if (!wc) { |
5453 | btrfs_free_path(path); | |
cb1b69f4 TI |
5454 | err = -ENOMEM; |
5455 | goto out; | |
38a1a919 | 5456 | } |
2c47e605 | 5457 | |
f3e3d9cc QW |
5458 | /* |
5459 | * Use join to avoid potential EINTR from transaction start. See | |
5460 | * wait_reserve_ticket and the whole reservation callchain. | |
5461 | */ | |
5462 | if (for_reloc) | |
5463 | trans = btrfs_join_transaction(tree_root); | |
5464 | else | |
5465 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5466 | if (IS_ERR(trans)) { |
5467 | err = PTR_ERR(trans); | |
5468 | goto out_free; | |
5469 | } | |
98d5dc13 | 5470 | |
0568e82d JB |
5471 | err = btrfs_run_delayed_items(trans); |
5472 | if (err) | |
5473 | goto out_end_trans; | |
5474 | ||
83354f07 JB |
5475 | /* |
5476 | * This will help us catch people modifying the fs tree while we're | |
5477 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5478 | * dropped as we unlock the root node and parent nodes as we walk down | |
5479 | * the tree, assuming nothing will change. If something does change | |
5480 | * then we'll have stale information and drop references to blocks we've | |
5481 | * already dropped. | |
5482 | */ | |
5483 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5484 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5485 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5486 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 5487 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 5488 | path->slots[level] = 0; |
bd681513 | 5489 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5490 | memset(&wc->update_progress, 0, |
5491 | sizeof(wc->update_progress)); | |
9f3a7427 | 5492 | } else { |
9f3a7427 | 5493 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5494 | memcpy(&wc->update_progress, &key, |
5495 | sizeof(wc->update_progress)); | |
5496 | ||
6702ed49 | 5497 | level = root_item->drop_level; |
2c47e605 | 5498 | BUG_ON(level == 0); |
6702ed49 | 5499 | path->lowest_level = level; |
2c47e605 YZ |
5500 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5501 | path->lowest_level = 0; | |
5502 | if (ret < 0) { | |
5503 | err = ret; | |
79787eaa | 5504 | goto out_end_trans; |
9f3a7427 | 5505 | } |
1c4850e2 | 5506 | WARN_ON(ret > 0); |
2c47e605 | 5507 | |
7d9eb12c CM |
5508 | /* |
5509 | * unlock our path, this is safe because only this | |
5510 | * function is allowed to delete this snapshot | |
5511 | */ | |
5d4f98a2 | 5512 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5513 | |
5514 | level = btrfs_header_level(root->node); | |
5515 | while (1) { | |
5516 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 5517 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 5518 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5519 | |
2ff7e61e | 5520 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5521 | path->nodes[level]->start, |
3173a18f | 5522 | level, 1, &wc->refs[level], |
2c47e605 | 5523 | &wc->flags[level]); |
79787eaa JM |
5524 | if (ret < 0) { |
5525 | err = ret; | |
5526 | goto out_end_trans; | |
5527 | } | |
2c47e605 YZ |
5528 | BUG_ON(wc->refs[level] == 0); |
5529 | ||
5530 | if (level == root_item->drop_level) | |
5531 | break; | |
5532 | ||
5533 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5534 | path->locks[level] = 0; |
2c47e605 YZ |
5535 | WARN_ON(wc->refs[level] != 1); |
5536 | level--; | |
5537 | } | |
9f3a7427 | 5538 | } |
2c47e605 | 5539 | |
78c52d9e | 5540 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5541 | wc->level = level; |
5542 | wc->shared_level = -1; | |
5543 | wc->stage = DROP_REFERENCE; | |
5544 | wc->update_ref = update_ref; | |
5545 | wc->keep_locks = 0; | |
0b246afa | 5546 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5547 | |
d397712b | 5548 | while (1) { |
9d1a2a3a | 5549 | |
2c47e605 YZ |
5550 | ret = walk_down_tree(trans, root, path, wc); |
5551 | if (ret < 0) { | |
5552 | err = ret; | |
20524f02 | 5553 | break; |
2c47e605 | 5554 | } |
9aca1d51 | 5555 | |
2c47e605 YZ |
5556 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5557 | if (ret < 0) { | |
5558 | err = ret; | |
20524f02 | 5559 | break; |
2c47e605 YZ |
5560 | } |
5561 | ||
5562 | if (ret > 0) { | |
5563 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5564 | break; |
5565 | } | |
2c47e605 YZ |
5566 | |
5567 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5568 | wc->drop_level = wc->level; |
5569 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5570 | &wc->drop_progress, | |
5571 | path->slots[wc->drop_level]); | |
5572 | } | |
5573 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5574 | &wc->drop_progress); | |
5575 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
5576 | |
5577 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5578 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5579 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5580 | ret = btrfs_update_root(trans, tree_root, |
5581 | &root->root_key, | |
5582 | root_item); | |
79787eaa | 5583 | if (ret) { |
66642832 | 5584 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5585 | err = ret; |
5586 | goto out_end_trans; | |
5587 | } | |
2c47e605 | 5588 | |
3a45bb20 | 5589 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5590 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5591 | btrfs_debug(fs_info, |
5592 | "drop snapshot early exit"); | |
3c8f2422 JB |
5593 | err = -EAGAIN; |
5594 | goto out_free; | |
5595 | } | |
5596 | ||
a22285a6 | 5597 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
5598 | if (IS_ERR(trans)) { |
5599 | err = PTR_ERR(trans); | |
5600 | goto out_free; | |
5601 | } | |
c3e69d58 | 5602 | } |
20524f02 | 5603 | } |
b3b4aa74 | 5604 | btrfs_release_path(path); |
79787eaa JM |
5605 | if (err) |
5606 | goto out_end_trans; | |
2c47e605 | 5607 | |
ab9ce7d4 | 5608 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5609 | if (ret) { |
66642832 | 5610 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5611 | err = ret; |
79787eaa JM |
5612 | goto out_end_trans; |
5613 | } | |
2c47e605 | 5614 | |
76dda93c | 5615 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5616 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5617 | NULL, NULL); | |
79787eaa | 5618 | if (ret < 0) { |
66642832 | 5619 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5620 | err = ret; |
5621 | goto out_end_trans; | |
5622 | } else if (ret > 0) { | |
84cd948c JB |
5623 | /* if we fail to delete the orphan item this time |
5624 | * around, it'll get picked up the next time. | |
5625 | * | |
5626 | * The most common failure here is just -ENOENT. | |
5627 | */ | |
5628 | btrfs_del_orphan_item(trans, tree_root, | |
5629 | root->root_key.objectid); | |
76dda93c YZ |
5630 | } |
5631 | } | |
5632 | ||
a3cf0e43 QW |
5633 | /* |
5634 | * This subvolume is going to be completely dropped, and won't be | |
5635 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
5636 | * commit transaction time. So free it here manually. | |
5637 | */ | |
5638 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
5639 | btrfs_qgroup_free_meta_all_pertrans(root); | |
5640 | ||
8c38938c | 5641 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 5642 | btrfs_add_dropped_root(trans, root); |
8c38938c | 5643 | else |
00246528 | 5644 | btrfs_put_root(root); |
d29a9f62 | 5645 | root_dropped = true; |
79787eaa | 5646 | out_end_trans: |
3a45bb20 | 5647 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5648 | out_free: |
2c47e605 | 5649 | kfree(wc); |
5caf2a00 | 5650 | btrfs_free_path(path); |
cb1b69f4 | 5651 | out: |
d29a9f62 JB |
5652 | /* |
5653 | * So if we need to stop dropping the snapshot for whatever reason we | |
5654 | * need to make sure to add it back to the dead root list so that we | |
5655 | * keep trying to do the work later. This also cleans up roots if we | |
5656 | * don't have it in the radix (like when we recover after a power fail | |
5657 | * or unmount) so we don't leak memory. | |
5658 | */ | |
897ca819 | 5659 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5660 | btrfs_add_dead_root(root); |
2c536799 | 5661 | return err; |
20524f02 | 5662 | } |
9078a3e1 | 5663 | |
2c47e605 YZ |
5664 | /* |
5665 | * drop subtree rooted at tree block 'node'. | |
5666 | * | |
5667 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5668 | * only used by relocation code |
2c47e605 | 5669 | */ |
f82d02d9 YZ |
5670 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5671 | struct btrfs_root *root, | |
5672 | struct extent_buffer *node, | |
5673 | struct extent_buffer *parent) | |
5674 | { | |
0b246afa | 5675 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5676 | struct btrfs_path *path; |
2c47e605 | 5677 | struct walk_control *wc; |
f82d02d9 YZ |
5678 | int level; |
5679 | int parent_level; | |
5680 | int ret = 0; | |
5681 | int wret; | |
5682 | ||
2c47e605 YZ |
5683 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5684 | ||
f82d02d9 | 5685 | path = btrfs_alloc_path(); |
db5b493a TI |
5686 | if (!path) |
5687 | return -ENOMEM; | |
f82d02d9 | 5688 | |
2c47e605 | 5689 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5690 | if (!wc) { |
5691 | btrfs_free_path(path); | |
5692 | return -ENOMEM; | |
5693 | } | |
2c47e605 | 5694 | |
b9447ef8 | 5695 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5696 | parent_level = btrfs_header_level(parent); |
67439dad | 5697 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5698 | path->nodes[parent_level] = parent; |
5699 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5700 | ||
b9447ef8 | 5701 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5702 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5703 | path->nodes[level] = node; |
5704 | path->slots[level] = 0; | |
bd681513 | 5705 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5706 | |
5707 | wc->refs[parent_level] = 1; | |
5708 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5709 | wc->level = level; | |
5710 | wc->shared_level = -1; | |
5711 | wc->stage = DROP_REFERENCE; | |
5712 | wc->update_ref = 0; | |
5713 | wc->keep_locks = 1; | |
0b246afa | 5714 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5715 | |
5716 | while (1) { | |
2c47e605 YZ |
5717 | wret = walk_down_tree(trans, root, path, wc); |
5718 | if (wret < 0) { | |
f82d02d9 | 5719 | ret = wret; |
f82d02d9 | 5720 | break; |
2c47e605 | 5721 | } |
f82d02d9 | 5722 | |
2c47e605 | 5723 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5724 | if (wret < 0) |
5725 | ret = wret; | |
5726 | if (wret != 0) | |
5727 | break; | |
5728 | } | |
5729 | ||
2c47e605 | 5730 | kfree(wc); |
f82d02d9 YZ |
5731 | btrfs_free_path(path); |
5732 | return ret; | |
5733 | } | |
5734 | ||
6d07bcec MX |
5735 | /* |
5736 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5737 | * space_info. takes mirrors into account. |
6d07bcec | 5738 | */ |
633c0aad | 5739 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5740 | { |
32da5386 | 5741 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5742 | u64 free_bytes = 0; |
5743 | int factor; | |
5744 | ||
01327610 | 5745 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5746 | if (list_empty(&sinfo->ro_bgs)) |
5747 | return 0; | |
5748 | ||
5749 | spin_lock(&sinfo->lock); | |
5750 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5751 | spin_lock(&block_group->lock); |
5752 | ||
5753 | if (!block_group->ro) { | |
5754 | spin_unlock(&block_group->lock); | |
5755 | continue; | |
5756 | } | |
5757 | ||
46df06b8 | 5758 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5759 | free_bytes += (block_group->length - |
bf38be65 | 5760 | block_group->used) * factor; |
6d07bcec MX |
5761 | |
5762 | spin_unlock(&block_group->lock); | |
5763 | } | |
6d07bcec MX |
5764 | spin_unlock(&sinfo->lock); |
5765 | ||
5766 | return free_bytes; | |
5767 | } | |
5768 | ||
2ff7e61e JM |
5769 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5770 | u64 start, u64 end) | |
acce952b | 5771 | { |
2ff7e61e | 5772 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5773 | } |
5774 | ||
499f377f JM |
5775 | /* |
5776 | * It used to be that old block groups would be left around forever. | |
5777 | * Iterating over them would be enough to trim unused space. Since we | |
5778 | * now automatically remove them, we also need to iterate over unallocated | |
5779 | * space. | |
5780 | * | |
5781 | * We don't want a transaction for this since the discard may take a | |
5782 | * substantial amount of time. We don't require that a transaction be | |
5783 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5784 | * to ensure that we're not discarding chunks that were released or |
5785 | * allocated in the current transaction. | |
499f377f JM |
5786 | * |
5787 | * Holding the chunks lock will prevent other threads from allocating | |
5788 | * or releasing chunks, but it won't prevent a running transaction | |
5789 | * from committing and releasing the memory that the pending chunks | |
5790 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5791 | * transaction and hold the commit root sem. We only need to hold |
5792 | * it while performing the free space search since we have already | |
5793 | * held back allocations. | |
499f377f | 5794 | */ |
8103d10b | 5795 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5796 | { |
8103d10b | 5797 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5798 | int ret; |
5799 | ||
5800 | *trimmed = 0; | |
5801 | ||
0be88e36 JM |
5802 | /* Discard not supported = nothing to do. */ |
5803 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5804 | return 0; | |
5805 | ||
52042d8e | 5806 | /* Not writable = nothing to do. */ |
ebbede42 | 5807 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5808 | return 0; |
5809 | ||
5810 | /* No free space = nothing to do. */ | |
5811 | if (device->total_bytes <= device->bytes_used) | |
5812 | return 0; | |
5813 | ||
5814 | ret = 0; | |
5815 | ||
5816 | while (1) { | |
fb456252 | 5817 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5818 | u64 bytes; |
5819 | ||
5820 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5821 | if (ret) | |
fee7acc3 | 5822 | break; |
499f377f | 5823 | |
929be17a NB |
5824 | find_first_clear_extent_bit(&device->alloc_state, start, |
5825 | &start, &end, | |
5826 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 5827 | |
c57dd1f2 QW |
5828 | /* Check if there are any CHUNK_* bits left */ |
5829 | if (start > device->total_bytes) { | |
5830 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5831 | btrfs_warn_in_rcu(fs_info, | |
5832 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
5833 | start, end - start + 1, | |
5834 | rcu_str_deref(device->name), | |
5835 | device->total_bytes); | |
5836 | mutex_unlock(&fs_info->chunk_mutex); | |
5837 | ret = 0; | |
5838 | break; | |
5839 | } | |
5840 | ||
53460a45 NB |
5841 | /* Ensure we skip the reserved area in the first 1M */ |
5842 | start = max_t(u64, start, SZ_1M); | |
5843 | ||
929be17a NB |
5844 | /* |
5845 | * If find_first_clear_extent_bit find a range that spans the | |
5846 | * end of the device it will set end to -1, in this case it's up | |
5847 | * to the caller to trim the value to the size of the device. | |
5848 | */ | |
5849 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5850 | |
929be17a | 5851 | len = end - start + 1; |
499f377f | 5852 | |
929be17a NB |
5853 | /* We didn't find any extents */ |
5854 | if (!len) { | |
499f377f | 5855 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5856 | ret = 0; |
499f377f JM |
5857 | break; |
5858 | } | |
5859 | ||
929be17a NB |
5860 | ret = btrfs_issue_discard(device->bdev, start, len, |
5861 | &bytes); | |
5862 | if (!ret) | |
5863 | set_extent_bits(&device->alloc_state, start, | |
5864 | start + bytes - 1, | |
5865 | CHUNK_TRIMMED); | |
499f377f JM |
5866 | mutex_unlock(&fs_info->chunk_mutex); |
5867 | ||
5868 | if (ret) | |
5869 | break; | |
5870 | ||
5871 | start += len; | |
5872 | *trimmed += bytes; | |
5873 | ||
5874 | if (fatal_signal_pending(current)) { | |
5875 | ret = -ERESTARTSYS; | |
5876 | break; | |
5877 | } | |
5878 | ||
5879 | cond_resched(); | |
5880 | } | |
5881 | ||
5882 | return ret; | |
5883 | } | |
5884 | ||
93bba24d QW |
5885 | /* |
5886 | * Trim the whole filesystem by: | |
5887 | * 1) trimming the free space in each block group | |
5888 | * 2) trimming the unallocated space on each device | |
5889 | * | |
5890 | * This will also continue trimming even if a block group or device encounters | |
5891 | * an error. The return value will be the last error, or 0 if nothing bad | |
5892 | * happens. | |
5893 | */ | |
2ff7e61e | 5894 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5895 | { |
32da5386 | 5896 | struct btrfs_block_group *cache = NULL; |
499f377f JM |
5897 | struct btrfs_device *device; |
5898 | struct list_head *devices; | |
f7039b1d | 5899 | u64 group_trimmed; |
07301df7 | 5900 | u64 range_end = U64_MAX; |
f7039b1d LD |
5901 | u64 start; |
5902 | u64 end; | |
5903 | u64 trimmed = 0; | |
93bba24d QW |
5904 | u64 bg_failed = 0; |
5905 | u64 dev_failed = 0; | |
5906 | int bg_ret = 0; | |
5907 | int dev_ret = 0; | |
f7039b1d LD |
5908 | int ret = 0; |
5909 | ||
07301df7 QW |
5910 | /* |
5911 | * Check range overflow if range->len is set. | |
5912 | * The default range->len is U64_MAX. | |
5913 | */ | |
5914 | if (range->len != U64_MAX && | |
5915 | check_add_overflow(range->start, range->len, &range_end)) | |
5916 | return -EINVAL; | |
5917 | ||
6ba9fc8e | 5918 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5919 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 5920 | if (cache->start >= range_end) { |
f7039b1d LD |
5921 | btrfs_put_block_group(cache); |
5922 | break; | |
5923 | } | |
5924 | ||
b3470b5d DS |
5925 | start = max(range->start, cache->start); |
5926 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
5927 | |
5928 | if (end - start >= range->minlen) { | |
32da5386 | 5929 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 5930 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 5931 | if (ret) { |
93bba24d QW |
5932 | bg_failed++; |
5933 | bg_ret = ret; | |
5934 | continue; | |
1be41b78 | 5935 | } |
676f1f75 | 5936 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5937 | if (ret) { |
93bba24d QW |
5938 | bg_failed++; |
5939 | bg_ret = ret; | |
5940 | continue; | |
1be41b78 | 5941 | } |
f7039b1d LD |
5942 | } |
5943 | ret = btrfs_trim_block_group(cache, | |
5944 | &group_trimmed, | |
5945 | start, | |
5946 | end, | |
5947 | range->minlen); | |
5948 | ||
5949 | trimmed += group_trimmed; | |
5950 | if (ret) { | |
93bba24d QW |
5951 | bg_failed++; |
5952 | bg_ret = ret; | |
5953 | continue; | |
f7039b1d LD |
5954 | } |
5955 | } | |
f7039b1d LD |
5956 | } |
5957 | ||
93bba24d QW |
5958 | if (bg_failed) |
5959 | btrfs_warn(fs_info, | |
5960 | "failed to trim %llu block group(s), last error %d", | |
5961 | bg_failed, bg_ret); | |
0b246afa | 5962 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5963 | devices = &fs_info->fs_devices->devices; |
5964 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5965 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5966 | if (ret) { |
5967 | dev_failed++; | |
5968 | dev_ret = ret; | |
499f377f | 5969 | break; |
93bba24d | 5970 | } |
499f377f JM |
5971 | |
5972 | trimmed += group_trimmed; | |
5973 | } | |
0b246afa | 5974 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5975 | |
93bba24d QW |
5976 | if (dev_failed) |
5977 | btrfs_warn(fs_info, | |
5978 | "failed to trim %llu device(s), last error %d", | |
5979 | dev_failed, dev_ret); | |
f7039b1d | 5980 | range->len = trimmed; |
93bba24d QW |
5981 | if (bg_ret) |
5982 | return bg_ret; | |
5983 | return dev_ret; | |
f7039b1d | 5984 | } |